摘要
针对现有虚拟线圈车流量统计算法准确度较差,容易产生误检以及错检问题的现状,提出一种基于YOLOv3的多车道车流量计数及车辆跟踪方法。首先通过特征提取网络对输入的图像提取特征,预测图像位置和类别概率值;接着比较在相邻两帧图像中检测到的车辆位置,根据相邻两帧图像中车辆标记框的中心点是否在同一点来判断这两帧中的车辆是否为同一辆车,从而达到跟踪的目的;最后利用设定的检测线和线框位置关系,得到每一车道上的车流量。该方法能够实现对车辆的跟踪及对任意车道上的车流量进行统计。实验结果表明,在车辆跟踪及车流量统计上,解决了传统运动目标检测算法中车辆目标区域粘连导致检测跟踪不准确以及虚拟线圈算法对多车道车流量检测的局限性的问题,检测的车流量准确率高于虚拟线圈算法。
In view of the current accuracy of the current virtual coil traffic statistics algorithm,it is easy to produce false detection and error detection.,multi-lane traffic flow statistics and vehicle tracking method based on YOLOv3 was proposed.Firstly,some features were extracted from the input image through the feature extraction network,the image position and the category probability value were predicted.Then vehicle positions detected in the adjacent two images were compared,and whether the vehicles in the two frames are the same vehicle was determined according to the center points of the vehicle marking frames in the adjacent two frames of images are at the same point,thus achieving the purpose of tracking.The traffic volume of each lane was obtained by the set detection line and the positional relationship of the wireframe.The method enables tracking of the vehicle and statistics of traffic flow in any lane.The experimental results show that in the vehicle tracking and traffic flow statistics,the detection tracking inaccuracy caused by the adhesion of the vehicle target area in the traditional moving target detection algorithm and the limitation detection of the multi-lane traffic flow virtual coil algorithm are solved.The accuracy is higher than the virtual coil algorithm.
作者
汪辉
高尚兵
周君
周建
张莉雯
Wang Hui;Gao Shangbing;Zhou Jun;Zhou Jian;Zhang Liwen(College of Computer and Software Engineering,Huaiyin Institute of Technology,Huaian 223001,China;Jiangsu Mobile Internet of Things Technology Engineering Laboratory,Huaian 223001,China)
出处
《国外电子测量技术》
2020年第2期42-46,共5页
Foreign Electronic Measurement Technology
关键词
YOLO算法
车辆跟踪划分
车流量统计算法
交通事件检测
图像处理
YOLO algorithm
vehicle tracking and division
traffic flow statistics algorithm
traffic event detection
image processing
作者简介
汪辉,本科,主要研究方向为计算机视觉,E-mail:brewin@foxmail.com;通信作者:高尚兵,博士,副教授,主要研究方向为图像处理、虚拟现实、模式识别,E-mail:luxiaofen_2002@126.com
