Traffic Sign Detection Algorithm Based on Improved YOLOv5 Framework
Aiming at the low precision of the conventional YOLOv5 algorithm in recognizing traffic signs,a new YOLOv5 model is pres-ented.Firstly,the coordinate attention mechanism is introduced into the backbone network,considering both channel information and po-sition information related to direction,thus improving the accuracy of detection.The DIoU_NMS is introduced,which takes into account the center point and retains more rectangular boxes,improving the recognition accuracy of obscuring overlapping traffic signs.Finally,in order to improve the small target detection ability,a small target detection head based on three output detection is added.The improved YOLOv5s algorithm is trained on the traffic sign dataset of TT100K(Tsinghua-Tencent 100K),and the experimental results show that mAP value of the improved model is 86.96％after training,which is 2.79％over that of the original YOLOv5.Meanwhile,a comparison is made between the proposed algorithm and other major algorithms,and the results indicate that the proposed method has better perfor-rmed in traffic sign recognition.
万方数据
维普
