A robust low cost approach for real time car positioning in a smart city using Extended Kalman Filter and evolutionary machine learning

A smart city is emerging as an application of information and communication technologies to mitigate the problems generated by the urban population growth. One of the smart city solutions is to establish an efficient fleet management relating to the use of a fleet of vehicles (e.g., ambulances and police vehicles). The most basic function in a fleet management system is the real time vehicle tracking component. This component is usually the Differential Global Positioning System (DGPS) or the integration of Global Positioning System (GPS) and Inertial Navigation Systems (INS). To predict the position, the Extended Kalman Filter (EKF) is generally applied using the sensor's measures and the GPS position as a helper. However, the DGPS high cost solution still suffers from GPS satellite signals loss due to multipath errors and the INS require more complex computing. Furthermore, the EKF performance depends on the vehicle dynamic variations and may quickly diverge because of environment changes (i.e, GPS failures by obstructions from building and trees). In this paper, we present a robust low cost approach using EKF and neural networks (NN) with Genetic Algorithm (GA) to reliably estimate the real time vehicle position using GPS enhanced with low cost Dead Reckoning (DR) sensors. While GPS signals are available, we train the NN with GA on different dynamics and outage times to learn the position errors so we can correct the future EKF predictions during GPS signal outages. We obtain empirically an improvement of up to 95% over the simple EKF predictions in case of GPS failures.