Multi-Epoch 3D-Mapping-Aided Positioning using Bayesian Filtering Techniques

Qiming Zhong, Paul D. Groves

Abstract: In dense urban areas, conventional GNSS does not perform satisfactorily, sometimes resulting in errors of tens of metres. This is due to the blocking, reflection and diffraction of GNSS satellite signals by obstructions such as buildings and moving vehicles. The 3D mapping data of buildings can be used to predict which GNSS signals are line-of-sight (LOS) and which are non-line-of-sight (NLOS). These data have been shown to greatly improve GNSS performance in urban environments. Location-based services typically use single-epoch positioning, while all pedestrian and vehicle navigation applications use filtered solutions. Filtering can reduce the impact of noise-like errors on the position solution. Kalman filtering-based solutions have been adopted as one of the standard algorithms for GNSS navigation in many different products, and particle filtering has been demonstrated by several research groups. This paper mainly investigates the performance of different filtering algorithms combined with 3D-mapping-aided (3DMA) techniques. In addition to the Kalman filter and particle filter, the grid filter is also considered. In contrast to a particle filter, the hypotheses of a grid filter are uniformly distributed (forming a grid), but with different likelihoods, which better fits the physics of the problem. At the same time, this allows the current UCL’s single-epoch 3DMA GNSS positioning algorithm to be easily extended to multi-epoch situations. This paper then compares the performance of these continuous positioning algorithms in urban environments. The datasets used for testing include pedestrian and vehicle navigation data, covering two main application scenarios that often appear in cities. Pedestrian navigation data is static, and was collected in the City of London using a u-blox EVK M8T GNSS receiver. The vehicle navigation data consists of GPS and Galileo measurements, collected in Canary Wharf by a trials van with a Racelogic Labsat 3 GNSS front end. Subsequently, these data are fed into several single- and multi-epoch filtering algorithms, including single-epoch conventional GNSS, single-epoch 3DMA GNSS, conventional extended Kalman Filter (EKF), conventional particle filter (PF), 3DMA GNSS particle filter (PF), and 3DMA GNSS grid filter (GF). The results show that filtering has a greater impact on the results of mobile positioning with significant movement compared to static positioning. In vehicle tests, the conventional multi-epoch GNSS algorithms improve positioning accuracy by more than 40% compared to single-epoch GNSS, whereas in static positioning they deliver a limited improvement. 3DMA GNSS significantly improves positioning accuracy in the denser environments, but provides little benefit in more open areas. The 3DMA GNSS techniques and the filtering algorithms benefit each other. The former provides the latter with a better position solution at the measurement update step, while the latter in turn repays the former with a better initial position and a smaller search area. In vehicle tests at Canary Wharf, the 3DMA GNSS filtering reduces the overall solution error by approximately 50% and 40% compared to the single-epoch 3DMA GNSS and filtered conventional GNSS, respectively. Thus, multi-epoch 3DMA GNSS filtering should bring maximum benefit to mobile positioning in dense environments. The results from both datasets also confirm that the performance of 3DMA GNSS particle filtering and grid filtering are similar in terms of positional accuracy. In terms of efficiency, 3DMA GNSS grid filtering uses fewer particles to achieve the same coverage of the search area as particle filtering.
Published in: Proceedings of the 34th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2021)
September 20 - 24, 2021
Union Station Hotel
St. Louis, Missouri
Pages: 195 - 225
Cite this article: Updated citation: Published in NAVIGATION: Journal of the Institute of Navigation
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In