Abstract
Road boundary detection on the roads without lane markings is one of the key technologies for advanced driver assistance systems. In particular, curb boundaries are important because curbs mean the physical boundary for the vehicle. This paper presents a method of detecting the curb boundary using an on-board stereo camera which can provide dense depth maps. The proposed method utilizes both the 3D road structure and the image texture. First, difference in level between streets and sidewalks is extracted from gradients of the dense disparity images at the short range. Next, the template image is set based on the position of the difference in level, and then the curb boundary at the long range is searched for using the template matching techniques. Some experiments in a road environment have confirmed the effectiveness of the proposed method.
