Evaluation of Flight Parameters in UAV-based 3D Reconstruction for Rooftop Infrastructure Assessment

TL;DR

This paper tackles rooftop infrastructure assessment via UAV-based photogrammetry and seeks to optimize flight parameters to balance reconstruction accuracy with flight efficiency. Using a controlled study over a multi-segment rooftop, it varies ground sampling distance (GSD) and image overlap (OL) and validates photogrammetric models against ground-truth point clouds from UAV-based LiDAR and terrestrial laser scanning (TLS). The results show that a GSD of $0.75-1.26$ cm with aic 85% OL yields high model accuracy while reducing image captures and flight time, offering actionable guidance for autonomous rooftop surveys. The work also demonstrates UAV-based LiDAR as a robust ground-truth source for photogrammetry and outlines implications for planning efficient, large-scale rooftop inspections.

Abstract

Rooftop 3D reconstruction using UAV-based photogrammetry offers a promising solution for infrastructure assessment, but existing methods often require high percentages of image overlap and extended flight times to ensure model accuracy when using autonomous flight paths. This study systematically evaluates key flight parameters-ground sampling distance (GSD) and image overlap-to optimize the 3D reconstruction of complex rooftop infrastructure. Controlled UAV flights were conducted over a multi-segment rooftop at Queen's University using a DJI Phantom 4 Pro V2, with varied GSD and overlap settings. The collected data were processed using Reality Capture software and evaluated against ground truth models generated from UAV-based LiDAR and terrestrial laser scanning (TLS). Experimental results indicate that a GSD range of 0.75-1.26 cm combined with 85% image overlap achieves a high degree of model accuracy, while minimizing images collected and flight time. These findings provide guidance for planning autonomous UAV flight paths for efficient rooftop assessments.

Figures (6)

• Figure 1: LiDAR point cloud of Ellis Hall, Kingston, Ontario with labeled roof sections and height above ground level (AGL).
• Figure 2: Recall scalar field for a cloud-to-cloud comparison of the UAV-based LiDAR and TLS point clouds.
• Figure 3: At the same OL, Flight 6 produces a more accurate model than Flight 4. Thin-walled surfaces feature less occlusions, seen in the exhaust ventilator on the left, and complex geometry was represented better, seen in the walkway on the right.
• Figure 4: RGB (left), precision (middle), and recall (right) scalar fields for three of the test flights. The model from Flight 2 data produced large regions of high error (> 10 cm), and was unable to recreate simple building geometry. With higher overlap percentages, the error becomes primarily concentrated around regions of high detail, and the lower roof sections with poorer lighting and greater obstruction. Note that the rectangular area of high error on the observatory dome was due to rotation of the telescope prior to collection of the GT, however this is consistent across all test flights.
• Figure 5: Precision at threshold values up to 6 cm for roof sections B, C, and D (left to right). For rooftops with less features, such as Roof C, the models remain fairly similar across different threshold values. As rooftop complexity is introduced, models begin to spread out.