Traffic object detection based on the improved yolov5

Nowadays, the detection of traffic objects is an essential key link in the fields of automatic driving and intelligent transportation, which is related to people's driving safety. A traffic object detection algorithm based on improved YOLOv5s is proposed to address the issues of missed detection, false detection, and low recognition accuracy in current traffic object recognition. Use deformable convolution (DCNv2) to replace the standard convolution (Conv) of the backbone network to improve the ability of convolution to understand the scene globally.Add an attention mechanism (SimAM) to the backbone to make the network focus on more important feature information. Then the loss function of YOLOv5s is improved to Wise IoU to focus on the prediction regression of ordinary quality anchor frame. Finally, the detection head in the model is replaced with a Decoupled Head to better solve the contradiction between the classification task and the regression task, and speed up the convergence of the model. The experimental results show that the average accuracy of the improved algorithm is average mAP@0.5: 0.95 is 64.0%, which is 4.6 percentage points higher than the original YOLOv5s, and the detection speed has reached 232.6 FPS, which can meet the actual detection needs.