Cande: A Model for Predicting the Risk of Campus Violence in an Edge Intelligent Computing Architecture

Currently, target detection technology is widely used in various research fields. For example, through the analysis of surveillance videos, it is possible to detect whether engineering personnel are wearing safety helmets and masks and whether there are illegal road occupations in open markets. Among these studies, the method of first uploading the collected videos to the central server and then conducting centralized analysis accounts for a high proportion. However, this method will bring a new server load and network transmission load. Based on the improved YOLOv5, this paper proposes a campus intelligent detection model, Cande (Campus Intelligence Detection Model), for risk prediction of campus violence incidents. The model we propose analyzes video on the edge side. The entire process does not increase the storage space capacity or bring new load to network transmission. Experimental results show that the average accuracy of our proposed model reaches $96.58 \%$, and the average detection speed reaches 33 frames/second. This experimental result shows that the proposed model can provide adequate technical support for risk prediction of school violence incidents.