Mapmatching

Mapmatching is the process of aligning raw positioning data to a digital road network. It aims to infer the path taken by a moving object by snapping observations to the most plausible sequence of road segments, particularly when measurements are noisy, sparse, or imprecise.

Inputs are timestamped location observations from GNSS receivers, sometimes augmented by inertial data. The output is a trajectory on the road network: a sequence of connected edges with aligned positions and a confidence estimate for each match.

Approaches are generally geometric, topological, or probabilistic. Geometric methods snap points to nearby edges; topological methods enforce network connectivity and plausible transitions; probabilistic methods, especially hidden Markov models, treat road segments as hidden states and GPS points as observations, using algorithms like Viterbi or particle filters to infer the most likely path.

The road network is modeled as a directed graph with attributes for edges, turns, and speeds. Preprocessing includes map cleaning, setting a snapping radius, and aligning data with the network. Some systems also use map hierarchies or sensor fusion to improve accuracy and scalability.

Key challenges include GPS errors, urban canyons, sparse sampling, complex intersections, and imperfect map data. Evaluation typically reports edge- or path-level accuracy and position error against ground truth.

Applications include turn-by-turn navigation, traffic state estimation, and mobility analytics for planning. Ongoing work addresses real-time mapmatching, multi-vehicle consistency, and robust methods for low-sampling traces.