Multi-Drone 3D Reconstruction of Architectural and Structural Entities

• Daniel Morais IME
• Fabio Suim Chagas University of South-Eastern Norway
• Thiago Baldivieso IME
• Gabriela Dias IME
• Paulo Rosa IME

Resumo

In this paper, we analyze the results of 3D reconstructions of buildings and monuments using a multi-drone sys-tem with predefined parameters. The drones follow an elliptical flight path with the same average speed in opposite directions and different altitudes for data acquisition of RGB videos. We set the drone cameras at a specific angle to avoid occlusion areas. Then, for 3D reconstructions, we used Convolutional Neural Networks techniques, specifically NeRF and Gaussian Splatting, in two different case studies: a simulation environment and a field test. As a result, we obtained detailed and visually accurate 3D models.

Palavras-chave: Solid modeling, Three-dimensional displays, Accuracy, Buildings, Education, Neural radiance field, Software, Robots, Drones, Videos, 3D Reconstruction, CNN models, Gaussian Splatting, NeRF, Flight Path, Multi UAVs