CREATING THREE-DIMENSIONAL GAME LEVELS BASED ON THE RECONSTRUCTION OF AERIAL PHOTOGRAPHS
DOI:
https://doi.org/10.34185/1991-7848.itmm.2026.01.068Keywords:
game development, aerial photograph, three-dimensional game level, Unity, game level generationAbstract
This paper focuses on automated generation of three-dimensional game levels based on the reconstruction of a single aerial photograph, as well as development of the corresponding method and software. The method combines monocular depth estimation using Depth Anything V2, semantic object segmentation using Segment Anything Model 3, and specialized geometry extraction algorithms for objects of four classes – buildings, roads, trees, and water objects. The proposed method also employs a coordinate transformation system that ensures a unified transition from the two-dimensional coordinate system of the image to the three-dimensional coordinate system
of the game level.
The software implementation is based on a microservice architecture with five containerized Docker services, providing logical separation of components and system scalability. Python and C# were selected as the programming languages, and Unity was used as the target game engine. The proposed approach enables full automation of the process – from image uploading to obtaining a ready-to-use game level.
References
Newlands C., Zauner K.-P. Procedural Generation and Rendering of Realistic, Navigable Forest Environments: An Open-Source Tool. Computer Graphics Forum. 2022. URL: https://doi.org/10.48550/arXiv.2208.01471.
Sketch2Scene: Automatic Generation of Interactive 3D Game Scenes from User's Casual Sketches / Y. Xu та ін. arXiv preprint arXiv:2408.04567. 2024.
Berrezueta-Guzman S., Koshelev A., Wagner S. From Reality to Virtual Worlds: The Role of Photogrammetry in Game Development. 2025 IEEE Gaming, Entertainment, and Media Conference (GEM), м. Kaohsiung, Taiwan, 16–18 лип. 2025 р. 2025. С. 1–6. URL: https://doi.org/10.1109/gem66882.2025.11155764.
.RealFusion 360° Reconstruction of Any Object from a Single Image / L. Melas-Kyriazi та ін. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), м. Vancouver, BC, Canada, 17–24 черв. 2023 р. 2023. URL: https://doi.org/10.1109/cvpr52729.2023.00816.
Depth Anything V2 / L. Yang та ін. Advances in Neural Information Processing Systems. 2024. URL: https://doi.org/10.48550/arXiv.2406.09414.
Carion N. SAM 3: Segment Anything with Concepts. 2025. URL: https://doi.org/10.48550/arXiv.2511.16719.
