Monocameral visual recognition of Marcus hand postures for personal robotic assistants

Postures recognition is the first step for gestures tracking of an artificial or a natural hand. We show visuo-motor tracking of mobile hand configurations, which is based on symbolic representations able to support biomechanical and perceptual information relative to evolving postures. After recognition, we have a virtual skeleton to identify simulated artificial hands. Such postures identification is adapted to an artificial Marcus hand developed as a human prosthesis. Nevertheless, for the complex mechatronic device, this symbolic representation allows the visual identification and tracking of hand points of interest, such as tactile sensors and finger joints. In this way, a feedback for the perception-action cycle is obtained to improve the man-machine interaction in personal robotics, with special regard to the assistance of disabled people.