On predicting vehicle tracks

GPS navigators track vehicle location in real-time. A related problem is tracking a vehicle's path after the travel. This often is a requirement for fleet management and also for ensuring correct payment of road user charges. Such path prediction uses data stored during the vehicle's travel. The problem then is to ensure storing minimum amount of data, while maximizing the path prediction accuracy: storing large volumes of data will enable better path prediction accuracy while sparse data collection and storage may reduce the prediction accuracy. In addition, processing large volumes of data and/or taking into account a large number of factors may reduce the performance of path prediction. Consequently, an approach that minimizes the storage and data requirements while not compromising path prediction accuracy is required. This paper proposes three techniques to estimate the most likely vehicle path based on vehicle location, direction, speed and the road network information. The first is called hybrid location updating method, which uses both time and vehicle direction change as thresholds to decide when to update vehicle information. The second is the bearing-distance hybrid map-matching algorithm, which maps the GPS location data onto the road network by using both heading difference and the distance between the collected vehicle data and its nearby roads. Finally, the map-matched point linking algorithm finds all the vehicle paths between the start and the end points, and then validates the most likely vehicle path by using total traveled distance, total direction change, vehicle speed and road network topology. Experimental evidence clearly shows that these techniques improve the vehicle tracking system accuracy and efficiency. Furthermore, these methods give reasonable estimations of the most likely vehicle path while using minimal updated point data.