Abstract
In this paper, we propose a robust approach to register two coarse binary volume models. The presented method is based on a combination of maximizing the spatial cross-correlation and an extended version of the iterative closest point algorithm (ICP). During an iterative optimization the cross-correlation, i.e. the section volume of both models, is maximized and the mean squared distance between surface points of the first model and their respective closest surface points on the second model is minimized simultaneously by varying the six parameters of an arbitrary spatial movement. The proposed method has been applied to merge different, independently created binary volume models of a rigid object. Each model is created by a shape from silhouette approach using a set of views captured by a single camera that observes the object rotating on a turntable. Since rotation around a single axis does not allow the camera to examine the entire object surface, different models, captured with varying orientations of the object on the turntable, have to be integrated to obtain a complete 3D model. Thus, merging two or more sequentially created coarse volume models is an essential technique to obtain high quality 3D models of real objects with a simple turntable setup.