Recognition of simplecurved surfaces from 3 D surface

This paper is concerned with the robust identiication of patches from simple geometric surfaces such as planes, spheres, and cylinders. This is done from 3D data acquired using a range scanner, along with principal quadric estimates at each point. Properties of the Gaussian and mean curvature (e.g., their sign) at points on surfaces of the target class are used to reject all points that could not be on such a surface. Each remaining point is then mapped to the parameters of the unique surface in its class that contains it, if it were on such a surface. Region growing and clustering techniques are used to identify those points actually on a surface of the target class and to extract the surface parameters.