Research On Group Abnormal Event Detection Based On Surveillance Video

The use of intelligent monitoring equipment to detect abnormal crowd behavior can effectively prevent or deal with mass incidents in a timely manner.In the existing research on crowd anomaly detection methods,the detection types of abnormal behaviors are often relatively single,and it is impossible to detect behaviors such as knife fights,arson,and abnormal crowd gathering in surveillance videos,and most algorithms have a large amount of calculation and poor real-time performance,the early warning information cannot be generated in time.Therefore,the detection method of abnormal crowd behavior based on surveillance video has become an important research direction.This paper proposes a method for detecting abnormal crowd behavior based on surveillance video,and designs and implements the software.The specific work is as follows:（1）Build a target detection model based on a lightweight network.In order to meet the speed and accuracy requirements of the model for the detection of abnormal behaviors of knife fighting,arson and smoking in surveillance videos,this paper introduces the Mobilenet and Ghostnet lightweight networks based on the YOLOv4 algorithm.The lightweight network is replaced with the backbone feature network of YOLOv4 and experiments are carried out.The experimental results show that the parameter quantity of the YOLOv4 model based on the lightweight network structure decreases significantly.Among them,the number of parameters of Mobilenetv2-YOLOv4 is reduced by 83% compared with the original network,and the detection speed is increased by 57% compared with the original network.After the model is lightweight,its parameter quantity and detection speed have been greatly improved,but the robustness of the model has been significantly reduced.（2）A target detection method based on attention-aware and lightweight network is proposed.In order to improve the robustness of the lightweight target detection model,this paper introduces an attention mechanism to improve the model’s ability to extract features.Three attention mechanisms,SENet,ECANet and CBAM,are mainly introduced for experiments.The experimental results show that in the model with the three attention mechanisms added,the mean average detection accuracy is improved.The average detection accuracy of the model with the CBAM attention module added is 84%,which is 2% higher than the model without the attention mechanism.（3）A mass abnormal event detection system is proposed and implemented in software.Based on the improved lightweight target detection model,this paper detects knife-wielding,arson and smoking behaviors in surveillance videos,and combines the MCNN crowd counting algorithm to detect crowd density to achieve abnormal crowd behavior from the two dimensions of abnormal target and crowd density.test.In terms of software implementation,it analyzes the relevant functional requirements,uses Py Qt5 to design the interface,and realizes the function of mass event detection and early warning.The research results of this paper can be applied to the field of public security,and provide technical support for the prevention and handling of sudden group abnormal events.