Abstract: | Shadow matching is a core component of urban localization with the Global Navigation Satellite System (GNSS). Shadow matching provides a general formulation to match line-of-sight (LOS) detections from GNSS signal features with LOS predictions from a 3D city map. Past work has focused on improving LOS detection from signal features and integrating shadow matching with other urban localization techniques. Although there has been work comparing map quality, typical map representations can only handle buildings in the environment. However, Neural City Maps can better capture the user’s environment, including buildings, vegetation, complex facades, and temporary structures that impact GNSS signals. Therefore, we extend our prior work on NLOS prediction with Neural City Maps to handle the more difficult problem of prediction GNSS shadows. We use a new ray accumulation feature, scene cropping, and logistic mapping to provide GNSS shadow predictions that can be aggregated with shadow matching. We further develop a method that exploits GPU parallelization for high-resolution GNSS shadow predictions with a Neural City Map, enabling localization predictions with many candidate user locations. We demonstrate that all four features provide broadly consistent GNSS shadows. However, the features differ in the clarity of the predictions and sensitivity to regions that are not well-observed at training time. |
Published in: |
Proceedings of the 37th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2024) September 16 - 20, 2024 Hilton Baltimore Inner Harbor Baltimore, Maryland |
Pages: | 2067 - 2079 |
Cite this article: | Neamati, Daniel, Partha, Mira, Gupta, Shubh, Gao, Grace, "Neural City Maps for GNSS Shadow Matching," Proceedings of the 37th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2024), Baltimore, Maryland, September 2024, pp. 2067-2079. https://doi.org/10.33012/2024.19736 |
Full Paper: |
ION Members/Non-Members: 1 Download Credit
Sign In |