Range image fusion for object reconstruction and modeling

A complete pipeline of data fusion from range images for a 3D object reconstruction and modeling is presented. The proposed approach includes multi-view registration, data integration, smoothing, and resampling. Firstly, the range images taken from multiple views are registered through a set of translation and rotation matrices whose coefficients are carefully pre-calculated. Then, a definition and two criteria to overlap elimination are provided as the foundation together with kd-tree data structure and nearest neighbor searching technique for data integration. A surface-based smoothing filter and a reliable resampling method, called the ball-travelbased resampling, are also given for surface quality improvement and data size reduction. All the operations manipulate range images directly without additional preprocessing operation, such as meshing or implicit surface function calculation to each range image, and thus provide a straightforward way to fuse any 3D data. The approach is applied to various range data sets of objects with different geometry shapes. The experimental results demonstrate the efficiency and applicability of the proposed method.