Extraction And Visual Analysis Of Urban Traffic Patterns Based On Trajectory Data

This paper considers transportation topics as a reflection of urban travel in a city.By mining this travel data,valuable information can be obtained to assist with urban planning,transportation and other important social applications.The common topic extraction used in such transportation applications is limited to single-level topic extraction based on a topic model.The number of topics must be specified in advance.Such single-level topic extraction strategies limit the spatial and temporal scale of user exploration of the data.This paper proposes a novel,interactive hierarchical traffic topic exploration strategy which is vastly superior to these traditional systems.This hierarchical traffic topic extraction and exploration strategy is derived from improvements implemented on the H-NMF method.This strategy does not require the number of topics to be specified in advance,displaying traffic topics supporting a range of different space-time scales for user interaction.Each of these traffic topics contains varying spatiotemporal characteristics that users typically find it difficult to understand the resultant topic semantics.This paper proposes a method to extract general spatiotemporal features of the topic.What’s more,the proposed system supports methods based on improved density clustering to extract topic flow distribution at particular instances.Based on this novel algorithm,the research has implemented and designed a hierarchical transportation topic exploration visualization system,with the system effectiveness verified through case analysis.The main work of the paper is as follows:(1)Interactive hierarchical traffic topic exploration method: This paper proposes a hierarchical traffic topic extraction and exploration strategy based on the H-NMF algorithm.Starting with high level space-time traffic topics,the novel algorithm allows the user to navigate through the system,choosing low-level space-time scale topics of interest for investigation.(2)Extraction method of general spatio-temporal features of traffic topic: This paper extracts three features(topic ID,time and space)from traffic topic data in order to construct a three-dimensional tensor.General spatio-temporal features of the topic are then extracted from this through tensor decomposition.(3)An algorithm for extracting key areas of the topic based on density clustering: A new algorithm,based on the density clustering algorithm is developed,making it suitable for clustering spatial distributions of traffic topics.In order to view the results of regional clustering,this research also proposes a filtering algorithm which allows the user to filter the important clusters based on their selection criteria.(4)Visual analysis system for hierarchical traffic topic mining: the visualization system for the hierarchical traffic topic exploration system proposed in this paper includes the following views: an interactive topic hierarchical exploration and segmentation view;two views displaying the topic pattern,namely the topic pattern spatial distribution view and the topic pattern time distribution view and a set of views to assist users in analyzing topics(including topic detail view,topic hour flow chart,flow distribution comparison view and POI view).