Improved traffic sign detection algorithm based on improved YOLOv8s

Abstract

Aiming at the problem of low accuracy in current traffic sign detection, a new traffic sign detection model YOLOV8-CO with high accuracy was proposed based on YOLOv8 algorithm in this paper. CCMA attention mechanism was introduced and C2f module was replaced by C2fO module. The global average pooling strategy was used to replace the traditional fully connected layer, and the corresponding relationship between feature maps and categories was forced to better conform to the convolutional structure. CIoU and YOLOv8 original loss function were selected to calculate the loss value and train the model. The detection accuracy of traffic signs is improved effectively and overfitting is avoided. TT100K data set and COCO data set were used to verify the validity of the algorithm. The experimental results show that on TT100K, mAP@0.5 increases by 2.2%, mAP@0.7 increases by 2.4%, mAP@0.5:0.95 increases by 1.6%; Improved 0.7% on the COCO dataset, mAP@0.7 improved 0.6%, and mAP@0.5:0.95 improved 0.7%.