Segmentation of range images via data fusion and morphological watersheds

Abstract

As in 2-D (in two-dimensional) computer vision, segmentation is one of the most important processes in 3-D (three-dimensional) vision. The recent availability of cost-effective range imaging devices has simplified the problem of obtaining 3-D information directly from a scene. Range images are characterized by two principal types of discontinuities: step edges that represent discontinuities in depth and roof (or trough) edges that represent discontinuities in the direction of surface normals. A Gaussian weighted least-squares technique is developed for extracting these two types of edges from range images. Edge extraction is then followed by data fusion to form a single edge map that incorporates discontinuities in both depth and surface normals. Edge maps serve as the input to a segmentation algorithm based on morphological watersheds. It is demonstrated by extensive experimentation, using synthetic and real range image data, that each of these three processes contributes to yield rugged and consistent segmentation results.