Search Results

The 2009 U.S. National Household Travel Survey (NHTS) contains detailed data on individual vehicle trips. This paper demonstrates several useful statistics from the NHTS concerning how people use their vehicles, such as how far they drive, where they go, how long they stay, and their sequence of destinations. These statistics, in turn, are potentially useful for vehicle design, vehicle use simulation, navigation algorithms, interpreting GPS data, and the placement of electric vehicle charging stations.

This paper develops and tests algorithms for predicting the end-to-end route of a vehicle based on GPS observations of the vehicle's past trips. We show that a large portion a typical driver's trips are repeated. Our algorithms exploit this fact for prediction by matching the first part of a driver's current trip with one of the set of previously observed trips. Rather than predicting upcoming road segments, our focus is on making long term predictions of the route. We evaluate our algorithms using a large corpus of real world GPS driving data acquired from observing over 250 drivers for an average of 15.1 days per subject. Our results show how often and how accurately we can predict a driver's route as a function of the distance already driven.

This paper describes an algorithm for making short-term route predictions for vehicle drivers. It uses a simple Markov model to make probabilistic predictions by looking at a driver's just-driven path. The model is trained from the driver's long term trip history from GPS data. We envision applications including driver warnings, anticipatory information delivery, and various automatic vehicle behaviors. The algorithm is based on discrete road segments, whose average length is 237.5 meters. In one instantiation, the algorithm can predict the next road segment with 90% accuracy. We explore variations of the algorithm and find one that is both simple and accurate.

Map matching determines which road a vehicle is on based on inaccurate measured locations, such as GPS points. Simple algorithms, such as nearest road matching, fail often. We introduce a new algorithm that finds a sequence of road segments which simultaneously match the measured locations and which are traversable in the time intervals associated with the measurements. The time constraint, implemented with a hidden Markov model, greatly reduces the errors made by nearest road matching. We trained and tested the new algorithm on data taken from a large pool of real drivers.

This paper presents a novel method for predicting the location of a driver's destination during the drive. Such a prediction can be used to help decide which information to automatically present to the driver, depending on where the driver is going. The prediction is based on the common intuition that drivers tend to chose efficient routes. We quantify this preference for efficiency probabilistically based on a database of driving trips we gathered with GPS receivers. We show how to use this probability along with a map of driving times to compute the probability of any candidate destination. Our tests show that halfway through the drive, we can predict the destination to within about 10 km, and at three quarters of the way, the error drops to about 3 km.