Multi-Camera Collision Detection allowing for Object Occlusions

Abstract

A multi-camera-based collision detection system is presented. We describe the computation of global collision information for the entire surveilled workspace based on local collision information extracted from camera images. If there are known occlusions (e.g., by the robot), the system is able to recover object collision information by fusing multiple camera images. The algorithm presented is part of the safety system of a prototype realizing human-robot-cooperation.