Multi-Model Fusion Vehicle Supervision System Based on Deep Learning

This paper presents a hybrid vehicle monitoring system for closed campus environments. The system leverages the strengths of YOLO and Faster R-CNN object detection algorithms to enable accurate vehicle detection and tracking across multiple camera views. By using Intersection over Union (IoU) technology to track vehicles in different regions of interest (ROI) and surveillance video frames, it dynamically updates vehicle count information for various areas, generating a vehicle distribution heat map for convenient visual analysis of campus traffic management. A congestion warning is issued when the number of vehicles in an area reaches the preset threshold. Additionally, the system includes a license plate detection module that accurately recognizes license plates by combining STN, CNN, LSTM, and CTC network models, recording license plates when speeding vehicles are detected. Experimental results show that the improved vehicle detection hybrid algorithm achieves an increase of 11.36 percentage points in mAP and has a significant improvement in processing speed. The enhanced network model demonstrates desirable results in license plate detection, with a nearly 2% increase in detection rate and reduced processing time. Overall, the system exhibits superior performance.