Abstract: A lane classification algorithm is presented. Firstly, LDA（Linear Discriminant Analysis） is adopted to convert road images to gray images directly, which can distinguish the lanes and roads better. Secondly, LSD（Line Segment Detector） is applied to detect straight lines of gray images to give the main direction of the lane by proper selection. On the basis of that, the pixels are selected within the grayscale range of the lane. Pixels in far vision are fitted by parabola while pixels in near vision use linear fitting. In the meantime, the lanes are marked by classification. Finally, the detecting results are verified by the continuity of video sequence. Experimental results show that the proposed method has good effects on the detection of curves and straight lines. The processing speed is 10 f/s while the frame rate of the original video is 15 f/s, which basically meets the real-time requirements.

Key words: Linear Discriminant Analysis（LDA）, Line Segment Detector（LSD）, linear-parabola model, lane classification, continuity of video sequence

摘要： 提出一种车道线分类检测算法。首先采用LDA对道路图像进行有针对性的灰度化，以便更好地区分车道线与道路。采用LSD算法检测灰度图像中的直线部分并确定车道线的方向。在此基础上，选取符合车道线灰度范围内的像素点。对远距离的像素点采用抛物线拟合，近距离的像素点采用直线拟合。同时，将检测到的车道线进行虚线实线的分类标记。最后结合视频序列的连续性对检测结果进行反向验证。实验结果证明，提出的方法对直道弯道检测均有很好的效果。算法的处理速度为每秒10帧左右，采用的测试视频的帧率为每秒15帧，基本满足实时性的要求。

关键词: 线性判别分析（LDA）, 线段检测器（LSD）, 直线-抛物线模型, 车道线分类, 视频序列连续性