Anomaly Event Detection And Analysis Based On Taxi Trajectories Data

Trajectory data produced in the city can reveal the hidden facts related to urban dynamic and human movement behavior,while data mining,machine learning and a series of data utilization and knowledge discovery methods,provide strong support to push forward the study of trajectory data in various fields.Trajectory data analysis is essentially geographic information processing,developing technically the analytical ability of GIS.Life in the city is not changeless,and occasionally there are some factors such as events' influence that cause different crowd behavior from the usual mobile patterns.Trajectory data record the regular travel footprint of urban population,and at the same time,it also implies the abnormal information about human's travel behavior,reflects the effects events caused on travel,and indirectly records the events' process.Detection and analysis of outliers can help to understand the inherent laws of anomalies and make more reasonable anomaly handling decisions,thereby reducing the loss caused by abnormal.In this paper,according to the two kinds of trajectory forms—the curve and mobile space-time range,we focus on the extraction of abnormal trajectory patterns and the event detection and analysis in the city from the taxi trajectory data.The results obtained can provide reference information for city intelligent transportation construction and public safety management.The main research work of this paper is as follows:1)The discovery of taxi anomaly trajectory pattern.Regarding trajectory as curve,we investigate the trajectory data of taxi moving between the same starting and destination in the city.To increase the interpretability of trajectory anomaly,the abnormal state of trajectory's temporal or spatial characteristics is determined by the trajectory's travel time or curve's spatial characteristics,so the trajectory can be divided into four categories: standard,temporal outlier,spatial outlier and spatio-temporal outlier.In the trajectory analysis,the trajectory,according to the angle,is divided into a plurality of approximate trajectory segments,and their local features are compared by the linear Hausdorff distance.Then we present the trajectory's spatial similarity judgment rules with constraints from trajectory length.Finally,the density clustering method is used to detect the trajectories with abnormal spatial features,and the K times standard deviation criterion is applied to separate the anomaly trajectory with the travel time exceeding the threshold value.The experimental results show that this method can be used to dig out some information from the trajectory,such as the route selection behavior,the personalized route,the abnormal location and the traffic section.2)Events detection and analysis.In this paper,we summarize the taxi trajectory by OD data,and utilize the trip volume reflected by the drop-off data in the taxi trajectory to describe the regional travel dynamic,with a single region and single timespan as the smallest spatio-temporal unit of perception of the city anomaly dynamic,and detects a large number of anomalies(called elemental events here)from the taxi trajectory data based on the likelihood ratio test method(LRT).Then a new idea of the event analysis on the trajectory data is proposed: the spatio-temporal characteristic of elemental events and the evolution process of events are analyzed,the events' impacts on city life are assessed according to the change range of trip volume in the elemental events,and continuous spatio-temporal information of events is analyzed by using the method similar to DBSCAN algorithm proposed to extract complex spatio-temporal events from massive elemental events,making a meaningful discovery.3)The discovery of abnormal aggregation behavior.The trip volume that reflects regional travel dynamic cannot reveal the difference of the travel purposes,so the concept of aggregation behavior is put forward based on the location points group.In this paper,the event that is represented as abnormal aggregation behavior is regarded as a special event,which has a higher degree of vigilance than other events.Points group with accumulation pattern is estimated more reasonably with improved average k-nearest neighbor index,providing the mathematics foundation to find aggregation behavior;to further analyze the aggregation behavior,the location feature vector and the cosine similarity measure model of points group are constructed;the method of analyzing the aggregation behavior pattern based on clustering is proposed,contributing to a more in-depth interpretation of the regional crowd travel rules;to discover the abnormal aggregation behavior from the regional drop-off data,the LRT based and distance based method is presented,obtaining the occurrence time and position's spatial distribution of special events,which helps block safety management and rational allocation of resources in the city.