Research On Moving Object Trajectories Outlier Detection Algorithms

In the field of data mining for moving object, anomaly detection is a hot topic. Based on the development of anomalous trajectory detection of moving object, this paper summarizes a wide range of solutions nowadays and represents the disadvantages of them. This paper also researches new solutions of trajectory anomaly detection of moving object aiming to improve effect, improve efficiency and optimize parameters. The main work and contribution of this dissertation are summarized as follows:Firstly, DBTOD algorithm for anomalous trajectory detection based on density is proposed to make up for the disadvantage of TRAOD algorithm that it is unable to detect anomalous defects when the trajectory is local and dense. DBTOD uses partition-and-detect framework. It calculates local density and then local anomalous factor of each trajectory line. If local anomalous factor exceeds a particular threshold, the trajectory line is considered anomalous. DBTOD is able to detect both anomalous sub-trajectories and anomalous local trajectories. Experiments show that DBTOD improves the effect of TRAOD.Secondly, a new trajectory feature parameter TD is introduced to describe the static distribution of trajectory to improve efficiency of trajectory anomaly detection. Meanwhile, a novel effective trajectory anomaly detection algorithm CFTOD is presented based on trajectory direction feature. CFTOD contains three stages, feature extracting, coarse detection and fine detection. In the stage of coarse detection, some anomalous trajectory whose direction changes acutely is detected based on TD parameter. Trajectory searching space is reduced in the stage of fine detection, which speeds up the detection. Experiments show that CFTOD improves the efficiency of TRAOD.Thirdly, there is a disadvantage in TRAOD algorithm that it is sensitive to input parameters and a number of attempts are needed to reach a satisfied result, because TRAOD uses Hausdorff distance to measure the distance of trajectory lines. This paper proposes a novel method that detects anomalous points in each trajectory and determines whether it is anomalous by calculating the anomalous trajectory rate composed by anomalous points. Besides, an anomalous trajectory detection algorithm Trajps_TOD based on SR tree index is presented. It uses Euclidean distance to measure the distance of trajectory points. Experiments show that Trajps_TOD detects anomalous trajectory efficiently, decreases the number of input parameters and optimizes parameter selection.