A Study of Obstacle Recognition Method for Mobile Robot for Rehabilitation Assistance Using Image and Distance Information

초록

EMG signal is generally used for control source of a prothetic arm which can substitute for damaged and missing parts of an arm. EMG signal recognition has been performed by using Artificial Neural Network. But this method neglect dynamic aspects of EMG signal and show slots recognition rate. This paper introduce EMG signal classifier compensates these problems. Materials md Methods We suggest classifier consist of Hidden-Markov Model (HMM) algorithm and Multi-La yer Perceptron (MLP). In addition, Genetic Algorithm(GA) was also used as a pre-processor of the MLP. HMM enable classifier to consider dynamic properties of the EMG signals and GA make it more faster. We considered EMG signals extracted from a subject aged 20 as a piece-wisely connected 6 primitive motion signal. The feature extracted from these ENG signals. such as LPC, IAV, ZCN. are classified using sugg ested HMM-MLP hybrid classifier. Results Three experiments were performed. When only HMM was applied, it showed 77.3% of recognition ra te. A combination of HMM and MLP was utilized for the same task. This experiment, Back-propagation was used, presented 87.2% of recognition rate. At the third experiment, classifier composed of HMM-GA-MLP shooed 90% of recognition rate and took half time than that of previous one. Conclusion Experimental results showed that this approach could apply to recognize the EMG signal as a new efficient classifier. The suggested HMM and 3A-MLP hybrid classifier has given a higher classifica tion accuracy by about 13% Moreover. it is about 2 times faster than other one.