Research On Mining Urban Attractive Areas And Popular Routes Based On Spatiotemporal Feature

Trajectory is a footprint formed by the continuous changes of the spatial position of the moving object over time.In recent years,with the development of Internet of things and mobile computing technology,Global Positioning System(GPS)and other forms of location acquisition technology have enabled us to collect massive spatial trajectory data of moving objects,such as people,vehicles and animals.Such a large number of trajectories provided us an unprecedented opportunity to automatically discover useful knowledge and patterns,which in turn offered the support for decision in various fields(e.g.urban service).In particular,the discovery of valuable information from large-scale taxi trajectory data has aroused an increasing interest in the applications based on taxi trajectory data.Taxi,providing personalized and efficient point-to-point service,is a popular form of public transportation in modern cities.Compared with other transportation services such as buses and subways,taxi services are more convenient for passengers.Due to the high flexibility of routes and operating time,at the same time,they provide reliable information on the taxi trajectories,which reflect the behavior patterns of both passengers and taxi drivers.On the one hand,by extracting the passengers' pick-up/drop-off data from the taxi trajectory data and clustering them,we can find some potential attractive areas in the city,which sufficiently reflect the passengers' mobility patterns and the spatiotemporal distributions of taxi-passenger demand.It is very significant for the dispatching of taxis and the planning of bus lines.On the other hand,the trajectory clustering can be exploited to discover the popular routes among attractive areas,and these trajectory clusters directly reflect the high traffic roads and the motion of the taxis during the running process,which is of great value to traffic management and mobility pattern analysis.In this paper,we utilize a large-scale GPS data set generated by over 10,000 taxis in a period of two weeks in Chongqing,China,aiming to explore the attractive areas and popular routes in the city.Firstly,this paper establishes the method of preprocessing trajectory data.Secondly,this paper proposes a new method based on grid density to mining attractive areas.Finally,in order to further discover the popular routes among attractive areas,this paper presents a trajectory clustering algorithm based on spatiotemporal similarity.More specifically,the main contributions of this paper are as follows:(1)According to the spatiotemporal characteristics of taxi trajectory data,this paper establishes a complete method of preprocessing trajectory data,which includes reducing dimensionality,filtering noise data,cleaning redundant data,extracting pick-up/drop-off data,map matching,and trajectory compression.The experimental results show that this method can extract the available data from the trajectory data and improve the accuracy of the original data.(2)This paper proposes a new clustering algorithm based on grid density,named GScan.In this method,firstly,the grid cells are divided from the trajectory data space and then the spatial points are mapped to grid cells,and the hot grid cells can be extracted by setting the threshold.At last,through merging reachable hot grid cells,the attractive areas in the city can be found.Based on taxis' pick-up/drop-off data of Chongqing,experiments and analysis are carried out.The parameters in the method are discussed,and a method of setting the parameters in the experiment is given.At the end of the paper,the spatial and temporal distribution of the attractive areas is presented in Chongqing,the travel behaviors of people in Chongqing can be discussed.(3)Based on the results of attractive areas,this paper presents a spatiotemporal trajectory clustering(ST-TCLUS)algorithm and uses it to discover the popular routes among these areas.In this method,the space similarity of the trajectory is improved and the effect of time similarity on the clustering results is taken into account.Compared with the existing trajectory clustering method,the experiments show that the clustering results are more accurate and representative.