Fast and Accurate Prediction of the Destination of Moving Objects

Companies and organizations that track moving objects are interested in predicting the intended destination of these moving objects. We develop a formal model for destination prediction problems where the agent (Predictor) predicting a destination may not know anything about the route planning mechanism used by another agent (Target) nor does the agent have historical information about the target's past movements nor do the observations about the agent have to be complete (there may be gaps when the target was not seen). We develop axioms that any destination probability function should satisfy and then provide a broad family of such functions guaranteed to satisfy the axioms. We experimentally compare our work with an existing method for destination prediction using Hidden Semi-Markov Models (HSMMs). We found our algorithms to be faster than the existing method. Considering prediction accuracy we found that, when the Predictor knows the route planning algorithm the target is using, the HSMM method is better, but without this assumption our algorithm is better.