The Automatic Sensing and Analysis of 3-D Surface Points from Visual Scenes.

摘要

Described are the design and implementation of a new range-measuring sensing device and an associated software algorithm for constructing surface descriptions of arbitrary three-dimensional objects from single or multiple views. The sensing device, which measures surface points from objects in its environment, is a computer-controlled, random-access, triangulating rangefinder with a mirror-deflected laser beam and revolving disc detectors. The algorithm processes these surface points and generates, in a deterministic fashion, complete surface descriptions of all encountered objects. In its processing, the algorithm also detects parts of objects for which there is insufficient data, and can supply the sensing device with the control parameters needed to successfully measure the uncharted regions. The rsulting object descriptions are suitable for use in a number of areas, such areas as computer graphics where the process of constructing object definitions has heretofore been very tedious. Together with the sensing device, this approach to object description can be utilized in a variety of scene analysis and pattern recognition applications which involve interaction with 'real world', three-dimensional objects.