Implementation aspects of a real time workspace monitoring system

To provide advanced robotic systems with the capability to share their workspace with humans, they must be equipped with an adequate security system. We propose a vision based security system that monitors an invisible but clearly defined, dynamically adjustable safety zone around the robot for intruding objects. A texture based stereo vision method is used, not requiring a time consuming correspondence search. Objects are detected when they penetrate the separation surface between the two workspaces. To detect objects in the separation surface, one camera image is transformed with respect to the separation surface, as if it would have been taken by the other camera, then both images are compared. Corresponding regions belong to objects on the separation surface. We aim at a hardware implementation to achieve real time video processing. We present results of investigations for the implementation of the correlation and image transformation in hardware and propose a suitable hardware architecture.<
>