抄録
For environmental recognition with cameras, a stereo vision system is frequently used. The accuracy of 3D reconstruction of a standard stereo vision system is limited due to a pixel size on camera's image plane. To enhance the accuracy, this paper proposes a camera with translational and rotational motion mechanism. Such a camera system can take numerous images by utilizing the mechanism. To make quantization error smaller, we get the average of reconstructed positions of feature points. Furthermore, wrong correspondences between an image-pair can be removed through statistical procedure. This paper discusses the effect of a proposed algorithm by applying it to real images. From the results, it is shown that the proposed algorithm can enhance the accuracy of 3D reconstruction.
