Research on YOLOv4 Traffic Sign Detection Algorithm Based on Deep Separable Convolution

In the field of autonomous vehicles and intelligent transportation, the environment is perceived by the cameras around the driving vehicle, and the traffic signs are recognized. The vehicle responds to the recognition results in a timely manner. In order to meet the real-time requirements in autonomous vehicles, this paper proposed an improved YOLOv4 lightweight target detection network. The original YOLOv4 backbone extraction network is improved through depth separable convolution, and a new backbone extraction network is obtained. The experimental results show that the mAP difference between the improved YOLOv4 model and the original YOLOv4 model based on the CSTD traffic sign dataset is only 0.88%, but the detection speed is increased by nearly three times, and the number of model parameters is reduced to a certain extent.