Deep learning optimization positioning algorithm based on UWB/IMU fusion in complex indoor environments

Abstract

With the rapid development of high-precision positioning systems, the demand for accurate positioning in complex indoor environments is growing. In complex indoor environments, single indoor positioning technologies such as ultra-wideband(UWB), light detection and ranging(LiDAR), wireless network technology(Wi-Fi), and Bluetooth(BLE) are easily affected by environmental factors such as indoor multipath effects and non-line-of-sight(NLOS), resulting in reduced positioning accuracy. In order to address these limitations, the fusion of two or more ranging sensors is usually used to overcome the limitations of a single positioning method, but multi-source data often introduces nonlinear errors and dynamic drifts during the integration process, which restricts the further improvement of its positioning performance. In this study, we proposed an optimization algorithm(CNN-LSTM-DEKF) that integrates convolutional neural networks and long short-term memory networks (CNN-LSTM) and embeds a distributed extended Kalman filter(DEKF) to improve the positioning performance of UWB/IMU fusion systems in complex indoor environments. The algorithm makes full use of CNN to extract spatial features, LSTM to model time series dependencies, and combines DEKF to achieve dynamic suppression of sensor noise and state estimation optimization. Experimental results show that the root mean square error (RMSE) and mean absolute error(MAE) of the proposed algorithm in a typical office environment are reduced to 0.205 m and 0.192 m respectively, and it exhibits better stability and robustness in non-line-of-sight scenarios, verifying its feasibility and superiority in practical applications.