Object identification for robotic applications using expert systems
Abstract
The objective of this project is to develop an intelligent machine vision system for robotic applications to identify engineering tools and components; The imaging system consists of a 2-D digital camera and an ultrasonic range sensor attached to the robot end effector. Images of the target objects are captured by the camera. The images are then processed to remove the signal noise and to extract the object boundary; One major objective is to develop object feature descriptions which are invariant to scaling, translation or orientation. Efficient data reduction to an array of fewer than 25 numbers is achieved by the use of Fourier and regional descriptors. One of the array elements, object thickness, is determined directly from ultrasound range measurement; An expert system was successfully developed to classify the objects based on their descriptors. The knowledge base consists of rules for searching and pattern matching. The sensors were integrated to form a working vision system for the PUMA 500 robot. The performance of the vision system was tested with a set of objects. The expert system was found to be efficient, successful, and reliable in identifying all tested objects even with signal noise being present.