Action recognition system for security monitoring

Abstract

In terms of security detection, security personnel often need to find potential dangerous events in time, although there are many monitoring equipment that can monitor behavior at any time. However, the real-time performance of monitoring equipment is not strong enough. Therefore, this paper proposes a novel action detection method suitable for violent action detection. In this method, the common camera is used as the data acquisition equipment, and the human skeleton structure is obtained by using the algorithm of object detection and pose estimation. Finally, it is recognized by graph convolution neural network. In addition, TensorRT acceleration technology is used in object detection and pose estimation algorithm, which further improves the real-time performance of action recognition and detection algorithm. Through the detection of the actual environment, it shows that the algorithm has good real-time performance and can be widely used in security monitoring.