地面无人平台车道线快速检测算法试验研究

Experimental Research on Algorithm for Fast Lane Detection of UGVs

针对地面无人平台视觉导航问题,提出一种改进的道路快速检测算法。改进水平均值投影法划分道路和背景区域,结合边缘检测算子和最大类间方差法构成双阈值法,对道路区域图像进行二值化处理。利用先验知识改进霍夫变换,在路面存在阴影和噪声干扰的条件下,准确检测车道标识线,动态预测划分感兴趣区域。采用菱形搜索法进行车道线跟踪,融合初始检测和后续跟踪两层算法,循环处理道路图像序列。试验表明,此算法具有良好实时性和鲁棒性,满足地面无人平台高速行驶要求。

An improved algorithm of fast lane detection is proposed for the visual navigation of unmanned ground vehicle( UGV). The mean horizontal projection of pixels' gray values is improved to divide an image into the road and background areas. A double-threshold method is generated by the edge detection operator and the largest between-class variance( Otsu algorithm) in order that the gray scale of road region can be converted into binary image. The algorithm improves Hough transform using the transcendental knowledge,delineates the dynamically predicted region of interest,and introduces a diamond search algorithm for tracking lane line. It can well and truly detect lane marking even if there are some interference factors,such as shadow and yawp,in the road,processes the sequences of road images by means of recycledly using the initial testing and subsequent tracking modules. The experimental result shows that this algorithm which is of high efficiency and robustness can meet the fast driving requirement of unmanned ground vehicle.