TrailScan is a HAWK Research Project funded by Sattelmuehle Foundation
TrailScan is QGIS Plugin designed to apply a deep learning model to detect and segment skid trails (logging trails) from Airborne Laser Scanning (ALS) data. In Central European mixed and deciduous forests, permanently designed skid trail networks are used by ground-based forest operations, e.g. harvesters, forwarders and other heavy vehicles.
ALS data can be collected using aircraft, helicopters, or UAVs equipped with LiDAR sensors. To ensure compatibility with TrailScan, we recommend the following:
- Format: LAS or LAZ format with a valid georeference (otherwise, the coordinate reference system of the QGIS project will be applied)
- Point Classification: Follow the standard LAS/LAZ classification scheme.
- Ground points (class 2) are required for preprocessing, vegetation points need to be labeled in at least one different class.
- If ground classification is missing, it can be generated using PDAL or equivalent tools.
- Point Density:
- Minimum: 5 points/m²
- Recommended point density: 10-20 points/m² (ensures reliable DTM, CHM, LRM, and VDI derivation, while higher point densities may increase processing time significantly)
Tested ALS Data Sources (Germany)
| Region | Data Portal | Notes |
|---|---|---|
| Bavaria | Geoportal Bayern | Direct LAZ downloads available |
| Hessia | Geodaten Online Hessen | Free LAZ data via "Produkte", limited to 1km² per download |
| Thuringia | Geoportal Thüringen | Direct LAZ downloads available |
| Saxony | Geoportal Sachsen | Direct LAZ downloads available |
| Brandenburg | GeoBasis Brandenburg | Direct LAZ downloads available |
| Rhineland-Palatinate | GeoShop RLP | Point clouds split into terrain (lpg) and objects (lpo) - merge in QGIS |
| NRW | OpenGeodata.NRW | Direct LAZ downloads available |
- Input: ALS point clouds in
.lazor.lasformat are opened directly in QGIS. - The preprocessing tool converts the point cloud into a 4-band georeferenced raster image:
- Band 1: Digital Terrain Model (DTM)
- Band 2: Canopy Height Model (CHM)
- Band 3: Micro-Relief Model (MRM)
- Band 4: Vegetation Density Index (VDI)
- All raster values are normalized to a range between 0 and 1, which is why the output is named the Normalized File.
- Input: the Normalized File generated in step 1.
- Download the TrailScan model here and save it on your local computer:
- Add the path to the TrailScan Model file saved on your computer.
- The TrailScan model processes the Normalized File and produces a Trailmap.
Trailmap is a prediction raster with values between 0 and 1:
0= no skid trail>0= probability of a skid trail (the higher the value, the more likely a trail is present).
Quick Start (Windows Users)
-
Install QGIS via the OSGeo4W Network Installer, e.g. via https://qgis.org/download
-
Choose "Quick Installation" – this should automatically include Python and PDAL
-
If you want to check the PDAL version used by QGIS, you might open the OSGeo4W Shell (from the Start Menu) and run:
pdal --version
Notes
-
Do not use the standard Windows Command Prompt or PowerShell for PDAL checks unless you installed PDAL system-wide.
-
On Linux, you may simply use:
pdal --version -
Currently we do NOT recommend the use of TrailScan on MacOS, see Issue #15
In QGIS, navigate to 'Plugins' --> 'Manage and Install Plugins...'
- Search for 'TrailScan' and click on 'Install Plugin'
- -Aditional Python packages are installed automatically with the plugin installation:
numpyscipylaspylazrsrasterioonnxruntime
Important: You may need to restart QGIS after TrailScan Plugin installation.
- CPU: Multi-core processor (Intel i7/i9 or AMD Ryzen recommended).
- RAM: Minimum 16 GB (32 GB or more recommended for large point clouds).
- GPU: Not required, but ONNX Runtime can optionally leverage GPU acceleration if supported drivers are installed.