Self-learning human body behavior recognition and anomaly detection method

Abstract

The invention discloses a self-learning human body behavior recognition and anomaly detection method, which comprises the following steps of: generating a salient region vector in a monitoring video,and updating a salient region in a timing self-learning manner; for the same salient region, calculating a state feature vector and an action value of the current frame according to the human body behavior state of the current frame, a human body behavior state prediction value and an award value after the current frame is transferred to the next frame, and an action in the state; executing the current target network, calculating the current human body behavior action value, and updating the target action network and the target value network; and after the network parameters converge or meet the maximum number of iterations, counting the weighted sum of the feedback reward value of the current action network and the human body behavior action value of the target value network within the time T to obtain a behavior abnormality level. The method disclosed by the invention has the advantages of low complexity, real-time performance, high detection efficiency and high detection accuracy.

Description

technical field [0001] The invention belongs to the technical field of video image recognition and detection, in particular to a self-learning method for human body behavior recognition and abnormal detection. Background technique [0002] At present, there are many human behavior recognition algorithms on RGB images and pose estimation. Among them, RGB video algorithms mainly extract spatiotemporal features from video frames / optical flow, mainly including, [0003] Dense trajectories and motionboundary descriptors for actionrecognition[J].Heng Wang,Alexander Cordelia Schmid, Cheng-Lin Liu. International Journal of Computer Vision, Springer Verlag, 2013, 103(1), pp.60-79. [0004] Action Recognition with Trajectory-Pooled Deep-Convolutional Descriptors[J]. Limin Wang, Yu Qiao, Xiaoou Tang, CVPR2015, 2015, 4305-4314. [0005] The problem with this approach is that the extracted video features are disturbed by background environment, lighting changes, and appearance changes...

AI Technical Summary

Problems solved by technology

[0014] (3) Once there is a sample that is different from the training library, or when a new human behavior is added, the labeling operation needs to be re-labeled, which is expensive. It belongs to the tutor-style supervised learning algorithm and cannot complete the self-learning process.

[0015] It can be seen that the existing human behavior recognition and anomaly detection methods have certain defects.

Images