Automated classification of gait patterns based on time-frequency analysis

A general scheme for the automated classification of gait patterns based on time-frequency analysis was proposed to discriminate acceleration signals characterized by high dimension, non-linearity, strong coupling and high time-varying acquired under different terrains and motion patterns of lower limbs. A three-axis acceleration sensor was mounted on a crus to acquire acceleration signals in the sagittal, coronal and cross-sectional planes separately. By using a 5-order Daubechies wavelet base, the features were extracted from time-series acceleration signals and further dimensionally reduced by employing linear discrimination analysis (LDA). The reduced features were classified by the decision tree and the support vector machine (SVM). From experimental results, both classifiers can achieve the high classification accuracy ratio over 90% and for the specified gait the ratio can be up to 100%, indicating the rationality and effectiveness of the proposed methods for feature extraction and dimension reduction.