Abstract: A robust lane recognition method for intelligent vehicles was proposed based on machine vision herein to solve the problem that the traditional lane recognition algorithm was difficult to recognize in the complex road environments. Firstly, in order to eliminate noise interference and improve the efficiency of feature detection, an adaptive region of interest (ROI) calculation method was designed, which could adaptively separate the lane region from the non lane region according to different conditions. Secondly, to improve the environmental adaptability of the algorithm, the detection operator of improved partition angle was used to detect the lane line features, and the multi-color threshold processing was used to deal with the lane images. Finally, DBSCAN clustering and NURBS curve were used to fit the lane line after changing the view angle, and random sampling consistency method was used to optimize the lane line model to filter out mismatches. Experimental results show that the algorithm may effectively identify lane lines under various road conditions.

Key words: machine vision, lane line identification, intelligent vehicle, feature detection, robustness

摘要： 针对目前传统车道线识别算法在复杂道路环境中识别困难的问题，提出了一种基于机器视觉的智能车辆鲁棒车道线识别方法。为消除噪声干扰以及提高特征检测效率，设计了一种自适应道路感兴趣区域(ROI)计算方法，针对不同情况的车道可自适应地将车道区域与非车道区域分离。对待检测目标采用改进划分角度的检测算子进行车道线特征检测，同时对车道图像有针对性地采用多色域阈值处理，以提高算法的环境适应性。对转换视角后的车道线采用DBSCAN聚类及NURBS曲线进行拟合，并利用随机抽样一致法优化车道线模型以滤除误匹配。实验结果表明，该算法可有效识别出各种道路工况下的车道线。

关键词: 机器视觉, 车道线识别, 智能车辆, 特征检测, 鲁棒性