KNOWLEDGE-BASED VISUAL PART IDENTIFICATION AND LOCATION IN A ROBOT WORKCELL

This paper presents a method for identifying and precisely locating parts visible in the workspace of a robot workcell through analysis of single or multiple perspective images. Automatic integration of any number of arbitrary viewpoints permits detection of objects occluded from certain views, as well as providing varying image resolution and wider coverage of the visible workspace. Analysis is possible in the presence of image noise such as glare or shadow from poor lighting conditions, or structural errors such as missing or obscured object features. Automated visual detection of camera locations avoids precise camera positioning or setup time and permits use of a movable or roving camera. The system is capable of processing multiple camera input in a few seconds on a SUN 3/160 workstation, without the use of additional image processing hardware and can be easily ported to various computer systems.