From 3D line segments to objects and spaces

An intermediate representation of stereo image data in terms of 3D line segments is used to extract visible surfaces and their parameters. Methods and algorithms for recovering planar, cylindrical, conical, and spherical surfaces are described, and some test results are presented. The essence of the approach is testing of small sets of 3D line segments for compatibility with a particular surface type. Maximal sets of segments of supporting different surface are then identified. Some of the algorithms involve a novel use of the dual space representation. In the domain of polyhedral scenes initially restricted to blocklike objects and spaces, the planar surfaces are combined, using connectivity, to create 3D boxes, that correspond either to simple (i.e. convex) objects or spaces, or to convex parts, which are further combined to create composite objects and spaces