Multi-Agent Event Detection Based Onmotion Information

Multi-agent event recognition activity is a highly challenging task because of the difficulties in detecting humans in a crowd, which composes of various types of motion, background, appearance, illumination and clutter. This paper presents a method to detect multi-agent activity such as snatch theft using motion information. These motion based descriptors are composed of motion vector flow and directional motion histogram to capture the interaction between two persons in a video sequence. The simulation results show that the proposed feature vector has successfully detected snatch theft activity up to 90% accuracy, which corresponds to less than 10 % error rate.