Image Matching of Holographic Mapping Drones Based on Multiple Mechanisms

Drone images have the characteristics of low overlap, large format, and large data volume. The extracted features are massive and uneven, which poses a challenge for image matching tasks. This article proposes a Multi LoFTR method. Firstly, a multi matching architecture is designed, which introduces the idea of local feature matching multiple times. In the newly added matching process, the feature reconstruction module has been improved. Effectively improving the repeatability and accuracy of effective features within the region and reducing computational complexity. Finally, upgrade the fine-grained matching paradigm to suppress fuzzy matching while being compatible with input image resolution differences. In the testing of the drone specific dataset, the algorithm was able to maintain a matching accuracy of 90.93% in challenging scenarios such as low texture, repetitive structure, and complex terrain, providing good support for 3D reconstruction.