一种基于形态学特征的车道线识别方法

蔡英凤 高 力 孙晓强 陈 龙 王 海

China Mechanical Engineering - - 中国机械工程 -

A Lane Identification Method Based on Morphological Features

CAI Yingfeng GAO Li SUN Xiaoqiang CHEN Long WANG Hai

Institute of Automotive Engineering,Jiangsu University,Zhenjiang,Jiangsu,212000 Abstract: The existing lane detection algorithm mainly used edge information to extract lane fea⁃ tures,and the algorithm that generated feature points through the contrast of adjacent pixels was easily affected by many external factors,and the detection results were easily disturbed,thus a new feature ex⁃ traction algorithm was proposed. This algorithm improved the robustness of the algorithm in complex sit⁃ uations by calculating the tensor rotation of the gray structure in the region and selecting the largest changing trend pixel as the feature point. A new lane line feature extraction algorithm was used to extract feature points in interest areas,then feature points were selected,and Hough transform was used to fit them. After the lane lines were obtained , the dashed line and the solid line were distinguished by the co⁃ ordinate variances of the feature points. The algorithm was tested by driving pictures of about 15 500 frames in different time periods. The results show that the detection method may detect the lane line well under various environments,the correct rate under the sunny conditions is as 99.18%,the correct rate under the rainy conditions is as 97.19%,and the correct rate under the worn pavement conditions is as 94.72%,the correct rate under the night conditions is as 97.62%.

Key words: feature point;Hough transform;ridge algorithm;dashed line and solid line

引言目前,无人驾驶技术吸引了人们极大的兴趣。根据车道线对车辆定位是无人驾驶技术的基本功能。从车道线的位置可以推测出车辆自身的位置、航向等信息。陈无畏等 提出一种基于边缘点

[] 1投影的车道线快速识别算法,在目标搜索区域内,用边缘信息提取车道特征。何鹏等 通过背景减

[] 2

收稿日期: 2017-12-25

基金项目:国家自然科学基金汽车联合基金资助重点项目( U1664258, U1764257);国家自然科学基金资助项目( 61601203,61403172)

除和二值化方法得到车道线的特征点,使用Hough 变换和 Catmull ⁃ Rom 样条曲线确定近场和远场的车道线。王楠 利用白色在YCbCr彩色

[] 3

空间的Y分量和黄色在Cb分量的属性区分黄色和白色车道线,但是没有给出区分虚线和实线的方法。NIU等 使用逆透视映射来生成道路图像

[] 4

的俯视图,使用简化的Hough变换检测得到车道线。王海等 提出基于方向可变Haar特征和双曲

[] 5线模型的分布式车道线检测方法,采用方向可变Haar特征提取边缘特征点并拟合车道线模型。由

江苏大学汽车工程研究院,镇江, 212000

摘要:现有车道线检测算法主要用边缘信息提取车道特征,通过相邻像素点灰度的对比产生特征点,易受多种外部因素影响导致检测结果易受干扰,为此,提出了一种新的特征点提取算法。该算法通过计算区域内灰度各向结构张量的旋度,选择变化趋势最大的像素作为特征点,提高了算法在复杂情况下的鲁棒性。在兴趣区域采用新的车道线特征提取算法提取特征点,而后筛选特征点,并用Hough变换拟合。在求得车道线后,通过特征点坐标方差区分虚线和实线。通过约15 500帧不同时间段的车道图片对算法进行检验,结果表明:检测方法能很好地实现在多种环境下的车道线检测,在晴天工况下的正确率为99.18%,在雨天工况下的正确率为97.19%,在受损路面工况下的正确率为94.72%,在夜晚工况下的正确率为97.62%。

关键词:特征点; Hough变换;脊度量算法;虚实线

中图分类号: U467.3

DOI:10.3969/j.issn.1004⁃132X.2018.15.014 开放科学(资源服务)标识码(OSID) :

Newspapers in Chinese (Simplified)

Newspapers from China

© PressReader. All rights reserved.