Learning to Predict Driver Route and Destination Intent

For many people, driving is a routine activity where people drive to the same destinations using the same routes on a regular basis. Many drivers, for example, will drive to and from work along a small set of routes, at about the same time every day of the working week. Similarly, although a person may shop on different days or at different times, they will often visit the same grocery store(s). In this paper, we present a novel approach to predicting driver intent that exploits the predictable nature of everyday driving. Our approach predicts a driver's intended route and destination through the use of a probabilistic model learned from observation of their driving habits. We show that by using a low-cost GPS sensor and a map database, it is possible to build a hidden Markov model (HMM) of the routes and destinations used by the driver. Furthermore, we show that this model can be used to make accurate predictions of the driver's destination and route through on-line observation of their GPS position during the trip. We present a thorough evaluation of our approach using a corpus of almost a month of real, everyday driving. Our results demonstrate the effectiveness of the approach, achieving approximately 98% accuracy in most cases. Such high performance suggests that the method can be harnessed for improved safety monitoring, route planning taking into account traffic density, and better trip duration prediction