Illumination compensation for classification of hyperspektral 3d point clouds [Beleuchtungsausgleich zur klassifikation auf hyperspektral annotierten 3d-punktwolken]
Autor(en): | Wiemann, T. Igelbrink, F. Mitschke, I. Hertzberg, J. |
Stichwörter: | Hyperspectral imaging; Machine learning; Terrestrial laser scanning | Erscheinungsdatum: | 2021 | Herausgeber: | VDE VERLAG GMBH | Journal: | AVN Allgemeine Vermessungs-Nachrichten | Volumen: | 128 | Ausgabe: | 3 | Startseite: | 147 | Seitenende: | 157 | Zusammenfassung: | In terrestrial laser scanning different kinds of cameras can be used to annotate the collected 3D data with additional modalities. In this paper, we describe a novel laser scanning system that features a co-calibrated hyperspectral camera. This camera allows to capture high resolution spectral images over a large portion of the electromagnetic spectrum. The collected spectral images encode the reflectance of different materials visible in a captured scene. In contrast to remote sensing, in terrestrial scanning the lighting conditions may vary drastically during the capture of a single scan due to the current viewing direction of the laser scanner, leading to large differences in the measured intensities dependent on the respective viewing angle. In this paper, we discuss these problems and present first results to compensate such intensity differences to allow reliable classification of materials within the scans using established techniques from machine learning to segment 3D point clouds. © 2021, VDE VERLAG GMBH. All rights reserved. |
ISSN: | 00025968 | Externe URL: | https://www.scopus.com/inward/record.uri?eid=2-s2.0-85110488381&partnerID=40&md5=c8e16cd1910b809f1be33c3b5269fd76 |
Show full item record