Vehicle Positioning with Dynamic Recurrent Vehicular Pattern Learning

High precision navigation in urban environment is a fundamental requirement for modern transportation systems. Global Navigation Satellite System (GNSS) and Inertial Measurement Units (IMUs) are the typical sensors on commercial vehicles. In urban environments, however, shadowing and multi-path effects on satellite signals can severely degrade the localization performance of GNSS receivers. Navigation with IMUs, on the other side, requires a precise GNSS signal to correct the typical biases of low-cost IMUs. In order to mitigate these effects, we propose a processing scheme that leverages on recurrent patterns of repeated passages of vehicles in a given region to make the positioning system much more robust. An edge cloud is used to store past trajectory data to form the a-priori dynamical state within any pre-defined area. The estimated a-priori is passed to the incoming vehicles and is combined with their internal state to reduce the GNSS dependency. An experimental campaign validates the proposed scheme, and results are favourably compared with standard Kalman filter navigation.