基于YOLO车道多目标检测识别系统的研究  

Research on Lane Multi object Detection and Recognition System by YOLO

作　　者：陈晓东 Chen Xiao-dong(Zhangzhou Institute of Technology,Fujian Zhangzhou 363000)

机构地区：[1]漳州职业技术学院电子信息学院,福建漳州363000

出　　处：《内燃机与配件》2025年第5期11-13,共3页Internal Combustion Engine & Parts

基　　金：福建省中青年教师教育科研项目(科技类)(JAT210849)。

摘　　要：为了解决人眼视力范围有限、疏忽等问题所引发的碰撞等交通事故,采用计算机视觉进行目标智能检测和识别,目前深度学习算法是目标检测识别的热门研究方向,主要有YOLO(You Only Look Once)、VGG、SSD(Single Shot MultiBox Detector)、CornerNet等算法,其中YOLO网络容易训练和调整,运行速度快,可达到实时检测效果,针对车道多目标检测的时效性等问题,文中提出基于YOLO车道多目标检测识别系统。该系统可以实时地检测出车道目标及类型的概率,同时通过多次实验,设置合适置信度,提高识别准确性,对于车辆和行人识别准确率在0.85以上,如果训练样本库数量继续增加,准确率还会提高,有效避免因人为疏忽发生碰撞等问题,对智能车道偏离辅助系统有一定的应用价值。In order to solve traffic accidents such as collisions caused by limited visual range and negligence of the human eye,computer vision is used for intelligent object detection and recognition.Currently,deep learning algorithms are a popular research direction in object detection and recognition,mainly including YOLO,VGG,SSD,CornerNet and other algorithms.YOLO networks are easy to train and adjust,run fast,and can achieve real-time detection results,In response to issues such as the timeliness of lane multi object Detection,this paper proposes a lane multi target detection and recognition system by YOLO.This system can detect the probability of lane targets and types in real-time,and through multiple experiments,set appropriate confidence levels to improve recognition accuracy,with an accuracy rate of over 0.85,effectively avoiding issues such as collisions caused by human negligence.It has certain application value for intelligent lane departure assistance systems.

关 键 词：YOLO 多目标 检测 识别 车道 

分 类 号：TP317[自动化与计算机技术—计算机软件与理论] TP391[自动化与计算机技术—计算机科学与技术]