Research On 3D Reconstruction Technology Based On Computer Vision

The 3-D reconstruction technology of computer vision is a significant field of computer vision, and what's more, it can help realizing 3-D untouched measurement and fast reconstruction. This technology is a main way of 3-D reconstruction in the future because it is efficient and automatic.This dissertation mainly does research in Camera Calibration, Feature Extraction, and Stereo Matching.For camera calibration, this dissertation applies the theory of ANN (Artificial Neural Networks) calibration to practical application. We can complete a binocular calibration by employing ANN calibration once, and the intricate calculation has been avoided.For feature extraction, this dissertation studies Barnard algorithm for Edge Point extraction and SUSAN algorithm for Corner extraction, Afterward, I put forward a method to extract inflexions, straight lines and curves based on Freeman Differential Code. Thus the use of Freeman Code can be extended and there will be more approaches to finish stereo correspondence.For stereo matching, this dissertation introduces relaxed iterative algorithms to realize the correspondence of points extracted through feature extraction. And based on extracted Freeman lines, I put forward a stereo correspondence technic for the element of lines. From experimenting on this technic, it can be found that stereo corresponding on lines is faster than it is on points.At the end of the dissertation, the full steps of reconstructing an actual model are given and all requisite experiments are done in MATLAB.