Improved extraction of planar segments for scanned surfaces

The reverse engineering of a physical object often requires the production of a digital representation of the object. The object surface is scanned and a large number of points are obtained. These points are often organised in some way which provides adjacency information. However, in other cases, the result is an unstructured set of points, or point cloud. From such a point cloud a triangulated surface can be generated, so in all cases adjacency information can be obtained. Copies of the original object can be manufactured using the triangulated surface to define the shape. Often the original object is made up of parts of a number of simple geometric primitives and could be represented much more simply. Our aim is to segment a discretely represented surface into a small number of such simple geometric components using a 'region growing' approach. This paper describes an improved algorithm (with 'super triangles') for planar extraction, which requires only the data points and some form of adjacency information.