Robust depth estimation for mobile robot navigation

Abstract

The depth estimation of a scene is very useful for solving problems like mobile robot navigation. This paper presents a novel and robust depth estimation approach based on feature points. Only two feature points are selected from an object. They are at the same height'from a ground plane, which is different from all reported depth estimation methods. When a camera moves at arbitrary direction to the object, consecutive two images containing the two feature points are taken. The depth o i the two featore points before and after the camera moves can be computed at a time according to the image coordinates of them. A search algorithm is also presented to enhance the precision of the computed depth and the robustness of this estimation approach. It can automatically find more correct image coordinates of the two feature points based on the really detected image coordinates. Experiments results show that it is effective and can enhance greatly the precision of the computed depth.Index Terms -Depth estimation, image sequence, motion vision, mbot navigation 0-7803-8748-1/04/520.00 02004 IEEE.