Copycat hand - robot hand control with hand posture estimation with high-speed processing at high accuracy

摘要

Three conditions at least should be satisfied when you try to control a dexterous robot hand according to the result of human hand posture estimation: the estimation accuracy should be high enough to generate the same motions as those of the human hand and fingers; the processing speed should be short enough to be controlled in real time processing; and each processing time should be almost the same for stable robot control. Various hand posture estimation systems have been proposed. Some of them seem to accomplish a certain level in accuracy, but the sufficient processing speed has not been achieved. In this paper, the authors propose a hand posture estimation system with high-speed and at high accuracy with uniform processing time, by adopting Self-Organizing Map with the algorism of self-multiplication and self annihilation to the database to accomplish multi-step search. The number of data in each classes in each steps is almost the same. And the history was used and the target classes and data were limited to decrease the search space. A color high-speed camera was used in our system to be robust against light conditions. Our system accomplished the human hand posture estimation with 160 fps.