Road Target Detection Method Based on Improved YOLOv7-Tiny

This paper presents a new algorithm called YOLOv7-tiny detection algorithm based on the low accuracy of target detection in traffic safety and automatic driving, as well as incomplete target recognition caused by occlusion. By introducing a new attention mechanism into the head network, the network can capture image details and enhance its ability to express features. A new activation function is introduced, which improves the accuracy of target detection. It is shown that after optimizing YOLOv7-tiny network model, the accuracy increases by 3.22% and mAP @ 0.5: 0.95 by 2.36% compared to original model. Results show good performance in different detection environments. In practice, the improved algorithm is practical and accurate, which provides powerful support to traffic safety management and autonomous driving.