Application of YOLOv3 in road traffic detection

Abstract

In order to calculate the traffic volume of different models at traffic intersections, the problem of target classification of different models of car, bus and truck can not meet the real-time problem. A real-time detection method for traffic flow of different models at traffic intersections is proposed. Through the analysis and experiment of the YOLOv3 (you look only once) convolutional neural network model, the vehicle detection mAP (mean accuracy) value of different models is 87.06%, and the detection speed is 38 frames/s. The experimental results show that the method can effectively detect vehicles with different types of traffic intersections and realize real-time statistics of traffic intersection traffic.