摘要
针对现有车辆在行驶环境下检测算法精度低、易受背景环境干扰、难以检测密集小目标车辆和行人的问题,提出了一种基于YOLOv8n算法改进的车辆道路行驶检测算法。将一个P2小目标检测层添加到YOLOv8n算法模型,结合感受野注意卷积的优点重构C2f模块,并利用Focal Modulation模块替换SPPF模块改进算法模型,通过消融实验以及与原始算法进行实验对比分析。结果表明,改进的算法在Cityscapes数据集上检测平均精度达到了55.3%,总体平均精度提升2.7%,平均召回率提升3.7%,且在soda数据集上的检测效果优于其他检测算法,证明了改进算法在道路目标检测方面的有效性和卓越性。
Aiming at the problem that the existing vehicle detection algorithm in the driving environment has low accuracy,susceptible to background environment interference,and is difficult to detect dense small target vehicles and pedestrians,an improved vehicle road driving detection algorithm based on YOLOv8n algorithm was proposed.A P2 small target detection layer was added to the YOLOv8n algorithm model,and the C2f module was reconstructed by combining the advantages of Receptive-Field Attention Convolution.The Focal Modulation module was used to replace the SPPF module to establish an improved algorithm model.The ablation experiment and the comparative experiment with the original algorithm were analyzed.The results show that the average accuracy of the improved algorithm on the Cityscapes dataset reaches 55.3%,the overall average accuracy is improved by 2.7%,and the average recall rate is improved by 3.7%.The detection effect on the soda dataset is better than other mainstream detection algorithms,which proves the effectiveness and excellence of the improved algorithm in road target detection.
作者
罗勇
赵红
陈俊杰
丁晓云
张泽谦
刘亚坤
LUO Yong;ZHAO Hong;CHEN Junjie;DING Xiaoyun;ZHANG Zeqian;LIU Yakun(College of Mechanical and Electrical Engineering,Qingdao University,Qingdao 266071,China;Shenbang Intelligent Technology Group Co.,LTD.,Qingdao 266071,China)
出处
《青岛大学学报(工程技术版)》
2025年第1期41-48,共8页
Journal of Qingdao University(Engineering & Technology Edition)
基金
青岛市科技惠民示范专项(24-1-8-cspz-18-nsh)。
作者简介
第一作者:罗勇(1998-)男,硕士研究生,主要研究方向为深度学习与计算机视觉在无人驾驶中的应用;通信作者:赵红(1973-)女,博士,副教授,主要研究方向为车辆节能减排与新能源技术。Email:qdlizh@163.com。
