Binocular stereo vision plays an important role in three-dimensional data acquisition. This paper gives a general introduction to stereo vision and its implementation steps. This paper also covers the main process of binocular 3D laser scanner, which uses laser dots array as initiative light, stereo matching, 3D reconstruction and local registration. Stereo matching is the most impor-tant and difficult step. We explain the commonly used epipolar-constraint and ray-intersection constraint in stereo matching. The paper summarizes three methods of 3D reconstruction: the method based on solving over-determined linear equations, the method based on ray intersection and the method based on the disparity principle. We advanced a new method, based on disparity, of 3D reconstruction, when the two cameras have an inclination; this method needs the data the same to the common methods. This paper sums up some methods of local registration: ICP method (iterative closed points method) and a non-iterative method based on RANSAC, and we also compares the merit and fault of the two methods. |