Abstract: It is difficult to maintain a balance between accuracy and real-time performance of lane line recognition, thus a new lane line recognition method was proposed based on region division. Firstly, an improved OTSU algorithm was applied to segment the edge image; then, feature points in that edge image were extracted by using Progressive Probabilistic Hough Transform (PPHT) algorithm and fitted as a line by using Least Square Method (LSM). Finally, all fitted lines were judged and the possible lines were chosen by using an anti-interference algorithm. Comparative experiments were conducted with three other algorithms mentioned in the references. In addition, an evaluation model was put forward to assess the performance of the algorithms when dealing with 500 typical lane images. Meanwhile, by calculating the average overhead time on processing each frame of a 1 min 26 s video, the response time of the algorithm was evaluated. The experimental results show that three indexes including precision, recall rate and F value of the proposed algorithm are better than the comparison algorithm, and the proposed algorithm also meets the requirement of real-time processing.

Key words: lane line recognition, edge image, OTSU algorithm, Progressive Probabilistic Hough Transform (PPHT), Least Square Method (LSM)

摘要： 针对多数研究中车道线检测的准确性和实时性难以有效平衡的问题,提出了一种应用区域划分的车道线识别方法。首先通过改进的大津(OTSU)算法提取边缘图像,再在所得边缘图像的基础上,利用改进的概率霍夫变换(PPHT)提取车道标识线上的特征点,并采用最小二乘法(LSM)对特征点点集进行直线拟合,最后通过提出的路面干扰线规避算法检测所有拟合得到的直线段并筛选可能的车道线。在实验方面,引入三种算法作为对比,并利用提出的准确性评价模型对500幅典型道路场景图中的车道线识别结果进行评估,同时统计在处理一段长为1 min 26 s的道路视频时每帧图像序列的平均耗时。实验结果表明所提算法的查准率、查全率、F量测值均优于对比算法,且达到实时处理的要求。

关键词: 车道线识别, 边缘图像, 大津算法, 概率霍夫变换, 最小二乘法