RSS indoor localization and tracking with INS assistance

• Main Author: Wen, Kai
• Other Authors: Tan Soon Yim
• Format: Theses and Dissertations
• Language: English
• Published: 2019
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/103670; http://hdl.handle.net/10220/49981
• Institution: Nanyang Technological University

In recent years, various indoor localization and tracking systems have received a significant amount of attention, due to the increasing demand for location-aware applications in indoor environments. Wireless localization technologies, such as received signal strength-based (RSS-based) localization systems, have been widely used due to their low implementation cost. However, such systems suffer from poor localization accuracy due to multipath and non-line-of-sight (NLOS) propagation. Unlike wireless localization systems, inertial navigation system (INS) does not suffer from multipath and NLOS propagation issues. However, INS accuracy degrades as time elapses, because of accumulated noise in accelerometers and gyroscopes. Furthermore, the inaccurate estimation of inertial measurement unit (IMU) orientation leads to the incorrect projection of gravitational acceleration in accelerometers, which causes large localization errors. Several fusion systems, such as INS with RSS, INS with mapping, etc., have been explored as a way of mitigating these issues. Existing INS-based techniques use an INS zero velocity update (ZUPT) with extended Kalman filter (EKF) to mitigate orientation and localization error. ZUPT requires users to physically attach the IMU to one foot to leverage the zero velocity measurement and continuously calibrate the IMU. To further improve localization accuracy, particle filtering (PF) is used to model the nonlinear system, and maps are used to confine particle movement. The placement of the IMU to utilize ZUPT, the computation complexity of PF, and the uncertain availability of accurate maps all make the fusion system challenging for implementation. This thesis presents a novel weighted RSS indoor localization and tracking scheme that uses INS projection. The proposed scheme does not require the IMU to be physically attached to the user’s foot, and mapping information is not required. The scheme also reduces computation time versus PF. Simulation results of the proposed scheme show that it significantly outperforms both conventional INS and an RSS fusion scheme using EKF. This thesis derives a closed form equation to analyze the noise of acceleration along the navigation frame, which leads to errors in stand-alone INS positioning estimations. The results are used to calculate appropriate weight for the proposed scheme in later chapters. For the case when there is perfect knowledge of IMU orientation with respect to the navigation frame coordinates, this thesis proposes a novel weighted nonlinear RSS localization and tracking scheme that integrates INS positioning estimation projections. Due to the cumulative effect of noise on the gyroscope, INS-based localization performance deteriorates rapidly even when the initial orientation is known. Thus, an algorithm is proposed to continuously mitigate the effect of noise on gyroscopes. Simulation results obtained from three different scenarios show that when fused with INS projection, the proposed RSS localization scheme outperforms conventional fusion schemes that use EKF by up to 56%. To develop a more practical scheme that does not require perfect knowledge of IMU orientation with respect to the navigation frame coordinates, an algorithm is proposed to estimate the initial orientation during the calibration stage. To further mitigate noise from the IMU, another algorithm is proposed to continuously mitigate the effect of noise on accelerometers. Once all of these proposed noise mitigation algorithms and initial calibrations are incorporated, the simulation results obtained from three different scenarios show that the proposed RSS localization scheme fused with INS projection significantly outperforms conventional fusion schemes that use EKF by up to 65% (82%) for 2D (3D) environments. The proposed scheme was tested on different routes at a real environment, and the results show that its localization is up to 53% (62%) more accurate than the other two localization schemes in 2D (3D) environments.