A Tensor Based on Optical Flow for Global Description of Motion in Videos

Abstract

Motion is one of the main characteristics that describe the semantic information of videos. In this work, a global video descriptor based on orientation tensors is proposed. This descriptor is obtained by combining polynomial coefficients calculated for each image in a video. The coefficients are found through the projection of the optical flow on Legendre polynomials, reducing the dimension of per frame motion estimations. The sequence of coefficients are then combined using orientation tensors. The global tensor descriptor created is evaluated by a classification of the KTH video database with a SVM classifier.

FASTensor: A tensor framework for spatiotemporal description

“…Trajectories Relationship Modeling [24] 95.6 [25] 97.4 [26] 98.2 [27] 94.8* [28] 95.3 [29] 94.5 [30] 93.9 [31] 94.2 [32] 94.5 [33] 93.8 Tensor [34] 94.2 Our approaches [18] 93.3 [20] 93.2 [19] 92.5 [21] 92.0 [22] 87.8 [23] 86.6…”

“…Holly-wood2 dataset is the most challenging, and has been collected from Hollywood movies. Table III 47.7 [28] 59.9 Tensor [36] 59.5 [34] 57.6 Our approaches [19] 40.3 [20] 40.3 [21] 34.0 [22] 15.0…”

A Variable Size Block Matching Based Descriptor for Human Action Recognition

“…Several distinct techniques have been employed to extract motion, like optical flow [2] and 3D gradients [3]. But so far, block matching has not been thoroughly explored as an action recognition tool, even though it has been used in an array of applications where motion is a relevant feature.…”

Initial-state invariant Binet-Cauchy kernels for the comparison of Linear Dynamical Systems

“…The average recognition rate achieved by this method was 72%. Another recent global optical-flow feature based approach was proposed by Mota et al [46] that gave recognition rates in the range of 70% to 86% on the KTH dataset. Unfortunately, all global-feature based representations do not fare well against the best reported local-feature based result on the entire (all four scenarios) KTH database, e.g., 98.1% in [47].…”

“…The results in Table V follow the trends that we have observed for other datasets: in general hybrid metrics perform better than dynamics-only metrics, and Martin and Binet-Cauchy determinant kernel perform the best. The best performing result of 53.14% is achieved by using the Binet-Cauchy [45] 72 * Mota et al [46] 70-86 * Sadanand et al [47] 98.1 * [50] 38.8 Laptev et al [43] 47.9 Sadanand et al [47] 57.9 determinant kernel. The UCF dataset is relatively new and has only recently started to gain the attention of researchers.…”