Simplified Gaussian and mean curvatures to range image segmentation

This paper describes a new method of detecting 3D convex surfaces from range data using two simplified Gaussian and mean curvatures. Many methods for feature detection from range images are based on the signs of the Gaussian and mean curvatures. Usually, range image regions are classified into one of eight basic surface types. In this paper, a direct method of detecting 3D convex surfaces is proposed using the signs of simplified Gaussian and mean curvatures based on classical differential geometry analysis, under the assumption that a range image surface can be modeled by a Monge patch surface. It is shown that the simplified Gaussian and mean curvatures and usual ones are compared on their different mathematical behaviors from the theory of differential geometry. Experimental results on real range data are presented