An Automated System for Image-to-Vector Georeferencing

Although modern imagery usually has latitude and longitude or similar coordinates to provide true world location, much potentially useful imagery lacks this information. This includes historical imagery, as well as modern imagery that may have lost its locational information through inappropriate processing. Its provision through the process of georeferencing is still primarily a manual procedure. This paper proposes an efficient, fully- automated solution for massively asymmetric image-to-vector georeferencing whereby an image of a relatively small geographic area is automatically located relative to a substantially larger vector map base. For control points, road intersections are automatically extracted from high resolution aerial or satellite imagery using the new concepts of reference circles and central pixels. For automated control point pair identification between image and vector map, an invariant point pattern matching approach is proposed based on shape invariance for similarity transforms and invariant area ratios for affine transforms. The matching algorithm necessitates only a small subset of image points and requires no additional information beyond pixel and map coordinates. Further, it tolerates inaccurate, missing and spurious points, and provides high performance with linear scalability. A final step performs transformation verification, globalization and optimization based upon an Iterative Closest Point Greedy algorithm. Experimentation shows that images covering a few city blocks with as few as 6 to 17 extracted road intersection points can be efficiently and correctly located using the road network of Dallas County, Texas, with over 80,000 intersections.