High-accuracy Range Image Generation by Fusing Binocular and Motion Stereo Using Fisheye Stereo Camera

This paper proposes a method that fuses two stereo measurements: binocular and motion stereo. The method is implemented for a fisheye stereo camera. The two stereo measurements have difficult characteristics. In case of the binocular stereo, the length of the baseline is usually small due to the installation conditions to a robot or a car. On the other hand, in the case of the motion stereo, the baseline is the distance the camera moves between two consecutive frames, and thus, the length of the baseline is usually longer than binocular stereo. In addition, the area-based approach is used for binocular stereo and the feature-based approach is used for motion stereo in the corresponding point search. In summary, the two stereo measurements have differences in the length of the baseline and the accuracy of the corresponding point search. Therefore, their fusion is studied to improve the stereo measurement. A bilateral-like filter, which is a weighted averaging method, is proposed for the fusion of the two measurements. In motion stereo, the proposed method is verified by outdoor experiments.