基于投影变换的快速行道线检测

Fast Lane Line Detection Based on Project

行道线检测是无人车辆自主驾驶系统中的一项关键技术。尽管行道线检测方法层出不穷,但是由于车载系统的局限性导致算法的时间复杂度不能过高,因此限制了许多算法在车辆嵌入式设备上的应用。为此,论文提出了一种快速行道线检测方法。该方法采用基于积分图的投影变换提取行道线候选点,然后通过基于方向过滤的霍夫变化方法粗定位候选行道线,最后利用消失点估计、对比度过滤等过滤器对行道线进行精定位。基于积分图的投影变换可以高效率地计算多尺度上的投影变换,同时给每个像素点建立局部投影变换,使得该像素点与整幅图的亮暗无关,提高了检测的鲁棒性,可有效解决遮挡问题。实验表明,对于典型路面图像,在i7处理器上中达到了10 ms每帧的速度。

Lane detection is a key technology in the automated vehicle system. There are many methods of lane detection,butthe limitation of vehicle system leads to the time complexity of the algorithm is not too high,so it limits the application of many algo-rithms in vehicle embedded devices. In this paper,a fast lane detection method is proposed. Firstly this method extracts the candi-date points in the lane image by projection transformation based on SAT. Then paper uses the Hough transform based on directionalfiltering to extract the line,its just a coarse positioning. Finally,paper uses the vanishing point estimation and contrast filter to lo-cate the line. Projection transformation based on SAT can improve the execution efficiency of the algorithm on multi scale. Using a lo-cal projection transformation to each pixel can make the pixel point is not responsible to the brightness of whole image,so it im-proves the robustness of detection and solves the problems in the shadow of lane. Experimental results show that 10 ms per frame inthe i7 CPU environment in typical pavement images.