Three gait patterns recognition with ground reaction force using support vector machine

This paper presents that support vector machine (SVM) is used to classify three gait patterns: level walking, stair ascent and stair descent based on ground reaction force (GRF). The recognition process consists of three stages: i) a three layers wavelet packet analysis is used for feature extraction, with which squared and standard deviation of decomposition coefficients compose features; ii) with attributes distribution analysis of features (similar to principal component analysis), the feature set is cut down to contain six ones; and iii) the optimal feature set is input to the SVM classifier. Six healthy subjects were tested wearing the designed shoes. As a result, the accuracy is around 83.3%. Improved performance of the classifier was evident in ideal situation i.e. a constant walking velocity, no tilting et al., accuracy of SVM is nearly 100%. These results suggest that SVM system can function as an efficient classifier for recognition of three gait patterns.