Bayesian Forecasting Based Anomaly Detection For Public Places And Individual Trajectory

The economy in our country is continually developing after opening up and economicreform. However, the infrastructure of public safety is considerable fragile, which leads tomany safety accidents, especially the public disorder. Monitoring some specific locationsis an important measurement for the prevention of public disorder. With the developmentof the Internet of things, the monitor system can provide real-time and acute informationof locations. But, this huge mass of data has not been well exploited in the abnormitydetection and prediction. The criteria employed in existing methods are static giventhresholds and traces.For the monitoring places, the abnormity detection of the number of specific locationsand individual mobile trace has been studied in this thesis. Different from the criteriausing in traditional methods, which lack of the consideration of individual differences, thecriterion for abnormity detection is to compare with object normal states obtained from theanalysis of historical information. The Bayesian prediction method is employed, whoseparameters dynamical change according to object characteristics and observed data.For the abnormity detection of locations, the periodicity and malconformationresulting from time factors have been considered in the model of the changes of humannumbers. The parameters in the model change according to real-time data and the safetyrange of the human number of specific locations is dynamical setting.In the abnormity detection of individual traces, a modified particle filter tracingmethod is proposed, which can be applied in road networks. Individual habits have beenincorporated into the tracing model. The synthesized information and characteristics ofindividual behaviors are utilized in detecting mobile objects. Also the standard trace forabnormity detection is obtained.