Traffic light detection based on depth improved YOLOV5

Traffic light detection is an important part of intelligent driving, which affects the driving safety of intelligent vehicles. However, the traffic light is small and the actual situation is complex, and the classical object detection algorithm cannot achieve good detection effect on small traffic lights. In view of this situation, an improved yolov5 model is proposed in this paper. Inspired by u2net, a backbone network more suitable for signal light detection is proposed, and ConvNextBlock module is introduced to improve the feature extraction ability of the model. Through these improvements, the model is more sensitive to the target and the ability to recognize the object of signal light is improved. Experiments show that the improved algorithm achieves an excellent level of 81.5% in the performance index of Map _ 0.5 on the Bosch traffic light data set. In order to test the robustness of the model, the depth and width of the network are compared with the original yolov5 model in S, M and L, and the effect is improved.