Digital Tree Twins: Detailed Reconstruction from Point Clouds using a Skeletonization Approach

Trees have a strong effect on the local wind climate. To better understand their impact, an accurate and detailed reconstruction of botanical trees into digital twins from terrestrial LiDAR scan point clouds is important. However, capturing the complex, multi-scale nature of tree structures poses significant challenges. Issues such as gaps in the model due to occlusion in the point cloud data and inaccuracies in branch thickness estimations — especially for smaller branches — are prevalent limitations. Most advanced reconstruction methods today, such as TreeQSM (Raumonen et al., 2013), have been primarily designed for forestry applications, such as volume and biomass estimation. However, numerical flow simulations pose additional requirements including the need for a closed and continuous surface.

This study introduces a different approach, building upon the work of Bærenzten et al., 2023, using tools from the field of computer graphics. The proposed method initially creates a graph from the point cloud by connecting nearby points. Subsequently, a highly detailed skeleton of the tree is generated using the so-called local separators approach (Bærenzten et al., 2021). Local separators are defined as collections of vertices that are contained within a sub-graph of the original graph. The removal of a local separator splits the sub-graph into multiple smaller sub-graphs. The branch diameters are subsequently determined using a hybrid method that blends data-driven estimates derived from the point cloud data with the Da Vinci rule for trees, which defines a relationship between the diameters of a mother branch and its daughter branches. Additionally, species-specific data obtained from direct diameter measurements is incorporated in the estimation process. The tree’s surface is then reconstructed by first generating an implicit representation from which a closed mesh is extracted as an iso-surface.

Through a parameter study, the two main parameters for the generation of the skeleton, as well as the two main parameters influencing the branch thickness estimation, were studied in detail. The algorithm effectively handles occlusion in the point cloud, producing fully connected branching structures. The combined approach notably enhances the branch thickness estimation compared to using only one approach. We demonstrate the robustness of the method by applying it to three trees of very different dimensions, complexities, and point cloud characteristics and outline how the finally reconstructed tree will be used in atmospheric flow simulations.

 

 

References

Raumonen, P., Kaasalainen, M., Åkerblom, M., Kaasalainen, S., Kaartinen, H., Vastaranta, M., Holopainen, M., Disney, M., & Lewis, P. (2013). Fast Automatic Precision Tree Models from Terrestrial Laser Scanner Data. Remote Sensing, 5, 491-520. https://doi.org/10.3390/rs5020491

Bærentzen, J. A., Villesen, I. B., & Dellwik, E. (2023). Reconstruction of a Botanical Tree from a 3D Point Cloud. In E. Christiani, M. Falcone, & S. Tozza (Eds.), Mathematical Methods for Objects Reconstruction: From 3D Vision to 3D Printing (Vol. 54, pp. 103-120). Springer. https://doi.org/10.1007/978-981-99-0776-2\_4

Bærentzen, A., & Rotenberg, E. (2021). Skeletonization via Local Separators. ACM Transactions on Graphics, 40(5), Article 187. https://doi.org/10.1145/3459233