Automatically and accurately matching objects in geospatial datasets

Identification of the same object represented in diverse geospatial datasets is a fundamental problem in spatial data handling and a variety of its applications. This need is becoming increasingly important as extraordinary amounts of geospatial data are collected and shared every day. Numerous difficulties exist in gathering information about objects of interest from diverse datasets, including different reference systems, distinct generalizations, and different levels of detail. Many research efforts have been made to select proper measures for matching objects according to the characteristics of involved datasets, though there appear to have been few if any previous attempts to improve the matching strategy given a certain criterion. This paper presents a new strategy to automatically and simultaneously match geographical objects in diverse datasets using linear programming, rather than identifying corresponding objects one after another. Based on a modified assignment problem model, we formulate an objective function that can be solved by an optimization model that takes into account all potentially matched pairs simultaneously by minimizing the total distance of all pairs in a similarity space. This strategy and widely used sequential approaches using the same matching criteria are applied to a series of hypothetical point datasets and real street network datasets. As a result, our strategy consistently improves global matching accuracy in all experiments.