High-Precision Vehicle Positioning in Urban Canyons Utilizing Photorealistic 3-D Maps and Factor Graph Optimization
TL;DR
A novel factor graph optimization (FGO) framework that pioneers the use of accessible, textured photorealistic 3-D maps to enhance vehicle positioning, and proposes a registration-based centroid clustering matching matching (RCCM) method to accurately associate landmark features derived from 3-D point clouds with those in the photorealistic 3-D maps.
Abstract
High-precision vehicle positioning is a fundamental prerequisite for road-infrastructure maintenance. However, traditional global navigation satellite system and inertial navigation system (GNSS/INS) fusion solutions suffer from severe accuracy degradation in urban canyons due to signal outages and multipath effects. While integrating prior maps is a common strategy, existing methods predominantly rely on vector-based high-definition (HD) maps, which are highly complex and costly to acquire. To overcome the limitation, this article proposes a novel factor graph optimization (FGO) framework that pioneers the use of accessible, textured photorealistic 3-D maps to enhance vehicle positioning. First, to overcome the computational complexity associated with raw photorealistic 3-D maps, we develop a comprehensive processing pipeline, thereby facilitating efficient landmark feature extraction and subsequent matching. Second, we propose a registration-based centroid clustering matching (RCCM) method to accurately associate landmark features derived from 3-D point clouds with those in the photorealistic 3-D maps. Furthermore, novel matching and confidence factors are designed and integrated into the map-aided FGO framework, termed global navigation satellite system (GNSS)/inertial navigation system (INS)/DLMC FGO. Extensive real-world experiments were conducted in the highly challenging, dense urban canyons of Hong Kong to rigorously evaluate the proposed system. Quantitative results demonstrate that our method achieves a mean horizontal error of 0.233 m and a mean vertical error of 0.141 m, yielding a 56.4 % reduction in horizontal root-mean-square error (RMSE) and a 57.6 % decrease in maximum trajectory drift compared to conventional GNSS/INS FGO. These findings validate the system’s exceptional robustness and accuracy, offering a highly reliable and cost-effective solution for smart city infrastructure management.