基于机器视觉的道路上前方多车辆探测方法研究

本文提出了一种基于车辆特征的方法识别和跟踪前方的车辆。首先，利用车辆底部存在阴影的特征，在图像中获得可能的车辆存住区域。然后，通过分形维数计算车辆的纹理特征，排除非车辆区域。最后，利用车辆的边缘信息，通过投影变换方法，对候选区域内的车辆进行定位。此外，利用NMI特征法对定位的车辆进行确认。该算法对不同环境和光照条件下的车辆图片进行了测试，以及在高速公路上进行了实时跟踪，具有较好的鲁棒性和可靠性。

Study on Method of Multiple Vehicle Detection and Tracking Based on Vision

This paper describes vehicle detection and tracking method based on vehicle feature. First, some AOI regions of vehicles are obtained by vehicle underneath shadow. Second, the no-vehicle regions are filtered by texture feature of AOI based on DBC. At last, the edge information is used to located vehicles by project transform, and we can locate the vehicle in the potential vehicle region. At the same time, a verification algorithm is proposed to check the correctness of vehicle based on NMI feature, It has a good result to locate vehicle in different illumination and road environment.