Research on the Influence of the Depth and Width of YOLOv5 Network Structure on Traffic Signal Detection Performance

Abstract

In order to achieve accurate and fast recognition of traffic signals, this paper compared and evaluated the detection performance of four YOLOv5 models from the detection speed and detection accuracy, and analyzed the influence of the depth and width of different network structures on the detection performance of traffic signals. Firstly, this paper introduces the basic principle of YOLOv5 algorithm and analyzes the structure characteristics of YOLOv5s, YOLOv5m, YOLOv5l and YOLOv5x network models. Secondly, the source of the dataset, the construction method and the target category contained in the image are described briefly. After the algorithm training and testing, the corresponding detection model is obtained. Finally, the recognition speed, confusion matrix, recall rate, accuracy rate and other evaluation indexes are compared and analyzed. Experimental results show that the recognition performance of YOLOv5m model is better than the other three models when considering the detection speed and detection accuracy.