Quality Enhancement of 3D Models Reconstructed By RGB-D Camera Systems

Abstract

Low-cost RGB-D cameras like Microsoft's Kinect capture RGB data for each vertex while reconstructing 3D models from objects with obvious drawbacks of poor mesh and texture qualities due to their hardware limitations. In this thesis we propose a combined method that enhances geometrically and chromatically 3D models reconstructed by RGB-D camera systems. Our approach utilizes Butterfly Subdivision and Surface Fitting techniques to generate smoother triangle surface meshes, where sharp features can be well preserved or minimized by different Surface Fitting algorithms. Additionally the global contrast of mesh textures is enhanced by using a modified Histogram Equalization algorithm, in which the new intensity of each vertex is obtained by applying cumulative distribution function and calculating the accumulated normalized histogram of the texture. A number of experimental results and comparisons demonstrate that our method efficiently and effectively improves the geometric and chromatic quality of 3D models reconstructed from RGB-D cameras.