A Semantic-Associated Factor Graph Model for LiDAR-Assisted Indoor Multipath Localization

In indoor environments where Global Navigation Satellite System (GNSS) signals are entirely blocked, wireless signals such as 5G and Ultra-Wideband (UWB) have become primary means for high-precision positioning. However, complex indoor structures lead to significant multipath effects, which severely constrain the improvement of positioning accuracy. Existing indoor positioning methods rarely link environmental semantic information (e.g., wall, column) to multipath error estimation, leading to inaccurate multipath correction-especially in complex scenes with multiple reflective objects. To address this issue, this paper proposes a LiDAR-assisted multipath estimation and positioning method. This method constructs a tightly coupled perception-positioning framework: first, a semantic-feature-based neural network for reflective surface detection is designed to accurately extract the geometric parameters of potential reflectors from LiDAR point clouds; subsequently, a unified factor graph model is established to multidimensionally associate and jointly infer terminal states, virtual anchor (VA) states, wireless signal measurements, and LiDAR-perceived reflector information, enabling dynamic discrimination and utilization of both line-of-sight (LOS) and non-line-of-sight (NLOS) paths. Experimental results demonstrate that the root mean square error (RMSE) of the proposed method is improved by 32.1% compared to traditional multipath compensation approaches. This research provides an effective solution for high-precision and robust positioning in complex indoor environments.