Research Of Key Problems Of Three-dimensional Reconstruction

With the development of computer science, the three-dimensional reconstruction technology has been applied extensively. The three dimensional reconstruction combines a number of computer vision technologies. A complete reconstruction system is composed of image acquisition, camera calibration, image preprocessing, stereo matching, reconstruction, and post-processing. Every step mentioned above is critical to the reconstruction results. Thus, several critical technologies for three dimensional reconstructions are researched in this thesis.Firstly, the camera calibration is discussed in this thesis. Various camera imaging distortions are analyzed, and a complete camera model is given. Also, the object-based calibration method is improved. Setting ellipse template as a calibration object, this thesis describes space ellipse and its model of center imaging, calibrates and extracts the ellipse center. The Tsai calibration approach, based on the ellipse center calibration, is given, which uses calibrated ellipse center to carry on the Tsai calibration. The experimental results prove this method could fulfill higher calibration accuracy.Moreover, image pre-processing method based on stereo matching is studied. The outlines of image are extracted through color threshold, median filtering, refinement and expansion technologies, in order to set up the foundation for subsequent stereo matching process. Images are corrected by Fusiello A algorithm, to reduce the computational burden of stereo matching.Furthermore, both SIFT features extraction and matching are demonstrated. The multi-scale image technology, image pyramids and scale space generation techniques are discussed. The complete implemental process is described. Then, the procedure is fulfilled experimentally and acquired complete results.Finally, the algorithm for depth image acquisition based on profile characteristics is researched. Parallax measurement principles are described, and the relationship between recovery and three-dimensional reconstruction is established. The three dimensional depth images are acquired, after the established depth profile images by angle matching method, are linearly interpolated.