Font Size: a A A

Research On Spatiotemporal Anomaly Detection Algorithms For Moving Object Trajectory

Posted on:2021-03-14Degree:MasterType:Thesis
Country:ChinaCandidate:H R PeiFull Text:PDF
GTID:2428330626458568Subject:Computer system architecture
Abstract/Summary:PDF Full Text Request
With the continuous development and wide application of positioning technology,mobile sensor equipment and wireless communication technology,a large number of moving objects can be tracked.As a result,a large number of trajectory data are stored in the database.It has always been the focus of research to analyze these spatiotemporal trajectory data and extract valuable information.Trajectory anomaly detection is an important research branch in the field of trajectory data mining.Trajectory anomaly detection is widely used in urban traffic,disaster weather warning,social public security management and animal habits research.In this thesis,the moving object's spatiotemporal trajectory data is taken as the research object,and the abnormal trajectory mining is taken as the research object.The main work of this thesis includes the following aspects:(1)Trajectory structure anomaly detection based on feature entropyStarting from the characteristics of trajectory structure,this thesis fully considers the comprehensiveness of the description of trajectory characteristics.At the same time,the thesis uses feature entropy to weight the trajectory features,explores the way to measure the importance of the trajectory features,and proposes the trajectory structure anomaly detection algorithm based on feature entropy.It solves the problem that the current distance based trajectory anomaly detection algorithm is mainly based on the shape of trajectory space,does not fully consider the internal feature information of trajectory,and ignores the importance of internal features.Experiments with multiple datasets show that the algorithm can find the anomaly from trajectory space shape and internal feature attributes.It can find obvious abnormal trajectory and its segments comprehensively,which makes the detection results more practical.(2)Unsupervised trajectory anomaly detection based on deep representationThis thesis deeply explores the representation of trajectory features,uses feature sequence to represent trajectory,and weaken the spatial attributes of trajectory.At the same time,this thesis fully mine the hidden features of the trajectory,integrate the lowdimensional shallow features and high-dimensional deep features of the trajectory,comprehensively describe the characteristics of the trajectory,and propose an unsupervised trajectory anomaly detection algorithm based on the deep representation.It makes the feature representation of trajectory get rid of the limitation of spatial attributes,and effectively solves the problem that the current trajectory anomaly detection algorithm which not easy finding some hidden features and high-dimensional combined features pays too much attention to shape,density,speed and other macro features of trajectory.In this experiment,multiple data sets are used to verify the feasibility and effectiveness of the algorithm with high F1 score.(3)Design and implementation of trajectory anomaly detection prototype systemBased on the idea of object-oriented and the design idea of modular development,combined with the theoretical research and practical operation of this subject,a prototype system of trajectory space-time anomaly detection for mobile objects is designed and developed.The system can demonstrate the experimental process of this research,test and analyze the research content and results of this project,verify the effectiveness and feasibility of the algorithm,and provide practical basis for further large-scale application of the research results of this thesis.In this thesis,there are 50 figures,4 tables and 115 references.
Keywords/Search Tags:anomaly detection, trajectory division, clustering, structural feature, autoencoder
PDF Full Text Request
Related items