Improving low cost sensor based vehicle positioning with Machine Learning

Current fleet management solutions rely on real time vehicle information to efficiently resolve transportation problems. In this study, a novel robust approach based on combining the Extended Kalman Filter (EKF) with Machine Learning techniques, Neural Networks or Support Vector Machines, is introduced to improve the accuracy of vehicle position estimation and circumvent the EKF limitations. The proposed solution guarantees also a low cost by using the Global Positioning System enhanced with Dead Reckoning integrated sensors. To verify our approach, extensive simulation tests are conducted on field data sets and very promising progress is obtained in the estimated vehicle position.