Object Recognition Using Vision and Touch

Abstract

A system is deacribed that integrates vision and tactile sensing in a robotica environment to perform objcct recognition tub. It uaes multiple sensor aystema (active touch and pusive stereo vision) to compute three dimensional primitives that can be matched against a model data bue of complex curved surtace objccts containing holes and cavitica. The low Ievel sensing elcments provide local surtace and feature matches which an: eonstrained by relational criteria embeddcd in the models. Onee a model has been invoked, a verilication procedure establishes conlidence measures for a correct recognition. The three dimen· sional nature of the sensed data makca the matehing process more robust u does the system'a ability to sense visually occluded areu with touch. The model is hierarchic in nature and allowa matehing at different Ievels to provide support or inhibition for recognition.