Anomaly Detection Method Of Spatiotemporal Trajectory Data Considering Semantics

Trajectory data is a time series data stream generated by various positioning devices,which describes the position changes of moving objects within a certain time range,reflects the behavior patterns of moving objects,and contains rich space-time characteristics and semantic characteristics.Therefore,trajectory data mining has become the focus of attention in the field of big data analysis.Trajectory anomaly detection is one of the important research content in the field of trajectory data mining,which aims to dig out a small amount of data sets and different trajectories.It can be seen that anomalies are not equal to noise,anomalies may arise from different mechanisms,and the occurrence of anomalies may be accompanied by interesting phenomena,so it has attracted a large number of scholars to conduct research on trajectory data abnormal detection.For existing trajectory abnormal detection methods,the space-time characteristics of the trajectory are mainly considered,which can detect the abnormal trajectory under the constraints of time and space characteristics,but ignore the effects of the multi-dimensional semantic characteristics of the trajectory and cannot achieve abnormal trajectory detection under semantic characteristics.This article takes care of the time,space and semantic characteristics of trajectory data,combined with the theory of detection of abnormal detection of trajectory data,to achieve trajectory abnormal detection under multi-dimensional semantic characteristics.The main research content is as follows:(1)Detection of trajectory noise point based on Kalman filtering.The trajectory data has trajectory point noise problems such as trajectory point drift,because trajectory data is affected by various factors during the sampling process.Based on the principle of Kalman filtering,this thesis proposes a trajectory noise point detection method,which fully considers the motion state vector of trajectory points.Based on the movement trend of trajectory,it finds noise data that does not conform to the trajectory motion trend and eliminate the point noise to improve data quality.(2)Detection of abnormal points of trajectory with spatiotemporal semantic constraints.For traditional anomaly detection of trajectory point,the detection method mainly considers spatial characteristics and ignores semantic characteristics.This thesis maps the trajectory point into a multi-dimensional semantic space.The distance of the trajectory points is measured based on the multi-dimensional semantic features,and then the local outlier factor of each trajectory point is calculated.An outlier detection method with temporal and spatial semantic constraints is proposed to provide support for further mining of trajectory data.(3)Trajectory anomaly detection based on m-iBOAT algorithm.For the traditional trajectory anomaly detection method,only the spatiotemporal characteristics are considered,which causes problems that the test results are inaccurate and difficult to explain.This thesis proposes the m-iBOAT algorithm,which measures the similarity of trajectory grid cells from the perspective of multi-dimensional semantic features.Then,the support degree and abnormal score of the trajectory are calculated,and the abnormal trajectory is identified to realize the abnormal trajectory detection under the constraint of multi-dimensional semantic features.