Abstract: Aiming at the problems of high computational cost, high equipment cost, and low robustness of existing indoor personnel gait recognition methods, this paper proposes a highly robust indoor personnel gait recognition method based on Channel State Information (CSI), called WiKown. This method uses the fast Fourier transform to set an energy indicator to monitor the occurrence of walking behavior. The collected CSI gait data are filtered and de-noised. Then, the characteristic values are extracted in the form of the sliding window and the observation sequence is established. Finally, Gaussian Mixture Model (GMM) is used to superimpose and fit the sequence, and then Hidden Markov Model (HMM) is used to calculate the probability of observation sequence to generate the gait parameter model. The method can effectively identify the gait information. In the real multi-person environment of corridors, laboratories, and halls, the average recognition rate of the WiKown method for a single persons gait can reach 92.71%. The experimental results show that, compared with the decision tree, dynamic time structuring, and long-duration memory network methods, this method can effectively identify gait information and improve the recognition accuracy and robustness.

Key words: gait recognition, channel state information, fast Fourier transform, Gaussian mixture model, hidden Markov model