基于改进YOLOv5的电动摩托车驾驶人头盔检测方法

Helmet Detection Method of Electric Motorcycle Driver Based on Improved YOLOv5

随着经济的快速发展及低碳环保出行方式的普及,电动摩托车投入量逐年上升,但由此带来的安全隐患也随之上升.针对传统的人工检查骑手是否规范佩戴安全帽耗时、耗力且存在漏检等问题,提出一种基于改进YOLOv5的头盔检测算法.首先,针对摩托车头盔大小尺寸不一的问题,使用K-means++算法重新设计初始锚框,增加了网络收敛速度;其次引入坐标注意力机制(Coordinate Attention),增强网络学习特征的表达能力;最后,引入α-IoU损失函数提高目标检测精度.实验表明,改进的YOLOv5模型的mAP达到98.83%,比YOLOv5的平均精度提升了5.29%,符合在道路复杂环境下对电动摩托车驾驶人头盔检测的要求.

With the rapid development of economy and the popularity of low-carbon and environmentally friendly travel modes, the amount of electric motorcycle investment is increasing year by year, but the resulting safety risks are also increasing.Aiming at the problems of time consuming, energy consuming and missing detection in the traditional manual inspection whether riders wear helmets in a standard way, a helmet detection algorithm based on improved YOLOv5 was proposed.Firstly, aiming at the problem of different sizes of motorcycle helmets, K-means++ algorithm was used to redesign the initial anchor frame to increase the convergence speed of the network.Secondly, Coordinate Attention mechanism was introduced to enhance the expression ability of network learning features.Finally, the α-IoU loss function is introduced to improve the target detection accuracy.The experimental results show that the mAP of the improved YOLOv5 model reaches 98.83%,which is 5.29% higher than the average accuracy of YOLOv5,which meets the requirements of helmet detection for electric motorcycle drivers in complex road environment.