Classification Algorithm for Human Walking Gait Based on Multi-sensor Feature Fusion

The accurate gait mode classification is the prerequisite and basis for judging movement intention and making controlling strategies of lower limb exoskeleton robot. In our research, we developed a wearable multi-sensor human gait information acquisition system. A kind of the human walking gait classification algorithm was established to collect human lower limb movement information. The algorithm could classify four walking gait modes, including standing, horizontal walking, up the stairs and down the stairs, which combined time domain and frequency domain characteristics. The acceleration and the angular acceleration signal was extracted by windowing, and the time domain feature (mean, variance, rms, max, minimum) and the frequency domain feature (the peaks of the fast Fourier transform (FFT)) were extracted from a single time window. And the SVM classifier with different kernel functions was used for classification and identification. The results showed that the four different gait modes could be distinguished by the feature extraction method combining the time domain with the frequency domain. When the window size was 300 ms, the average recognition rate of the Polynomial kernel function classifier reached to 96.956%.