Efficiently mining outliers from trajectories of unrestraint movement

With rapid development of GPS and wireless techniques, there accumulates a huge volume of trajectory data with long path in many applications. Thus, detecting outliers from trajectory data has become an attractive and interesting research topic. Like pattern matching, current researches on detecting outliers from trajectory data mainly focus on comparing trajectory's shape. This paper proposes a new framework of efficiently mining outliers from trajectory data, which were produced by the objects that move on unrestraint environment. Firstly, according to trajectory's characteristics, a distance computation method is designed, which is derived from the idea of Minimum Hausdoff Distance under Translation, which is used in pattern matching. This distance function not only considers the directory and the velocity of objects movement besides the shape, but also the costs of this distance function could be reduced sharply by R-Tree. Extensive experimental results demonstrate the efficiency and effectiveness of the proposed framework for trajectory outlier detection.