Reconstruction of Real-World Objects via Simultaneous Registration and Robust Combination of Multiple Range Images

Range finder devices are quite useful to measure the geometry of real-world objects. However, the reconstruction of large and complicated objects often requires the acquisition of multiple range images showing different views of the object. With our reconstruction method, the relative orientations of all range images are determined via simultaneous registration of all range images. The registration process is based on a least-squares approach where a distance metric between the overlapping range images is minimized. Registration errors are not accumulated, and it is even possible to reconstruct large objects from an arbitrary number of small range images. A resolution hierarchy accelerates the registration substantially. In order to exploit redundancies, the overlapping surfaces are optimally adapted to each other. The processed range images are converted to a signed distance function, which is based on the idea that no part of the object can lie between the scanner and the measured surface in any view. According to the signed distance function an intermediate volumetric model is sculptured out and polygonalized with isosurface techniques. The accuracy of the generated mesh is improved by moving its vertices onto the surface implicitly defined by the registered range images.