A MLP-BiGRU Model-Based Dynamic Displacement Estimation Method for Multistate Indoor Positioning

To achieve accurate multi state 3D indoor positioning, a dynamic displacement estimation (DDE) method based on Multilayer Perceptron-Bidirectional Gated Recurrent Unit (MLP-BiGRU) is proposed and integrated into the error-state extended Kalman filter (ES-EKF). MLP-BiGRU uses self-built IMU and plantar pressure dataset to extract more overlooked features for dual task classification of four gaits and six phases. The DDE method estimates the leg length and inter step displacement in different gaits through the displacement of the center of plantar pressure (COP) between each phase, and uses them as ES-EKF observations to assist zero velocity update (ZUPT) to correct positioning errors. The experimental results show that the proposed DDE method achieves a relative position accuracy of 0.51%, which decreased by 80.27% compared to the method without using DDE.