Self-calibration for Active Automotive Stereo Vision

In this paper, we present the extrinsic self-calibration of an active stereo rig with independently moving cameras. The proposed self-calibration recovers all extrinsic parameters up to scale and accounts for mechanical inaccuracies in the camera setup. Our algorithm is based on two criteria: the minimization of the reprojection error of tracked image points and the epipolar constraint. The self-calibration is implemented as a robust iterated extended Kalman filter to allow continuous operation. Special care is taken to minimize physically relevant geometric errors instead of algebraic cost functions. The proposed algorithm is evaluated on synthetic and real-life imagery and allows initial extrinsic calibration as well as continuous tracking of the stereo camera parameters