A Badly Calibrated Camera in Ego-Motion Estimation, Propagation of Uncertainty

This research was supported by Region Rhöne Alpes with the program TEMPRA -- East Europe between l'INPG Grenoble and the Czech Technical University, the Czech Ministry of Education grant No. VS 96049, grants 102/97/0480, 102/97/0855 and 102/95/1378 of the Grant Agency of the Czech Republic and European Union grant Copernicus CP941068, and the grant of the Czech Technical University No 3097472.International audienceThis paper deals with the ego-motion estimation (motion of the camera) from two views. When we want to estimate the ego-motion we have to find correspondences and we need a calibrated camera. In this paper we solve the problem how to propagate known camera calibration errors into the uncertainty of the motion parameters. We present a linear estimate of the uncertainty of the motion parameters based on the uncertainty in the calibration parameters. We show that the linear estimate of the motion parameters uncertainty is very stable and useful up to 20 % noise in camera calibration even with a noise in correspondences