Recognition of large work objects in difficult industrial environments

This paper presents a method for recognizing the very large work objects in difficult industrial environments. An automatic industrial robot workcell where the single production is manufactured requires fast and flexible recognition of work objects. The purpose of the workcell is to manufacture moulds and billets for prototype and short series castings so the requirements for flexibility and reliability are high. To solve this problem, we have examined a stereo-vision-based object localization system which has several features suitable for actual robotic workcells. In this paper, we also propose an effective subpixel edge estimation method for improving the accuracy of 3-D measurement by the stereo vision. Then, we provide experimental results on localizing a large work object to demonstrate the performance of the system in a difficult industrial environment.