Feature detection and tracking for extraction of crowd dynamics

Extraction of crowd dynamics from video is the fundamental step for automatic detection of abnormal events. However, it is difficult to obtain sufficient performance with object tracking due to occlusions and insufficient resolution of the objects in the scene. As a result, optical flow or feature tracking methods are preferred in crowd videos. These applications also require algorithms to work in real-time. In this work, we investigated the applicability and performance of feature detection and tracking algorithms in crowd videos. The algorithms that were tested in this paper include Scale Invariant Feature Transform (SIFT) and Speeded-Up Robust Features (SURF) as well as relatively newer approaches Binary Robust Independent Elementary Features (BRIEF) and Oriented Fast and Rotated Brief (ORB). These algorithms have been tested with videos having different crowd densities and comparative results of their accuracy and computational performance have been reported. The results show that BRIEF is computationally faster than the others, allowing real-time operation, and comparable with other algorithms regarding matching accuracy.