Performance enhancement using sensor data fusion for an indoor localization system based on ultra-wideband
Indoor localization utilizing wireless communication is rising in significance due to the potential wide range of services that it can provide. This thesis presents an approach for performance enhancement of an indoor localization system based on ultra-wideband (UWB) using sensor data fusion. Data fusion is performed by an algorithm based on Kalman filter in a loosely-coupled manner, i.e., the position estimates computed firstly by the UWB-subsystem are later fused with the motion acceleration delivered by the inertial sensor unit (IMU). Thus, this approach is independent from the positioning method of the UWB-subsystem. The system is implemented in real-time on the hardware platform, which stems from previous projects at Fraunhofer Institute for Embedded Systems and Communication Technologies ESK (Fraunhofer ESK), including one target node (tag) and four reference nodes (anchors). We applied the multilateration approach utilizing Time Difference of Arrival (TDOA) for the UWB-based localization and deployed the inertial sensor unit MPU9250 with integrated digital motion processor (DMP) to process the motion information. Both the multilateration approach and the fusion algorithm are performed on a windows PC, while the inertial data processing takes place on the target node. Since no extra information from the target is required, the enhanced localization system proposed in this thesis is self-contained and thus, generally applicable for any target (e.g., robots, pedestrian and vehicles) in indoor environments. The positioning accuracy in real-time is experimentally evaluated. In line-of-sight conditions, the accuracy, assessed by root-mean-square errors (RMSE), is on average 0.11 m with and 0.29 m without data fusion for a robot moving at 0.5 m/s along a rectangular of 1 m x 3.8 m. In tests with different trajectories and robot speeds, the results verify an accuracy improvement of 49% - 62% under LOS conditions. In the case of non-line-of-sight (NLOS) conditions, the obtained results show that the localization error (RMSE) decreases from 0.81 m to 0.49 m, which yields a performance improvement of 39.5% due to sensor data fusion. This significant performance enhancement provides convincing potential of the UWB-based localization in various emerging markets, such as autonomous robots and location-based services.
München, TU, Master Thesis, 2018