Research On Key Technology Of 3D Vision Reconstruction Based On Static Scenes

The technology of three-dimensional (3D) vision reconstruction is one of the research focuses on the field of computer vision. According to relevant theories and algorithms in stereo matching, a series of images of objects and scenes shot by image acquisition equipment are employed to calculate the depth information of the objects in the images. And finally the 3D information of the objects are acquired within the scenes through the method of 3D reconstruction.On account of building disaster rescue background and computer stereo vision technology, the interior scenarios of building are displayed on real-time in the form of 3D visualization model according to the data received from the binocular video acquisition nodes distributed throughout the buildings. Thus through 3D panoramic view, rescuers are able to accurately and intuitively figure out the distributions of people in every part of the building and the disaster spread, and therefore make the fastest, safest and most effective plans of rescuing and dredging, ensuring that the rescues carry out successfully.Based on the learning of basic knowledge and analysis of a large number of relevant references, this thesis carries out research on three aspects:camera calibration and image correction, stereo matching of image pairs, and multi-view 3D reconstruction.(1) Classic camera calibration algorithm is applied to calibrate the binocular camera. And according to the obtained parameters, the images are rectified through the nonlinear distortion correction algorithm presented in this thesis. Then the epipolar line in the image pairs is corrected by the epipolar line correction algorithm presented in this thesis.(2) By analyzing and comparing the advantages and disadvantages of conventional feature-based and region-based stereo matching algorithms respectively, this thesis propose dynamic programming stereo matching algorithm under the constraints of feature points and mean-pyramid which programs the region-matching search range dynamically. Through this algorithm, a more precise disparity map can be obtained after stereo matching of image pairs.(3) Scenes from two different viewpoints are reconstructed respectively and ICP algorithm is adopted to compute the relation matrix between two pairs of 3D point-clouds. According to this relation, the points are converted to the same coordinate system so as to improve the reconstruction effect and expand the scope of reconstruction.The distortion correction algorithm, epipolar rectification algorithm and stereo matching algorithm presented in this thesis are validated by a large number of experiments to be of great accuracy and strong generality for static scenes in different situations.