Comparison of GNSS LOS/NLOS Labeling Techniques Based on a 3D model and Sky View Imagery for Soft Mobility and Vehicle Data in Urban Areas

The distinction between Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) signals is crucial for improving the positioning accuracy of the Global Navigation Satellite System (GNSS), especially in urban environments where obstacles affect signal quality. Traditional classification methods rely on 3D city models and ephemeris data, but are unable to account for vegetation, street furniture and pedestrian obstacles. Recently, Sky View Image (SVI)-based techniques using fisheye cameras and deep learning have emerged as an alternative for a more accurate representation of the environment.This study compares the performance of 3D model-based and SVI-based LOS/NLOS classification methods and investigates the impact of different mobility types — vehicles, e-scooters and pedestrians — on GNSS signal quality. We collected GNSS and fisheye image data in three urban environments and applied two classification techniques: a 3D city model-based approach that uses building footprints and heights, and an SVI-based method that uses sky masks generated using deep learning. The comparison shows that both methods perform similarly well in vehicle scenarios, while discrepancies occur in soft-mobility contexts due to missing elements in the 3D model. Pedestrians and e-scooter users consistently have a lower view of the sky and more NLOS conditions compared to vehicles.The results suggest that SVI-based methods provide a more detailed representation of the environment, which is particularly beneficial for soft mobility users. Conventional GNSS signal quality indicators show weaker linear correlations with LOS classifications in pedestrian and e-scooter scenarios, highlighting the need for mobility-specific morphological indicators.