Research On Distributed Detection Algorithm Of Anomaly In Spatio-Temporal Trajectories Of Vehicles

With widespread use of Internet of Vehicle(IoV)technologies for traffic safety and management,a large number of spatio-temporal trajectories of commercial vehicles are integrated into the real-time monitoring system.Traffic status can be evaluated by analyzing these trajectories in real time.Appropriate measures can be taken in advance in the case of any anomaly in order to guarantee operational efficiency of enterprises,protect property and personal security and avoid possible losses.To this end,a solution is needed to efficient processing of vehicle trajectories and correct detection of anomaly from trajectories.The work in this thesis is in part supported by Science & Technology Pillar Program of Hubei Province named "Key Technology and Application of Collaborative Monitoring of Logistics Vehicles Based on Internet of Vehicles"(2014BAA146),and by the Nature Science Foundation of Hubei Province named "Vehicle Safety Perception and Collaborative Monitoring Technology Based on Internet of Vehicles"(2015CFA059).Clustering of vehicle trajectories and detection of abnormal trajectories is studied to support effective monitoring of vehicles.Contributions of this thesis are as follows.(1)After fully considering the motion features of vehicles,a method is proposed to segment trajectories and measure similarities.A novel algorithm for distributed clustering of trajectories is designed.Common pattern among the spatio-temporal trajectories cannot be identified without clustering the information on the same motion pattern.How to segment trajectories and measure similarity is the prerequisite of cluster analysis.The traditional method for extraction of feature points causes the loss of much information.In order to address this problem,various local features of vehicles are taken into account during trajectory segmentation and similarity measurement,including vehicle location,speed,angle,petrol consumption and trajectory shape.A strategy for trajectory segmentation based on multiple motion features and a method for similarity measurement based on trajectory structure are proposed.In order to improve the performance,the principle of the CluStream model is adopted to propose an algorithm for distributed clustering of stream-typed trajectories.The clustering efficiency of trajectories is enhanced through synergy among modules of priori knowledge,local clustering,global clustering and data distribution.(2)A new abnormal trajectory detection algorithm is proposed,and it is implemented via Storm and verified through real experiment.Due to the need for frequency calculation of distance and density of neighborhood,the anomaly detection algorithm usually faces heavy computational loads.Hence,the clustering results are used to prune the datasets,resulting in improved algorithm efficiency and real-time performance.The Storm simulation results demonstrate the ability of the proposed framework for detection of abnormal trajectories to tolerate heavy data streams.The proposed algorithm is thus suited for real-world applications.To sum up,detection of anomaly in the spatio-temporal trajectories is studied with support from relevant projects.A framework for anomaly detection is proposed,laying foundation for future development in intelligent transport system.