A Novel Indoor Tracking Method Based on IMM-MLKF With CSI Measurements

Abstract

Indoor target tracking based on Wi-Fi signals entails the integration of both the target motion model and the measurement model. However, a single motion model is inadequate to match the changing motion state of the target at each moment, and linearization of the measurement model may introduce additional errors. To tackle these problems, we propose a localization method called IMM-MLKF. For the measurement equation, we adopt the fingerprinting model to perform maximum likelihood estimation on the target, obtaining the mean and covariance matrix of the target state distribution. This method eliminates the error introduced by linearization. For the motion equation, multiple models are used for parallel computation. The likelihood function values generated by each model from the measurements are used as their model confidences, enabling the combining model to match the motion of the target. We validated the method in two different scenarios using channel state information as the fingerprint feature. Our results show that compared with similar Bayesian filtering methods based on interactive multiple models, this method has superior tracking accuracy. Additionally, this method's operating efficiency is higher than particle filtering methods.