On Predicting and Compressing Vehicular GPS Traces

Many vehicular safety applications rely on vehicles periodically broadcasting their position information and location trace. In very dense networks, such safety messaging can lead to offered traffic loads that saturate the shared wireless medium. One approach to address this problem is to reduce the frequency of location update messages when the movements of a vehicle can be predicted by nearby vehicles. In this paper, we study how predictable vehicular locations are, given a Global Positioning System trace of a vehicles recent path. We empirically evaluate the performance of linear and higher degree polynomial prediction algorithms using about 2500 vehicle traces collected under city and highway driving conditions. We find that linear polynomial prediction using the two most recent known locations performs best. Also, traces with a time granularity of 0.2s are highly predictable in low speed city environments, and a location update rate of 1Hz may suffice to represent city vehicular movements. Lastly, the paper also evaluates compression of different time-granularity traces using line simplification and polynomial interpolation techniques to reduce message sizes.