| This work deals with aiding and developing upper body rehabilitation engineering methods for stroke victims. In most rehabilitation cases, only partial success is accomplished after long training sessions, and the patient is left with a limited range of motion. The aim of this research is to develop a training tool utilizing an augmented reality device (ARWED) that addresses the rehabilitation of human hand and forearm motion. For this purpose, the research is accomplished in three steps. The first step focuses on the development of real – time sEMG classification methods where ten classification methods have been chosen based on their different characteristics. The second step explains the development of real – time Artificial Neural Network (ANN) models based on the signal classifications. This part consists of training the ANN model based on real – time classification. The third step discusses the development of identification algorithms for identifying motion intend using real – time ANN models by using Simulink™ block model.The results indicate preferences for the preferred classifier: Slope Sign Change (SSC), Waveform Length (WL), Zero Crossing (ZC), and Wilson Amplitude (WAMP) which can help in the design of the signal processing for real – time ANN implementation.
Key Words: Rehabilitation, surface Electromyography (sEMG) classification, real – time Artificial Neural Network (ANN), Backpropagation, Levenberg – Marquardt(LM), Arduino Mega 2560, Simulink™, MATLAB®, Motion identification, Slope Sign Change (SSC), Waveform Length (WL), Zero Crossing (ZC), and Wilson Amplitude (WAMP), Neural Network training. |