SUMMARYWhen applied to rapidly moving objects with complex trajectories, the information-rate limitation imposed by video-camera frame rates impairs the effectiveness of structured-light techniques in real-time robot servoing. To improve the performance of such systems, the use of fast infra-red proximity detectors to augment visual guidance in the final phase of target acquisition was explored. It was found that this approach was limited by the necessity of employing a different range/intensity calibration curve for the proximity detectors for every object and for every angle of approach to complex objects. Consideration of the physics of the detector process suggested that a single log-linear parametric family could describe all such calibration curves, and this was confirmed by experiment. From this result, a technique was devised for cooperative interaction between modalities, in which the vision sense provided on-the-fly determination of calibration parameters for the proximity detectors, for every approach to a target, before passing control of the system to the other modality. This technique provided a three hundred percent increase in useful manipulator velocity, and improved performance during the transition of control from one modality to the other.
EyeKeys: A Real-Time Vision Interface Based on Gaze Detection from a Low-Grade Video Camera