Automatic 3D scan view registration using 6D scene differentials

Abstract

Commonly the determination of one coordinate system relative to another is based on the equivalence of scenes measured in the two frames. If 3D scene data is sufficiently accurately available in both frames, the calibration in 6D configuration space can even be carried out fully automatically. However, in some applications, no scene data is explicitly available, but merely differentials in 6D space are known. One important application is automatic registration of 3D scan views, using data from an external placement measurement device, but without using markers. In this paper we show that from two linearly independent scene differentials, each measured relative to two frames, a calibration of the frames can be achieved. The problem can be reduced to minimization of a bivariate function of two real parameters. We have investigated the conditions in which the method is sufficiently accurate. We report on this investigation and describe the application of the technique to quick 3D scanning of hand-held objects.