Creation of a 3D surface-point cloud from photographs taken by a non-RTK UAV to track beech-canopy changes in mountainous areas: Evaluation of clarity and accuracy when used in combination with a defined shooting method and marker points

• Published in: Japanese Journal of Conservation Ecology p. 2234
• Main Authors: Yamane, Masanobu, Suzuki, Toru, Amemiya, Tamotsu
• Format: Journal Article
• Language: Japanese
• Published: The Ecological Society of Japan 2024
• Subjects: aerial photography; broad-leaved forest; leaf feeding; SfM; shooting method
• ISSN: 1342-4327; 2424-1431
• DOI: 10.18960/hozen.2234

To monitor the condition of beech canopies in a reproducible, recordable, and labour-efficient manner, we investigated flight planning, aerial photography, and structure-from-motion (SfM) processing techniques to generate orthoimages with accurate positioning using non-real-time kinematic (RTK) unmanned aerial vehicles (UAVs) in mountainous regions where Internet access is restricted. The automated UAV flight maintained a consistent altitude. The data encompassed a 20-ha region on the summits of Mt. Tanzawa and Mt. Hirugatake in the Tanzawa Mountains, and were acquired during July and August of 2021 and 2022, with a ground resolution of 2 cm/pixel. The directly downward photos had a parallel perspective with overlap and side-wrap percentages of 80% and 60%, respectively. The oblique-perspective images had a parallel view, with the camera lens tilted 20° or 30° forward, accompanied by overlap and side-wrap percentages of approximately 40% and 30%, respectively. Ground control points (GCPs) and verification points were established using the precise point positioning (PPP)-RTK method with a dual-frequency global navigation satellite system (GNSS). A virtual reference station received correction signals via the quasi-zenith Michibiki satellite. The GCPs were located in areas with an unobstructed view of the sky, while the validation points were placed at terrestrial features, such as on a staircase or bench along a mountain trail. The study involved manipulating aerial photographs and using GCPs for SfM processing, with variable SfM processing quality. The objective was to compare the clarity and positional accuracy of the orthoimages and a three-dimensional (3D) surface-point cloud obtained through SfM processing. High-quality SfM processing and GCP correction using both nadir and oblique imagery produced colour orthoimages and a 3D surface-point cloud with spatial coordinate accuracy within 0.3 m. The resulting images had satisfactorily clear views of exposed branches and missing canopy material, ensuring dependable canopy identification and tracking. This method proved effective for monitoring and comparing individual trees over time, although some issues remain, such as lens calibration and identifying the optimal GCP locations.