交叉口入口路段车流量及车辆换道率检测研究

Study of traffic flow and vehicle lane-changing rate detecting at intersection entrance

针对交叉口入口路段交通视频中存在车辆遮挡严重、频繁走停以及频繁换道插队等特点,提出了一种获取交叉口停车线后大视野范围内车流量以及车辆换道率的检测方法。首先采用多级虚拟特征线生成多级时空图,对车辆进行快速检测和分割,获得车辆候选区域;然后对车辆候选区域内局部特征点进行初步分组和跟踪,并根据相同组内特征点运动趋势相似性来修正分组,解决车辆遮挡问题,用于检测车辆换道率;最后将多级时空图与特征点跟踪相结合,进行相互反馈,实现对车辆准确分割和鲁棒跟踪,避免车辆行驶中走停的影响。实验结果表明,通过该方法能实时准确地获取大视野范围内交叉口入口路段车流量和车辆换道率的交通参数。

In view of the serious vehicle occlusion in the traffic video,stop-and-go driving,vehicles changing lanes and jumping the queue,this paper presented a method to obtain some important parameters,such as traffic flow and vehicle lane-changing rate,in large-area outdoor traffic environment behind the stop line. Firstly, it used multiple virtual characteristic lines to generate muhiple time-spatial images, and quickly obtained the detection and segmentation results of vehicles. Furthermore,it acquired the candidate regions. Secondly, it tracked and grouped the local feature points in the candidate region to implement vehicles tracking. Simultaneously, based on the movement trend similarity of the feature points in the same group to revise groupings,the problem of the occlusion could be solved for detecting the lane-changing rate. Thirdly, combining multiple time-space images with the tracking of feature points to achieve mutual feedback, it acquired the accurate segmenting and robust tracking of vehicles, meanwhile the influence of stop-and-go driving could be avoided. Experiments show that traffic parameters, such as traffic flow and vehicle lane-changing rate, can be obtained accurately using the proposed method in large-area outdoor.