基于视频的车辆瞬时停车延误检测

为了给交通灯控制提供更准确的信息，提出了瞬时停车延误的基本概念和量度方法．利用视觉颜色聚类算法，通过计算道路与背景的颜色距离实现图像中车辆与背景分离，从而有效地识别车辆．以颜色距离最大值所在位置标定车辆位置，再通过帧间对比判定车辆的运行状态，最后获得引道内所有车辆的瞬时停车延误总量．对相同的场景记录，由3个独立的测试员利用人工目测方法获得瞬时停车延误总量的目测值，与算法获得的程序测量值进行比对．11个灯时的实验表明，人工目测成功率为90．9％，有90．0％的程序测试值和目测值之间的误差满足要求．

Instantaneous Stop Delay Detection Based on video

To offer more information to the traffic lights controller, the concept of the instantaneous stop delay is defined and the corresponding measuring method is developed. The color clustering algorithm is adopted to separate the vehicles and the background from images effectively, thus all vehicle positions are identified and recorded by the maximum color distance, finally the stop delay in each frame is evaluated by comparing the differences between frames. During 11 light cycles, the success rate of manual results gets 90. 9% where 90.0% meet the error requirement compared with the computer results.