基于改进YOLOv8s模型的电动车骑乘人员头盔佩戴检测  

Detection of Helmet Wearing of Electric Bicycle Riders Based on Improved YOLOv8s Model

作　　者：袁宇乐 汤文兵[1] YUAN Yule;TANG Wenbing(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《湖北民族大学学报（自然科学版）》2024年第3期355-360,367,共7页Journal of Hubei Minzu University:Natural Science Edition

基　　金：国家自然科学基金项目(52374154)。

摘　　要：针对电动车头盔检测模型易受天气、视角等因素影响,存在漏检、误检、精度低和实时检测效率差等问题,基于原始YOLO第8小版(you only look once version 8 small, YOLOv8s)模型进行改进,提出改进YOLOv8s模型。主干特征提取网络选用轻量级的香草网络(vanilla network, VanillaNet)模块,颈部网络采用轻量级的上采样算子内容感知特征重组(content-aware reassembly of features, CARAFE)模块,增加160像素×160像素的极小目标检测层(tiny object detection layer, tiny)模块并修改损失函数为多尺度预测距离交并比(multi-scale prediction distance intersection over union, MPDIoU)。为验证优化模块的有效性,采用消融实验并对比模型改进前后的差异。结果表明,改进YOLOv8s模型平均精确率均值达95.6%,检测速度提升至102帧/s,检测精度有明显提升且延时有所降低。改进YOLOv8s模型能够在实际场景中有效检测电动车骑乘人员的头盔佩戴情况,对于减少人身伤害、提升道路安全和优化智能交通系统具有重要作用。To address the issues of missed,false detection,low precision,and poor efficiency of real-time detection in electric bicycle helmet detection models caused by factors such as weather and viewing angles,an improved you only look once version 8 small(YOLOv8s)model was proposed based on original YOLOv8s model.The lightweight vanilla network(VanillaNet)module was selected for the backbone feature extraction network,and the lightweight upsampling operator content-aware reassembly of features(CARAFE)module was adopted for the neck network.A 160 pixel×160 pixel tiny object detection layer(tiny)module was added,and the loss function was modified to multi-scale prediction distance intersection over union(MPDIoU).Ablation experiments were conducted to verify the effectiveness of the optimization modules,and the differences before and after model improvement were compared.The results showed that the improved YOLOv8s model achieved a mean average precision of 95.6%,and the detection speed increased to 102 frames/s,significantly improving detection accuracy and reducing latency.The improved YOLOv8s model was able to effectively detect helmet-wearing of electric bicycle riders in real-world scenarios,playing a crucial role in reducing personal injuries,enhancing road safety,and optimizing intelligent transportation systems.

关 键 词：深度学习 目标检测 YOLOv8s VanillaNet CARAFE 极小目标检测层 MPDIoU 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] U495[自动化与计算机技术—计算机科学与技术]