Gait and Posture Analysis Method Based on Genetic Algorithm and Support Vector Machines with Acceleration Data

[abstFig src='/00280003/18.jpg' width=""300"" text='The result of parameters optimization by GA' ] The support vector machine (SVM) we propose for automated gait and posture recognition is based on acceleration. Acceleration data are obtained from four accelerators attached to the human thigh and lower leg. In the experiment, volunteers take part in four gaits and postures, i.e., sitting, standing, walking and ascending stairs. Acceleration data that are preprocessed include normalization, a wavelet filter and dimension reduction. We used the SVM and a neural network to analyze the data processed. Simulation results indicate that SVM parameters C and g selected by a genetic algorithm (GA) are more effective for gait and posture analysis when compared to the parameter C and g selected by a grid search. The overall classification precision of the four gaits and postures exceeds 90.0%, and neural network simulation results indicate that the SVM using the GA is preferable for use in analysis.