Framework Study on Behavior Understandings Based on Posture and Location State Transition for Railway Station Security

Behavior understanding is an essential technique for automated surveillance. In this paper, we propose methods for recognition of pedestrian states and construct behavior understanding framework for railway station security with them. The framework consists of occlusion robust object tracking, location context estimation, and pedestrian posture recognition. Pedestrian behavior is defined as state transition in our framework. Occlusion robustness of our tracking method gives accurate objects’ trajectories and enables recognition of pedestrian postures when occlusion occurs. We performed experiments to evaluate the state estimation methods and confirmed the feasibility of the framework by using realistic videos.