# An Annotated Multi-Platform Laser Scanning Dataset (MLS, ULS, TLS) for Semantic Segmentation of Point Clouds in Complex Forest Environments

---
**Authors:** Felix Rocco Matzke, Anette Eltner

**Contributors:** Robert Krüger, Hanne Hendrickx, Oliver Grothum, Isabell Raab, Almut Jäschke

**Corresponding author:** Felix Rocco Matzke

**Contact information:** felix_rocco.matzke@tu-dresden.de, TU Dresden

---
## Overview

This dataset contains five annotated LiDAR point clouds acquired in the Richterschlüchte area, Saxon Switzerland, Germany, in March 2023, using three laser scanning platforms: mobile laser scanning (MLS), uncrewed aerial vehicle laser scanning (ULS), and terrestrial laser scanning (TLS). All point clouds are manually annotated with three semantic classes: trunk, ground, and miscellaneous. The MLS point cloud is partitioned into training, validation, and test subsets, while the TLS and ULS point clouds are provided to assess generalization across varying sensing systems. The dataset is intended for developing and testing deep learning approaches for the semantic segmentation of point clouds in complex forest environments.


## Study Area

The study area is located in the southern part of the Saxon Switzerland National Park, in the Richterschlüchte region. The data cover forested areas with scattered rock formations. Parts of the area were affected by the large-scale wildfires of 2022. The combination of partial fire disturbance and complex terrain makes the scene a challenging environment for point cloud semantic segmentation.


## Dataset Files

The dataset consists of five point clouds in LAZ format:

* mls_train.laz - MLS point cloud, training subset

* mls_validation.laz - MLS point cloud, validation subset

* mls_test.laz - MLS point cloud, test subset

* uls.laz - ULS point cloud

* tls.laz - TLS point cloud

with the following point attributes:

* MLS: Classification

* ULS: RGB, Classification, Intensity, Return Number, Number of Returns, GPS Time, Scan Angle Rank

* TLS: Classification, Intensity, Return Number, Number of Returns, GPS Time, Amplitude, Reflectance, Deviation


**Classification**

trunk (0), ground (1), and miscellaneous (2)


**Coordinate reference system**

horizontal: ETRS89 / UTM zone 33N (EPSG:25833)

vertical: Ellipsoidal heights on the GRS80 ellipsoid (not heights above sea level)


## Dataset Statistics

The table below lists the total number of points per point cloud and the
relative class distribution. The class distribution differs considerably
between platforms, reflecting the different viewing geometries.

| Dataset        | Total points | Trunk (%) | Ground (%) | Miscellaneous (%) |
|----------------|-------------:|----------:|-----------:|------------------:|
| mls_train      |    5,242,067 |      20.3 |       51.0 |              28.7 |
| mls_validation |      929,334 |      21.4 |       41.7 |              36.9 |
| mls_test       |    1,165,025 |       7.4 |       38.5 |              54.1 |
| tls            |    5,396,608 |      18.7 |       24.6 |              56.6 |
| uls            |   12,775,143 |       1.7 |       57.2 |              41.1 |


## Acquisition and Processing

The MLS dataset was collected along a hiking trail in the Richterschlüchte, and the ULS and TLS datasets contain sections of the nearby ridge.

* MLS: GeoSLAM ZEB handheld scanner; processed with GeoSLAM Connect v1.7 (LiDAR-inertial SLAM).

* ULS: DJI Zenmuse L1 mounted on DJI Matrice 300 RTK at 80 m altitude above start position, 6.3 m/s flight speed, 85% cross-overlap, scan rate of 160 kHz in repetitive scanning mode, up to three returns per emitted pulse (triple return); processed with DJI Terra.

* TLS: RIEGL VZ-400i; 250,000 measurements/s; vertical angle range 30°–130°; processed with RiSCAN PRO.

* Labeling: manual annotation in CloudCompare v2.13.2.


## License and Citation

Please cite this dataset using DOI: https://doi.org/10.25532/OPARA-1576 and adhere to the CC BY 4.0 license (https://creativecommons.org/licenses/by/4.0/).

The point clouds were annotated and used within the project RTG-AirMetro (2947), funded by the German Research Foundation (DFG, grant number 508591287).

A related manuscript using this data is currently under review.