Low-cost sensors aided vehicular position prediction with partial least squares regression during GPS outage

Vehicular position prediction is very important in intelligent transport systems (ITS), and the requirements of accuracy for position prediction have been significantly increasing in recent years. In this paper, we focus on designing a more low-cost and convenient method which can operate during GPS outages. In order to better deal with the position prediction during the lack of GPS signals, we introduce a windowed partial least squares regression (WPLSR) approach where vehicle position information from the low-cost sensors was used. Moreover, the window is adjustable and it reduces the step of regression in WPLSR algorithm. The sensor data outside the window that has nothing to do with the latest position prediction is eliminated. Therefore, the position accuracy can be improved significantly. Finally, the proposed method is evaluated by using road experiments from real urban areas. Compared with the conventional techniques such as PLSR and extended Kalman filter combined with an interactive multimodel (IMM-EKF), the results show that WPLSR presents the higher position accuracy especially during the GPS outages.