On-Machine Visual Modeling System With Object Recognition
Xiaodong Tian, Hanyu Deng, Kazuo Yamazaki, Makoto Fujishima, M. Mori · 2005
This paper presents an on-machine modeling system that tries to bridge the gap between the design and the machining. This system is able to build a comprehensive solid model of the CNC machining workspace after the workpiece and fixtures have been installed onto the working table. This solid model can be used for simulation to enhance its credibility. For this purpose, one prototype of a 3D visual modeling system is proposed and designed. In order to accurately calibrate CCD cameras upon the absolute coordinate frame of the machining center, a practical calibration method is presented at first. To segment the target part and extract its 2D features on the captured images, the techniques of Image Decomposition and a modified Standard Hough Transform (SHT) are designed. Using these 2D features, the 3D visual stereovision system, powered by a designed feature matching engine, is capable of obtaining the 3D features of the target part. Furthermore, the part has been identified by the object recognition technology. This recognition includes part recognition and pose recognition. In the part recognition, the part is recognized and an initial pose transform is obtained. Using this initial pose transform, the pose optimization method, named as Dual Iterative Closest Lines (DICL), is designed to locate the optimum position and orientation of the solid model of the recognized part. Finally, this modeling system is tested on a machining center. The experimental result indicates the innovation and feasibility of the proposed modeling system.