A Method of Video Loitering Detection Based on Dynamic Programming

Aim at performance downgrade of loitering detection in the case of motion target being motionless of short duration or multiple targets having crisscross motion, furthermore, loitering detection pays more attention to trajectory of motion target than itself, so this paper proposed a method of video loitering detection based on dynamic programming. First, we extract characteristic points from every frame of video, and find matched characteristic points for them in last frame by using self-adaption SSDA. Next, these characteristic points need to be associated with trajectory based on dynamic programming. At last, every trajectory is evaluated by evaluation function which is proposed in this paper. If the evaluation function value of some trajectory is greater than threshold which is set in advance, this one is considered as loitering trajectory. Simulation results show that this method has good applicability, strong robustness and its performance is stabilized and accurate.