Pattern-Mining Conflation Approach for Enriching Crowdsourcing Road Networks with POIs

ZHANG Yunfei,YANG Bisheng

2014-01-01

Abstract:Presently, Navigation and Location based services (LBS) are moving towards the period of dynamic and holographic navigation, demanding a navigation system integrates all kinds of information (e.g., traffic and non-traffic) and makes a reasonable and high efficient traveling route for individuals. On the other hand, the popularity of Volunteered Geographic Information (VGI) yielded new means for humans to participate in creating and editing geographic information. By collaborative mapping, large amount of trajectories and geo-tagged data, named crowdsourcing geospatial data, are increasingly obtained, providing a crucial data source for enriching road maps and associated POI database. Nevertheless, due to the participation of non-experts, the heterogeneities of internal and external data representations reduce its availabilities. Hence, conflation technologies were developed to bridge the gap between heterogeneous data and many advanced approaches for matching road networks or building footprints were proposed. However, few studies paid attention on conflating crowdsourcing road network and POIs with the preservation of spatial relations and semantic data consistency. As a result of different geometric types, new challenges occur to conflation of crowdsourcing road network and POIs. This paper thus proposes a pattern-mining approach for conflating crowdsourcing road networks and POIs, resulting in enrichment of crowdsourcing road networks. The proposed approach tries to mine and extract the common patterns in road networks and POIs to build their correspondence. The strokes and intersections of road networks are firstly identified and simplified to compose a generalized skeleton graph. Then, a k-means clustering algorithm is utilized to match POIs with the road strokes. All POIs are partitioned into k road strokes and the linear pattern of POIs of the same stroke is characterized as a principle curve by a polygonal algorithm. The skeleton graph of POIs is thus composed of the principle curves of all POI partitions. Two generated skeleton graphs of POIs and road networks are then matched to recognize the corresponding points. Spatial adjustment based on the corresponding points is performed to rectify crowdsourcing road networks and POIs geometrically. A semantic-edit module is also developed to detect their semantic inconsistencies and amend the attribute information. Three experimental datasets in Wuhan, Shanghai and Beijing were selected to valid the proposed method. The road networks are downloaded from OpenStreetMap and the POI points are obtained by crowdsourcing or authorized agencies. The experiments show that the spatial relations between crowdsourcing road networks and POIs are mostly corrected after conflation. As the correspondences between POIs and road networks are determined by a k-means clustering, the semantic inconsistencies between them are detected and the attribute information is mutually enriched. Additionally, several geo-tagged data are conflated to the same reference road networks and a conflated POI data are refined by eliminating the duplicate records.