Rigorous and extensive accuracy assestment of automatically classified LiDAR data: a case study in the city of Milan, Italy
LiDAR data filtering has been an active research area for nearly thirty years and continues to present significant challenges due to the increasing density of acquired LiDAR data. This study analysed aerial LiDAR data from Milan, characterised by a density of 20-30 pts/m<sup>2</sup>. Ini...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2025-07-01
|
| Series: | The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| Online Access: | https://isprs-archives.copernicus.org/articles/XLVIII-G-2025/1007/2025/isprs-archives-XLVIII-G-2025-1007-2025.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | LiDAR data filtering has been an active research area for nearly thirty years and continues to present significant challenges due to the increasing density of acquired LiDAR data. This study analysed aerial LiDAR data from Milan, characterised by a density of 20-30 pts/m<sup>2</sup>. Initiated in the summer of 2022, the survey included aerial and terrestrial surveys. Aerial LiDAR data was captured at a minimum of 20 pts/m<sup>2</sup> using the sensor Leica CityMapper-2S, while mobile data were acquired using Cyclomedia’s MMS system across 2555 km of roads. The ALS dataset includes Milan’s provincial territory and features unclassified and automatically classified point clouds in an industrial environment. Two areas, San Siro and Città Studi, were selected to create a ground truth without automated methods. We created a comprehensive ground truth dataset to validate our filtering method through valid and well-known algorithms like TerraSolid, and that obtained in an industrial environment where expert users applied these algorithms under time constraints. Our classification achieved 95.8% accuracy in the San Siro area and 94.6% in the Città Studi district, while the classification of the industrial environment obtained 93.7% and 88.9%, respectively. In the future, we intend to refine parameters to improve automatic classification accuracy and extend the process to other areas in Milan, integrating deep learning algorithms within the MATLAB environment. |
|---|---|
| ISSN: | 1682-1750 2194-9034 |