ALGORITHM FOR MATCHING 3-D LINE SEGMENTS WITH APPLICATION TO MULTIPLE-OBJECT MOTION ESTIMATION.

The algorithm segments the objects that move differently in space and identifies corresponding line features for each object. The matching is accomplished in two stages. First, corresponding line features which allow a consistent interpretation of rotation are identified by a tree-search procedure. The pairings generated in the first stage are then pruned by a Hough-like procedure such that a 3-D translation would bring the (rotated) corresponding line features to overlap. The algorithm can also be applied to object recognition, robot navigation, and object model construction problems by incorporating other object properties or nongeometrical object attributes in the matching process. Results of running the algorithm on experimental data are given.