GPS Location Data Accuracy via Hidden Markov Model Map Snapping

The patent extracts and utilizes pre-existing map data (polyline features representing streets) as external reference information to improve GPS accuracy. By comparing GPS locations against known map features, the system achieves better accuracy without relying solely on computationally intensive filtering algorithms. The map data serves as an external knowledge base that constrains and refines location estimates.

The patent introduces map polyline features as an intermediary between raw GPS data and final location estimates. Instead of directly processing GPS measurements through complex filters, the system uses map features as a mediator to constrain possible locations to those consistent with known street networks, reducing the computational search space and improving efficiency.

The patent merges multiple sources of information: GPS measurements, map polyline features, emission probabilities (likelihood of observing GPS data given map features), and transition probabilities (movement constraints between locations). By combining these diverse information sources in a unified probabilistic framework, the system achieves globally optimal solutions that consider all constraints simultaneously, rather than sequentially as in traditional filtering approaches.

The patent transforms the location estimation problem from directly processing GPS coordinates to estimating probabilities of being at various map locations. By changing the parameter space from raw coordinates to probability distributions over map features, and by incorporating transition probabilities that model movement constraints, the system achieves more reliable and globally optimal solutions.