For Static Trajectory Outlier Detection Algorithm

With the rapid development of the Internet,the trajectory data obtained by mobile devices also shows explosive growth.The huge amount of trajectory data not only brings important research value,but also poses serious challenges in data processing and data mining.Confronted with such a large trajectory data set,how to mine useful information from these data has become very important.A typical trajectory data analysis task is to examine the trajectory set,mining the moving rules and behavior patterns of moving objects.Once the abnormal behavior or pattern is detected,the cause of it should be further analyzed to reduce the loss caused by it.Sometimes the analysis of anomalies can produce more unknown information and provide decision-making for future emergencies.In order to detect the trajectory data more effectively,this thesis proposes two anomaly detection algorithms.One is Local Density Trajectory Outlier Detection,LDTRAOD in short and the other is Time-dependent Popular Routes Based Trajectory Outlier Detection,TPRO in short.LDTRAOD algorithm is based on the traditional TRAOD algorithm and improves.When the distance between segments are being calculated,the number of segments in the neighborhood is added.The definition of local density is the combination of them.As a result,though the trajectories are too dense,the impact on the test results can be avoided.Then,the method that a judgement about normolity is based on local anomaly factor is more practical.Meanwhile,in order to solve the problem of large amount of computation and large time consumption when calculating the distance of trajectory segments in the algorithm,it is necessary to establish a grid index.The method only retrieves the trajectory segment set in the neighborhood,avoids the distance calculation of the global trajectory segments and improves trajectory outlier detection of the running efficiency and scalability.The TPRO algorithm first establishes Time-dependent Community transfer Graph to record the number of trajectories passing through each community in different time.With the help of Time-dependent Community transfer Graph,TPRO algorithm can efficiently query in a certain period of time the user defined within the K of the most popular routes,calculate the distance deviation between each trajectory and the popular route.TPRO also defines a formula based on the distance of time,it not only calculates the space position of the distance,but also takes into account the time parameter in distance calculation.Considering the distance with time,the results of outlier detection can be more accurate.This thesis uses the the Atlantic hurricane,Beijing taxi data as experiment data,compared the two proposed algorithms with traditional algorithm in outlier detection accuracy and running time.As is proved in the experiments,LDTRAOD and TPRO are more efficient and more accurate than traditional TRAOD.Meanwhile,in different scale trajectory data sets,LDTRAOD algorithm and TPRO algorithm,compared with traditional TRAOD,can display high retrieval efficiency in different size datasets.