Three-Dimensional Sensing and Interpretation
Gerald J. Agin, Martin J. Uram, Peter T. Highnam · 1985
TIhe m1ain goal or this project is to produce a representational schemna that can beC used to compare shape dc riptions or solid three-dimensional objects.Wc arc working toward a demonstraion to show recognition and determination of position and orientation of solid objects on a thIbCtop, using a sensor consisting of a light-stripe projector and camera mounted in the hand of a robot manipulator.This report presents the results or a number or activities dircctcd toward that goal: calibration and use or the range finder, a tactile sensor for exploring a three-dimecnsional domain, a system for identifying pipes in a bin, a method for * assessing the similarity of cross-section curves, and pose cluster matching.a ncw technique for shape matching.* choosing local features that are invariant to viewing angle, and o incorporating as many local features as possible in our prototype models, without regard to redundancy or overlap of multiple feature descriptions.Matching can proceed on the basis of whatever subset of features correspond between image and prototype, without regard to "completeness" of any description or set of features.V., * system, we can bring the speed up, but whenever we try out new motions of the arm, it would be prudent to keep the motions as slow as possible.Another factor is the verbose communication format dictated by VAL's interaction.In order to K specify a new position and orientation to the PUMA, about 200 characters of communication are required over the serial line.between our LSI-11 and the PUMA controller.We are examining the interaction and modifying our procedures to reduce the communication requirements.The long-term solution to this aspect is to upgrade our PUMA to use VAL-Il.