Walking Pattern Recognition Based On Surface Electromyography

With the continuous improvement of living standard and the rapid development of science and technology, prosthesis, rehabilitation training system and exoskeleton for the disabled, sickness and old are being further studied. In order to ensure the safety, stability, comfortableness and intelligence of the external mechanical device, the research of walking pattern recognition is particularly critical. s EMG signals can reflect motion intentions better than motion information, which is a remarkable advantage. The issue extracted time domain feature of the lower limb s EMG at the beginning of a gait, proposed the Support Vector Machine-K Nearest Neighbor(SVM-KNN) algorithm based on threshold segmentation and recognized walking pattern. This paper recognized five walking patterns, including level-ground walking, upstairs, downstairs, upslope and downslope. The research contents are as follows:Firstly, select lower limb muscle group and collect s EMG signals of the five typical walking patterns. Comparing the information of different gaits, the method of moving windows is proposed to process gluteal muscle s EMG to determine the initial moment for feature extraction. Feature extraction and feature vector construction are followed.Then, propose SVM-KNN algorithm based on threshold segmentation. The method is used to recognize feature vectors of different muscle combinations. The recognition rate is above 94%. The classification method also used to recognizes continuous road condition combinations and good results are obtained.Finally, construct a demo system for on-line recognition based on Lab VIEW. The system gets recognition result before a gait was completed. The effectiveness of SVM-KNN algorithm based on threshold segmentation is verified by the system.