复杂交通环境下二轮机动车乘员头盔检测算法

针对现有二轮机动车乘员头盔检测算法在目标密集分布、随机遮挡等情况下效果较差且难以在边缘设备上应用的问题，制作了具有针对性的数据集，对比现有模型后，以YOLOv7为参考提出一种复杂交通环境下二轮机动车乘员头盔检测算法.首先，采用EfficientNet-B3作为主干网络，可提高特征提取能力且更为轻量化；其次，将增大感受野模块(RFB)引入特征融合结构中，以增大模型感受野，提升小目标头盔检测能力；最后，在检测头嵌入SimAM机制，在不增加参数的前提下提高算法精度.结果表明：相较于YOLOv7,文中算法的准确率、召回率和平均准确率分别提高了2.84%,2.26%和3.26%,参数量和运算量分别为YOLOv7的33.1%,23.5%,可实现当前主流模型算法的最佳检测性能和效率；在NVIDIA Jetson Nano开发板上的处理速度达到47.58 F·s^-1,可满足边缘设备部署需求.

Helmet Detection Algorithm of Two-Wheeled Motor Vehicle Occupant in Complex Traffic Environment

The existing helmet detection algorithms for two-wheeled motor vehicle occupant are less effective in the case of dense object distribution and random occlusion, and difficult to apply on edge devices. To address this problem, a targeted dataset is created and a helmet detection algorithm of two-wheeled motor vehicle occupant in complex traffic environment is proposed using YOLOv7 as a reference after comparing existing models. Firstly, EfficientNet-B3 is used as the backbone network to improve the feature extraction capability and make it more lightweight. Secondly, receptive field block(RFB)is introduced into the feature fusion structure to increase the receptive field of the model, and improve the detection capability of small target helmets. Finally, SimAM mechanism is embedded in the detection head to improve the accuracy of the algorithm without increasing the number of parameters. The results show that compared to YOLOv7, the accuracy, recall rate and average accuracy of the proposed algorithm have been improved by 2.84%, 2.26% and 3.26% respectively, and the number of parameters and operations are 33.1% and 23.5% of YOLOv7 respectively, achieving the best detection performance and efficiency of current mainstream model algorithms. The processing speed on the NVIDIA Jetson Nano development board reaches 47.58 frames per second, which can meet the requirements of edge device deployment.