Discovering motion patterns in traffic videos using improved Group Sparse Topical Coding

Analyzing motion patterns in traffic videos can directly lead to generate some high-level descriptions of the video content. In this paper, an unsupervised method is proposed to automatically discover motion patterns occurring in traffic video scenes. For this purpose, based on optical flow features extracted from video clips, an improved Group Sparse Topical Coding (GSTC) framework is applied for learning of semantic motion patterns. Then, each video clip can be sparsely represented as a weighted sum of learned patterns which can further be employed in very large range of applications. Compared to the original GSTC, the proposed improved version of GSTC selects only a small number of relevant words for each topic and hence provides a more compact and precise representation of topic-word relationships. Moreover, in order to deal with large-scale video analysis problems, we develop the online algorithm for improved GSTC, which can not only deal with large video corpora but also can deal with dynamic video streams. Experimental results show that our proposed approach finds accurately the motion patterns and gives a meaningful representation for the video.