Detection of Abnormal Crowd Behavior Based on Graph Convolutional Neural Network
Xiang Zhou, Ruliang Xiao · 2022
With the increasing demand for public safety precautions, the accuracy and timeliness of abnormal crowd behavior detection is gradually improving. Therefore, with the development of the technology for the abnormal crowd behavior, it’s even more demanding on how to effectively prevent the occurrence of accidents and eliminate the impact of disasters on the society. This article first introduces the current applications of abnormal behavior detection and the research and development of related technologies. Second, it proposes a method for detecting abnormal crowd behavior based on Graph Convolutional Neural Network (GCN). This method first proposes to effectively and accurately locate the abnormal behavior characteristics through the extraction of abnormal behavior characteristics from the auto-encoder of the surveillance video. With the construction of the graph classification model based on SAGPool, the classification of normal and abnormal behavior is carried out. Finally, experimental data show that data displayed with this method meet the expected requirements.