复杂环境下的夜间视频车辆检测*

分析了夜间复杂交通场景的特点,提出了应用于夜间交通信息采集的HLEPT（headlight extraction,pai-ring and tracking）算法。该算法包含车灯提取算法和配对跟踪规则,并结合先配对车灯后跟踪其轨迹和先跟踪车灯后配对其轨迹两种方法,对车流量、车速等交通信息进行统计。实验表明,HLEPT算法复杂度低,具有良好的实时性、鲁棒性,良好环境下其检测率达到96%以上;即使在雨夜路面有车灯倒影的交通拥挤路段,也能达到88%的检测率。

Nighttime video vehicle detection in complex environment

This paper analyzed the features of nighttime complex traffic scenes, and proposed the HLEPT algorithm to collect nighttime traffic information. It contained a headlight extraction algorithm and regulations of pairing and tracking, integrating with two methods: pre pairing headlights and post tracking trajectories, pre tracking headlights and post pairing trajectories, and finally measured the traffic flow and velocity. Experiment results indicate that this algorithm is robust with low computational complexity and real time performance, and its detection ratio reaches above 96% in fine condition, 88% in a traffic jam at rainy night when inverted reflections of lights exist.