GIS and landmark-aided autonomous vehicle localization for navigation continuity in urban corridors

Reliable localization is essential for autonomous vehicle navigation in urban corridors where Global Navigation Satellite System (GNSS) positioning can become unavailable or unreliable. This study presents a Geographic Information System (GIS) and landmark-aided route-constrained localization framework. The framework tracks vehicle progress along a known route using the incremental distance traveled between successive time steps. Sparse landmarks are used to correct accumulated errors in route progress. Two methods are evaluated for estimating incremental distance: a Long Short-Term Memory (LSTM) model and a Kalman Filter (KF). The KF achieved an incremental-distance mean absolute error of 0.0148 m, compared with 0.0213 m for the LSTM model, while their complete map-referenced localization errors were comparable at 3.45 m and 3.41 m, respectively. Component ablation demonstrated that mean error reduced from 32.63 m for unconstrained dead reckoning to 22.06 m with the route constraint, 6.05 m with landmark correction, and 3.45 m with both components. Landmark-event perturbations showed that correction timing and association integrity strongly affect localization performance. Over the complete experimental route, the proposed framework maintained meter-level localization accuracy without using GNSS position updates. These results demonstrate the potential of lightweight route constraints and sparse landmark corrections to provide practical navigation continuity for autonomous vehicles when GNSS positioning is unavailable.