Potential features for object identification by robots

Abstract

Object identification is very important for Robotic Manipulation of objects. A study on potential has been done. Three prime techniques of features extraction have been analyzed: Vision, Touch and Destruction of materials (NDE). The role of perceptual organization in aiding object identification is also discussed. A list of features has been obtained for each technique. Evaluation, based on cost of computation and accuracy of computation, of techniques for features extraction is presented. Some sample object identification systems have been designed using Classification Expert System Maker (CESM). Use of D-matrix (distiguishability matrix) is emphasized to get the optimal feature set used to generate a classification tree. The classification tree is transferred into CESM knowledge base to obtain an expert system. A comprehensive multisensor system for object identification is proposed.