尾随运动车辆识别与超车检测方法研究

针对尾随车辆由于类型、尺寸、颜色、环境等非确定因素导致识别不稳定、不可靠的问题,提出基于时空域纵向投影跟踪滤波和相对运动检测的车辆识别与超车行为检测算法。首先在对比度空间对序列图像进行定向滤波，采用概率Hough变换估计车辆横向边缘，通过时间域投影方法过滤环境噪声，得到尾随车辆跟踪轨迹，同时使用光流法对序列图像特征角点进行跟踪检测，通过目标运动方向矢量将车辆和背景予以区分。在此基础上，采用车辆置信度函数对上述两种方法的识别结果进行综合决策，提出基于行为检测的超车检测算法。试验结果表明，该方法在结构化道路环境中能够快速准确提取近视场运动尾随车辆，对环境干扰和目标车辆类型变化具有较强的免疫能力。  

Algorithm for Following Vehicle Motion Detection and Overtaking Warning

As type,color,size and other uncertainties are likely to cause unstable and unreliable for following vehicle detection algorithm,this paper proposed a novel cost-effective system,based on longitudinal projection in time-space domain and relative motion analysis.The algorithm used directional filter in contrast space and then estimated vehicle edge with probabilistic Hough transform.According to time-domain projection,following vehicle tracking trajectory can be obtained.Furthermore,this paper used optical flow to track characteristic corners,in other to separate target vehicle from background.On the basis of above,the algorithm integrated detection results of the above two methods,and then an overtaking warning algorithm was proposed based on behavior detection.The test results show that the method can extract myopia campaign following vehicle quickly and accurately in the structural road,and it also has a strong immunity to environmental interference and type change. 