基于消失点检测与分段直线模型的车道线识别

Lane Detection Combining Piecewise Linear Model with Vanish Point Detection

提出一种基于消失点检测与分段直线模型的车道线识别方法。首先对高速公路图像预处理,采用基于消失点检测设定动态感兴趣区域(ROI)的方法,截掉预处理图像中包含天空等的无用信息,然后将动态感兴趣区域(ROI)均分为n部分,采用改进的极角约束Hough变换检测直线段,最后利用分段直线模型拟合车道线。实验结果表明,不管是直道还是弯道,分段直线模型都有很好的拟合效果,继承了Hough变换鲁棒性强、实时性好的优点,避免了曲线模型复杂、计算量大的缺点,同时使直弯道识别统一化。平均一帧图像的处理时间能满足实时性的要求。

This paper presents a lane detection method combining vanish point detection with linear model. Initially, highway RGB images were preprocessed, and then we set dynamic region piecewise of interest （ROI） based on vanish point detection of each frame to intercept off section containing sky and some useless information. Next, ROI were divided into n sections and we used polar angle constraint Hough transform to detect straight-line segments of each section coming along with piecewise linear model fitting lanes. Experimental results of image sequences show piecewise linear model which inherit the robust and real-time merit of Hough transform and avoid the complicated and heavy computation demerit of curve model have an excellent fitting effect both straight and curve. Besides, the piecewise linear model unify straight and curve lane recognition. The mean processing speed of each image meets real-time requirements.