Abstract
An efficient incremental algorithm for integrating overlapping, registered range images acquired from different points of view is presented. Every new range image, which is represented by a 3D triangular mesh, is merged with a second triangular mesh which represents the current reconstructed model. The merging process consists of three stages. The first stage finds the area of overlap between the new 3D mesh and the reconstructed model. Triangles that belong to the area of overlap of the coarsest mesh are removed. In the second stage, the boundary created after removing the previous triangles, along with its adjacent boundary from the unaltered (fine) mesh, are projected to a reference plane and triangulated with a 2D constrained Delaunay algorithm. Finally, the last stage projects the previous 2D mesh back to the 3D space, leading to a 3D mesh that sews the new range image to the current reconstructed model. The algorithm is suitable for integrating range images acquired with different spatial resolutions and containing one or more separate objects.