基于视觉的车辆侧后方道路图像检测技术.pdf

( ) 计算机工程与设计 Computer Engineering and Design 2800 2010,31 12 计算机工程与设计 Computer Engineering and Design 多媒体技术 基于视觉的车辆侧后方道路图像检测技术 刘志强， 宋世亮， 汪 澎， 仲晶晶 ( 江苏大学 汽车与交通工程学院，江苏镇江2 12013) 摘 要：针对车辆在换道过程中易发生碰撞的问题，提出了一种针对车辆侧后方包括盲点区域在内的道路图像的检测算 法。该算法对车辆侧后方车道线分为内外两侧进行分析，解决了因摄像机安放位置所产生的感兴趣区域不对称问题。利用 基于梯度特征的阴影检测法对接近的车辆进行识别，在此基础上，通过单目视觉测距和两车的几何位置关系计算两车的距 离。实验结果表明，该方法可以准确地识别本车后方的车道线和车辆，并且当两车距离在55 米以内时，系统可以有效的对车 距进行估计。 关键词：换道; 盲点区域; 感兴趣区域; 梯度特征; 单目视觉 中图法分类号：TP391.4 1 文献标识码：A 文章编号：1000-7024 (2010) 12-2800-04 Vision based for road image detection of rear vehicle LIU Zhi-qiang, SONG Shi-liang, WANG Peng, ZHONG Jing-j ing (Department of Vehicle and Traffic Engineering, Jiangsu University, Zhenj iang 2 12013, China) Abstract ：Aiming at the problems of the vehicle collision that happened in lane changing, a method of detecting the road area behind the vehicle including side blind zone is proposed. The method divides lane marks side-back into two parts inside and outside to research sequentially. Therefore, it solves the problems that the left and right regions of interest (ROI) are dissymmetrical which is caused by the position of the camera. The closing vehicles are identified through shadow detection which is based on the gradient characteristics in the lane area. Based on these, the distance between two vehicles is computed based on their geometrical position and monocular vision. Experiment show that the methods can detect the lane marks and vehicles exactly, meantime the distance between two vehicles is estimated effectively when their real distance was within 55meters. Key words ：lane change; side blind zone; regions of interest; gradient characteristics; monocular vision