Traffic Sign Detection Algorithm Based On Improved YOLOv4
Xiru Wu, Haozhe Cao · Journal of Physics Conference Series · 2022
Abstract This paper proposes a traffic sign detection algorithm using improved lightweight YOLOv4 in order to solve the problems of large parameters quantity, poor real-time performance, and low accuracy for the unmanned driving system. Firstly, we use the MobileNetV2 network to replace the original feature extraction network in YOLOv4 to improve the detection speed by reducing the number of model parameters. Secondly, we add an attention mechanism to the backbone network to effectively extract feature information. Trim the network layer of the prediction part and add residual structure, so that feature reuse enhancement can avoid the gradient disappearing when the network layer is deeper. The size of the improved network model is reduced by 79.5%. The experimental results show that the average accuracy of the network model is 2.5% lower than that of the YOLOV4 network, and the detection speed is 58.8% higher than that of the YOLOv4 network.