基于YOLOv7-Tiny的交通标识检测算法研究

Research on Traffic Sign Detection Algorithm Based on YOLOv7-Tiny

摘要: 目标检测是智能驾驶系统中的重要组成部分，其中交通标识属于小目标检测，在图像中所占的像素比极少，识别难度大，且种类众多，数据样本不均衡，因此本文提出一种基于YOLOv7-tiny的交通标识检测方法。在YOLOv7-tiny算法中增加目标检测层，使网络更加关注小目标，改善目标检测的效果。引入TAM (Triplet Attention Module)三重注意力机制，通过使用一个三分支结构计算注意力权重，提高小目标检测能力。在TT100K交通标识数据集上，mAP达到84.23%，相较于YOLOv7-tiny提高了4.21%。试验结果表明，该方法对于复杂环境下的交通标识具有更好的检测性能，更能满足实际需求。

Abstract: Object detection is an important part of intelligent driving system, among which traffic signs are small target detection, accounting for a very small pixel ratio in the image, which is difficult to iden-tify, with numerous types and unbalanced data samples. Therefore, a traffic sign detection method based on YOLOv7-tiny is proposed in this paper. Add the object detection layer to the YOLOv7-tiny algorithm to make the network pay more attention to small targets and improve the effect of target detection. The Triplet Attention Module (TAM) is introduced to improve the detection ability of small targets by using a three-branch structure to capture interdimensional interactions and calculate attention weights. In TT100K traffic sign data set, mAP reaches 84.23%, 4.21% higher than YOLOv7-tiny. The test results show that this method has better detection performance for traffic signs in a complex environment and can meet the actual demand.