基于机器视觉的多车道线检测和分类算法  

Multiple lane line detection and classification algorithm based on machine vision

作　　者：朱贤臻 魏霞[1] 黄德启[1] ZHU Xianzhen;WEI Xia;HUANG Deqi(School of Electrical Engineering,Xinjiang University,Urumqi 830017,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830017

出　　处：《现代电子技术》2022年第23期49-54,共6页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(51468062)。

摘　　要：为解决机器视觉在多车道线检测时出现车道线漏识别、误识别、分类难和准确率低等问题,提出一种检测多车道线和快速分类的方法。首先获取原始图像的感兴趣区域并使用Gamma校正增强多车道线特征信息;然后使用多相位Gabor滤波器叠加滤波图像提取车道线纹特征,获得其边缘梯度;接着基于改进带多参数Hough变换提取出直线线段,基于累加角度值的直方图和双截距约束法实现多车道线快速分类;最后只需提取出线段上少量关键点,通过最小二乘法拟合出完整的车道线,生成动态有效的检测区域,减小后续检测时间。实验结果表明:该设计算法在Tusimple数据集与实际拍摄拥挤、夜间环境的道路中进行验证,识别准确率分别为95.4%,96.22%和95.22%,高于其他常用方法;该算法充分利用车道线多个特征信息,不易受检测环境变化带来的影响。In view of the lane line recognition omission,false recognition,difficult classification and low accuracy rate in multiple lane line detection by machine vision,a method for detecting multiple lane lines and fast classification of lane line is proposed.The region of interest(ROI)of the original image is obtained and the feature information of multi⁃lane lines is strengthened by Gamma correction.And then,the lane line features are extracted to obtain their edge gradients by superimposing the filtering image with multi⁃phase Gabor filter.Furthermore,the straight line segments are extracted based on the improved band multi⁃parameter Hough transform,and the fast classification of multi⁃lane lines is realized on the basis of histogram of accumulated angle values and double intercept constraint method.Only a small number of key points on the line segments are required to be extracted to fit the complete lane line by least squares method and generate a dynamic and effective detection area,so as to shorten the subsequent detection time.The experiment show that the recognition accuracy rate of the algorithm is 95.4%,96.22%and 95.22%respectively when the algorithm is validated in the Tusimple dataset and in the actual filming of congested roads and roads at night,which is higher than other commonly⁃used methods.The multiple feature information of lane lines are utilized sufficiently in the algorithm,so the algorithm is unsusceptible to the change of detection environment.

关 键 词：机器视觉 车道线检测 GAMMA校正 GABOR滤波器 Hough变换 直方图 截距约束 

分 类 号：TN911.73-34[电子电信—通信与信息系统] U471.15[电子电信—信息与通信工程]