Precise and robust large-shape reproduction using uncalibrated vision

The paper presents an alternative to CNC machining, based on a proposed method for creating very large and precise replicas of prototype shapes using material-removal and/or material-adding mechanisms that need not be calibrated. The method is based upon the creation of a ldquovirtual moldrdquo, which is created in the two-dimensional image planes of two or more stationary, uncalibrated cameras. The cameras are placed so as to view portions of the prototype surface and remain stationary while multiple-beam laser pointers cast their light toward the prototype shape and the resulting laser-spot reflections are registered and matched among the cameras. Large numbers of images can be acquired with slightly offset pan and tilt angles of laser-pointer-bearing units, resulting in collected laser-spot-centroid indications that may number in the thousands. The spot-centroid coordinates in camera space are matched in such a way that individual, physical spot-centroid indications are registered in memory in accordance with correspondence among those cameras able to view any one particular spot. This forms the basis of the ldquovirtual moldrdquo. Subsequent reproduction of the shape, as with a CNC machine, relies upon action of a robot to remove material from a blank. This blank is placed so as to occupy the region where the prototype had been during laser spot casting. Camera-space manipulation is used to guide the robot in consecutive passes of material removal, with laser spots cast upon each newly revealed intermediate surface and reflected up into the original cameras. The paper presents the use of a FANUC M16-iB industrial robot, in order to obtain experimental evidence on how the original surface regions can be replicated with extremely high precision despite no need for calibration either of the cameras or the manipulator.