摘要
针对车道线检测不能满足实时性与鲁棒性要求的问题,提出了一种新的车道线检测方法。基于R,G和B三原色在灰度图像中所占比例的多样性、车道线与道路的亮度差,将黄色、白色像素作为优先像素处理。首先通过图像的形态学礼帽算法去除大量噪声,再经最大类间方差法(OTSU)将图像二值化,最终通过轮廓的筛选标注车道线,后续视频帧采用卡尔曼滤波追踪处理,确定新的感兴趣区域。本算法大大减少数据计算量,提高了处理效率,同时正确提取感兴趣区域,提高了算法的鲁棒性,降低车道线检测的误检率。
Aiming at the defect of lane detection in meeting both real-time performance and robustness,a novel lane detection algorithm is proposed. Based on the diverse proportion of the red,green and blue three primary colors in gray image and the brightness difference between lane markings and road surface,yellow and white pixels are selected as the high priority pixels to be processed. A great number of noises are removed first with top hat algorithm in image morphological operations. Then image binaryzation is conducted by using OTSU algorithm. Finally the lane markings are fixed by profile screening,the subsequent video frames are tracked by Kalman filter,and the new region of interest is determined. With the algorithm proposed,the computation efforts are greatly reduced,the operation efficiency is enhanced,the ROIs are correctly extracted,with the robustness of the algorithm raised and false detection rate reduced.
出处
《汽车工程》
EI
CSCD
北大核心
2016年第2期200-205,220,共7页
Automotive Engineering
基金
湖南省自然科学基金(14JJ3055)
国家863计划项目(2012AA111802)资助
