The Performance Analysis of an AKF-based INS/GNSS Integration Scheme with Vehicular and Vertical Constraints in Urban Environment

• Main Authors: You-LiangChen, 陳侑良
• Other Authors: Kai-Wei Chiang
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/23v5f8

碩士 === 國立成功大學 === 測量及空間資訊學系 === 106 === Global Navigation Satellite Systems (GNSS) integrated with Inertial Navigation System (INS) are widely applied to improve the reliability for navigation. Global Navigation Satellite Systems (GNSS) integrated with Inertial Navigation System (INS) has the complementary characteristics to overcome the drawbacks for each sensor so that the integrated system provides superior performance. The advantage of GNSS is the higher positioning accuracy, but it decreases easily with the worse light-of-sight visibility to the satellites. On the other hand, the benefit of INS is self-contained and independency of external signal. Nevertheless, the accuracy of INS degrades rapidly because of the nonlinear error and noises from inertial sensors including accelerometers and gyros. GNSS is usually used to update the estimates from INS as well as minimize the drifts of inertial measurements over time. Most importantly, the INS bridges the gap of losing GNSS signals in harsh environments such as tunnel, urban area and indoor parking. It is common to use Extended Kalman Filter (EKF) to fuse the heterogeneous data, and loosely-coupled (LC) integration is a simpler GNSS/INS architecture which has two EKF algorithms. However, the output of the first EKF in GNSS will stop functioning when the number of the satellites in view is less than four. Then, the errors in the position and velocity solutions provided by the first EKF in GNSS are time-correlated, which might cause the instability of the second EKF for navigation. The positioning error of INS/GNSS integration is also influenced by velocity error and attitude error. The motion of the land vehicle will not jump of the ground or slide on the ground under normal circumstances. Thus, the specific vehicular motions become the constraints for the land vehicle navigation. In this research, zero velocity update (ZUPT), zero integrated heading rate (ZIHR) and non-holonomic constraint (NHC) are evaluated for land vehicle application. Furthermore, it is well known that the accuracy of height from INS/GNSS is weaker then horizontal positions; therefore, barometer which estimates the height above the sea level based on the measurement of atmospheric pressure is commonly used for improving the height accuracy. By using barometer, the height constraint is added to improve the accuracy of INS/GNSS integration. Besides, a LC INS/GNSS integration scheme using Adaptive Kalman Filter (AKF) as the core estimator are implemented in this research. Due to the priori uncertainty from the measurement or the dynamic model, AKF has the capability to reduce the fault caused by the suboptimal of EKF. The significant task for AKF is the tuning algorithm of the measurement covariance matrix (R) or the dynamic model covariance matrix (Q) adaptively. In this study, the innovation-based and residual-based adaptive estimations of the measurement matrix are used for the improvement of Kalman filter. In order to validate the performance of LC INS/GNSS integration scheme with AKF and EKF, the experimental scenarios are conducted in downtown area where multipath signal is severe or the satellite geometry is bad. The test and reference platform, low-tactical grade INS and high-tactical grade INS, together with the geodetic GNSS antenna/receiver and barometer were mounted on the top of a land vehicle. Analysing the performance of AKF with the adaptive measurement covariance matrix is focused on this research. In addition, not only the vertical constraint from barometer but also the velocity constraints from vehicle are added into AKF. The proposed integration scheme can provide more stable solutions with the vertical and velocity constraints. The results display around 55% / 35% improvements of maximum errors for three dimensional filtered/smoothed positioning errors in the average cases.