Depth Image-based 3d Reconstruction Techniques

3D reconstruction have two important performance indicator, model accuracy and automation, this thesis aimed at improve geometric accuracy and the degree of automation for structure light based 3d scanner. Structure light scanner use a projector to project coded structure pattern to objective, and a calibrated camera snap the correspond image, and then apply image analyze technologies to reconstruct the range image, and then register and integrate these range images to build the model.3D Reconstruction system prototype acquired a face scan objects depth images models, and then require many artificial operation to register these scan results, made it hard to be full automatic, this thesis apply high-precision digital rotator to solve the problem, the corresponding issue is how to joint the scanner coordinate frame and the rotator coordinate frame, the paper adopted Circle Fit and Straight line Fit to deal with it, so that we can get the precise registered depth images by combine the depth image and rotate angle.Precision of Reconstruction system mainly constrained by system calibration, the paper analyze the major factor about precision and find that the most serious stuff is cameras and the projectors lens geometric distortions, this thesis build several deform mathematical models, and undistorted two-dimensional image, thereby improving the accuracy of 3D data.Another work lies in the integration of registered and undistorted depth images. There exists two categories of integration, point cloud integration and mesh integration, which is similar in nature, this thesis will demonstrate these two technologies and compare their respective results.Our reconstruction system has been applied in several fields, culture relic digitization and reverses engineering for instance, and the performance is rather good.