For traffic data extraction at nighttime, a headlights-based traffic data collection system is designed and implemented. Firstly, an automatic lane detection algorithm is designed to extract lane lines and the camera is calibrated. Secondly, headlights candidates are extracted by an adaptive segmentation algorithm. Thirdly, the headlights candidates are paired and grouped by spatial information. Then real headlights are located by rule-based reasoning to generate vehicle hypotheses. Finlly, these potential targets are tracked over frames by linear searching combined with nearest neighbor rules and shape matching. A Kalman filter is integrated into tracking module to handle occlusions. The spatial continuity extracted from tracking process is used to confirm vehicles' presence. The results of experiments demonstrate that the system is effective and robust for real-time vehicle detection and tracking at nighttime.%针对夜间交通环境的特点,设计和实现了一种基于车灯的交通流视频检测系统.首先,提出一种夜间车道线检测算法,提取车道线并标定摄像机参数.接着,采用一种自适应阈值分割算法提取候选车灯连通域,并利用空间距离信息配对和分组属于同一辆车的连通域,根据规则集定位车灯,建立车辆假设.然后,通过线性搜索,结合最近邻准则和形状属性匹配在帧间关联车辆假设.对于部分和全部遮挡的情况,结合Kalman滤波器处理.根据跟踪信息的连续性,确认车辆存在并保存跟踪轨迹.实验表明,算法的复杂度低,能够在夜晚多种交通环境下实时检测和跟踪车辆,误检和漏检率低,并且对遮挡情况具有一定的鲁棒性.
展开▼
