Anomaly Detection Model for Key Places Based on Improved YOLOv5

In recent years, key places such as underground stations and train stations, which are crowded and highly mobile, have become key targets for abnormal behaviour such as violence by some extremists or violent elements. The public safety risks in key places cannot be ignored, and the need to detect abnormal behaviour in key places is urgent in order to protect the personal safety of the people in such key places. When abnormal people and abnormal events occur in key places, timely detection and early warning are required to prevent and protect the safety of the people in a timely manner. Therefore, a real-time anomaly detection system based on the improved YOLOv5 key place video is proposed for such key places with dense personnel, intricate and complex identities, low accuracy of anomalous behaviour detection and slow detection speed. The method improves the target recognition effect by improving the loss function and optimising the resolution. Test results show that under the same training conditions, the improved YOLOv5 network has a significantly higher correct rate of anomalous behaviour detection and a faster detection speed of anomalous behaviour compared with the original YOLOv5 network.