Abnormal Behavior Recognition for Human Motion Based on Improved Deep Reinforcement Learning

Abstract

Recognizing abnormal behavior recognition (ABR) is an important part of social security work. To ensure social harmony and stability, it is of great significance to study the identification methods of abnormal human motion behavior. Aiming at the low accuracy of human motion ABR method, ABR method for human motion based on improved deep reinforcement learning (DRL) is proposed. First, the background image is processed in combination with the Gaussian model; second, the background features and human motion trajectory features are extracted, respectively; finally, the improved DRL model is constructed, and the feature information is input into the improvement model to further extract the abnormal behavior features, and the ABR of human motion is realized through the interaction between the agent and the environment. The different methods were examined based on UCF101 data set and HiEve data set. The results show that the accuracy of human motion key point acquisition and posture estimation accuracy is high, the proposed method sensitivity is good, and the recognition accuracy of human motion abnormal behavior is as high as 95.5%. It can realize the ABR for human motion and lay a foundation for the further development of follow-up social security management.