Refinement of Existing Road Vector Layers Through High Resolution Satellite Images

Abstract

Abstract Geospatial data originate from different sources. Integrating multiple geospatial datasets which originate from different sources can provide insights and capabilities that are not possible from a single dataset. However, positional misalignments can occur during this integration process resulting in a requirement of an alignment approach to conflate or integrate different geospatial data types while reducing positional inconsistencies. In this dissertation, two novel alignment approaches are proposed to address the problem of misalignment between geospatial data in road vector layer and satellite imagery. The first approach, focuses on the distance ratio and angle between the provided control points in the two layers for the alignment process. Whereas, the second approach uses the concept of Bézier curves for alignment. The first three steps of these two approaches are the same. While the two approaches differ in the final step based on the concept used for the alignment process. The initial step is to decompose the road vector layer into a set of points. In the second step, the satellite imagery and the vector layer are transformed to the same coordinate reference system. Third and the most important step is providing control points from the satellite imagery and vector layer. Final step is to carryout the alignment. A novel evaluation model is proposed to evaluate the accuracy of the two proposed approaches. The proposed approaches are evaluated against the existing Piecewise Linear Rubber sheeting method on the Sri Lankan road vector data on four test scenarios consist of road segments with different misalignments. The research shows that the proposed bézier approach and distance ratio and angle approach has an average misalignment reduction percentage of 75.8% and 73.6% respectively whereas the piecewise linear rubber-sheeting method has an average misalignment reduction percentage of 68.6%.