Multi-Scale Spatio-Temporal Aggregation And Dynamic Visualization Of Massive Trajectory Dataset With Road Network Constraints

With the rapid development of Internet of Things,Internet of Vehicles,and other technologies,the trajectory data of moving objects collected by GPS has exploded extensively.Thus there is of great urgency to use interactively visual analysis to help explore the inherent characteristics and potential patterns in the big trajectory dataset.However,due to its characteristics of massive volume,spatio-temporal correlation,and high-dimensional attributes,there are still many problems during visual analysis,i.e,low rendering efficiency and poor visual effects.Most of existing solutions use preaggregation methods to improve rendering efficiency.However,the pre-aggregation methods tend to treat the trajectory data as a set of discrete points in space and divide them with a uniform grid or administrative polygons,which would ignore the spatial constraints from the road network structure,and also neglect the spatial sequence information of trajectory data.To address these problems,this paper proposed a multi-scale spatio-temporal aggregation model that takes into account road network constraints.Firstly,the model maps trajectory sequence into discrete segments following the road network structure.In the spatial dimension,the mapping results of the trajectory are then aggregated based on the road network level by level.In the temporal dimension,the results are grouped from finer to coarser granularity,i.e.hour-day-week-month.For better data organization,this paper builds a 4D data cube model(Road * Hour * Day * Month),which can represent the correlation of trajectory segments and multiple temporal granularities simultaneously.With this 4D structure,users can directly choose different temporal granularities according to practical needs without online real-time aggregation.And we can choose three of them for aggregation,thus achieving a multi-dimensional aggregation solution under different perspectives.Furthermore,HBase is selected as the distributed storage of aggregated trajectory data due to high scalability of preprocessing capability and storage capacity.Accordingly,this paper designed an efficient spatio-temporal query algorithm which can exploit the built spatial index to quickly locate the target road segments,and efficiently conducts the temporal filter to access the dat.Finally,a dynamic visualization system of trajectory data has been implemented for change detection and pattern exploration in the traffic sequences,which uses multi-view linkage display,i.e.,lines,bars,charts and heat maps and also supports dynamic animation visualization.The Chengdu taxi trajectory dataset published by Didi are chosen to evaluate the performance of the proposed model and the dynamic visualization system.The experimental results show that the multi-scale spatio-temporal aggregation model does not change dramatically but increases slowly with the expansion of the query window size.The system can maintain a fast response speed when the time span is changed at different spatial levels.During dynamic playback,we found that caching partial data with pre-fetching can accelerate the playback progress compared with other two methods,and ensure smooth playback.