The Visualization And Analysis Of Traffic Data Stream Based On Topic Modeling

Traffic problem such as road congestion,travel difficulties has become a common problem in many cities.It is very important to improve the traffic condition by putting forward some targeted traffic management measures that come from analyzing the traffic data to find the traffic patterns.In our work,we focus on two types of particularly important urban traffic data sets: taxi GPS data and surveillance devices data.These two kinds of data can obtain detail and initial trajectories from taxis and all vehicles respectively.However,these trajectories always have large quantity and are time-varying,which makes their analysis extremely complicated.In addition,establishing a method to visualize and explore the analysis results interactively still remains challenging.In this paper,we utilize text mining techniques and create a new method for effective knowledge discovery of trajectories.The key point of our work lies in that we propose a Latent-Dirichlet-Allocation(LDA)based approach to the discovery of underlying traffic topic regions from vehicle trajectories captured by GPS installed in taxis or surveillance devices installed along roadsides.We treat vehicle trajectories as documents and values of different traffic features,such as locations,direction,speeds,vehicle types,Points of Interest(POI)(e.g.,restaurant and shops)and others as the corresponding words.After applying the LDA model,we obtain a list of traffic topic regions with combined features values,in which the different feature values are clustered with probabilistic assignments.The chatic trajectories can be expressed by the probability vector under traffic topic regions and then cluster trajectories to explore the traffic condition.Meanwhile,we design the corresponding visualization methods for understanding the analysis results and further exploration of the traffic patterns.Then,we integrate the visualizations into a system prototype.It includes Map View,Word Cloud View,TreeMap View and Matrix-Table View.Finally,we conduct a real case based on the traffic data in Wenzhou City.We investigate a large taxi trajectory data set acquired from 3,744 taxis in a city and surveillance devices data containing approximately 157 surveillance devices and 750,000 moving vehicles.The case demonstrates the effectiveness of both our proposed approach and the visualizations.