基于改进YOLOv5s的交通标志识别算法

Traffic sign recognition algorithm based on improved YOLOv5s

交通标志的自动识别对汽车的安全行驶具有重要意义。针对现有交通标志识别算法存在识别精度低、速度慢的问题,该文提出了一种基于改进YOLOv5s的交通标志识别算法。引入MobileNetv3主干网络,将RFB模块与ECA-Net模块相结合,在不提高网络计算量的情况下,确保更大范围内聚焦有效特征;在特征融合中采用AFF模块,将注意力从同层融合扩展到跨层区域;采用Matrix NMS筛选候选框,以提升模型检测速度。在中国交通标志数据集CCTSDB上的验证结果表明,该算法识别精度为96%,速度为48帧/s,在多种环境下对目标的识别能力得到增强,可以满足交通标志实时识别的需要。

Automatic identification of traffic signs is of great significance to the safe driving of vehicles. Aiming at the problems of low accuracy and slow speed of existing traffic sign recognition algorithms, a traffic sign recognition algorithm based on improved YOLOv5s is proposed in this paper. The MobileNetv3 backbone network is introduced, and the RFB module is combined with ECA-Net module to ensure the focus of effective features in a larger range without increasing the amount of network computation. AFF module is used in feature fusion to extend the attention from same-layer fusion to cross-layer region. Matrix NMS is used to screen candidate boxes to improve the speed of model detection. The validation results on CSUST Chinese traffic sign detection benchmark(CCTSDB) show that the proposed algorithm has a recognition accuracy of 96% and a speed of 48 frames per second, which can enhance the target recognition ability in various environments and meet the needs of real-time traffic sign recognition.