Improved motion description for action classification

Even though the importance of explicitly integrating motion characteristics in video descriptions has been demonstrated by several recent papers on action classification, our current work concludes that adequately decomposing visual motion into dominant and residual motions, i.e., camera and scene motion, significantly improves action recognition algorithms. This holds true both for the extraction of the space-time trajectories and for computation of descriptors. We designed a new motion descriptor – the DCS descriptor – that captures additional information on local motion patterns enhancing results based on differential motion scalar quantities, divergence, curl, and shear features. Finally, applying the recent VLAD coding technique proposed in image retrieval provides a substantial improvement for action recognition. These findings are complementary to each other and they outperformed all previously reported results by a significant margin on three challenging datasets: Hollywood 2, HMDB51, and Olympic Sports as reported in Jain et al. (2013). These results were further improved by Oneata et al. (2013), Wang and Schmid (2013), and Zhu et al. (2013) through the use of the Fisher vector encoding. We therefore also employ Fisher vector in this paper, and we further enhance our approach by combining trajectories from both optical flow and compensated flow. We as well provide additional details of DCS descriptors, including visualization. For extending the evaluation, a novel dataset with 101 action classes, UCF101, was added.