Accurate pose estimation from a single perspective view

Abstract

The problem of guiding a mobile robot in the 3D world, using a set of known targets conveniently positioned at each of its docking stations, is discussed. A sequence of images is acquired along the way from one station to another. Each image in the sequence is processed, and motion clues are extracted to correct the 3D position and orientation of the robot caused by motion slip and inaccuracy in the mechanical assembly and platform holding the robot. Using the rigidity of the target with respect to a fixed coordinate system, the authors provide a set of equations from which a unique solution to the camera pose is determined. Tests of this solution on synthetic data are presented; they show that the analytic method is very accurate.< >