Intelligent Trajectory Classification for Improved Movement Prediction

We treat the problem of movement prediction as a classification task. We assume the existence of a (gradually populated/trained) knowledge base and try to compare the movement pattern of a certain object with stored information in order to predict its future locations. A conventional prediction scheme would suffer from potential noise in movement patterns. Such noise (typically manifested as small-random deviations from previously seen patterns): 1) negatively impacts the prediction capability (accuracy) of the classification system and 2) oversizes the knowledge base (i.e., the storage needs become excessive). We try to alleviate such shortcomings through the use of optimal stopping theory (OST) and the introduction of a very specific movement prediction work-flow. OST relaxes the classification task so that slightly different patterns can be treated as similar. Moreover, the underlying knowledge base is kept as concise as possible by retaining those patterns with limited spatial variance. The performance assessment and comparison to other schemes reveals the superiority of the proposed system.