Traffic state estimation with Bayesian Networks at extremely low V2X penetration rates

In this paper the concept of Bayesian Networks (BN) is applied to the problem of traffic data acquisition by data fusion. Two wireless communication based sensors are used as data sources: IEEE 802.15.1 Bluetooth and IEEE 802.11p V2X (vehicle to vehicle and vehicle to infrastructure). Via V2X so called cooperative awareness messages (CAM) are received, which provide information on vehicle location and speed. For Bluetooth only the presence of a Bluetooth device can be detected. Currently and in the near future a low amount of road users is expected to be equipped with V2X. Therefore the rate of V2X vehicles is very low (≈1%). The penetration rate of Bluetooth devices is much higher. Approximately 3% to 50% of all road users can be detected and re-identified with a Bluetooth scanning device. Bluetooth detectors have notably been used for traffic management purposes for years, e.g. for obtaining journey times, but they have not been applied for speed estimation so far. The approach of this paper is to provide vehicle count data and vehicle speed by fusing Bluetooth data at moderate and V2X data at low penetration rates. The challenging task is to obtain accurate speed estimation data. By applying BNs for this purpose, we show the robustness of this stochastic fusion engine. It is capable of reaching speed RMSEs between 2 and 5 m/s and enhance the completeness of the state estimation by 35% by fusing 1% V2X with 30% Bluetooth. The investigations are made on the basis of simulation.