One of the 3-D vision tasks is to create a full 3-D model of an object. Only the partial range images from different view points can be obtained by real range finders. it is possible to get partially overlapped rang images. The views differ by Euclidean motion. This motions have to be known for combining the views. This article describes a method for registration of range images. Our algorithm is based on matching of differential structures of the surface. Results of this algorithm are used as first estimate in the distance minimization algorithm -- Iterative Closest Reciprocal Points (ICRP). Such approach produces more accurate results and solves some problems of ICRP algorithm.

The sequence of steps that allow to recover volumetric 3D CAD models of a 3D object from a set of scanned data (range images) is described. The novelty of the approach is twofolds: (i) the whole chain of providing usable technology for industry was closed, (ii) segmentation of the triangulated surface into analytic quadric patches is used to considerably improve the obtained 3D models of a typical industrial objects. The range finder is used to capture range images from different viewpoints. The triangulated surface is constructed over each range image. The data sets are reduced by decimation of triangular meshes. Surfaces are registered into a common object centered coordinate system and data are fused into a tetrahedralization of the points by -shape. The tetrahedralization is segmented into explicitly model paths. The beautifying step improves fused data by moving the projected points onto the recovered models of surfaces. Such data enable direct reconstruction a full ...