Abnormal Behavior Detection Based on Optical Flow Trajectory of Human Joint Points
Yimin Dou, Fudong Cai, Jinping Li, Wei Cheng · 2019
Detection of abnormal behavior of pedestrians in public places is always one of the key issues for public security. Video surveillance is an effective approach to address the issue. Considering the characteristics of multi-person poses, we propose an effective and practical method to detect the abnormality based on the optical flow trajectory of joint points for each human body. There are 4 basic steps in the proposed method: firstly, estimate the posture of each individual in the crowd to determine the corresponding joint points; secondly, compute the optical flow field of the joint points for each person; thirdly, making use of the trajectory constraints for the computed optical flow field, extract feature vectors after noise removal and vector trajectories synthesis of the joint points of each pedestrian; finally, use SVM (Support Vector Machine) to determine abnormal behavior. Experimental results show this method can effectively detect the abnormal behavior of pedestrians in the crowd.