UAV localisation under linear mapping for vision-based navigation

In the process of building vision based navigation systems, UAV self-localisation by image registration against a database is a challenging issue. The issue arises from the requirement to register a scene from an aerial image with a geo-referenced image. This is typically done by matching features across multiple regions in the aerial image in the presence of registration errors. A statistical approach is used here to incorporate all available information. Specifically, we derive a weighted least squares approach to provide the optimal affine transformation to match the scenes. The performance of the proposed algorithm is compared with existing approaches using simulated examples.