Edward J. Altman
Real Time Hand Motion Detection and Recognition
Abstract:Low level visual techniques for hand motion detection and hand gesture recognition are typically performed on a series of static images and are computationally intensive. The dynamic nature of hand motion and the requirements for real time processing place further demands upon visual sensing. Simple hand gestures such as pointing, grasping, or releasing may, however; be viewed as temporal patterns of spatially coherent motion. The spatial and temporal coherency of the motion pattern is due to the systematic changes in the relationships among the individual fingers and the hand. The common motion of the fingers toward the center of the hand is one example of coherent motion. Dynamical systems which mimic the coherency of hand motion provide an alternative to traditional computer vision approaches for hand gesture recognition.
A completely new approach to hand motion detection and recognition is proposed in this project. This approach is based upon analog computation and uses dynamical systems concepts to design a novel visual sensor which can detect and recognize a fixed repertoire of dynamic hand motions. The real time performance of this sensor is guaranteed by the fact that the dynamical systems can be directly implemented as electronic circuits. The design of this hand motion sensor comes in three parts: (a) mathematical analysis, (b) simulation of the dynamics, and (c) analog implementation. Research at the beginning of the project will focus on the design, analysis, and simulation of the proposed sensor. Later work will focus on the analog implementation.