基于机器视觉的道路上前方多车辆探测方法研究  

A Study on Multiple Vehicle Detection Based on Computer Vision

作　　者：顾柏园[1] 王荣本[1] 郭烈[1] 余天洪[1] 

机构地区：[1]吉林大学交通学院,长春130025

出　　处：《汽车工程》2006年第10期902-905,共4页Automotive Engineering

摘　　要：提出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,在图像中确定可能存在的车辆区域。然后,通过分形维数计算该区域的纹理特征,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NM I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试验,具有较好的鲁棒性和可靠性。A front vehicle detection and tracking method based on vehicle feature is presented. First, the possible vehicle regions are obtained by detecting vehicle underneath shadow. Then, the non-vehicle regions are filtered from the area of interest by texture feature based on differential box counting. And the edge information of vehicle is used to position the vehicles by projection transformation. Finally, the positioned vehicles are verified based on normalized mutual information feature. The results of detection and tracking tests show that the method has good robustness and reliability.

关 键 词：计算机视觉 车辆探测 阴影和边缘 纹理特征 NMI特征法 

分 类 号：U495[交通运输工程—交通运输规划与管理]