Session A5: Sensor-Fusion for GNSS-Challenged Navigation

Improving GNSS Performance with Fish-eye Camera Integration and Robust Kalman Filter
Arunima Das, ANavS and Technical University of Munich; and Patrick Henkel, ANavS and Technical University of Munich
Location: Beacon A

Precise Point Positioning (PPP) with GNSS/ INS tight coupling is attractive for achieving a both accurate and reliable absolute positioning. The inertial measurements are helpful to overcome short GNSS outages and enable a far higher output rate.
However, in urban environments, the GNSS measurements are often affected by severe multipath and even non-line of sight (NLOS) satellite signals with lower carrier to noise power ratios (C/N0), which introduce measurement outliers. These outliers degrade the positioning accuracy and robustness of the estimation process. The optimized estimation strongly depends on the updated measurement noise characteristics along with the assumed Gaussian noise model.
This work aims on detecting and reducing the weight of such outlier measurements on the position solution. Thus, the stochastic noise model is improved for the estimation and, therefore, the accuracy of the estimated solution also improves in GNSS-challenged environments.
To achieve this goal, one approach is the combination of fish-eye camera imagery with GNSS and Inertial Navigation System (INS) measurements to detect NLOS satellites and the dynamic weighting of the measurements in the measurement noise covariance matrix.
The second approach is the implementation of a Robust Kalman Filter (RKF) to de-weight the outlier measurements based on post-fit residuals for optimising and improving the state estimation performance.
The performance evaluation, using real-world data, demonstrates significant accuracy improvement for both approaches: For the sensor fusion solution, a 0.9 m reduction in the maximum error value along with a reduction of the Median value by 40 cm are observed. The RKF solution show a considerable improvement of 0.7 - 0.8 m for the maximum error value showing comparability with the sensor fusion approach. The evaluation, therefore, shows that both the approaches effectively improve the positioning accuracy and precision offering robust solutions in urban scenarios with presence of NLOS satellites and outlier measurements.