CAD-Based Off-Line Planning for Active Vision

摘要

A novel active-vision system for 3-D object recognition was developed. The vision system combines object pre-marking and active sensing. Therein, an object is modeled by a few of its 2-D perspective (standard) views, each with a corresponding viewing axis defined by a marker placed on the object. The operation of the active-vision system is characterized by its off-line planning and on-line recognition phases. The primary objective of this article is to address two issues in regard to the former: optimal object pre-marking and optimal camera placement. The optimal object pre-marking problem is defined as the determination of the minimum number of markers, and their best locations on a given set of objects to yield maximum "distinctiveness" for the 2-D standard-views defined by these markers. The optimal camera-placement problem targets minimization of camera movements in detecting the markers. Following a brief description of the proposed on-line recognition process, the techniques developed for the off-line planning stage are described in this article. For experimental verification purposes, a prototype of the proposed active-vision system was also implemented and is described herein.