Target modelization and filter generation for 3D pattern recognition by optical correlation

This work presents a three-dimensional scanner system based on stereovision techniques to generate plane views of an object from an arbitrary viewpoint. These views are used as the reference templates in an optical correlator system designed to recognize the object by optical means. The stereo-matching algorithm we have developed is based on intensity criteria. In order to obtain the complete description of an object, we have to perform scans from several viewpoints. So, in general, each data set (range map) is in its own coordinate system. We have implemented a modified iterative closest point algorithm in order to register the collection of range maps into a common object-centered coordinate system. Based on the information of the 3D model we generate Fourier-space filters of an arbitrary 2D view, obtained by an implementation of a view-based rendering algorithm. The filters have been evaluated by a simulation of a VanderLugt architecture.