Lightweight relative map-matching with slope-aware multi-level ground constraints for accurate and robust urban localization

Accurate and robust localization is essential for high-precision mobile mapping and autonomous navigation. Despite advances in LiDAR-, visual-, and inertial-based odometry, achieving long-term, drift-free localization in complex urban environments remains challenging. Map-based localization offers global consistency without GNSS reliance by registering online LiDAR scans to pre-built maps. However, absolute scan-to-map matching remains computationally expensive and highly sensitive to initialization and map discrepancies. These challenges are amplified with low-resolution LiDAR sensors from mobile mapping systems on sloped or multi-layered roads, where sparse vertical geometry exacerbates vertical drift and reduces localization reliability. To address these limitations, this work offers a new perspective on map-based localization by introducing a computationally efficient multi-layer localization framework that harmonizes local odometry with innovative map-derived constraints. We proposed a novel map-guided differential motion factor that uses a pre-built queried map as a relative-motion regularizer rather than an absolute pose anchor and injects a map-derived inter-keyframe motion constraint as a binary between-factor into the factor graph with an alignment-reliability adaptive covariance to down-weight partial-overlap and degenerate matches. This design substantially reduces computational cost while maintaining comparable accuracy and robustness of absolute map-aided localization approaches. However, reliance solely on our relative map constraints cannot entirely suppress vertical drift, particularly on sloped or multi-layered terrain. To overcome this, we introduce a slope-aware, multi-layer adaptive ground segmentation module, followed by a corresponding ground-constraint factor to strengthen vertical observability across frames. The fusion of these modules achieves decimeter-level, drift-resilient localization with 0.85 m RMSE, while reducing runtime by over 50% compared to absolute scan-to-map localization, providing an effective accuracy–efficiency trade-off for localization performance in complex urban environments.