Wind during terrestrial laser scanning of trees: simulation-based assessment of effects on point cloud features and leaf-wood classification
LiDAR point cloud data of trees is often affected by wind-induced movements. This leads to misalignments between overlapping point clouds and distortions in the merged representation. Understanding these wind effects is crucial since they affect downstream tasks like tree parameter quantification an...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Chapter/Article Conference Paper |
| Language: | English |
| Published: |
10. Juli 2025
|
| In: |
ISPRS Geospatial Week 2025 "Photogrammetry & remote sensing for a better tomorrow ...", 6-11 April 2025, Dubai, United Arab Emirates (UAE)
Year: 2025, Pages: 25-32 |
| DOI: | 10.5194/isprs-annals-X-G-2025-25-2025 |
| Online Access: | Resolving-System, kostenfrei, Volltext: https://doi.org/10.5194/isprs-annals-X-G-2025-25-2025 Verlag, kostenfrei, Volltext: https://isprs-annals.copernicus.org/articles/X-G-2025/25/2025 
|
| Author Notes: | William Albert, Hannah Weiser, Ronald Tabernig, and Bernhard Höfle |
| Summary: | LiDAR point cloud data of trees is often affected by wind-induced movements. This leads to misalignments between overlapping point clouds and distortions in the merged representation. Understanding these wind effects is crucial since they affect downstream tasks like tree parameter quantification and leaf-wood separation. In this study, we investigate the impact of wind during multi-station Terrestrial Laser Scanning (TLS) acquisition on tree structure and leaf-wood classification by simulating TLS acquisitions of trees in both static and windy conditions using the LiDAR simulator HELIOS++. To assess wind effects, we compare the geometric features of leaf and wood points in each scenario and validate our simulations with real tree point clouds acquired in windy conditions. Finally, we train a Random Forest classifier for leaf-wood segmentation on both static and dynamic data and evaluate their performance on both datasets. Our results highlight that two of the nine geometric features are statistically significant in differentiating leaf and wood in windy conditions, compared to six features in static conditions. When trained with static data, leaf-wood classification results drop by ca. 10% intersection over union and decrease by ca. 4% overall accuracy from static to dynamic conditions. Further, we demonstrate that increasing the number of scan positions (i.e. from using one to merging 6 point clouds per tree) reduces classification success by ca. 25% in static conditions and ca. 35% in windy conditions. Our findings emphasize the need to account for wind effects in leaf-wood classification. We show that training on dynamic data can slightly improve classification of dynamic data (ca. 2%) compared to training on static data. |
|---|---|
| Item Description: | Gesehen am 14.07.2025 |
| Physical Description: | Online Resource |
| DOI: | 10.5194/isprs-annals-X-G-2025-25-2025 |