Markerless online stereo calibration for a humanoid robot

Abstract

In this paper we present an online stereo calibration system that calibrates the eyes of a humanoid robot at a kinematic chain level using information from the embedded cameras and the motor encoders. The system calibrates the joints' offsets and mounting errors of the cameras using advanced filtering techniques based in the epipolar constraint. By calibrating the stereo system at a kinematic chain level we are able to continuously provide an accurate stereo reconstruction of the world with version and vergence of the eyes. Our method is able to converge even with large initial errors and is independent of camera movements - can operate with still or moving cameras. The filtering approach is essential for smoothing out errors arising from spurious image matches and converges in just a few seconds in case of sudden model changes. Experiments using the iCub robotic head are presented and illustrate the performance of the methodology as well as the advantages of using such an approach.