A Systematic Approach to Error Sources for the Evaluation and Validation of a Binocular Vision System for Robot Control

Abstract

The automatic control of production lines may be done in appropriate cases by robots controlled by digital image processing. CCD video cameras are used as optical sensors. When a two-dimensional image of a three-dimensional scene is generated the depth information is lost, however applications with a high degree of automation commonly require an exact recognition of three-dimensional objects and their positions. Therefore, binocular video cameras are used. For the application of image processing algorithms and subsequent robot control, a comprehensive calibrating procedure is required for the cameras and for the robot basis. An evaluation procedure of the corresponding software estimates only the reliability of the whole system, ignoring single effects of errors. In the paper presented, model dependent error sources are recorded systematically to find appropriate procedures for independent verification. It is presumed that no logical errors or errors due to computer internal process communications are present. This approach may be applied to similar complex model dependent software systems.