Multiresolution registration of 3D range images of free-form surfaces
Vania Conan, Pascal Aubry · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1997
We propose a new method for the registration of several views acquired by a range finder, for rapid prototyping applications. This method works well with free-form surfaces, and does not need an initial approximate relative positioning of the two data sets. The algorithm works along the following line: First approximate each data file with a multi-resolution quadric spline. This is achieved using fast filtering techniques. Second compute the gross rigid motion. This is achieved through a hypothesis/verification algorithm based on the multi-resolution quadric spline patches. This step minimizes a robust criterion, based on the least median of squares, which is tolerant to up to 50% of false matches. Last run an iterative closest point procedure. This local iterative scheme refines the gross rigid estimate, and gives the final registration. The method is tested on real 3D data. The front and side views of a human face are registered in order to build a more complete 3D model.