基于改进RANSAC算法的车道线识别方法  

Lane Marking Detection Technique Based on Improved RANSAC Algorithm

作　　者：樊超[1] 狄帅[1] 侯利龙[1] 徐静波[1] 

机构地区：[1]河南工业大学信息科学与工程学院,郑州450001

出　　处：《汽车工程》2014年第4期503-508,共6页Automotive Engineering

基　　金：河南省教育厅自然科学基金(14A510019);河南省自然科学基金(132300410392);河南省杰出青年基金(104100510008)资助

摘　　要：为在光照变化和车道线破损、阴影遮挡和污迹覆盖等路况下提高车道线识别算法的实时性和鲁棒性,在特征提取基础上提出一种改进的随机抽样一致(RANSAC)算法。首先使用对应掩模分别对特征提取后的左右车道线区域进行滤波得到特征点的随机抽样集,并从中随机抽取特征点初始化车道线参数;然后在余集中选取到初始模型距离小于阈值容限的特征点,构成一致集,并用其中的特征点重新构建新的车道线模型;最后删除新模型一致集内的所有特征点以及到新模型距离介于阈值容限和删除容限之间的点;按照上述步骤重复迭代数次后,选取最大一致集所计算的参数作为最佳车道线参数。各种路况下大量实验测试和对比分析的结果表明:改进的RANSAC算法在各种工况下均能实现车道线的正确识别,具有较好的鲁棒性;同时删除容限的引入使算法的迭代过程相互独立,在提高算法实时性的基础上更有利于得到最佳车道线参数。To improve the real-time performance and robustness of lane marking detection algorithm in the conditions of varying illumination and worn, shady and stain covered lane marks, etc, an improved random sample consensus （RANSAC） algorithm is proposed based on feature extraction. Firstly corresponding marks are used to filter the right and left lane marking areas respectively and obtain a random sample set of feature points, from which feature points are randomly selected to initiate the lane marking parameters. Then the feature points, whose distance to initial model is less than threshold allowance, are selected from the complimentary set to form a consensus set and feature points in consensus set are used to reconstruct a new lane marking model. Finally all feature points in the consensus set of new model and those whose distance to new model is between threshold and deleting allowances are deleted. With above procedure iterated several cycles, the parameters calculated from the maximum consensus set are selected as the best parameters of lane marking. The results of a large number of experiments in various road conditions and comparative analyses show that the improved RANSAC algorithm can achieve correct detection of lane markings in various conditions with good robustness, and the introduction of deleting allowance makes iteration procedure independent, more conducive to obtaining the best lane performance of algorithm. marking parameters while enhancing the real time

关 键 词：辅助驾驶系统 车道线识别 改进RANSAC算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] U463.6[自动化与计算机技术—计算机科学与技术]