Application of neural network models in the mapping of telecommunications infrastructure objects

• Autorzy: Lewandowski Bartosz , Wilk Łukasz , Zachar Paulina

Identyfikatory

DOI

10.14681/apcrs-2023-002

Warianty tytułu

PL

Zastosowanie modeli sieci neuronowych w odwzorowaniu obiektów infrastruktury telekomunikacyjnej

• Języki publikacji: EN PL

Abstrakty

EN

Neural Radiance Fields (NeRF) is another 3D reconstruction method developed in recent years using artificial intelligence. This paper focuses on the study of object reconstruction using NeRF in the representation of objects such as telecommunication masts. Experiments were conducted using the Mega-NeRF model and two models (Nerfacto and Nerfacto-big) provided by the Nerfstudio framework on a UAV dataset. Various models and training parameters were tested, and the results were compared with reference data obtained from UAV photogrammetry and TLS laser scanning. The final analysis of the accuracy of the point clouds generated by the NeRF models indicated that they were of similar quality to the reference data, with slight differences in density and accuracy for different models and settings. The potential of NeRF methods for reconstructing 3D objects was demonstrated, especially in the context of mapping telecommunications masts, while noting the challenges associated with training parameters and the specifics of the analyzed object.

PL

Neural Radiance Fields (NeRF) to kolejna rozwijana w ostatnich latach metoda rekonstrukcji 3D wykorzystująca sztuczną inteligencję. W artykule skupiono się na badaniach odwzorowania obiektów za pomocą NeRF w reprezentacji obiektów takich jak maszty telekomunikacyjne. Przeprowadzono eksperymenty z wykorzystaniem modelu Mega-NeRF oraz dwóch modeli (Nerfacto i Nerfacto-big) udostępnionych przez Nerfstudio na zbiorze danych UAV. Przetestowano różne modele i parametry treningowe, a wyniki były porównywane z danymi referencyjnymi pozyskanymi z fotogrametrii UAV oraz skaningu laserowego TLS. Ostateczna analiza dokładności chmur punktów wygenerowanych przez modele NeRF wskazała na ich zbliżoną jakość do danych referencyjnych, z niewielkimi różnicami gęstości i dokładności dla różnych modeli i ustawień. Wykazano potencjał metod NeRF do rekonstrukcji obiektów 3D, zwłaszcza w kontekście odwzorowania masztów telekomunikacyjnych, jednocześnie zauważając wyzwania związane z parametrami treningowymi i specyfiką analizowanego obiektu.

Słowa kluczowe

EN

Nerf; neural networks; artificial intelligence; telecommunication mast; 3D models

PL

Nerf; sieci neuronowe; sztuczna inteligencja; maszt telekomunikacyjny; modele 3D

• Wydawca: Zarząd Główny Stowarzyszenia Geodetów Polskich
• Czasopismo: Archiwum Fotogrametrii, Kartografii i Teledetekcji
• Rocznik: 2023
• Tom: Vol. 35
• Strony: 39--66
• Opis fizyczny: Bibliogr. 16 poz.

Twórcy

autor

Lewandowski Bartosz

• Department of Photogrammetry, Remote Sensing and Spatial Information Systems, Faculty of Geodesy and Cartography, Warsaw University of Technology

autor

Wilk Łukasz

• Department of Photogrammetry, Remote Sensing and Spatial Information Systems, Faculty of Geodesy and Cartography, Warsaw University of Technology

autor

Zachar Paulina

• Department of Photogrammetry, Remote Sensing and Spatial Information Systems, Faculty of Geodesy and Cartography, Warsaw University of Technology

Bibliografia

• 1. Balloni, E., Gorgoglione, L., Paolanti, M., Mancini, A., & Pierdicca, R. (2023). Few shot photogrammetry: a comparison between NeRF and MVS-SfM for the documentation of cultural heritage. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 155-162.
• 2. Condorelli, F., Rinaudo, F., Salvadore, F., & Tagliaventi, S. (2021). A comparison between 3D reconstruction using nerf neural networks and MVS algorithms on cultural heritage images. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 43, 565-570.
• 3. Croce, V., Caroti, G., De Luca, L., Piemonte, A., & Véron, P. (2023). Neural radiance fields (NeRF): review and potential applications to digital cultural heritage. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 453-460.
• 4. Kniaz, V. V., Knyaz, V. A., Bordodymov, A., Moshkantsev, P., Novikov, D., & Barylnik, S. (2023). Double Nerf: Representing Dynamic Scenes as Neural Radiance Fields. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 115-120.
• 5. Martin-Brualla, R., Radwan, N., Sajjadi, M. S., Barron, J. T., Dosovitskiy, A., & Duckworth, D. (2021). Nerf in the wild: Neural radiance fields for unconstrained photo collections. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 7210-7219.
• 6. Mazzacca, G., Karami, A., Rigon, S., Farella, E. M., Trybala, P., & Remondino, F. (2023). NeRF for heritage 3D reconstruction. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 1051-1058.
• 7. Mildenhall, B., Srinivasan, P. P., Tancik, M., Barron, J. T., Ramamoorthi, R., & Ng, R. (2021). Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 65(1), 99-106.
• 8. Murtiyoso, A., & Grussenmeyer, P. (2023). Initial assessment on the use of state-of-the-art NeRF neural network 3d reconstruction for heritage documentation. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 1113-1118.
• 9. Murtiyoso, A., Markiewicz, J., Karwel, A. K., & Kot, P. (2023). Investigation on the Use of NeRF for Heritage 3D Dense Reconstruction for Interior Spaces. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences-ISPRS Archives, 48(1/W3-2023), 115-121.
• 10. Müller, T., Evans, A., Schied, C., & Keller, A. (2022). Instant neural graphics primitives with a multiresolution hash encoding. ACM Transactions on Graphics, 41(4), 1-15.
• 11. Nex, F., Zhang, N., Remondino, F., Farella, E. M., Qin, R., & Zhang, C. (2023). Benchmarking the extraction of 3D geometry from UAV images with deep learning methods. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 123-130.
• 12. Palestini, C., Basso, A., & Perticarini, M. (2022). Machine Learning as AN Alternative to 3D Photomodeling Employed in Architectural Survey and Automatic Design Modelling. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 191-197.
• 13. Tancik, M., Casser, V., Yan, X., Pradhan, S., Mildenhall, B., Srinivasan, P. P., ... & Kretzschmar, H. (2022). Block-nerf: Scalable large scene neural view synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 8248-8258).
• 14. Tancik, M., Weber, E., Ng, E., Li, R., Yi, B., Wang, T., ... & Kanazawa, A. (2023). Nerfstudio: A modular framework for neural radiance field development. In ACM SIGGRAPH 2023 Conference Proceedings (pp. 1-12).
• 15. Turki, H., Ramanan, D., & Satyanarayanan, M. (2022). Mega-nerf: Scalable construction of large-scale nerfs for virtual fly-throughs. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 12922-12931.
• 16. Vandenabeele, L., Häcki, M., & Pfister, M. (2023). Crowd-sourced surveying for building archaeology: the potential of structure from motion (SfM) and neural radiance fields (NERF). The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 1599-1605.