EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
2094. Geometry and deep learning-based methods comparison for segmentation of trees in aerial laser scan point clouds
Invited abstract in session TC-28: Advancements of OR-analytics in statistics, machine learning and data science 6, stream Advancements of OR-analytics in statistics, machine learning and data science.
Tuesday, 12:30-14:00Room: 065 (building: 208)
Authors (first author is the speaker)
| 1. | Povilas Treigys
|
| Institute of Data Science and Digital Technologies, Vilnius University | |
| 2. | Kasparas Karlauskas
|
| Institute of Data Science and Digital Technologies, Image and Signal Analysis Group, Vilnius University |
Abstract
Laser scanning technology has facilitated advanced analysis of complex outdoor environments by capturing 3D point clouds. Identifying points within a point cloud corresponding to individual objects of a particular class, semantic instance segmentation is of great relevance in numerous fields. In the context of forest point clouds, in particular, semantic instance segmentation proves valuable for forestry applications such as tree inventorization and infrastructure contexts such as transmission line corridor surveys, where the positioning of trees is used for safety considerations. While recent developments in deep learning algorithms for point cloud processing have yielded solutions for semantic instance segmentation in terrestrial laser scans of forests, applying such methods to lower spatial resolution aerial laser scans still needs to be explored. Compared to their terrestrial counterparts, aerial laser scans present an attractive prospect for large-scale forest analysis. This paper introduces a benchmark dataset for forest semantic instance segmentation created from publicly available aerial laser scans. The dataset is an evaluation platform for standard geometry-based classical tree instance segmentation algorithms. Additionally, it is used to evaluate a deep learning model for tree instance segmentation. A comparison between the different methods is made in the context of aerial laser scans of forests.
Keywords
- Artificial Intelligence
- Machine Learning
- Computer Science/Applications
Status: accepted
Back to the list of papers