Research on UWB/Inertial Guided Fusion Localization Based on Convolutional Neural Network

Ultra-Wide Band (UWB) signals in indoor environments might suffer from positive deviation errors due to the influence of Non-Line-of-Sight (NLOS), which leads to inaccurate positioning results. To address this problem, we propose an Error State Kalman Filter (ESKF) fusion algorithm based on Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM), named as C-L-ESKF. Firstly, we use CNNLSTM to identify the NLOS observations, and then weighted data pre-processing is applied to reduce the influence of NLOS signals on the location result. Moreover, in order to mitigate the accumulated errors of inertial systems over time, the processed UWB and Inertial data are tightly coupled via the ESKF method, which improves the positioning accuracy further. Our experimental results on real data show that, compared with the ESKF method alone, our proposed approach yields a significant performance improvement in terms of positioning accuracy by 18.93%.