cvpr cvpr2013 cvpr2013-88 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Giuseppe Ottaviano, Pushmeet Kohli
Abstract: Traditional video compression methods obtain a compact representation for image frames by computing coarse motion fields defined on patches of pixels called blocks, in order to compensate for the motion in the scene across frames. This piecewise constant approximation makes the motion field efficiently encodable, but it introduces block artifacts in the warped image frame. In this paper, we address the problem of estimating dense motion fields that, while accurately predicting one frame from a given reference frame by warping it with the field, are also compressible. We introduce a representation for motion fields based on wavelet bases, and approximate the compressibility of their coefficients with a piecewise smooth surrogate function that yields an objective function similar to classical optical flow formulations. We then show how to quantize and encode such coefficients with adaptive precision. We demonstrate the effectiveness of our approach by com- paring its performance with a state-of-the-art wavelet video encoder. Experimental results on a number of standard flow and video datasets reveal that our method significantly outperforms both block-based and optical-flow-based motion compensation algorithms.
Reference: text
sentIndex sentText sentNum sentScore
1 it Abstract Traditional video compression methods obtain a compact representation for image frames by computing coarse motion fields defined on patches of pixels called blocks, in order to compensate for the motion in the scene across frames. [sent-3, score-0.765]
2 This piecewise constant approximation makes the motion field efficiently encodable, but it introduces block artifacts in the warped image frame. [sent-4, score-0.659]
3 In this paper, we address the problem of estimating dense motion fields that, while accurately predicting one frame from a given reference frame by warping it with the field, are also compressible. [sent-5, score-0.711]
4 We introduce a representation for motion fields based on wavelet bases, and approximate the compressibility of their coefficients with a piecewise smooth surrogate function that yields an objective function similar to classical optical flow formulations. [sent-6, score-1.119]
5 We demonstrate the effectiveness of our approach by com- paring its performance with a state-of-the-art wavelet video encoder. [sent-8, score-0.386]
6 Experimental results on a number of standard flow and video datasets reveal that our method significantly outperforms both block-based and optical-flow-based motion compensation algorithms. [sent-9, score-0.549]
7 Introduction Most modern video compression algorithms fall into the category of hybrid video encoders that work by using previously decoded frames and some side information (explicitly added by the encoder) to make a prediction for the current frame. [sent-11, score-0.516]
8 The difference between the prediction and the frame, called residual orprediction error, is then encoded separately to correct the prediction. [sent-12, score-0.433]
9 On one hand a better prediction implies a more compact residual encoding; on the other hand, the side information must be kept as compact as possible to avoid its encoding cost outweighing the benefits of the more accurate prediction. [sent-13, score-0.381]
10 com @mi The largest part of such side information consists of a × motion field, which allows to compensate for the motion of the camera and the objects in the scene across consecutive frames, hence forming a motion compensated prediction. [sent-16, score-0.682]
11 Given a pair of images I0 and I1, a (dense) motion field u is a field of per-pixel motion vectors describing how to warp pixels from I1to form a new image I1(u), which we refer to as I1warped with u. [sent-17, score-0.856]
12 Ideally, we would associate to each pixel in the image the motion vector that minimizes the residual; however such a field may contain more information than the image itself: a field for n pixels has 2n degrees of freedom. [sent-19, score-0.609]
13 In its basic version, a fixed block size (say 16 16) is chosen and a motion vector is associated to eac1h6 b ×lo c16k. [sent-22, score-0.357]
14 In the above definition this is equivalent to requiring that the motion field is constant within the blocks. [sent-23, score-0.448]
15 This piecewise constant flow approximation makes the motion field efficiently encodable, but it introduces block artifacts in the decoded image frame. [sent-24, score-0.745]
16 In this paper we address the problem of estimating dense motion fields which, while accurately predicting one frame from a given reference frame by warping it with the field, are also compressible. [sent-25, score-0.711]
17 We introduce a new representation for motion fields as linear combinations of a given basis. [sent-26, score-0.347]
18 The computation of the basis coefficients can be posed as a global piecewise-smooth optimization problem which resembles classical optical flow formulations, optimizing for both compressibility and residual magnitude. [sent-27, score-0.601]
19 We focus on wavelet bases, which loosely generalize block-based algorithms, and whose orthogonality simplifies the optimization. [sent-30, score-0.312]
20 Our results reveal that our wavelet motion fields outperform both block-based and optical-flow-based motion compensation techniques. [sent-32, score-1.069]
21 Related Work As mentioned before, block-based algorithms induce motion fields that are likely to be discontinuous at block boundaries, thus introducing in the predicted image, and in-turn in the residual, discontinuities that are visually noticeable and expensive to encode. [sent-35, score-0.52]
22 Different block sizes in a block motion compensation algorithm give different accuracy/compactness trade-offs. [sent-41, score-0.69]
23 In VBMC, each block can be segmented into smaller sub-blocks, with the segmentation encoded as side information, and a different motion vector is encoded for each sub-block. [sent-45, score-0.543]
24 Compact representations of dense motion fields are explored in [12], where the authors use a DCT-based encoder for the field, in [15] where a multiscale approach similar to our wavelet decomposition is used, and in [8], which uses a quad-tree-like hierarchical representation. [sent-57, score-0.89]
25 In what follows, we show that by using a penalty modeled after the actual entropy encoder of the field, it is possible to obtain fields that can be encoded in significantly fewer bits, compared to the fields obtained with the smoothness penalty, without sacrificing prediction quality. [sent-60, score-0.698]
26 For a motion field to be feasible we constrain its motion vectors inside the image rectangle, i. [sent-63, score-0.66]
27 o Wf neo ctaaltli othne we eoxf tfeeandsi bIl to ethldes continuous rectangle [0, w − 1] [0, h − 1] by interpolation (in our implementation we use b[0ic,uhb −ic) 1 a]nd define the image I warped with the motion vector u as I(u), formally I(u)i,j = Ii+ui0,j ,j+ui1,j (for instance, I(0) = I). [sent-68, score-0.256]
28 This notation allows us to write the residual of I0 and I1 under the motion field u as I0 − I1 (u). [sent-69, score-0.641]
29 Representation and coding cost Field representation We represent a motion field u by its coefficients α in a linear basis represented by a matrix W, so that u = Wα and α = W−1u. [sent-72, score-0.663]
30 The coefficients α are 1In fact, modern optical flow algorithms favor fields with sharp edges at object boundaries, which would be harder to compress. [sent-73, score-0.433]
31 222222555200 lossily encoded using a quantizer and an entropy coder. [sent-75, score-0.287]
32 This is not dissimilar to lossy transform coding for images, used for example in DCT coders such as JPEG and wavelet encoders such as JPEG2000. [sent-76, score-0.455]
33 be the set of feasible fields with integer coefficients in the? [sent-79, score-0.298]
34 for the field, we wish to minimize the residual subject to the budget min? [sent-87, score-0.258]
35 agrangian multiplier, which trades off bits of the field encoding for residual magnitude. [sent-102, score-0.638]
36 the horizontal and vertical components of the field are transformed independently with the same transform matrix, and that W? [sent-113, score-0.263]
37 is an orthogonal separable multilevel wavelet transform, so we can write W−1 = WT. [sent-114, score-0.345]
38 levels, which represent the detail at each level of the recursive wavelet decomposition, and in the separable 2D case each level (except the first) can be further divided into 3 subbands, which correspond to horizontal, vertical and diagonal detail. [sent-116, score-0.405]
39 A comprehensive account of multilevel wavelet decompositions can be found in [13]. [sent-117, score-0.312]
40 VBMC and the Haar Wavelet Here we show that the motion fields obtained with Variable-Block Motion Compensation can be represented sparsely in the Haar wavelet basis, suggesting that other wavelet bases may be good choice for representing motion as well. [sent-119, score-1.242]
41 In the first approximation level of the Haar decomposition, each coefficient corresponds to the average of the motion field inside a macroblock (modulo normalization constants). [sent-121, score-0.457]
42 Each level of splitting of the blocks adds a constant number of coefficients to the corresponding detail level in the Haar decomposition, which correspond to the difference of the vectors in the sub-blocks. [sent-122, score-0.311]
43 VBMC represents the segmentation explicitly, and encodes the difference of the vector with the median of the neighboring motion vectors, to exploit the local coherency of the field. [sent-125, score-0.248]
44 In the wavelet representation the segmentation is implicitly encoded in the set of non-zeros of the coefficients, while the local coherency is exploited by the recursive encoding of averages and differences: if neighboring blocks have similar field, the difference coefficient will be small. [sent-126, score-0.599]
45 This is the same principle that makes wavelet bases suitable to represent natural images. [sent-127, score-0.366]
46 Reference frame Current frame Absolute difference Detail Haar motion field Haar prediction Haar residual Haar prediction detail Sym5 motion field Sym5 prediction Sym5 residual Sym5 prediction detail Figure 1. [sent-128, score-1.938]
47 Second and third row: motion fields, prediction and residual obtained with Haar and Sym5 wavelets. [sent-130, score-0.557]
48 We first start by noting that to encode the coefficienbtsit so(f· )WTu we need to encode both the positions of the non-zero coefficients (this term is significant if the representation is sparse) and the sign and magnitude of the quantized coefficients. [sent-134, score-0.261]
49 2) with integer coefficients in the transformed basis, hence already quantized, and let nb be the number of coefficients in the subband b and mb the number of non-zeros. [sent-136, score-0.568]
50 =ffi c0i]ents cloostg we assume that the number of bits needed to encode a coefficient α can be bounded by γ1 log(|α| + 1) + γ2 ; this is true for many universal codes forl integers, s 1u)ch + as the Gamma codes [7] used in our entropy encoder. [sent-142, score-0.276]
51 Logarithmic (and in general concave) penalties are known to encourage sparse solutions [2]; in fact the motion fields we obtain have very few non-zero coefficients. [sent-162, score-0.347]
52 Thcisa can sbeet u tose ∞ful t iof c we wtraanint to obtain a locally constant motion field, by discarding the higher-resolution subbands. [sent-165, score-0.252]
53 In the Haar case discarding the last two levels of the wavelet decomposition is equivalent to imposing a minimum block size of 4 4. [sent-166, score-0.51]
54 Given a field estimate u0 we perform a first-order Taylor expansion of I1(u) at u0, giving a linearized data term ? [sent-169, score-0.263]
55 To do this we follow the standard approach of dividing the coefficients into small square blocks and assigning an uniform quantizer qk with dead-zone [4] to each block k, which means that if a coefficient α is located in block k the integer value sgn(α) ? [sent-243, score-0.701]
56 It remains to be decided what quantizer qk to assign to each block k. [sent-248, score-0.271]
57 222222555422 Following again Rate-Distortion Optimization [17], a widely adopted strategy is to fix a component-wise distortion metric D on the coefficients to be encoded, for example squared difference, and optimize over q = (q1, . [sent-249, score-0.317]
58 Since each block can be optimized separately, the running time is linear in the number of blocks and quantizer choices. [sent-258, score-0.307]
59 By setting a strict bound on the average distortion (for example less than quarter-pixel precision) the quantized field can be made close enough to the real-valued field. [sent-267, score-0.359]
60 The ideal distortion we would like to optimize when quantizing a motion field is the warping error ? [sent-272, score-0.68]
61 For this reason we derive a coefficient-wise surrogate distortion metric that approximates the warping error. [sent-282, score-0.395]
62 The squared linearized warping error can then be written as eTWT∇I[u]2W e˜. [sent-314, score-0.261]
63 We give an intuitive explanation of why this is the case: most vectors of the wavelet basis are localized in space, with most of their energy concentrated in a small number of pixels; hence in the matrix a large part of the energy is concentrated around the diagonal. [sent-316, score-0.415]
64 Since W is a multi-level wavelet transform, we do ∇nIot[ uh]aWve the matrix in explicit form, but an algorithm that computes the linear operator instead. [sent-321, score-0.312]
65 However it is easy to see that if the wavelet has constant support, and the number of levels ? [sent-322, score-0.347]
66 The Symlet wavelets are orthogonal, compactly supported, and almost symmetric; moreover, as most wavelets used in signal processing, they are continuous, hence any finite approximation of a signal by Symlet wavelets is continuous, regardless of the number of coefficients used. [sent-331, score-0.513]
67 On the other hand, the Haar wavelet is discontinuous, thus it produces significant discontinuities in sparse approximations of continuous signals. [sent-332, score-0.312]
68 An example is shown in Figure 1: the Haar basis exhibits the same artifacts as block-based methods, while field and prediction obtained with Sym5 are smooth. [sent-333, score-0.382]
69 1) the log nb term suggests to use increments of 2 per level, because at each level the number of coefficients increase by a factor of 4; hence for the 6 levels we used (2, 4, 6, 8, ∞, ∞) in all the experiments. [sent-336, score-0.305]
70 We give infinite weight 4to, 6th,e8 l,a∞st ,t∞wo) levels both to control the sparsity, and also because we estimate the field only on the luma component and use the same field on the chroma components; constraining the field to be locally smooth reduces the risk of overfitting to the luminance. [sent-337, score-0.695]
71 Field and residual encoder To evaluate experimentally the effectiveness of the method in a video compression setting we implemented a complete video encoder. [sent-338, score-0.633]
72 This requires implementing an encoder for the residual image; we use 222222555533 × again a wavelet transform on the residual and then the same quantizer/entropy coder to encode the coefficients of both the field and the residual. [sent-339, score-1.457]
73 As quantizer/entropy coder we use a simplified version of Dirac’s residual coder: the coefficients are split into blocks (we use 16 16 blocks) and each block is assigned an uniform quantizer q. [sent-340, score-0.784]
74 The entropy coder is the same context-based binary arithmetic coder used by Dirac, but we use a smaller number of contexts for predicting zeros and signs based on the neighboring coefficients already encoded. [sent-342, score-0.453]
75 Unlike Dirac, for simplicity no coefficient prediction or zero prediction based on the parent subband are performed. [sent-343, score-0.367]
76 Finally, the wavelet used in the decomposition is Sym5, instead of the biorthogonal wavelets used in Dirac. [sent-344, score-0.53]
77 Experiments Adaptive quantization We evaluated the quantization algorithm described in Section 5 on the motion field of the first pair of frames of four of the videos used in the experiments. [sent-346, score-0.576]
78 The curves in Figure 2 show the PSNR of the actual warping error, measured as PSNR(I1 (u) , I1( u˜q)) at different bit sizes (obtained by varying λquant), using the standard L22 distortion metric and the weighted distortion metric Σ-L22 described above. [sent-347, score-0.469]
79 By optimizing for Σ-L22 instead of L22 the field can be encoded in roughly half the bits at the same quality . [sent-348, score-0.425]
80 Furthermore, the rate-distortion curve obtained with the weighted distortion is significantly closer to linear, showing that the surrogate distortion is highly correlated with the actual warping error. [sent-349, score-0.475]
81 Experiments on the Middlebury dataset As an early synthetic benchmark, we used the grayscale image pairs of the Middlebury dataset [1] and measured the quality of the prediction against the compressed field size in bits, comparing with Horn-Schunck, as reported in Figure 3. [sent-350, score-0.34]
82 Also, on almost all pairs, adaptive quantization significantly reduces the encoded field bitrate. [sent-354, score-0.349]
83 0 Motion field kbits Warping error PSN R: highway Motion field kbits Motion field kbits Warping error PSN R: sintel Motion field kbits Figure 2. [sent-378, score-1.374]
84 Rate-distortion curves of the field between the first two frames of the four video sequences used for evaluation. [sent-379, score-0.351]
85 The y axis shows the PSNR between the reference warped with the continuous field and with the quantized field. [sent-380, score-0.349]
86 Video compression evaluation We compared our implementation against Dirac, a state-of-the-art wavelet video codec; its simplicity allowed us to implement a very similar encoder, so we can compare the contribution of the motion compensation component. [sent-381, score-0.88]
87 As in the Dirac encoder, no explicit rate ×× control is performed; instead λ for the motion estimation and the λquant for field and residual quantization are fixed in advance and remain constant across the frames. [sent-383, score-0.736]
88 The first frame of the sequence is encoded with no prediction, then each subsequent frame is predicted from the previous decoded frame. [sent-385, score-0.299]
89 Test sequences We performed our experiments on eight color (YUV420) sequences from a database of standard video compression test sequences [20]. [sent-388, score-0.272]
90 They combine a variety of (camera and scene) motion types, and have different frame resolutions, 352 288 for foreman, news, highway, bus, flower, an3d5 5fo2o ×tb 2a8l8l, 832 352 for sintel, and 704 576 for soccer. [sent-389, score-0.29]
91 Reference frame Current frame Absolute difference L1Log-WQ motion field L1Log-WQ prediction L1Log-WQ residual 5024 btis HS motion field PSNR: 38. [sent-396, score-1.357]
92 8 HS prediction HS residual 5048 btis PSNR: 35. [sent-397, score-0.385]
93 It should be noted that the decoding complexity is only marginally higher than block-based motion compensation, as just a wavelet transform and a pixel-level warping are needed. [sent-405, score-0.72]
94 This may be a source of inefficiency in sequences with large occlusions, because the coefficients of the motion vectors in occluded areas are encoded anyway. [sent-411, score-0.515]
95 Another limitation is the single reference frame constraint: multiple reference frames greatly improve the efficiency of video encoders. [sent-412, score-0.314]
96 The terms t and lcan be again transformed in a wavelet basis and quantized/encoded. [sent-423, score-0.386]
97 Variable size block matching motion compensation with applications to video coding. [sent-452, score-0.624]
98 A new motion compensation [11] [12] [13] [14] [15] [16] [17] [18] [19] method for image sequence coding using hierarchical grid interpolation. [sent-497, score-0.481]
99 An optical flow based motion compensation algorithm for very low bit-rate video coding. [sent-511, score-0.618]
100 Multiscale modeling and estimation of motion fields for video coding. [sent-530, score-0.421]
wordName wordTfidf (topN-words)
[('wavelet', 0.312), ('wtu', 0.246), ('residual', 0.228), ('motion', 0.217), ('field', 0.196), ('haar', 0.193), ('compensation', 0.193), ('vbmc', 0.179), ('encoder', 0.173), ('warping', 0.156), ('dirac', 0.144), ('block', 0.14), ('coefficients', 0.137), ('bits', 0.136), ('bitrate', 0.134), ('psnr', 0.131), ('fields', 0.13), ('wavelets', 0.115), ('prediction', 0.112), ('coder', 0.112), ('foreman', 0.112), ('kbits', 0.112), ('distortion', 0.111), ('hs', 0.102), ('subband', 0.099), ('surrogate', 0.097), ('encoded', 0.093), ('quantizer', 0.089), ('compression', 0.084), ('log', 0.083), ('sintel', 0.079), ('blocks', 0.078), ('video', 0.074), ('frame', 0.073), ('coding', 0.071), ('psn', 0.069), ('optical', 0.069), ('luma', 0.067), ('quant', 0.067), ('symlet', 0.067), ('linearized', 0.067), ('flow', 0.065), ('highway', 0.063), ('reference', 0.062), ('quantization', 0.06), ('entropy', 0.06), ('compressibility', 0.06), ('decoded', 0.06), ('decomposition', 0.058), ('nb', 0.054), ('bases', 0.054), ('news', 0.053), ('middlebury', 0.053), ('quantized', 0.052), ('linearization', 0.048), ('mb', 0.047), ('biorthogonal', 0.045), ('btis', 0.045), ('encodable', 0.045), ('hornschunck', 0.045), ('lossily', 0.045), ('coefficient', 0.044), ('frames', 0.043), ('basis', 0.042), ('qk', 0.042), ('encoding', 0.041), ('chroma', 0.04), ('warped', 0.039), ('circuits', 0.038), ('squared', 0.038), ('sequences', 0.038), ('compressible', 0.037), ('encoders', 0.037), ('trades', 0.037), ('encode', 0.036), ('bus', 0.036), ('constant', 0.035), ('transform', 0.035), ('flower', 0.033), ('separable', 0.033), ('discontinuous', 0.033), ('bmc', 0.033), ('modern', 0.032), ('grayscale', 0.032), ('transformed', 0.032), ('sparsity', 0.032), ('objective', 0.032), ('arithmetic', 0.032), ('artifacts', 0.032), ('metric', 0.031), ('detail', 0.031), ('hence', 0.031), ('integer', 0.031), ('sullivan', 0.031), ('coherency', 0.031), ('linearize', 0.031), ('budget', 0.03), ('vectors', 0.03), ('diagonal', 0.029), ('bit', 0.029)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999905 88 cvpr-2013-Compressible Motion Fields
Author: Giuseppe Ottaviano, Pushmeet Kohli
Abstract: Traditional video compression methods obtain a compact representation for image frames by computing coarse motion fields defined on patches of pixels called blocks, in order to compensate for the motion in the scene across frames. This piecewise constant approximation makes the motion field efficiently encodable, but it introduces block artifacts in the warped image frame. In this paper, we address the problem of estimating dense motion fields that, while accurately predicting one frame from a given reference frame by warping it with the field, are also compressible. We introduce a representation for motion fields based on wavelet bases, and approximate the compressibility of their coefficients with a piecewise smooth surrogate function that yields an objective function similar to classical optical flow formulations. We then show how to quantize and encode such coefficients with adaptive precision. We demonstrate the effectiveness of our approach by com- paring its performance with a state-of-the-art wavelet video encoder. Experimental results on a number of standard flow and video datasets reveal that our method significantly outperforms both block-based and optical-flow-based motion compensation algorithms.
Author: Won Hwa Kim, Moo K. Chung, Vikas Singh
Abstract: The analysis of 3-D shape meshes is a fundamental problem in computer vision, graphics, and medical imaging. Frequently, the needs of the application require that our analysis take a multi-resolution view of the shape ’s local and global topology, and that the solution is consistent across multiple scales. Unfortunately, the preferred mathematical construct which offers this behavior in classical image/signal processing, Wavelets, is no longer applicable in this general setting (data with non-uniform topology). In particular, the traditional definition does not allow writing out an expansion for graphs that do not correspond to the uniformly sampled lattice (e.g., images). In this paper, we adapt recent results in harmonic analysis, to derive NonEuclidean Wavelets based algorithms for a range of shape analysis problems in vision and medical imaging. We show how descriptors derived from the dual domain representation offer native multi-resolution behavior for characterizing local/global topology around vertices. With only minor modifications, the framework yields a method for extracting interest/key points from shapes, a surprisingly simple algorithm for 3-D shape segmentation (competitive with state of the art), and a method for surface alignment (without landmarks). We give an extensive set of comparison results on a large shape segmentation benchmark and derive a uniqueness theorem for the surface alignment problem.
3 0.20895801 244 cvpr-2013-Large Displacement Optical Flow from Nearest Neighbor Fields
Author: Zhuoyuan Chen, Hailin Jin, Zhe Lin, Scott Cohen, Ying Wu
Abstract: We present an optical flow algorithm for large displacement motions. Most existing optical flow methods use the standard coarse-to-fine framework to deal with large displacement motions which has intrinsic limitations. Instead, we formulate the motion estimation problem as a motion segmentation problem. We use approximate nearest neighbor fields to compute an initial motion field and use a robust algorithm to compute a set of similarity transformations as the motion candidates for segmentation. To account for deviations from similarity transformations, we add local deformations in the segmentation process. We also observe that small objects can be better recovered using translations as the motion candidates. We fuse the motion results obtained under similarity transformations and under translations together before a final refinement. Experimental validation shows that our method can successfully handle large displacement motions. Although we particularly focus on large displacement motions in this work, we make no sac- rifice in terms of overall performance. In particular, our method ranks at the top of the Middlebury benchmark.
4 0.15521537 59 cvpr-2013-Better Exploiting Motion for Better Action Recognition
Author: Mihir Jain, Hervé Jégou, Patrick Bouthemy
Abstract: Several recent works on action recognition have attested the importance of explicitly integrating motion characteristics in the video description. This paper establishes that adequately decomposing visual motion into dominant and residual motions, both in the extraction of the space-time trajectories and for the computation of descriptors, significantly improves action recognition algorithms. Then, we design a new motion descriptor, the DCS descriptor, based on differential motion scalar quantities, divergence, curl and shear features. It captures additional information on the local motion patterns enhancing results. Finally, applying the recent VLAD coding technique proposed in image retrieval provides a substantial improvement for action recognition. Our three contributions are complementary and lead to outperform all reported results by a significant margin on three challenging datasets, namely Hollywood 2, HMDB51 and Olympic Sports. 1. Introduction and related work Human actions often convey the essential meaningful content in videos. Yet, recognizing human actions in un- constrained videos is a challenging problem in Computer Vision which receives a sustained attention due to the potential applications. In particular, there is a large interest in designing video-surveillance systems, providing some automatic annotation of video archives as well as improving human-computer interaction. The solutions proposed to address this problem inherit, to a large extent, from the techniques first designed for the goal of image search and classification. The successful local features developed to describe image patches [15, 23] have been translated in the 2D+t domain as spatio-temporal local descriptors [13, 30] and now include motion clues [29]. These descriptors are often extracted from spatial-temporal interest points [12, 3 1]. More recent techniques assume some underlying temporal motion model involving trajectories [2, 6, 7, 17, 18, 25, 29, 32]. Most of these approaches produce large set of local descriptors which are in turn aggregated to produce a single vector representing the video, in order to enable the use of powerful discriminative classifiers such as support vector machines (SVMs). This is usually done with the bag- Figure 1. Optical flow field vectors (green vectors with red end points) before and after dominant motion compensation. Most of the flow vectors due to camera motion are suppressed after compensation. One of the contributions of this paper is to show that compensating for the dominant motion is beneficial for most of the existing descriptors used for action recognition. of-words technique [24], which quantizes the local features using a k-means codebook. Thanks to the successful combination of this encoding technique with the aforementioned local descriptors, the state of the art in action recognition is able to go beyond the toy problems ofclassifying simple human actions in controlled environment and considers the detection of actions in real movies or video clips [11, 16]. Despite these progresses, the existing descriptors suffer from an uncompleted handling of motion in the video sequence. Motion is arguably the most reliable source of information for action recognition, as often related to the actions of interest. However, it inevitably involves the background or camera motion when dealing with uncontrolled and re- alistic situations. Although some attempts have been made to compensate camera motion in several ways [10, 21, 26, 29, 32], how to separate action motion from that caused by the camera, and how to reflect it in the video description remains an open issue. The motion compensation mechanism employed in [10] is tailor-made to the Motion Interchange Pattern encoding technique. The Motion Boundary Histogram (MBH) [29] is a recent appealing approach to 222555555533 suppress the constant motion by considering the flow gradient. It is robust to some extent to the presence of camera motion, yet it does not explicitly handle the camera motion. Another approach [26] uses a sophisticated and robust (RANSAC) estimation of camera motion. It first segments the color image into regions corresponding to planar parts in the scene and estimates the (three) dominant homographies to update the motion associated with local features. A rather different view is adopted in [32] where the motion decomposition is performed at the trajectory level. All these works support the potential of motion compensation. As the first contribution of this paper, we address the problem in a way that departs from these works by considering the compensation of the dominant motion in both the tracking stages and encoding stages involved in the computation of action recognition descriptors. We rely on the pioneering works on motion compensation such as the technique proposed in [20], that considers 2D polynomial affine motion models for estimating the dominant image motion. We consider this particular model for its robustness and its low computational cost. It was already used in [21] to separate the dominant motion (assumed to be due to the camera motion) and the residual motion (corresponding to the independent scene motions) for dynamic event recognition in videos. However, the statistical modeling of both motion components was global (over the entire image) and only the normal flow was computed for the latter. Figure 1 shows the vectors of optical flow before and after applying the proposed motion compensation. Our method successfully suppresses most of the background motion and reinforces the focus towards the action of interest. We exploit this compensated motion both for descriptor computation and for extracting trajectories. However, we also show that the camera motion should not be thrown as it contains complementary information that is worth using to recognize certain action categories. Then, we introduce the Divergence-Curl-Shear (DCS) descriptor, which encodes scalar first-order motion features, namely the motion divergence, curl and shear. It captures physical properties of the flow pattern that are not involved in the best existing descriptors for action recognition, except in the work of [1] which exploits divergence and vorticity among a set of eleven kinematic features computed from the optical flow. Our DCS descriptor provides a good performance recognition performance on its own. Most importantly, it conveys some information which is not captured by existing descriptors and further improves the recognition performance when combined with the other descriptors. As a last contribution, we bring an encoding technique known as VLAD (vector oflocal aggregated descriptors) [8] to the field of action recognition. This technique is shown to be better than the bag-of-words representation for combining all the local video descriptors we have considered. The organization of the paper is as follows. Section 2 introduces the motion properties that we will consider through this paper. Section 3 presents the datasets and classification scheme used in our different evaluations. Section 4 details how we revisit several popular descriptors of the literature by the means of dominant motion compensation. Our DCS descriptor based on kinematic properties is introduced in Section 5 and improved by the VLAD encoding technique, which is introduced and bench-marked in Section 6 for several video descriptors. Section 7 provides a comparison showing the large improvement achieved over the state of the art. Finally, Section 8 concludes the paper. 2. Motion Separation and Kinematic Features In this section, we describe the motion clues we incorporate in our action recognition framework. We separate the dominant motion and the residual motion. In most cases, this will account to distinguishing the impact of camera movement and independent actions. Note that we do not aim at recovering the 3D camera motion: The 2D parametric motion model describes the global (or dominant) motion between successive frames. We first explain how we estimate the dominant motion and employ it to separate the dominant flow from the optical flow. Then, we will introduce kinematic features, namely divergence, curl and shear for a more comprehensive description of the visual motion. 2.1. Affine motion for compensating camera motion Among polynomial motion models, we consider the 2D affine motion model. Simplest motion models such as the 4parameter model formed by the combination of 2D translation, 2D rotation and scaling, or more complex ones such as the 8-parameter quadratic model (equivalent to a homography), could be selected as well. The affine model is a good trade-off between accuracy and efficiency which is of primary importance when processing a huge video database. It does have limitations since strictly speaking it implies a single plane assumption for the static background. However, this is not that penalizing (especially for outdoor scenes) if differences in depth remain moderated with respect to the distance to the camera. The affine flow vector at point p = (x, y) and at time t, is defined as waff(pt) =?cc12((t ) ?+?aa31((t ) aa42((t ) ? ?xytt?. (1) = + + = + uaff(pt) c1(t) a1(t)xt a2(t)yt and vaff(pt) c2(t) a3 (t)xt + a4(t)yt are horizontal and vertical components of waff(pt) respectively. Let us denote the optical flow vector at point p at time t as w(pt) = (u(pt) , v(pt)). We introduce the flow vector ω(pt) obtained by removing the affine flow vector from the optical flow vector ω(pt) = w(pt) − waff(pt) . (2) 222555555644 The dominant motion (estimated as waff(pt)) is usually due to the camera motion. In this case, Equation 2 amounts to canceling (or compensating) the camera motion. Note that this is not always true. For example in case of close-up on a moving actor, the dominant motion will be the affine estimation of the apparent actor motion. The interpretation of the motion compensation output will not be that straightforward in this case, however the resulting ω-field will still exhibit different patterns for the foreground action part and the background part. In the remainder, we will refer to the “compensated” flow as ω-flow. Figure 1 displays the computed optical flow and the ωflow. We compute the affine flow with the publicly available Motion2D software1 [20] which implements a realtime robust multiresolution incremental estimation framework. The affine motion model has correctly accounted for the motion induced by the camera movement which corresponds to the dominant motion in the image pair. Indeed, we observe that the compensated flow vectors in the background are close to null and the compensated flow in the foreground, i.e., corresponding to the actors, is conversely inflated. The experiments presented along this paper will show that effective separation of dominant motion from the residual motions is beneficial for action recognition. As explained in Section 4, we will compute local motion descriptors, such as HOF, on both the optical flow and the compensated flow (ω-flow), which allows us to explicitly and directly characterize the scene motion. 2.2. Local kinematic features By kinematic features, we mean local first-order differential scalar quantities computed on the flow field. We consider the divergence, the curl (or vorticity) and the hyperbolic terms. They inform on the physical pattern of the flow so that they convey useful information on actions in videos. They can be computed from the first-order derivatives of the flow at every point p at every frame t as ⎨⎪ ⎪ ⎪ ⎧hcdyuipvr1l2(p t) = −∂ u ∂ (yxp(xtp) +−∂ v ∂ v(px ypxt ) The diverg⎪⎩ence is related to axial motion, expansion scaling effects, the curl to rotation in the image plane. hyperbolic terms express the shear of the visual flow responding to more complex configuration. We take account the shear quantity only: shear(pt) = ?hyp12(pt) + hyp22(pt). (3) and The corinto (4) 1http://www.irisa.fr/vista/Motion2D/ In Section 5, we propose the DCS descriptor that is based on the kinematic features (divergence, curl and shear) of the visual motion discussed in this subsection. It is computed on either the optical or the compensated flow, ω-flow. 3. Datasets and evaluation This section first introduces the datasets used for the evaluation. Then, we briefly present the bag-of-feature model and the classification scheme used to encode the descriptors which will be introduced in Section 4. Hollywood2. The Hollywood2 dataset [16] contains 1,707 video clips from 69 movies representing 12 action classes. It is divided into train set and test set of 823 and 884 samples respectively. Following the standard evaluation protocol of this benchmark, we use average precision (AP) for each class and the mean of APs (mAP) for evaluation. HMDB51. The HMDB51 dataset [11] is a large dataset containing 6,766 video clips extracted from various sources, ranging from movies to YouTube. It consists of 51 action classes, each having at least 101 samples. We follow the evaluation protocol of [11] and use three train/test splits, each with 70 training and 30 testing samples per class. The average classification accuracy is computed over all classes. Out of the two released sets, we use the original set as it is more challenging and used by most of the works reporting results in action recognition. Olympic Sports. The third dataset we use is Olympic Sports [19], which again is obtained from YouTube. This dataset contains 783 samples with 16 sports action classes. We use the provided2 train/test split, there are 17 to 56 training samples and 4 to 11test samples per class. Mean AP is used for the evaluation, which is the standard choice. Bag of features and classification setup. We first adopt the standard BOF [24] approach to encode all kinds of descriptors. It produces a vector that serves as the video representation. The codebook is constructed for each type of descriptor separately by the k-means algorithm. Following a common practice in the literature [27, 29, 30], the codebook size is set to k=4,000 elements. Note that Section 6 will consider encoding technique for descriptors. For the classification, we use a non-linear SVM with χ2kernel. When combining different descriptors, we simply add the kernel matrices, as done in [27]: K(xi,xj) = exp?−?cγ1cD(xic,xjc)?, 2http://vision.stanford.edu/Datasets/OlympicSports/ 222555555755 (5) where D(xic, xjc) is χ2 distance between video xic and xjc with respect to c-th channel, corresponding to c-th descriptor. The quantity γc is the mean value of χ2 distances between the training samples for the c-th channel. The multiclass classification problem that we consider is addressed by applying a one-against-rest approach. 4. Compensated descriptors This section describes how the compensation ofthe dominant motion is exploited to improve the quality of descriptors encoding the motion and the appearance around spatio-temporal positions, hence the term “compensated descriptors”. First, we briefly review the local descriptors [5, 13, 16, 29, 30] used here along with dense trajectories [29]. Second, we analyze the impact of motion flow compensation when used in two different stages of the descriptor computation, namely in the tracking and the description part. 4.1. Dense trajectories and local descriptors Employing dense trajectories to compute local descriptors is one of the state-of-the-art approaches for action recognition. It has been shown [29] that when local descriptors are computed over dense trajectories the performance improves considerably compared to when computed over spatio temporal features [30]. Dense Trajectories [29]: The trajectories are obtained by densely tracking sampled points using optical flow fields. First, feature points are sampled from a dense grid, with step size of 5 pixels and over 8 scales. Each feature point pt = (xt, yt) at frame t is then tracked to the next frame by median filtering in a dense optical flow field F = (ut, vt) as follows: pt+1 = (xt+1 , yt+1) = (xt, yt) + (M ∗ F) | (x ¯t,y ¯t) , (6) where M is the kernel of median filtering and ( x¯ t, y¯ t) is the rounded position of (xt, yt). The tracking is limited to L (=15) frames to avoid any drifting effect. Excessively short trajectories and trajectories exhibiting sudden large displacements are removed as they induce some artifacts. Trajectories must be understood here as tracks in the spacetime volume of the video. Local descriptors: The descriptors are computed within a space-time volume centered around each trajectory. Four types of descriptors are computed to encode the shape of the trajectory, local motion pattern and appearance, namely Trajectory [29], HOF (histograms of optical flow) [13], MBH [4] and HOG (histograms of oriented gradients) [3]. All these descriptors depend on the flow field used for the tracking and as input of the descriptor computation: 1. The Trajectory descriptor encodes the shape of the trajectory represented by the normalized relative coor- × dinates of the successive points forming the trajectory. It directly depends on the dense flow used for tracking points. 2. HOF is computed using the orientations and magnitudes of the flow field. 3. MBH is designed to capture the gradient of horizontal and vertical components of the flow. The motion boundaries encode the relative pixel motion and therefore suppress camera motion, but only to some extent. 4. HOG encodes the appearance by using the intensity gradient orientations and magnitudes. It is formally not a motion descriptor. Yet the position where the descriptor is computed depends on the trajectory shape. As in [29], volume around a feature point is divided into a 2 2 3 space-time grid. The orientations are quantized ian 2to × ×8 b2i ×ns 3fo srp HacOe-Gti amned g g9r ibdi.ns T fhoer o oHriOenFt (awtioitnhs one a qdudainttiiozneadl zero bin). The horizontal and vertical components of MBH are separately quantized into 8 bins each. 4.2. Impact of motion compensation The optical flow is simply referred to as flow in the following, while the compensated flow (see subsection 2. 1) is denoted by ω-flow. Both of them are considered in the tracking and descriptor computation stages. The trajectories obtained by tracking with the ω-flow are called ω-trajectories. Figure 2 comparatively illustrates the ωtrajectories and the trajectories obtained using the flow. The input video shows a man moving away from the car. In this video excerpt, the camera is following the man walking to the right, thus inducing a global motion to the left in the video. When using the flow, the computed trajectories reflect the combination of these two motion components (camera and scene motion) as depicted by Subfigure 2(b), which hampers the characterization of the current action. In contrast, the ω-trajectories plotted in Subfigure 2(c) are more active on the actor moving on the foreground, while those localized in the background are now parallel to the time axis enhancing static parts of the scene. The ω-trajectories are therefore more relevant for action recognition, since they are more regularly and more exclusively following the actor’s motion. Impact on Trajectory and HOG descriptors. Table 1reports the impact of ω-trajectories on Trajectory and HOG descriptors, which are both significantly improved by 3%4% of mAP on the two datasets. When improved by ωflow, these descriptors will be respectively referred to as ω-Trajdesc and ω-HOG in the rest of the paper. Although the better performance of ω-Trajdesc versus the original Trajectory descriptor was expected, the one 222555555866 2. Trajectories obtained from optical and compensated flows. The green tail is the trajectory the current frame. The trajectories are sub-sampled for the sake of clarity. The frames are extracted Figure over every 15 frames with red dot indicating 5 frames in this example. DescriptorHollywood2HMDB51 BaseTrliaωnje- Tc(rtoarejrdpyreos[c2d9u]ced)54 7 1. 7 4% %2382.–89% BaseliHnωOe- (GHreOp [2rG9od]uced)4 451 . 658%%%2296.– 13%% Table 1. ω-Trajdesc and ω-HOG: Impact of compensating flow on Trajectory descriptor and HOG descriptors. achieved by ω-HOG might be surprising. Our interpretation is that HOG captures more context with the modified trajectories. More precisely, the original HOG descriptor is computed from a 2D+t sub-volume aligned with the corresponding trajectory and hence represents the appearance along the trajectory shape. When using ω-flow, we do not align the video sequence. As a result, the ω-HOG descriptor is no more computed around the very same tracked physical point in the space-time volume but around points lying in a patch of the initial feature point, whose size depends on the affine flow magnitude. ω-HOG can be viewed as a “patchbased” computation capturing more information about the appearance of the background or of the moving foreground. As for ω-trajectories, they are closer to the real trajectories of the moving actors as they usually cancel the camera movement, and so, more easier to train and recognize. Impact on HOF. The ω-flow impacts computation used as an input to HOF computation itself. Therefore, HOF can both types of trajectories (ω-trajectories both the trajectory and the descriptor be computed along or those extracted MethodHollywood2HMDB51 Table(ω2rHf.-alocO IwomkF)inpHgacOtFobf[2u9ωhsb]i:f-nlo ωgwot-hwωHOflFown5H 02 34O. 58291F% %descripto3 r706s38.:–1076% m%APfor Hollywood2 and average accuracy for HMDB5 1. The ω-HOF is used in subsequent evaluations. from flow) and can encode both kinds of flows (ω-flow or flow). For the sake of completeness, we evaluate all the variants as well as the combination of both flows in the descriptor computation stage. The results are presented in Table 2 and demonstrate the significant improvement obtained by computing the HOF descriptor with the ω-flow instead of the optical flow. Note that the type of trajectories which is used, either “Tracking flow” or “Tracking ω-flow”, has a limited impact in this case. From now on, we only consider the “Tracking ω-flow” case where HOF is computed along ω-trajectories. Interestingly, combining the HOF computed from the flow and the ω-flow further improves the results. This suggests that the two flow fields are complementary and the affine flow that was subtracted from ω-flow brings in additional information. For the sake of brevity, the combination of the two kinds of HOF, i.e., computed from the flow and the ω-flow using ω-trajectories, is referred to as the ω-HOF 222555555977 MethodHollywood2HMDB51 Tab(lerT3a.cIkmM inpBgMacHgωtBf-loH w [u2)s9in]gω f-lo wo MBH5 d42 e.052s7c% riptos:m34A90P.–3769f% orHllywood2 and average accuracy for HMDB5 1. DTerHasMjcBrOeblitpHGeFor4.ySumTωraw- frcilykto hw ionfgtheduωpCs-fcaolrtωmeiwNp-df/tl+Aωoutrw-finlogwthdesωcr- isTpc-fHtrMloaiOjrBpdswtGeHFosrc descriptor in the rest of this paper. Compared to the HOF baseline, the ω-HOF descriptor achieves a gain of +3.1% of mAP on Hollywood 2 and of +7.8% on HMDB51. Impact on MBH. Since MBH is computed from gradient of flow and cancel the constant motion, there is practically no benefit in using the ω-flow to compute the MBH descriptors, as shown in Table 3. However, by tracking ω-flow, the performance improves by around 1.3% for HMDB5 1 dataset and drops by around 1.5% for Hollywood2. This relative performance depends on the encoding technique. We will come back on this descriptor when considering another encoding scheme for local descriptors in Section 6. 4.3. Summary of compensated descriptors Table 4 summarizes the refined versions of the descriptors obtained by exploiting the ω-flow, and both ω-flow and the optical flow in the case of HOF. The revisited descriptors considerably improve the results compared to the orig- inal ones, with the noticeable exception of ω-MBH which gives mixed performance with a bag-of-features encoding scheme. But we already mention as this point that this incongruous behavior of ω-MBH is stabilized with the VLAD encoding scheme considered in Section 6. Another advantage of tracking the compensated flow is that fewer trajectories are produced. For instance, the total number of trajectories decreases by about 9. 16% and 22.81% on the Hollywood2 and HMDB51 datasets, respectively. Note that exploiting both the flow and the ω-flow do not induce much computational overhead, as the latter is obtained from the flow and the affine flow which is computed in real-time and already used to get the ω-trajectories. The only additional computational cost that we introduce by using the descriptors summarized in Table 4 is the computation of a second HOF descriptor, but this stage is relatively efficient and not the bottleneck of the extraction procedure. 5. Divergence-Curl-Shear descriptor This section introduces a new descriptor encoding the kinematic properties of motion discussed in Section 2.2. It is denoted by DCS in the rest of this paper. Combining kinematic features. The spatial derivatives are computed for the horizontal and vertical components of the flow field, which are used in turn to compute the divergence, curl and shear scalar values, see Equation 3. We consider all possible pairs of kinematic features, namely (div, curl), (div, shear) and (curl, shear). At each × ×× pixel, we compute the orientation and magnitude of the 2-D vector corresponding to each of these pairs. The orientation is quantized into histograms and the magnitude is used for weighting, similar to SIFT. Our motivation for encoding pairs is that the joint distribution of kinematic features conveys more information than exploiting them independently. Implementation details. The descriptor computation and parameters are similar to HOG and other popular descriptors such as MBH, HOF. We obtain 8-bin histograms for each of the three feature pairs or components of DCS. The range of possible angles is 2π for the (div,curl) pair and π for the other pairs, because the shear is always positive. The DCS descriptor is computed for a space-time volume aligned with a trajectory, as done with the four descriptors mentioned in the previous section. In order to capture the spatio-temporal structure of kinematic features, the volume (32 32 pixels and L = 15 frames) is subdivided into a spatio-temporal grid nofd s Lize = nx 5× f ny m×e nt, sw situhb nx =de ny =to 2a and nt = 3. These parameters ×hnave× × bneen fixed for the sake of consistency with the other descriptors. For each pair of kinematic features, each cell in the grid is represented by a histogram. The resulting local descriptors have a dimensionality equal to 288 = nx ny nt 8 3. At the video level, these descriptors are nenc×od end i×nto 8 a single vector representation using either BOF or the VLAD encoding scheme introduced in the next section. 6. VLAD in actions VLAD [8] is a descriptor encoding technique that aggregates the descriptors based on a locality criterion in the feature space. To our knowledge, this technique has never been considered for action recognition. Below, we briefly introduce this approach and give the performance achieved for all the descriptors introduced along the previous sections. VLAD in brief. Similar to BOF, VLAD relies on a codebook C = {c1, c2 , ...ck} of k centroids learned by k-means. bTohoek representation is ob}t oaifn ked c by summing, efodr b yea kch-m mveiasunasl. word ci, the differences x − ci of the vectors x assigned to ci, thereby producing a sv exct −or c representation oflength d×k, 222555556088 DMeBscHriptorV5 LH.A1o%Dlywo5Bo4d.O2 %F4V3L.3HA%MD B35B91.O7%F Taωbl-eDHM5rOCBa.FSjGPdHe+rsωfco-mMHaBOnFeofV54L2936A.51D% with5431ω208-.5T96% rajde3s42c97158,.ω3% -HOG342,58019ω.6-% HOF descriptors and their combination. where d is the dimension ofthe local descriptors. We use the codebook size, k = 256. Despite this large dimensionality, VLAD is efficient because it is effectively compared with a linear kernel. VLAD is post-processed using a componentwise power normalization, which dramatically improves its performance [8]. While cross validating the parameter α involved in this power normalization, we consistently observe, for all the descriptors, a value between 0.15 and 0.3. Therefore, this parameter is set to α = 0.2 in all our experiments. For classification, we use a linear SVM and oneagainst-rest approach everywhere, unless stated otherwise. Impact on existing descriptors. We employ VLAD because it is less sensitive to quantization parameters and appears to provide better performance with descriptors having a large dimensionality. These properties are interesting in our case, because the quantization parameters involved in the DCS and MBH descriptors have been used unchanged in Section 4 for the sake of direct comparison. They might be suboptimal when using the ω-flow instead of the optical flow on which they have initially been optimized [29]. Results for MBH and ω-MBH in Table 5 supports this argument. When using VLAD instead of BOF, the scores are stable in both the cases and there is no mixed inference as that observed in Table 3. VLAD also has significant positive influence on accuracy of ω-DCS descriptor. We also observe that ω-DCS is complementary to ω-MBH and adds to the performance. Still DCS is probably not best utilized in the current setting of parameters. In case of ω-Trajdesc and ω-HOG, the scores are better with BOF on both the datasets. ω-HOF with VLAD improves on HMDB5 1, but remains equivalent for Hollywood2. Although BOF leads to better scores for the descriptors considered individually, their combination with VLAD outperforms the BOF. 7. Comparison with the state of the art This section reports our results with all descriptors combined and compares our method with the state of the art. TrajectorCy+omHbOiGna+tHioOnF+ MDBCHSHol5 l98y.w76%o%od2H4M489.D02%B%51 All ω-descriptors all five compensated descriptors using combined62.5%52.1% Table 6. Combination of VLAD representation. WU*JVliaOnughreM tHaeolth. [yo2w9d87o] 256 0985. 37% SKa*duOeJinhrau tnegdMteHatlMa.h [ol1Dd.0B[91]25 24 609.8172% Table 7. Comparison with the state of the art on Hollywood2 and HMDB5 1 datasets. *Vig et al. [28] gets 61.9% by using external eye movements data. *Jiang et al. [9] used one-vs-one multi class SVM while our and other methods use one-vs-rest SVMs. With one-against-one multi class SVM we obtain 45. 1% for HMDB51. Descriptor combination. Table 6 reports the results obtained when the descriptors are combined. Since we use VLAD, our baseline is updated that is combination of Trajectory, HOG, HOF and MBH with VLAD representation. When DCS is added to the baseline there is an improvement of 0.9% and 1.2%. With combination of all five compensated descriptors we obtain 62.5% and 52.1% on the two datasets. This is a large improvement even over the updated baseline, which shows that the proposed motion compensation and the way we exploit it are significantly important for action recognition. The comparison with the state of the art is shown in Table 7. Our method outperforms all the previously reported results in the literature. In particular, on the HMDB51 dataset, the improvement over the best reported results to date is more than 11% in average accuracy. Jiang el al. [9] used a one-against-one multi-class SVM, which might have resulted in inferior scores. With a similar multi-class SVM approach, our method obtains 45. 1%, which remains significantly better than their result. All others results were reported with one-against-rest approach. On Olympic Sports dataset we obtain mAP of 83.2% with ‘All ω-descriptors combined’ and the improvement is mostly because of VLAD and ω-flow. The best reported mAPs on this dataset are Liu et al. [14] (74.4%) and Jiang et al. [9] (80.6%), which we exceed convincingly. Gaidon et al. [6] reports the best average accuracy of 82.7%. 8. Conclusions This paper first demonstrates the interest of canceling the dominant motion (predominantly camera motion) to make the visual motion truly related to actions, for both the trajectory extraction and descriptor computation stages. It pro222555556199 duces significantly better versions (called compensated descriptors) of several state-of-the-art local descriptors for action recognition. The simplicity, efficiency and effectiveness of this motion compensation approach make it applicable to any action recognition framework based on motion descriptors and trajectories. The second contribution is the new DCS descriptor derived from the first-order scalar motion quantities specifying the local motion patterns. It captures additional information which is proved complementary to the other descriptors. Finally, we show that VLAD encoding technique instead of bag-of-words boosts several action descriptors, and overall exhibits a significantly better performance when combining different types of descriptors. Our contributions are all complementary and significantly outperform the state of the art when combined, as demon- strated by our extensive experiments on the Hollywood 2, HMDB51 and Olympic Sports datasets. Acknowledgments This work was supported by the Quaero project, funded by Oseo, French agency for innovation. We acknowledge Heng Wang’s help for reproducing some of their results. References [1] S. Ali and M. Shah. Human action recognition in videos using kinematic features and multiple instance learning. IEEE T-PAMI, 32(2):288–303, Feb. 2010. [2] T. Brox and J. Malik. Object segmentation by long term analysis of point trajectories. In ECCV, Sep. 2010. [3] N. Dalal and B. Triggs. Histograms of oriented gradients for human detection. In CVPR, Jun. 2005. [4] N. Dalal, B. Triggs, and C. Schmid. Human detection using oriented histograms of flow and appearance. In ECCV, May 2006. [5] P. Dollar, V. Rabaud, G. Cottrell, and S. Belongie. Behavior recognition via sparse spatio-temporal features. In VS-PETS, Oct. 2005. [6] A. Gaidon, Z. Harchaoui, and C. Schmid. Recognizing activities with cluster-trees of tracklets. In BMVC, Sep. 2012. [7] A. Hervieu, P. Bouthemy, and J.-P. Le Cadre. A statistical video content recognition method using invariant features on object trajectories. IEEE T-CSVT, 18(1 1): 1533–1543, 2008. [8] H. J ´egou, F. Perronnin, M. Douze, J. S ´anchez, P. P ´erez, and C. Schmid. Aggregating local descriptors into compact [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] codes. IEEE T-PAMI, 34(9):1704–1716, 2012. Y.-G. Jiang, Q. Dai, X. Xue, W. Liu, and C.-W. Ngo. Trajectory-based modeling of human actions with motion reference points. In ECCV, Oct. 2012. O. Kliper-Gross, Y. Gurovich, T. Hassner, and L. Wolf. Motion interchange patterns for action recognition in unconstrained videos. In ECCV, Oct. 2012. H. Kuehne, H. Jhuang, E. Garrote, T. Poggio, and T. Serre. Hmdb: A large video database for human motion recognition. In ICCV, Nov. 2011. I. Laptev and T. Lindeberg. Space-time interest points. In ICCV, Oct. 2003. I. Laptev, M. Marzalek, C. Schmid, and B. Rozenfeld. Learning realistic human actions from movies. In CVPR, Jun. 2008. J. Liu, B. Kuipers, and S. Savarese. Recognizing human actions by attributes. In CVPR, Jun. 2011. D. Lowe. Distinctive image features from scale-invariant keypoints. IJCV, 60(2):91–1 10, Nov. 2004. M. Marzalek, I. Laptev, and C. Schmid. Actions in context. In CVPR, Jun. 2009. P. Matikainen, M. Hebert, and R. Sukthankar. Trajectons: Action recognition through the motion analysis of tracked features. In Workshop on Video-Oriented Object and Event Classification, ICCV, Sep. 2009. R. Messing, C. J. Pal, and H. A. Kautz. Activity recognition using the velocity histories of tracked keypoints. In ICCV, Sep. 2009. J. C. Niebles, C.-W. Chen, and F.-F. Li. Modeling temporal structure of decomposable motion segments for activity classification. In ECCV, Sep. 2010. [20] J.-M. Odobez and P. Bouthemy. Robust multiresolution estimation of parametric motion models. Jal of Vis. Comm. and Image Representation, 6(4):348–365, Dec. 1995. [21] G. Piriou, P. Bouthemy, and J.-F. Yao. Recognition of dynamic video contents with global probabilistic models of visual motion. IEEE T-IP, 15(1 1):3417–3430, 2006. [22] S. Sadanand and J. J. Corso. Action bank: A high-level representation of activity in video. In CVPR, Jun. 2012. [23] C. Schmid and R. Mohr. Local grayvalue invariants for image retrieval. IEEE T-PAMI, 19(5):530–534, May 1997. [24] J. Sivic and A. Zisserman. Video Google: A text retrieval approach to object matching in videos. In ICCV, pages 1470– 1477, Oct. 2003. [25] J. Sun, X. Wu, S. Yan, L. F. Cheong, T.-S. Chua, and J. Li. Hierarchical spatio-temporal context modeling for action recognition. In CVPR, Jun. 2009. [26] H. Uemura, S. Ishikawa, and K. Mikolajczyk. Feature tracking and motion compensation for action recognition. In BMVC, Sep. 2008. [27] M. M. Ullah, S. N. Parizi, and I. Laptev. Improving bag-offeatures action recognition with non-local cues. In BMVC, Sep. 2010. [28] E. Vig, M. Dorr, and D. Cox. Saliency-based space-variant descriptor sampling for action recognition. In ECCV, Oct. 2012. [29] H. Wang, A. Kl¨ aser, C. Schmid, and C.-L. Liu. Action recognition by dense trajectories. In CVPR, Jun. 2011. [30] H. Wang, M. M. Ullah, A. Kl¨ aser, I. Laptev, and C. Schmid. Evaluation of local spatio-temporal features for action recognition. In BMVC, Sep. 2009. [3 1] G. Willems, T. Tuytelaars, and L. J. V. Gool. An efficient dense and scale-invariant spatio-temporal interest point detector. In ECCV, Oct. 2008. [32] S. Wu, O. Oreifej, and M. Shah. Action recognition in videos acquired by a moving camera using motion decomposition of lagrangian particle trajectories. In ICCV, Nov. 2011. 222555666200
5 0.13689841 158 cvpr-2013-Exploring Weak Stabilization for Motion Feature Extraction
Author: Dennis Park, C. Lawrence Zitnick, Deva Ramanan, Piotr Dollár
Abstract: We describe novel but simple motion features for the problem of detecting objects in video sequences. Previous approaches either compute optical flow or temporal differences on video frame pairs with various assumptions about stabilization. We describe a combined approach that uses coarse-scale flow and fine-scale temporal difference features. Our approach performs weak motion stabilization by factoring out camera motion and coarse object motion while preserving nonrigid motions that serve as useful cues for recognition. We show results for pedestrian detection and human pose estimation in video sequences, achieving state-of-the-art results in both. In particular, given a fixed detection rate our method achieves a five-fold reduction in false positives over prior art on the Caltech Pedestrian benchmark. Finally, we perform extensive diagnostic experiments to reveal what aspects of our system are crucial for good performance. Proper stabilization, long time-scale features, and proper normalization are all critical.
6 0.12949592 319 cvpr-2013-Optimized Product Quantization for Approximate Nearest Neighbor Search
7 0.12470976 170 cvpr-2013-Fast Rigid Motion Segmentation via Incrementally-Complex Local Models
8 0.11523835 334 cvpr-2013-Pose from Flow and Flow from Pose
9 0.11503229 66 cvpr-2013-Block and Group Regularized Sparse Modeling for Dictionary Learning
10 0.11405347 369 cvpr-2013-Rotation, Scaling and Deformation Invariant Scattering for Texture Discrimination
11 0.11399329 10 cvpr-2013-A Fully-Connected Layered Model of Foreground and Background Flow
12 0.09929274 108 cvpr-2013-Dense 3D Reconstruction from Severely Blurred Images Using a Single Moving Camera
13 0.099264152 124 cvpr-2013-Determining Motion Directly from Normal Flows Upon the Use of a Spherical Eye Platform
14 0.095884413 113 cvpr-2013-Dense Variational Reconstruction of Non-rigid Surfaces from Monocular Video
15 0.090104342 79 cvpr-2013-Cartesian K-Means
16 0.089828305 345 cvpr-2013-Real-Time Model-Based Rigid Object Pose Estimation and Tracking Combining Dense and Sparse Visual Cues
17 0.08798676 453 cvpr-2013-Video Editing with Temporal, Spatial and Appearance Consistency
18 0.086520016 203 cvpr-2013-Hierarchical Video Representation with Trajectory Binary Partition Tree
19 0.085560933 344 cvpr-2013-Radial Distortion Self-Calibration
20 0.085393421 187 cvpr-2013-Geometric Context from Videos
topicId topicWeight
[(0, 0.196), (1, 0.08), (2, -0.034), (3, 0.024), (4, -0.06), (5, -0.003), (6, 0.02), (7, -0.116), (8, -0.09), (9, 0.043), (10, 0.051), (11, 0.11), (12, 0.082), (13, -0.031), (14, 0.124), (15, 0.074), (16, 0.061), (17, -0.032), (18, -0.016), (19, 0.01), (20, -0.055), (21, -0.047), (22, 0.055), (23, -0.053), (24, 0.016), (25, 0.074), (26, -0.046), (27, -0.011), (28, -0.015), (29, -0.071), (30, 0.014), (31, 0.023), (32, 0.04), (33, -0.013), (34, -0.108), (35, 0.064), (36, -0.053), (37, -0.092), (38, -0.045), (39, -0.0), (40, -0.088), (41, -0.003), (42, 0.037), (43, 0.009), (44, -0.002), (45, -0.071), (46, -0.006), (47, -0.073), (48, 0.022), (49, 0.043)]
simIndex simValue paperId paperTitle
same-paper 1 0.94839108 88 cvpr-2013-Compressible Motion Fields
Author: Giuseppe Ottaviano, Pushmeet Kohli
Abstract: Traditional video compression methods obtain a compact representation for image frames by computing coarse motion fields defined on patches of pixels called blocks, in order to compensate for the motion in the scene across frames. This piecewise constant approximation makes the motion field efficiently encodable, but it introduces block artifacts in the warped image frame. In this paper, we address the problem of estimating dense motion fields that, while accurately predicting one frame from a given reference frame by warping it with the field, are also compressible. We introduce a representation for motion fields based on wavelet bases, and approximate the compressibility of their coefficients with a piecewise smooth surrogate function that yields an objective function similar to classical optical flow formulations. We then show how to quantize and encode such coefficients with adaptive precision. We demonstrate the effectiveness of our approach by com- paring its performance with a state-of-the-art wavelet video encoder. Experimental results on a number of standard flow and video datasets reveal that our method significantly outperforms both block-based and optical-flow-based motion compensation algorithms.
2 0.72051305 244 cvpr-2013-Large Displacement Optical Flow from Nearest Neighbor Fields
Author: Zhuoyuan Chen, Hailin Jin, Zhe Lin, Scott Cohen, Ying Wu
Abstract: We present an optical flow algorithm for large displacement motions. Most existing optical flow methods use the standard coarse-to-fine framework to deal with large displacement motions which has intrinsic limitations. Instead, we formulate the motion estimation problem as a motion segmentation problem. We use approximate nearest neighbor fields to compute an initial motion field and use a robust algorithm to compute a set of similarity transformations as the motion candidates for segmentation. To account for deviations from similarity transformations, we add local deformations in the segmentation process. We also observe that small objects can be better recovered using translations as the motion candidates. We fuse the motion results obtained under similarity transformations and under translations together before a final refinement. Experimental validation shows that our method can successfully handle large displacement motions. Although we particularly focus on large displacement motions in this work, we make no sac- rifice in terms of overall performance. In particular, our method ranks at the top of the Middlebury benchmark.
3 0.65462488 124 cvpr-2013-Determining Motion Directly from Normal Flows Upon the Use of a Spherical Eye Platform
Author: Tak-Wai Hui, Ronald Chung
Abstract: We address the problem of recovering camera motion from video data, which does not require the establishment of feature correspondences or computation of optical flows but from normal flows directly. We have designed an imaging system that has a wide field of view by fixating a number of cameras together to form an approximate spherical eye. With a substantially widened visual field, we discover that estimating the directions of translation and rotation components of the motion separately are possible and particularly efficient. In addition, the inherent ambiguities between translation and rotation also disappear. Magnitude of rotation is recovered subsequently. Experimental results on synthetic and real image data are provided. The results show that not only the accuracy of motion estimation is comparable to those of the state-of-the-art methods that require explicit feature correspondences or optical flows, but also a faster computation time.
4 0.65124786 10 cvpr-2013-A Fully-Connected Layered Model of Foreground and Background Flow
Author: Deqing Sun, Jonas Wulff, Erik B. Sudderth, Hanspeter Pfister, Michael J. Black
Abstract: Layered models allow scene segmentation and motion estimation to be formulated together and to inform one another. Traditional layered motion methods, however, employ fairly weak models of scene structure, relying on locally connected Ising/Potts models which have limited ability to capture long-range correlations in natural scenes. To address this, we formulate a fully-connected layered model that enables global reasoning about the complicated segmentations of real objects. Optimization with fully-connected graphical models is challenging, and our inference algorithm leverages recent work on efficient mean field updates for fully-connected conditional random fields. These methods can be implemented efficiently using high-dimensional Gaussian filtering. We combine these ideas with a layered flow model, and find that the long-range connections greatly improve segmentation into figure-ground layers when compared with locally connected MRF models. Experiments on several benchmark datasets show that the method can re- cover fine structures and large occlusion regions, with good flow accuracy and much lower computational cost than previous locally-connected layered models.
5 0.64217192 59 cvpr-2013-Better Exploiting Motion for Better Action Recognition
Author: Mihir Jain, Hervé Jégou, Patrick Bouthemy
Abstract: Several recent works on action recognition have attested the importance of explicitly integrating motion characteristics in the video description. This paper establishes that adequately decomposing visual motion into dominant and residual motions, both in the extraction of the space-time trajectories and for the computation of descriptors, significantly improves action recognition algorithms. Then, we design a new motion descriptor, the DCS descriptor, based on differential motion scalar quantities, divergence, curl and shear features. It captures additional information on the local motion patterns enhancing results. Finally, applying the recent VLAD coding technique proposed in image retrieval provides a substantial improvement for action recognition. Our three contributions are complementary and lead to outperform all reported results by a significant margin on three challenging datasets, namely Hollywood 2, HMDB51 and Olympic Sports. 1. Introduction and related work Human actions often convey the essential meaningful content in videos. Yet, recognizing human actions in un- constrained videos is a challenging problem in Computer Vision which receives a sustained attention due to the potential applications. In particular, there is a large interest in designing video-surveillance systems, providing some automatic annotation of video archives as well as improving human-computer interaction. The solutions proposed to address this problem inherit, to a large extent, from the techniques first designed for the goal of image search and classification. The successful local features developed to describe image patches [15, 23] have been translated in the 2D+t domain as spatio-temporal local descriptors [13, 30] and now include motion clues [29]. These descriptors are often extracted from spatial-temporal interest points [12, 3 1]. More recent techniques assume some underlying temporal motion model involving trajectories [2, 6, 7, 17, 18, 25, 29, 32]. Most of these approaches produce large set of local descriptors which are in turn aggregated to produce a single vector representing the video, in order to enable the use of powerful discriminative classifiers such as support vector machines (SVMs). This is usually done with the bag- Figure 1. Optical flow field vectors (green vectors with red end points) before and after dominant motion compensation. Most of the flow vectors due to camera motion are suppressed after compensation. One of the contributions of this paper is to show that compensating for the dominant motion is beneficial for most of the existing descriptors used for action recognition. of-words technique [24], which quantizes the local features using a k-means codebook. Thanks to the successful combination of this encoding technique with the aforementioned local descriptors, the state of the art in action recognition is able to go beyond the toy problems ofclassifying simple human actions in controlled environment and considers the detection of actions in real movies or video clips [11, 16]. Despite these progresses, the existing descriptors suffer from an uncompleted handling of motion in the video sequence. Motion is arguably the most reliable source of information for action recognition, as often related to the actions of interest. However, it inevitably involves the background or camera motion when dealing with uncontrolled and re- alistic situations. Although some attempts have been made to compensate camera motion in several ways [10, 21, 26, 29, 32], how to separate action motion from that caused by the camera, and how to reflect it in the video description remains an open issue. The motion compensation mechanism employed in [10] is tailor-made to the Motion Interchange Pattern encoding technique. The Motion Boundary Histogram (MBH) [29] is a recent appealing approach to 222555555533 suppress the constant motion by considering the flow gradient. It is robust to some extent to the presence of camera motion, yet it does not explicitly handle the camera motion. Another approach [26] uses a sophisticated and robust (RANSAC) estimation of camera motion. It first segments the color image into regions corresponding to planar parts in the scene and estimates the (three) dominant homographies to update the motion associated with local features. A rather different view is adopted in [32] where the motion decomposition is performed at the trajectory level. All these works support the potential of motion compensation. As the first contribution of this paper, we address the problem in a way that departs from these works by considering the compensation of the dominant motion in both the tracking stages and encoding stages involved in the computation of action recognition descriptors. We rely on the pioneering works on motion compensation such as the technique proposed in [20], that considers 2D polynomial affine motion models for estimating the dominant image motion. We consider this particular model for its robustness and its low computational cost. It was already used in [21] to separate the dominant motion (assumed to be due to the camera motion) and the residual motion (corresponding to the independent scene motions) for dynamic event recognition in videos. However, the statistical modeling of both motion components was global (over the entire image) and only the normal flow was computed for the latter. Figure 1 shows the vectors of optical flow before and after applying the proposed motion compensation. Our method successfully suppresses most of the background motion and reinforces the focus towards the action of interest. We exploit this compensated motion both for descriptor computation and for extracting trajectories. However, we also show that the camera motion should not be thrown as it contains complementary information that is worth using to recognize certain action categories. Then, we introduce the Divergence-Curl-Shear (DCS) descriptor, which encodes scalar first-order motion features, namely the motion divergence, curl and shear. It captures physical properties of the flow pattern that are not involved in the best existing descriptors for action recognition, except in the work of [1] which exploits divergence and vorticity among a set of eleven kinematic features computed from the optical flow. Our DCS descriptor provides a good performance recognition performance on its own. Most importantly, it conveys some information which is not captured by existing descriptors and further improves the recognition performance when combined with the other descriptors. As a last contribution, we bring an encoding technique known as VLAD (vector oflocal aggregated descriptors) [8] to the field of action recognition. This technique is shown to be better than the bag-of-words representation for combining all the local video descriptors we have considered. The organization of the paper is as follows. Section 2 introduces the motion properties that we will consider through this paper. Section 3 presents the datasets and classification scheme used in our different evaluations. Section 4 details how we revisit several popular descriptors of the literature by the means of dominant motion compensation. Our DCS descriptor based on kinematic properties is introduced in Section 5 and improved by the VLAD encoding technique, which is introduced and bench-marked in Section 6 for several video descriptors. Section 7 provides a comparison showing the large improvement achieved over the state of the art. Finally, Section 8 concludes the paper. 2. Motion Separation and Kinematic Features In this section, we describe the motion clues we incorporate in our action recognition framework. We separate the dominant motion and the residual motion. In most cases, this will account to distinguishing the impact of camera movement and independent actions. Note that we do not aim at recovering the 3D camera motion: The 2D parametric motion model describes the global (or dominant) motion between successive frames. We first explain how we estimate the dominant motion and employ it to separate the dominant flow from the optical flow. Then, we will introduce kinematic features, namely divergence, curl and shear for a more comprehensive description of the visual motion. 2.1. Affine motion for compensating camera motion Among polynomial motion models, we consider the 2D affine motion model. Simplest motion models such as the 4parameter model formed by the combination of 2D translation, 2D rotation and scaling, or more complex ones such as the 8-parameter quadratic model (equivalent to a homography), could be selected as well. The affine model is a good trade-off between accuracy and efficiency which is of primary importance when processing a huge video database. It does have limitations since strictly speaking it implies a single plane assumption for the static background. However, this is not that penalizing (especially for outdoor scenes) if differences in depth remain moderated with respect to the distance to the camera. The affine flow vector at point p = (x, y) and at time t, is defined as waff(pt) =?cc12((t ) ?+?aa31((t ) aa42((t ) ? ?xytt?. (1) = + + = + uaff(pt) c1(t) a1(t)xt a2(t)yt and vaff(pt) c2(t) a3 (t)xt + a4(t)yt are horizontal and vertical components of waff(pt) respectively. Let us denote the optical flow vector at point p at time t as w(pt) = (u(pt) , v(pt)). We introduce the flow vector ω(pt) obtained by removing the affine flow vector from the optical flow vector ω(pt) = w(pt) − waff(pt) . (2) 222555555644 The dominant motion (estimated as waff(pt)) is usually due to the camera motion. In this case, Equation 2 amounts to canceling (or compensating) the camera motion. Note that this is not always true. For example in case of close-up on a moving actor, the dominant motion will be the affine estimation of the apparent actor motion. The interpretation of the motion compensation output will not be that straightforward in this case, however the resulting ω-field will still exhibit different patterns for the foreground action part and the background part. In the remainder, we will refer to the “compensated” flow as ω-flow. Figure 1 displays the computed optical flow and the ωflow. We compute the affine flow with the publicly available Motion2D software1 [20] which implements a realtime robust multiresolution incremental estimation framework. The affine motion model has correctly accounted for the motion induced by the camera movement which corresponds to the dominant motion in the image pair. Indeed, we observe that the compensated flow vectors in the background are close to null and the compensated flow in the foreground, i.e., corresponding to the actors, is conversely inflated. The experiments presented along this paper will show that effective separation of dominant motion from the residual motions is beneficial for action recognition. As explained in Section 4, we will compute local motion descriptors, such as HOF, on both the optical flow and the compensated flow (ω-flow), which allows us to explicitly and directly characterize the scene motion. 2.2. Local kinematic features By kinematic features, we mean local first-order differential scalar quantities computed on the flow field. We consider the divergence, the curl (or vorticity) and the hyperbolic terms. They inform on the physical pattern of the flow so that they convey useful information on actions in videos. They can be computed from the first-order derivatives of the flow at every point p at every frame t as ⎨⎪ ⎪ ⎪ ⎧hcdyuipvr1l2(p t) = −∂ u ∂ (yxp(xtp) +−∂ v ∂ v(px ypxt ) The diverg⎪⎩ence is related to axial motion, expansion scaling effects, the curl to rotation in the image plane. hyperbolic terms express the shear of the visual flow responding to more complex configuration. We take account the shear quantity only: shear(pt) = ?hyp12(pt) + hyp22(pt). (3) and The corinto (4) 1http://www.irisa.fr/vista/Motion2D/ In Section 5, we propose the DCS descriptor that is based on the kinematic features (divergence, curl and shear) of the visual motion discussed in this subsection. It is computed on either the optical or the compensated flow, ω-flow. 3. Datasets and evaluation This section first introduces the datasets used for the evaluation. Then, we briefly present the bag-of-feature model and the classification scheme used to encode the descriptors which will be introduced in Section 4. Hollywood2. The Hollywood2 dataset [16] contains 1,707 video clips from 69 movies representing 12 action classes. It is divided into train set and test set of 823 and 884 samples respectively. Following the standard evaluation protocol of this benchmark, we use average precision (AP) for each class and the mean of APs (mAP) for evaluation. HMDB51. The HMDB51 dataset [11] is a large dataset containing 6,766 video clips extracted from various sources, ranging from movies to YouTube. It consists of 51 action classes, each having at least 101 samples. We follow the evaluation protocol of [11] and use three train/test splits, each with 70 training and 30 testing samples per class. The average classification accuracy is computed over all classes. Out of the two released sets, we use the original set as it is more challenging and used by most of the works reporting results in action recognition. Olympic Sports. The third dataset we use is Olympic Sports [19], which again is obtained from YouTube. This dataset contains 783 samples with 16 sports action classes. We use the provided2 train/test split, there are 17 to 56 training samples and 4 to 11test samples per class. Mean AP is used for the evaluation, which is the standard choice. Bag of features and classification setup. We first adopt the standard BOF [24] approach to encode all kinds of descriptors. It produces a vector that serves as the video representation. The codebook is constructed for each type of descriptor separately by the k-means algorithm. Following a common practice in the literature [27, 29, 30], the codebook size is set to k=4,000 elements. Note that Section 6 will consider encoding technique for descriptors. For the classification, we use a non-linear SVM with χ2kernel. When combining different descriptors, we simply add the kernel matrices, as done in [27]: K(xi,xj) = exp?−?cγ1cD(xic,xjc)?, 2http://vision.stanford.edu/Datasets/OlympicSports/ 222555555755 (5) where D(xic, xjc) is χ2 distance between video xic and xjc with respect to c-th channel, corresponding to c-th descriptor. The quantity γc is the mean value of χ2 distances between the training samples for the c-th channel. The multiclass classification problem that we consider is addressed by applying a one-against-rest approach. 4. Compensated descriptors This section describes how the compensation ofthe dominant motion is exploited to improve the quality of descriptors encoding the motion and the appearance around spatio-temporal positions, hence the term “compensated descriptors”. First, we briefly review the local descriptors [5, 13, 16, 29, 30] used here along with dense trajectories [29]. Second, we analyze the impact of motion flow compensation when used in two different stages of the descriptor computation, namely in the tracking and the description part. 4.1. Dense trajectories and local descriptors Employing dense trajectories to compute local descriptors is one of the state-of-the-art approaches for action recognition. It has been shown [29] that when local descriptors are computed over dense trajectories the performance improves considerably compared to when computed over spatio temporal features [30]. Dense Trajectories [29]: The trajectories are obtained by densely tracking sampled points using optical flow fields. First, feature points are sampled from a dense grid, with step size of 5 pixels and over 8 scales. Each feature point pt = (xt, yt) at frame t is then tracked to the next frame by median filtering in a dense optical flow field F = (ut, vt) as follows: pt+1 = (xt+1 , yt+1) = (xt, yt) + (M ∗ F) | (x ¯t,y ¯t) , (6) where M is the kernel of median filtering and ( x¯ t, y¯ t) is the rounded position of (xt, yt). The tracking is limited to L (=15) frames to avoid any drifting effect. Excessively short trajectories and trajectories exhibiting sudden large displacements are removed as they induce some artifacts. Trajectories must be understood here as tracks in the spacetime volume of the video. Local descriptors: The descriptors are computed within a space-time volume centered around each trajectory. Four types of descriptors are computed to encode the shape of the trajectory, local motion pattern and appearance, namely Trajectory [29], HOF (histograms of optical flow) [13], MBH [4] and HOG (histograms of oriented gradients) [3]. All these descriptors depend on the flow field used for the tracking and as input of the descriptor computation: 1. The Trajectory descriptor encodes the shape of the trajectory represented by the normalized relative coor- × dinates of the successive points forming the trajectory. It directly depends on the dense flow used for tracking points. 2. HOF is computed using the orientations and magnitudes of the flow field. 3. MBH is designed to capture the gradient of horizontal and vertical components of the flow. The motion boundaries encode the relative pixel motion and therefore suppress camera motion, but only to some extent. 4. HOG encodes the appearance by using the intensity gradient orientations and magnitudes. It is formally not a motion descriptor. Yet the position where the descriptor is computed depends on the trajectory shape. As in [29], volume around a feature point is divided into a 2 2 3 space-time grid. The orientations are quantized ian 2to × ×8 b2i ×ns 3fo srp HacOe-Gti amned g g9r ibdi.ns T fhoer o oHriOenFt (awtioitnhs one a qdudainttiiozneadl zero bin). The horizontal and vertical components of MBH are separately quantized into 8 bins each. 4.2. Impact of motion compensation The optical flow is simply referred to as flow in the following, while the compensated flow (see subsection 2. 1) is denoted by ω-flow. Both of them are considered in the tracking and descriptor computation stages. The trajectories obtained by tracking with the ω-flow are called ω-trajectories. Figure 2 comparatively illustrates the ωtrajectories and the trajectories obtained using the flow. The input video shows a man moving away from the car. In this video excerpt, the camera is following the man walking to the right, thus inducing a global motion to the left in the video. When using the flow, the computed trajectories reflect the combination of these two motion components (camera and scene motion) as depicted by Subfigure 2(b), which hampers the characterization of the current action. In contrast, the ω-trajectories plotted in Subfigure 2(c) are more active on the actor moving on the foreground, while those localized in the background are now parallel to the time axis enhancing static parts of the scene. The ω-trajectories are therefore more relevant for action recognition, since they are more regularly and more exclusively following the actor’s motion. Impact on Trajectory and HOG descriptors. Table 1reports the impact of ω-trajectories on Trajectory and HOG descriptors, which are both significantly improved by 3%4% of mAP on the two datasets. When improved by ωflow, these descriptors will be respectively referred to as ω-Trajdesc and ω-HOG in the rest of the paper. Although the better performance of ω-Trajdesc versus the original Trajectory descriptor was expected, the one 222555555866 2. Trajectories obtained from optical and compensated flows. The green tail is the trajectory the current frame. The trajectories are sub-sampled for the sake of clarity. The frames are extracted Figure over every 15 frames with red dot indicating 5 frames in this example. DescriptorHollywood2HMDB51 BaseTrliaωnje- Tc(rtoarejrdpyreos[c2d9u]ced)54 7 1. 7 4% %2382.–89% BaseliHnωOe- (GHreOp [2rG9od]uced)4 451 . 658%%%2296.– 13%% Table 1. ω-Trajdesc and ω-HOG: Impact of compensating flow on Trajectory descriptor and HOG descriptors. achieved by ω-HOG might be surprising. Our interpretation is that HOG captures more context with the modified trajectories. More precisely, the original HOG descriptor is computed from a 2D+t sub-volume aligned with the corresponding trajectory and hence represents the appearance along the trajectory shape. When using ω-flow, we do not align the video sequence. As a result, the ω-HOG descriptor is no more computed around the very same tracked physical point in the space-time volume but around points lying in a patch of the initial feature point, whose size depends on the affine flow magnitude. ω-HOG can be viewed as a “patchbased” computation capturing more information about the appearance of the background or of the moving foreground. As for ω-trajectories, they are closer to the real trajectories of the moving actors as they usually cancel the camera movement, and so, more easier to train and recognize. Impact on HOF. The ω-flow impacts computation used as an input to HOF computation itself. Therefore, HOF can both types of trajectories (ω-trajectories both the trajectory and the descriptor be computed along or those extracted MethodHollywood2HMDB51 Table(ω2rHf.-alocO IwomkF)inpHgacOtFobf[2u9ωhsb]i:f-nlo ωgwot-hwωHOflFown5H 02 34O. 58291F% %descripto3 r706s38.:–1076% m%APfor Hollywood2 and average accuracy for HMDB5 1. The ω-HOF is used in subsequent evaluations. from flow) and can encode both kinds of flows (ω-flow or flow). For the sake of completeness, we evaluate all the variants as well as the combination of both flows in the descriptor computation stage. The results are presented in Table 2 and demonstrate the significant improvement obtained by computing the HOF descriptor with the ω-flow instead of the optical flow. Note that the type of trajectories which is used, either “Tracking flow” or “Tracking ω-flow”, has a limited impact in this case. From now on, we only consider the “Tracking ω-flow” case where HOF is computed along ω-trajectories. Interestingly, combining the HOF computed from the flow and the ω-flow further improves the results. This suggests that the two flow fields are complementary and the affine flow that was subtracted from ω-flow brings in additional information. For the sake of brevity, the combination of the two kinds of HOF, i.e., computed from the flow and the ω-flow using ω-trajectories, is referred to as the ω-HOF 222555555977 MethodHollywood2HMDB51 Tab(lerT3a.cIkmM inpBgMacHgωtBf-loH w [u2)s9in]gω f-lo wo MBH5 d42 e.052s7c% riptos:m34A90P.–3769f% orHllywood2 and average accuracy for HMDB5 1. DTerHasMjcBrOeblitpHGeFor4.ySumTωraw- frcilykto hw ionfgtheduωpCs-fcaolrtωmeiwNp-df/tl+Aωoutrw-finlogwthdesωcr- isTpc-fHtrMloaiOjrBpdswtGeHFosrc descriptor in the rest of this paper. Compared to the HOF baseline, the ω-HOF descriptor achieves a gain of +3.1% of mAP on Hollywood 2 and of +7.8% on HMDB51. Impact on MBH. Since MBH is computed from gradient of flow and cancel the constant motion, there is practically no benefit in using the ω-flow to compute the MBH descriptors, as shown in Table 3. However, by tracking ω-flow, the performance improves by around 1.3% for HMDB5 1 dataset and drops by around 1.5% for Hollywood2. This relative performance depends on the encoding technique. We will come back on this descriptor when considering another encoding scheme for local descriptors in Section 6. 4.3. Summary of compensated descriptors Table 4 summarizes the refined versions of the descriptors obtained by exploiting the ω-flow, and both ω-flow and the optical flow in the case of HOF. The revisited descriptors considerably improve the results compared to the orig- inal ones, with the noticeable exception of ω-MBH which gives mixed performance with a bag-of-features encoding scheme. But we already mention as this point that this incongruous behavior of ω-MBH is stabilized with the VLAD encoding scheme considered in Section 6. Another advantage of tracking the compensated flow is that fewer trajectories are produced. For instance, the total number of trajectories decreases by about 9. 16% and 22.81% on the Hollywood2 and HMDB51 datasets, respectively. Note that exploiting both the flow and the ω-flow do not induce much computational overhead, as the latter is obtained from the flow and the affine flow which is computed in real-time and already used to get the ω-trajectories. The only additional computational cost that we introduce by using the descriptors summarized in Table 4 is the computation of a second HOF descriptor, but this stage is relatively efficient and not the bottleneck of the extraction procedure. 5. Divergence-Curl-Shear descriptor This section introduces a new descriptor encoding the kinematic properties of motion discussed in Section 2.2. It is denoted by DCS in the rest of this paper. Combining kinematic features. The spatial derivatives are computed for the horizontal and vertical components of the flow field, which are used in turn to compute the divergence, curl and shear scalar values, see Equation 3. We consider all possible pairs of kinematic features, namely (div, curl), (div, shear) and (curl, shear). At each × ×× pixel, we compute the orientation and magnitude of the 2-D vector corresponding to each of these pairs. The orientation is quantized into histograms and the magnitude is used for weighting, similar to SIFT. Our motivation for encoding pairs is that the joint distribution of kinematic features conveys more information than exploiting them independently. Implementation details. The descriptor computation and parameters are similar to HOG and other popular descriptors such as MBH, HOF. We obtain 8-bin histograms for each of the three feature pairs or components of DCS. The range of possible angles is 2π for the (div,curl) pair and π for the other pairs, because the shear is always positive. The DCS descriptor is computed for a space-time volume aligned with a trajectory, as done with the four descriptors mentioned in the previous section. In order to capture the spatio-temporal structure of kinematic features, the volume (32 32 pixels and L = 15 frames) is subdivided into a spatio-temporal grid nofd s Lize = nx 5× f ny m×e nt, sw situhb nx =de ny =to 2a and nt = 3. These parameters ×hnave× × bneen fixed for the sake of consistency with the other descriptors. For each pair of kinematic features, each cell in the grid is represented by a histogram. The resulting local descriptors have a dimensionality equal to 288 = nx ny nt 8 3. At the video level, these descriptors are nenc×od end i×nto 8 a single vector representation using either BOF or the VLAD encoding scheme introduced in the next section. 6. VLAD in actions VLAD [8] is a descriptor encoding technique that aggregates the descriptors based on a locality criterion in the feature space. To our knowledge, this technique has never been considered for action recognition. Below, we briefly introduce this approach and give the performance achieved for all the descriptors introduced along the previous sections. VLAD in brief. Similar to BOF, VLAD relies on a codebook C = {c1, c2 , ...ck} of k centroids learned by k-means. bTohoek representation is ob}t oaifn ked c by summing, efodr b yea kch-m mveiasunasl. word ci, the differences x − ci of the vectors x assigned to ci, thereby producing a sv exct −or c representation oflength d×k, 222555556088 DMeBscHriptorV5 LH.A1o%Dlywo5Bo4d.O2 %F4V3L.3HA%MD B35B91.O7%F Taωbl-eDHM5rOCBa.FSjGPdHe+rsωfco-mMHaBOnFeofV54L2936A.51D% with5431ω208-.5T96% rajde3s42c97158,.ω3% -HOG342,58019ω.6-% HOF descriptors and their combination. where d is the dimension ofthe local descriptors. We use the codebook size, k = 256. Despite this large dimensionality, VLAD is efficient because it is effectively compared with a linear kernel. VLAD is post-processed using a componentwise power normalization, which dramatically improves its performance [8]. While cross validating the parameter α involved in this power normalization, we consistently observe, for all the descriptors, a value between 0.15 and 0.3. Therefore, this parameter is set to α = 0.2 in all our experiments. For classification, we use a linear SVM and oneagainst-rest approach everywhere, unless stated otherwise. Impact on existing descriptors. We employ VLAD because it is less sensitive to quantization parameters and appears to provide better performance with descriptors having a large dimensionality. These properties are interesting in our case, because the quantization parameters involved in the DCS and MBH descriptors have been used unchanged in Section 4 for the sake of direct comparison. They might be suboptimal when using the ω-flow instead of the optical flow on which they have initially been optimized [29]. Results for MBH and ω-MBH in Table 5 supports this argument. When using VLAD instead of BOF, the scores are stable in both the cases and there is no mixed inference as that observed in Table 3. VLAD also has significant positive influence on accuracy of ω-DCS descriptor. We also observe that ω-DCS is complementary to ω-MBH and adds to the performance. Still DCS is probably not best utilized in the current setting of parameters. In case of ω-Trajdesc and ω-HOG, the scores are better with BOF on both the datasets. ω-HOF with VLAD improves on HMDB5 1, but remains equivalent for Hollywood2. Although BOF leads to better scores for the descriptors considered individually, their combination with VLAD outperforms the BOF. 7. Comparison with the state of the art This section reports our results with all descriptors combined and compares our method with the state of the art. TrajectorCy+omHbOiGna+tHioOnF+ MDBCHSHol5 l98y.w76%o%od2H4M489.D02%B%51 All ω-descriptors all five compensated descriptors using combined62.5%52.1% Table 6. Combination of VLAD representation. WU*JVliaOnughreM tHaeolth. [yo2w9d87o] 256 0985. 37% SKa*duOeJinhrau tnegdMteHatlMa.h [ol1Dd.0B[91]25 24 609.8172% Table 7. Comparison with the state of the art on Hollywood2 and HMDB5 1 datasets. *Vig et al. [28] gets 61.9% by using external eye movements data. *Jiang et al. [9] used one-vs-one multi class SVM while our and other methods use one-vs-rest SVMs. With one-against-one multi class SVM we obtain 45. 1% for HMDB51. Descriptor combination. Table 6 reports the results obtained when the descriptors are combined. Since we use VLAD, our baseline is updated that is combination of Trajectory, HOG, HOF and MBH with VLAD representation. When DCS is added to the baseline there is an improvement of 0.9% and 1.2%. With combination of all five compensated descriptors we obtain 62.5% and 52.1% on the two datasets. This is a large improvement even over the updated baseline, which shows that the proposed motion compensation and the way we exploit it are significantly important for action recognition. The comparison with the state of the art is shown in Table 7. Our method outperforms all the previously reported results in the literature. In particular, on the HMDB51 dataset, the improvement over the best reported results to date is more than 11% in average accuracy. Jiang el al. [9] used a one-against-one multi-class SVM, which might have resulted in inferior scores. With a similar multi-class SVM approach, our method obtains 45. 1%, which remains significantly better than their result. All others results were reported with one-against-rest approach. On Olympic Sports dataset we obtain mAP of 83.2% with ‘All ω-descriptors combined’ and the improvement is mostly because of VLAD and ω-flow. The best reported mAPs on this dataset are Liu et al. [14] (74.4%) and Jiang et al. [9] (80.6%), which we exceed convincingly. Gaidon et al. [6] reports the best average accuracy of 82.7%. 8. Conclusions This paper first demonstrates the interest of canceling the dominant motion (predominantly camera motion) to make the visual motion truly related to actions, for both the trajectory extraction and descriptor computation stages. It pro222555556199 duces significantly better versions (called compensated descriptors) of several state-of-the-art local descriptors for action recognition. The simplicity, efficiency and effectiveness of this motion compensation approach make it applicable to any action recognition framework based on motion descriptors and trajectories. The second contribution is the new DCS descriptor derived from the first-order scalar motion quantities specifying the local motion patterns. It captures additional information which is proved complementary to the other descriptors. Finally, we show that VLAD encoding technique instead of bag-of-words boosts several action descriptors, and overall exhibits a significantly better performance when combining different types of descriptors. Our contributions are all complementary and significantly outperform the state of the art when combined, as demon- strated by our extensive experiments on the Hollywood 2, HMDB51 and Olympic Sports datasets. Acknowledgments This work was supported by the Quaero project, funded by Oseo, French agency for innovation. We acknowledge Heng Wang’s help for reproducing some of their results. References [1] S. Ali and M. Shah. Human action recognition in videos using kinematic features and multiple instance learning. IEEE T-PAMI, 32(2):288–303, Feb. 2010. [2] T. Brox and J. Malik. Object segmentation by long term analysis of point trajectories. In ECCV, Sep. 2010. [3] N. Dalal and B. Triggs. Histograms of oriented gradients for human detection. In CVPR, Jun. 2005. [4] N. Dalal, B. Triggs, and C. Schmid. Human detection using oriented histograms of flow and appearance. In ECCV, May 2006. [5] P. Dollar, V. Rabaud, G. Cottrell, and S. Belongie. Behavior recognition via sparse spatio-temporal features. In VS-PETS, Oct. 2005. [6] A. Gaidon, Z. Harchaoui, and C. Schmid. Recognizing activities with cluster-trees of tracklets. In BMVC, Sep. 2012. [7] A. Hervieu, P. Bouthemy, and J.-P. Le Cadre. A statistical video content recognition method using invariant features on object trajectories. IEEE T-CSVT, 18(1 1): 1533–1543, 2008. [8] H. J ´egou, F. Perronnin, M. Douze, J. S ´anchez, P. P ´erez, and C. Schmid. Aggregating local descriptors into compact [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] codes. IEEE T-PAMI, 34(9):1704–1716, 2012. Y.-G. Jiang, Q. Dai, X. Xue, W. Liu, and C.-W. Ngo. Trajectory-based modeling of human actions with motion reference points. In ECCV, Oct. 2012. O. Kliper-Gross, Y. Gurovich, T. Hassner, and L. Wolf. Motion interchange patterns for action recognition in unconstrained videos. In ECCV, Oct. 2012. H. Kuehne, H. Jhuang, E. Garrote, T. Poggio, and T. Serre. Hmdb: A large video database for human motion recognition. In ICCV, Nov. 2011. I. Laptev and T. Lindeberg. Space-time interest points. In ICCV, Oct. 2003. I. Laptev, M. Marzalek, C. Schmid, and B. Rozenfeld. Learning realistic human actions from movies. In CVPR, Jun. 2008. J. Liu, B. Kuipers, and S. Savarese. Recognizing human actions by attributes. In CVPR, Jun. 2011. D. Lowe. Distinctive image features from scale-invariant keypoints. IJCV, 60(2):91–1 10, Nov. 2004. M. Marzalek, I. Laptev, and C. Schmid. Actions in context. In CVPR, Jun. 2009. P. Matikainen, M. Hebert, and R. Sukthankar. Trajectons: Action recognition through the motion analysis of tracked features. In Workshop on Video-Oriented Object and Event Classification, ICCV, Sep. 2009. R. Messing, C. J. Pal, and H. A. Kautz. Activity recognition using the velocity histories of tracked keypoints. In ICCV, Sep. 2009. J. C. Niebles, C.-W. Chen, and F.-F. Li. Modeling temporal structure of decomposable motion segments for activity classification. In ECCV, Sep. 2010. [20] J.-M. Odobez and P. Bouthemy. Robust multiresolution estimation of parametric motion models. Jal of Vis. Comm. and Image Representation, 6(4):348–365, Dec. 1995. [21] G. Piriou, P. Bouthemy, and J.-F. Yao. Recognition of dynamic video contents with global probabilistic models of visual motion. IEEE T-IP, 15(1 1):3417–3430, 2006. [22] S. Sadanand and J. J. Corso. Action bank: A high-level representation of activity in video. In CVPR, Jun. 2012. [23] C. Schmid and R. Mohr. Local grayvalue invariants for image retrieval. IEEE T-PAMI, 19(5):530–534, May 1997. [24] J. Sivic and A. Zisserman. Video Google: A text retrieval approach to object matching in videos. In ICCV, pages 1470– 1477, Oct. 2003. [25] J. Sun, X. Wu, S. Yan, L. F. Cheong, T.-S. Chua, and J. Li. Hierarchical spatio-temporal context modeling for action recognition. In CVPR, Jun. 2009. [26] H. Uemura, S. Ishikawa, and K. Mikolajczyk. Feature tracking and motion compensation for action recognition. In BMVC, Sep. 2008. [27] M. M. Ullah, S. N. Parizi, and I. Laptev. Improving bag-offeatures action recognition with non-local cues. In BMVC, Sep. 2010. [28] E. Vig, M. Dorr, and D. Cox. Saliency-based space-variant descriptor sampling for action recognition. In ECCV, Oct. 2012. [29] H. Wang, A. Kl¨ aser, C. Schmid, and C.-L. Liu. Action recognition by dense trajectories. In CVPR, Jun. 2011. [30] H. Wang, M. M. Ullah, A. Kl¨ aser, I. Laptev, and C. Schmid. Evaluation of local spatio-temporal features for action recognition. In BMVC, Sep. 2009. [3 1] G. Willems, T. Tuytelaars, and L. J. V. Gool. An efficient dense and scale-invariant spatio-temporal interest point detector. In ECCV, Oct. 2008. [32] S. Wu, O. Oreifej, and M. Shah. Action recognition in videos acquired by a moving camera using motion decomposition of lagrangian particle trajectories. In ICCV, Nov. 2011. 222555666200
6 0.63717371 170 cvpr-2013-Fast Rigid Motion Segmentation via Incrementally-Complex Local Models
7 0.62016714 362 cvpr-2013-Robust Monocular Epipolar Flow Estimation
8 0.61231327 316 cvpr-2013-Optical Flow Estimation Using Laplacian Mesh Energy
9 0.61056238 137 cvpr-2013-Dynamic Scene Classification: Learning Motion Descriptors with Slow Features Analysis
10 0.59874958 203 cvpr-2013-Hierarchical Video Representation with Trajectory Binary Partition Tree
11 0.58899957 158 cvpr-2013-Exploring Weak Stabilization for Motion Feature Extraction
12 0.57907027 46 cvpr-2013-Articulated and Restricted Motion Subspaces and Their Signatures
13 0.57423097 118 cvpr-2013-Detecting Pulse from Head Motions in Video
14 0.57369441 369 cvpr-2013-Rotation, Scaling and Deformation Invariant Scattering for Texture Discrimination
15 0.56721503 283 cvpr-2013-Megastereo: Constructing High-Resolution Stereo Panoramas
16 0.56107372 333 cvpr-2013-Plane-Based Content Preserving Warps for Video Stabilization
17 0.54804844 113 cvpr-2013-Dense Variational Reconstruction of Non-rigid Surfaces from Monocular Video
18 0.51234382 84 cvpr-2013-Cloud Motion as a Calibration Cue
19 0.51179433 334 cvpr-2013-Pose from Flow and Flow from Pose
20 0.49908745 290 cvpr-2013-Motion Estimation for Self-Driving Cars with a Generalized Camera
topicId topicWeight
[(10, 0.103), (16, 0.026), (26, 0.284), (33, 0.295), (36, 0.012), (57, 0.014), (67, 0.046), (69, 0.045), (77, 0.011), (86, 0.017), (87, 0.056)]
simIndex simValue paperId paperTitle
1 0.96633607 423 cvpr-2013-Template-Based Isometric Deformable 3D Reconstruction with Sampling-Based Focal Length Self-Calibration
Author: Adrien Bartoli, Toby Collins
Abstract: It has been shown that a surface deforming isometrically can be reconstructed from a single image and a template 3D shape. Methods from the literature solve this problem efficiently. However, they all assume that the camera model is calibrated, which drastically limits their applicability. We propose (i) a general variational framework that applies to (calibrated and uncalibrated) general camera models and (ii) self-calibrating 3D reconstruction algorithms for the weak-perspective and full-perspective camera models. In the former case, our algorithm returns the normal field and camera ’s scale factor. In the latter case, our algorithm returns the normal field, depth and camera ’s focal length. Our algorithms are the first to achieve deformable 3D reconstruction including camera self-calibration. They apply to much more general setups than existing methods. Experimental results on simulated and real data show that our algorithms give results with the same level of accuracy as existing methods (which use the true focal length) on perspective images, and correctly find the normal field on affine images for which the existing methods fail.
2 0.96475887 280 cvpr-2013-Maximum Cohesive Grid of Superpixels for Fast Object Localization
Author: Liang Li, Wei Feng, Liang Wan, Jiawan Zhang
Abstract: This paper addresses a challenging problem of regularizing arbitrary superpixels into an optimal grid structure, which may significantly extend current low-level vision algorithms by allowing them to use superpixels (SPs) conveniently as using pixels. For this purpose, we aim at constructing maximum cohesive SP-grid, which is composed of real nodes, i.e. SPs, and dummy nodes that are meaningless in the image with only position-taking function in the grid. For a given formation of image SPs and proper number of dummy nodes, we first dynamically align them into a grid based on the centroid localities of SPs. We then define the SP-grid coherence as the sum of edge weights, with SP locality and appearance encoded, along all direct paths connecting any two nearest neighboring real nodes in the grid. We finally maximize the SP-grid coherence via cascade dynamic programming. Our approach can take the regional objectness as an optional constraint to produce more semantically reliable SP-grids. Experiments on object localization show that our approach outperforms state-of-the-art methods in terms of both detection accuracy and speed. We also find that with the same searching strategy and features, object localization at SP-level is about 100-500 times faster than pixel-level, with usually better detection accuracy.
3 0.95270103 281 cvpr-2013-Measures and Meta-Measures for the Supervised Evaluation of Image Segmentation
Author: Jordi Pont-Tuset, Ferran Marques
Abstract: This paper tackles the supervised evaluation of image segmentation algorithms. First, it surveys and structures the measures used to compare the segmentation results with a ground truth database; and proposes a new measure: the precision-recall for objects and parts. To compare the goodness of these measures, it defines three quantitative meta-measures involving six state of the art segmentation methods. The meta-measures consist in assuming some plausible hypotheses about the results and assessing how well each measure reflects these hypotheses. As a conclusion, this paper proposes the precision-recall curves for boundaries and for objects-and-parts as the tool of choice for the supervised evaluation of image segmentation. We make the datasets and code of all the measures publicly available.
4 0.9458673 440 cvpr-2013-Tracking People and Their Objects
Author: Tobias Baumgartner, Dennis Mitzel, Bastian Leibe
Abstract: Current pedestrian tracking approaches ignore important aspects of human behavior. Humans are not moving independently, but they closely interact with their environment, which includes not only other persons, but also different scene objects. Typical everyday scenarios include people moving in groups, pushing child strollers, or pulling luggage. In this paper, we propose a probabilistic approach for classifying such person-object interactions, associating objects to persons, and predicting how the interaction will most likely continue. Our approach relies on stereo depth information in order to track all scene objects in 3D, while simultaneously building up their 3D shape models. These models and their relative spatial arrangement are then fed into a probabilistic graphical model which jointly infers pairwise interactions and object classes. The inferred interactions can then be used to support tracking by recovering lost object tracks. We evaluate our approach on a novel dataset containing more than 15,000 frames of personobject interactions in 325 video sequences and demonstrate good performance in challenging real-world scenarios.
5 0.93372983 152 cvpr-2013-Exemplar-Based Face Parsing
Author: Brandon M. Smith, Li Zhang, Jonathan Brandt, Zhe Lin, Jianchao Yang
Abstract: In this work, we propose an exemplar-based face image segmentation algorithm. We take inspiration from previous works on image parsing for general scenes. Our approach assumes a database of exemplar face images, each of which is associated with a hand-labeled segmentation map. Given a test image, our algorithm first selects a subset of exemplar images from the database, Our algorithm then computes a nonrigid warp for each exemplar image to align it with the test image. Finally, we propagate labels from the exemplar images to the test image in a pixel-wise manner, using trained weights to modulate and combine label maps from different exemplars. We evaluate our method on two challenging datasets and compare with two face parsing algorithms and a general scene parsing algorithm. We also compare our segmentation results with contour-based face alignment results; that is, we first run the alignment algorithms to extract contour points and then derive segments from the contours. Our algorithm compares favorably with all previous works on all datasets evaluated.
6 0.91451055 311 cvpr-2013-Occlusion Patterns for Object Class Detection
same-paper 7 0.88662088 88 cvpr-2013-Compressible Motion Fields
8 0.88621998 353 cvpr-2013-Relative Hidden Markov Models for Evaluating Motion Skill
9 0.85174096 21 cvpr-2013-A New Perspective on Uncalibrated Photometric Stereo
10 0.84666348 465 cvpr-2013-What Object Motion Reveals about Shape with Unknown BRDF and Lighting
11 0.84068733 104 cvpr-2013-Deep Convolutional Network Cascade for Facial Point Detection
12 0.83201462 405 cvpr-2013-Sparse Subspace Denoising for Image Manifolds
13 0.83164024 96 cvpr-2013-Correlation Filters for Object Alignment
14 0.83058167 424 cvpr-2013-Templateless Quasi-rigid Shape Modeling with Implicit Loop-Closure
15 0.82709521 331 cvpr-2013-Physically Plausible 3D Scene Tracking: The Single Actor Hypothesis
16 0.82656372 317 cvpr-2013-Optimal Geometric Fitting under the Truncated L2-Norm
17 0.82621413 429 cvpr-2013-The Generalized Laplacian Distance and Its Applications for Visual Matching
18 0.82230306 77 cvpr-2013-Capturing Complex Spatio-temporal Relations among Facial Muscles for Facial Expression Recognition
19 0.8221451 194 cvpr-2013-Groupwise Registration via Graph Shrinkage on the Image Manifold