As a first step in the process of automatically building geometric models of articulated objects from multiple range images, we have developed an algorithm that segments range images into rigid subsets. Evidence for the presence of rigid object subcomponents can be determined by measuring the transformations that align surface points and patches from one range image to another. Here only our new work on surface patches is reported but we have previously reported successful segmentation using point correspondences. Surface patches belonging to the same subcomponent can be identified as a set of surface pairings which undergo a similar transformation. The algorithm has been tested on real range data and shown to work effectively.
One advantage of this approach is that, in addition to performing the required rigid part segmentation, an estimate of the transformation that aligns each subcomponent between images is determined. This will provide a valuable initial estimate for performing accurate surface registration later on.
In this paper, the algorithm has only been tested on planar surface patches, but the approach can be extended to many different classes of surface. We are currently experimenting with cylindrical surfaces and intend to extend the algorithm to other degenerate surfaces, for example spheres. We have already published results on a rigid part segmentation algorithm which works on non-developable surfaces and intend to integrate these algorithms together to provide a general purpose tool.
Anthony Ashbrook