Gesture recognition based on parallel hardware neural network implemented with stochastic logics

Abstract

A new method based on neural network using stochastic computing is presented for the recognition of human gesture. In the current gesture recognition study, most of the technologies require high hardware resources and power consumption. Considering gesture recognition algorithms, the power limitations of their complex systems have encouraged designers toward searching for a reconfigurable architecture, stochastic computing. For different neural networks with complex arithmetic operations, computation on stochastic bit streams costs fewer resources and performs very efficient in operation. The experimental results demonstrate that the stochastic neural network could recognize different hand gesture effectively and take less hardware area. Even more, it has good robustness to the different environments.