Coordinate georeferencing of geoimages using computer vision

Abstract

This paper presents the process of creating a model for automated georeferencing of geo- images using image matching technology, an increasingly popular computer vision concept that has not received proper attention in cartography to date. LoFTR (Local Feature Matching with Transformers), a new approach to find and match key points in pairs of images was used as a basis for the developed model. At the first stage, a model architecture and successive stages of input data processing were defined. At the second stage, the model was implemented as a program in accordance with the previously outlined steps. At the third stage, the resulting model was tested on several pairs of geoimages to further evaluate its effectiveness and applicability in various scientific tasks. Results show that the developed model provides a universal algorithm for automated georeferencing of geo-images, demonstrating high-quality results