This page contains the list of accepted presentations.You can use the controls in the side panel to order the list in various ways, and to toggle all the abstracts on or off, at once.

Downloadable material is attached to some of the presentations. The material is made available with the express consent of the respective presenters. All rights to the material are reserved by the the respective authors. If you want to, e.g., share, modify or build upon the work, please consult the respective authors for consent. Presentations with attached material are marked with the Downloadable tag.

If you are a presenter, you can view the presentation guidelines.

Wang, Di Towards an automated processing chain for 3D tree reconstructions from large scale TLS data Talk Downloadable
Di Wang1
(1) Aalto University, Finland

The advancements of tree Quantitative Structure Models (QSM) using Terrestrial Laser Scanning (TLS) data have enabled a new era of precise tree structure quantifications. Current QSM methods operate on the single tree scale, requiring plot-level TLS point clouds to be properly segmented and filtered. This requirement is particularly referring to the single tree segmentation and leaf-wood separation. Both processing steps are currently undertaken through laborious and time-consuming manual works. Several automated methods were respectively developed, but their performances were not assessed thoroughly on the accurate crown segmentation and leaf-wood separation.

In this contribution, we first introduce a unique synthetic TLS data set scanned over a highly realistic forest scene. Mesh models that contain more than two billion triangles represented a 50x50m plot of large deciduous trees with overlapping crowns. The simulated point cloud contained point-wise labels of tree IDs and leaf-wood components, serving as a unique reference set for future studies. Secondly, we present a new fully automatic algorithm that integrates the single tree segmentation and leaf-wood separation. The intuition was that these two steps could complement each other. Experiments showed that our novel method yielded an accuracy of 96.1% and 86.7% for detailed crown segmentation and leaf-wood separation, respectively.

Our study can be combined with QSM methods to frame a fully automatic processing chain for large scale 3D tree reconstructions using TLS data.