AI-ML driven map-matching algorithm for distinguishing highway and service road vehicular movements under GNSS degradation

Highway and service road differentiation is still a vital issue in vehicles localization by GNSS in both poor conditions like multipath interference, and parallel networked roads. The paper suggests an AI-based, ML-based hybrid map-matching framework, which combines probabilistic emission-transition modelling with supervised semantic road-type classification and altitude-conditioned correction. The framework solves the trajectory inference and road classification as a single optimization problem by including classification likelihood in probabilistic decoding. Spatial-temporal mobility attributes, road structure attributes, and 3D altitude data are used in combination to provide strength in complex urban settings. The 12,400 labelled vehicular trajectory samples were first evaluated in experimental conditions with a constant GNSS noise level of 5 m, 10 m, and 20 m. The overall classification accuracy of the proposed approach was 92.7% and the accuracy was 90.2% when subjected to severe conditions of 20 m degradation, which is 11.2 and 19.5 times higher than the HMM-based and geometric ones, respectively. The significance of the improvements in the performance is statistically validated with the help of confidence intervals, z-tests, and ANOVA. Also, the framework is real-time and has an average latency per update of 52 ms. The findings suggest that semantic learning and probabilistic map-matching can be integrated into a robust and scalable road hierarchy discrimination system in GNSS-degraded scenes, and can be used in intelligent transportation systems and connected vehicle technologies.