Methods For Congruent Conflation Of Multi-source Road Networks And POIs

Spatial data is an important national information resources for constructing digital earth and smart city. Maintaining the accuracy and currency of spatial data is the primary mission of establishing national fundamental geographic information databases. Due to specific requirements and standards, different agencies utilized various instruments to acquire massive heterogeneous spatial data. Heterogeneous data from multi sources occur to large discrepancies on precisions, scales, spatial relations, semantic representations, and data models, which leads to the contradiction of duplications in data collection and difficulties in information sharing. Hence, fast and efficiently integrating multi-source, multi-dimensional heterogeneous spatial data becomes a burning problem in Geographical Information Science (GIS) community.Presently, with the emerging and popularity of web 2.0 and mobile devices, the public citizens started to participate in creating and sharing geographic information, named crowdsourcing geospatial data, Volunteered Geographic Information (VGI) or User Generated Content (UGC). Crowdsourcing geospatial data generated by public users provides a timely and low cost means for collecting and updating fundamental geographic information, which has been widely applied in many areas, such as Intelligent Transportation System (ITS), smart city, environment monitoring, and emergency response. However, due to public participation and lacks of professional supervision, there are geometric, semantic and topological inconsistencies between crowdsourcing and professional geospatial data. Traditional conflation methods failed to approach crowdsourcing geospatial data with different scales, different accuracies, complex spatial relations and unstructured semantic description.Particularly, Point of Interest (POI) and road network are the basic data sources in navigation and Location Based Services (LBS). Retaining the accuracy and continual updating of POI and road data plays a significant role in improving the service quality of navigation and LBS. The dissertation thus concentrates on the particular issue of congruently conflating multi-source POI and road networks. The main contents are detailedly described as follows.(a) A probabilistic relaxation approach is subsequently proposed to identify the correspondences between two road networks. The proposed method calculates the distance, shape and length similarities between candidate matching roads detected by a buffering operation. The three similarities are integrated to determine an initial matching matrix, and then the compatibilities of neighboring roads are computed to heuristically update the initial matching matrix until it is convergent. Final, the 1:0,1:1,1:M and M:N matches are selected based on the convergent matching matrix.(b) Mining geometric pattern from heterogeneous data provides a feasible solution for improving the performance and efficiency of spatial data matching and harmonization. The dissertation attempts to propose semantic-based and geometric-based approaches to extract the geometric patterns from POIs. Two methods utilize the semantic knowledge or positional distribution separately to partition POIs to different clusters and delineate the geometric patterns of POI clusters as polylines to construct POI connectivity graph. The exploratory study of extracting geometric pattern from POIs indicate that POIs are pattern-related with the associated road networks, providing a novel and feasible solution for conflating multi-source POIs and road networks.(c) Based on the research of extracting geometric patterns from POIs, a pattern mining approach is presented to address the positional and semantic inconsistencies between heterogeneous POI and road networks. The proposed method mines the geometric patterns from POI and road networks to construct two skeleton graphs for two data. The controlling points are then recognized by matching the skeleton graphs of POI and road networks to align the POI points to the target road networks geometrically. The semantic inconsistencies between POI and road networks are also detected by comparing the road-related attributes between matched POI and roads to infer the road names for crowdsourcing road networks.(d) Several experimental data sets are selected to valid the efficiency and reliability of the proposed methods. The experimental results and precision evaluations indicate that the proposed methods can efficiently identify the correspondences between different road networks as well as different POIs and road networks. The detailed analysis and discussions about the algorithm parameters demonstrate that the proposed methods can achieve reliable results with different parameters and experimental data sets. Moreover, quality assessment of OpenSteetMap data is fulfilled based on the matching of OSM road networks and professional data. And the geometric adjustment and semantic enrichment are implement based on the positional and object correspondences between multi-dimensional heterogeneous POIs and road data.