Analysis And Application Of Abnormal Behavior Based On Spatial Temporal Graph Convolutional Network

With the development of artificial intelligence technology and the diversification of acquisition equipment,a huge amount of video image content has been created.Therefore,it is increasingly important to detect abnormal human behaviour in video and image content through artificial intelligence technology to improve the screening efficiency of relevant staff and promptly avoid the adverse consequences of current abnormal behaviour.In this paper,we study an abnormal behaviour detection algorithm based on spatial temporal convolutional network and design and implement an abnormal behaviour detection system to detect three kinds of abnormal behaviours and alert the abnormal behaviours,including "smoking","falling",and "fighting".The work of the thesis is as follows:(1)Aiming at the problems that exist in existing methods,such as the difficulty of modelling multi-layer semantic information,the high computational complexity of the model,and the neglect of the relationship between non-physical connection points,which leads to the reduction of model accuracy,this paper proposes an algorithm for abnormal behaviour detection based on improved spatial temporal graphical convolutional networks.In the design of the network model,the end-to-end detection with improved DETR,the spatial temporal graph convolution unit with fused multiheaded attention mechanism,and the spatial temporal graph convolution unit with a fused residual network is added respectively,and the loss function with fused dual branches and the optimizer with fused momentum is established to achieve accurate detection of human anomalous behaviors in the scene to be detected.The experimental results show that the average PCK of the improved algorithm is 93.2% and 92.7% on both FSD and MPII datasets respectively,which is 1.9% and 1.7% higher than that of the baseline algorithm.In addition,the corresponding model computation of the improved algorithm is about 1.7GFLOPS and the corresponding MACs are about6.4GMACs,which indicates that the improved algorithm has high recognition accuracy and good performance.(2)Aiming at the inefficiency and reduced accuracy of manual screening of video surveillance,an abnormal behavior detection system is designed and implemented to detect abnormal behaviors in street,campus,and classroom scenes,and to send out timely alerts via mobile phone SMS when abnormal behaviors occur to avoid possible adverse consequences of the current abnormal behaviors.