一种快速的基于边缘的道路检测算法  

A Fast Edge Based Unstructured Road Detection Algorithm

作　　者：杨文杰[1] 胡明昊[1] 杨静宇[1] 

机构地区：[1]南京理工大学计算机系,南京210094

出　　处：《计算机科学》2006年第5期257-260,共4页Computer Science

基　　金：国防科工委研究基金(K1703060113)

摘　　要：道路检测是机器人视觉导航中的重要环节。本文提出了一种新的非结构化道路检测算法。算法基于这样的一个基本假设,即道路的轮廓由一些具有一定长度的连续直线段组成。在一定的范围中通过计算边缘点,然后通过Hough 变换求得直线段,从而确定道路边界的位置。但是边缘图像中不仅仅包含道路的边缘,还有阴影,植物等其他的边缘。如果不消除这类边缘的影响,将大大削弱算法的效率和准确性。本文提出一种基于颜色和方差的规则有效地消除这类非道路的边缘。一旦当前帧中获取道路的边缘,通过限定下一帧中的感兴趣区域(ROI)减少搜索范围从而快速定位帧中的道路边缘,从而进一步提高算法的效率。大量实地(校内公园道路)实验验证了该算法的快速和鲁棒性。Road detection is one of the basic tasks for automatic guidance. A new approach to detect the unstructured road is proposed. It is based on the fundamental hypothesis that the contour of a road is made up of many connected line sections with some length. Here, line sections can be got by Hough Transform, which accumulates edge points in the image and determine the position of the road contour. But image edges include not only the road contour but also many other kinds of edges, such as edges of shadow, plant and so on. If these kinds of edges being considered, the algorithm＇s efficiency and accuracy will be impaired greatly. Some rules are used to eliminate the edges of non-road contour before accumulating. Besides when the road has been found in one frame, we can infer ROI （the region of interest） road to decrease the searching scope and fast locate the road edge in the latter frame, which can furthermore improve the algorithm＇s efficiency. Lots of experiments result shows that it is fast and robust for roads detection at a park of the eampus.

关 键 词：道路检测 道路跟踪 边缘过滤 Hough变换 机器人 视觉导航 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]