Terrain-based vehicle localization from real-time data using dynamical models

This paper describes a novel method for the location of road vehicles using vehicle pitch data obtained from on-board sensors. The method encodes the road map data using linear dynamical models, and then, during travel, identifies the vehicle location through continuous validation of the previously obtained linear models. The approach presented has several advantages over previous approaches in the literature, namely a smaller computational burden, a more definitive location estimate, and a simplified and more direct way of handling common types of noise. These benefits have the potential to both increase the speed of the localization and to reduce the implementation cost of terrain-based localization. The method is tested in simulation using real-world road data collected in State College PA, USA. Performance is demonstrated both in a noise-free and noisy environments, and a bound is shown on the convergence distance.