基于视觉的路口车辆排队长度检测方法  

Vehicle Queue Length Detection Method at Intersection Based on Vision

作　　者：沈振乾[1] 苗长云[1] 张芳[1] 

机构地区：[1]天津工业大学电子与信息工程学院,天津300387

出　　处：《计算机工程》2014年第4期218-222,共5页Computer Engineering

基　　金：国家自然科学基金青年科学基金资助项目(61102150)

摘　　要：现有基于视觉的交叉路口车辆排队长度测量方法,不能给出实际排队长度,且摄像机标定复杂。为此,提出基于连通区域的车辆排队长度检测算法及以停车线为参照物的摄像机快速标定方法。在停车线后设定停车检测区域,并利用相邻帧差法判断是否存在车辆停车。在排队检测区域内用最大类间方差法自动设定阈值并分割图像,以检测车辆存在。利用连通区域各种参数确定车辆所在的连通域并检测出队尾。运用针孔成像模型论证以停车线为参照物的摄像机标定方法。实验结果证明,该方法可以准确检测出队首和队尾像素,计算出实际长度,可满足实时性要求。Aiming at the problem that the existing methods cannot provide the actual length or have the complex camera calibration, so the vehicle queue length detection algorithm and the rapid camera calibration method reference to stop line are provided. It sets the parking detection region and the adjacent frame difference to determine the existence of the parking of vehicles behind the stop line, detects the existence of vehicles by image segmentation with automatic threshold setting based on Otsu method in queue detection region, and detects the connected domain where the vehicle locates and the queue tail by connected domain parameters. The camera calibration method reference to stop line is demonstrated using pinhole imaging model. Experimental results show that the method can accurately detect pixel value of the head and tail, effectively calculate the actual length and meet the real-time requirements.

关 键 词：视觉测量 排队长度检测 连通域 最大类间方差法 针孔成像 摄像机标定 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]