Multi-sensor fusion scheme for calibration-free stereo vision in a manufacturing workcell

This paper discusses the subject of detecting the position and orientation of a moving part, which is placed randomly on a rotating disc conveyor in a manufacturing workcell. Two major difficulties of applying machine vision based technology are the requirement of precise system calibration and processing speed. In this paper, a new multisensor fusion scheme is introduced to overcome these difficulties. Firstly, a new algorithm is developed for a calibration-free stereo vision system to determine the relative position and orientation of the part with respect to the disc conveyor. Secondly, a novel multisensor fusion approach is introduced, which integrates information from the visual sensors and encoder measurements on the rotating disk so that the task space position and orientation of the part can be determined. Finally, a "parallel-guided" tracking scheme is implemented for planning and control of a robotic manipulator in order to track and grasp the moving part. The experimental results clearly demonstrate the advantages of the proposed scheme.