Object detection in outdoor scenes by disparity map segmentation

Abstract

An approach for scene segmentation based on disparity map is proposed. The basic idea is to group features close in space into objects. The advantages of such an approach are that the search takes place in 2D space, instead of 3D space and that the uncertainty can be taken into account easily and uniformly. The authors show how to express a 3D proximity criterion in terms of disparity and how the problem of uncertainty is coped with efficiently. The result on a traffic sequence is presented.<>