Road surface estimation robust against vehicles' existence for stereo-based vehicle detection

Abstract

In parameters estimation for stereo-based vehicle detection, it is important to estimate the three-dimensional geometrical relationship between a stereo camera and a road surface. However, vehicles' existence in a field of view sometimes has a bad influence upon the accuracy of road surface estimation. Therefore, we propose a method for road surface estimation robust against vehicles' existence in a field of view. This method considers the difference between the distribution of feature points from a road surface and those from a vehicle in the spatio-disparity space. Therefore, road surface estimation can be done using those mostly from a road surface. We also confirm its effectiveness by experiments on field images.