Stereo calibration in vehicles

Abstract

In this paper we present a self-calibration approach that updates the extrinsic parameters and the focal lengths of a stereo vision sensor. We employ a recursive estimation algorithm based on an Extended Kalman Filter. To improve the self-calibration process, we introduce a robust innovation stage for the Kalman filter: A Least Median Squares estimator is employed to eliminate outliers and thus to achieve better performance. The algorithm gives promising results on experiments with synthetic and natural imagery.