Calibration of a free-form surface mirror in a stereo vision system

The presence of repetitive patterns in the domain of stereo vision results in multiple matching hypotheses. Choosing the wrong hypothesis leads to an incorrect determination of objects and calculation of their distances. Especially in vision-based driver assistance and safety systems a high accuracy in matching and distance calculation is vital. One approach to increase the robustness of matching is to add a third camera to the system. Another idea is to resolve the ambiguities over time. In this paper we present a different solution without additional hardware expense: The shiny varnished hood of the car provides a mirror which is in a fixed position relative to the cameras. This mirror displays additional views of the scene, providing the required information. The cameras and the respective camera-mirror systems need to be calibrated. The epipolar geometry and the disparity-distance function of the free-form surface mirror must be determined. After establishing this we can obtain the correct matching hypothesis.