Unstructured Road Edge Detection and Initial Positioning Approach based on Monocular Vision

This paper proposes two complementary road edge detection methods for unstructured roads. The methods can be used for visual navigation and intelligent control of mobile robots. One method is color-based, which is applicable to the circumstance that the road and off-road areas have obvious color difference; the other is based on threshold segmentation and can be applied even the road boundary is not clear. On the other hand, shadow has enormous impact on threshold segmentation method, but color-based method is not sensitive to it. These make the two methods complementary. Furthermore, on the basis of edge detection, inverse perspective projection transformation is carried out to calculate running state of the robots. Through experiments on mobile robot platform, verifications that the two proposed methods complement each other in different road environments well are made. The experiments also show that the methods achieve high accuracy in initial positioning and meet requirement of real-time.