Exploring NeRF for automated 3D phenotyping in greenhouse: a promising direction in shape measurement and analysis

3D plant phenotyping is crucial for understanding plant growth and environmental responses.
Traditional methods, like structure from motion (SfM) and light detection and ranging (LiDAR), have limitations in resolution, noise sensitivity, and computational efficiency.
We propose a novel, automated pipeline to plant phenotyping using neural radiance fields (NeRF), a deep learning technique that models continuous volumetric scene representations.
Incorporating a 6-DOF robotic manipulator, our method automates the data acquisition process, significantly reducing manual labor and human error.
NeRF overcomes the limitations of SfM and LiDAR by offering better handling of computation time, better results in greenhouse environment, and robustness on sparse input data.
Our work demonstrates the potential of NeRF as a valuable tool for plant phenotyping, advancing our understanding and quantification of plant morphology and growth.
This preliminary exploration lays the groundwork for future research and development in plant phenotyping and shape analysis, potentially benefiting from advancements in other cutting-edge techniques such as neural implicit surfaces.

International Symposium on New Technologies for Sustainable Greenhouse Systems: GreenSys2023

H.B. Choi, H.I. Lee, H.J. Lee, S.H. Park, T.S. Lee

3D phenotyping, neural radiance fields, neural implicit representation, point cloud, growth measurement

https://doi.org/10.17660/ActaHortic.2025.1426.2