Automated recognition of human gait pattern using manifold learning algorithm

In this paper, we investigated the application of the manifold learning algorithm in gait data analysis for the improvement of the gait classification performance. A manifold learning algorithm such as isometric feature mapping algorithm (ISOMAP) was firstly employed to perform nonlinear feature extraction for initiating the training set, and its effect on a subsequent classification was then tested in combination with learning algorithms such as support vector machines. The gait data including young and elderly participants were analyzed, and the experimental results demonstrated that the generalization performance of ISOMAP-SVM is an evidently improved performance compared to the traditional classifier for recognizing young-elderly gait patterns. Our work suggested that manifold learning algorithm can find the intrinsic low-dimensional manifold embedding in high-dimensional gait data, and obtain the `true' nonlinear gait features associated with human gait function change for improving the gait classification performance. The proposed technique has considerable potential for future clinical applications.