Abstract

This paper describes the recovery of wireframe object models from the edge data delivered by stereo vision systems. Edges are typically broken into segments, vertices are often absent, and the recovered geometry may be inaccurate, in addition data associated with surface texture and shadows is not distinguished from the objects edges. The work reported here is addressed at the problem of deriving from such data accurate object models representing geometrical and topological properties explicitly in the form of a Surface/Edge Vertex (SEV) graph. Such viewpoint invariant models will form the basis for automatic model acquisition, model matching, and geometric reasoning for motion planning, grasping etc.

© 1989 Optical Society of America