3D Reconstruction Of Anomalistic Objects Based On Stereo Vision

The goal of the research on stereo vision is to understand and recognize the object, so it is necessary and important to reconstruct the object from two or above 2D images to a 3D image in virtual environment. Influenced by the facts such as the shape characteristics of the object, illumination, surface material characteristics of the object and the parameters of the cameras, it is a reverse process to conclude the parameters above only by the gray-level of the image, and these questions are non-liner and the result is un-unique, so the process is ill-conditioned. The major difficulty of the stereo vision technology, therefore, lies in how to get a robust result.The paper has researched camera's calibration, image pre-processing (including image graying, image smoothing, image edge detection and image segmentation), stereo matching and 3d reconstruction, and the paper is focused on image segmentation and stereo matching, which can be described as follow:Image segmentation: presenting an optimized segmentation method based on 2D gray histogram compare with traditional method. Generally, the gray value of a certain pixel on the filtered image is almost the same as that of its neighborhood, so the area near the diagonal deserves more attention. According to this foreknowledge operation can be reduced and the noise pixels, defined as gray value of the pixel which is much different from its neighborhood, can be avoided. So the operation speed and accuracy of the optimized method is much more realistic and useful than traditional one.Stereo matching: integrating the merits of the two technologies of laser scanner and stereo vision. The paper presents the method "A Method of Image Matching Based on Laser Scanner Aided Stereo Vision". The method takes advantage of the good direction-sense of the laser and fast, untouched and all-around data acquisition of stereo matching. Aided epipolar constraint can lead to excellent matching results.Finally the theoretical analyses and the measurement results show that the method presented in this paper can partly improve the operation speed and measurement accuracy while simplifying measuring equipment and reducing measuring costs.