基于ARV-YOLO的交通标志检测算法

Traffic Sign Detection Algorithm Based on ARV-YOLO

摘要: 本文针对复杂道路场景中交通标志目标尺度小、尺度变化大以及背景干扰强等问题，提出一种基于改进YOLOv8n的交通标志检测方法。首先，在主干网络中引入RFCAConv模块，通过结合感受野建模与通道注意力机制增强网络对多尺度特征的表达能力；其次，借鉴YOLOv7的下采样，以多分支结构在降低特征分辨率的同时尽可能保留空间细节信息减小背景干扰；最后，构建ASF-P2多尺度特征融合模块，充分利用浅层高分辨率特征，提高模型对小尺度交通标志的检测能力。实验结果表明，ARV-YOLO模型在TT100K数据集上的mAP@0.5达到80.8%，较原YOLOv8n提升9.9%。同时，模型检测精度与召回率分别提升9.3%和7.1%，有效提升了交通标志检测中多尺度标志和小目标检测性能，解决了背景干扰强问题，具有较好的鲁棒性和应用价值。

Abstract: This paper addresses the challenges of small-scale traffic signs, large scale variations, and strong background interference in complex road scenarios, proposing a traffic sign detection method based on an improved YOLOv8n. First, an RFCAConv module is introduced into the backbone network to enhance the network’s ability to represent multi-scale features by combining receptive field modeling and channel attention mechanisms. Second, borrowing downsampling from YOLOv7, a multi-branch structure is used to reduce feature resolution while preserving spatial details and minimizing background interference. Finally, an ASF-P2 multi-scale feature fusion module is constructed to fully utilize shallow high-resolution features, improving the model’s ability to detect small-scale traffic signs. Experimental results show that the ARV-YOLO model achieves an mAP@0.5 of 80.8% on the TT100K dataset, a 9.9% improvement over the original YOLOv8n. Simultaneously, the model’s detection accuracy and recall are improved by 9.3% and 7.1%, respectively, effectively enhancing the detection performance of multi-scale signs and small targets in traffic sign detection, solving the problem of strong background interference, and demonstrating good robustness and application value.