Depth assessment by using qualitative stereo-vision

Abstract

Presents a method for assessing the distance between a mobile robot and an obstacle. The idea is to make use of stereo cameras that are mounted on a pan-tilt platform (the pan-tilt device itself is mounted on the mobile robot). When an obstacle is encountered, the pan-tilt device drives the stereo cameras to an orientation such that the left and right images of the obstacle are in a symmetrical position with respect to the vertical central line of the horizontally aligned stereo images. In this configuration, the distance between the left and right images of the obstacle is a function of the depth between the mobile robot and the obstacle. We qualitatively establish this relationship by using a look-up table or a 2D graph to avoid the need to do 3D reconstruction. Therefore, no calibration is needed to determine the intrinsic/extrinsic parameters of the stereo vision.