Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TreeDGS: Aerial Gaussian Splatting for Distant DBH Measurement

Belal Shaheen, Minh-Hieu Nguyen, Bach-Thuan Bui, Shubham, Tim Wu, Michael Fairley, Matthew David Zane, Michael Wu, James Tompkin

TL;DR

TreeDGS tackles the problem of estimating tree diameter at breast height (DBH) from stand-off UAV RGB imagery in pixel- and view-limited forests. It combines a SfM/OpenMVS initialization with 3D Gaussian Splatting (RaDe-GS) to produce a dense, opacity-informed surface representation, which is then densified into trunk-focused samples via depth-aware sampling and multi-view reliability cues. A trunk-segmentation step isolates stem geometry, and an opacity-weighted, slice-wise solid-circle RANSAC with height-wise taper regression yields accurate DBH estimates at h_BH, outperforming a UAV LiDAR baseline in field tests across 10 plots (RMSE 4.79 cm, MAE 3.70 cm). The approach demonstrates that dense, opacity-guided Gaussian splat geometry can enable accurate, low-cost aerial DBH measurement from RGB imagery, with potential extensions to joint trunk primitives and end-to-end supervision.

Abstract

Aerial remote sensing enables efficient large-area surveying, but accurate direct object-level measurement remains difficult in complex natural scenes. Recent advancements in 3D vision, particularly learned radiance-field representations such as NeRF and 3D Gaussian Splatting, have begun to raise the ceiling on reconstruction fidelity and densifiable geometry from posed imagery. Nevertheless, direct aerial measurement of important natural attributes such as tree diameter at breast height (DBH) remains challenging. Trunks in aerial forest scans are distant and sparsely observed in image views: at typical operating altitudes, stems may span only a few pixels. With these constraints, conventional reconstruction methods leave breast-height trunk geometry weakly constrained. We present TreeDGS, an aerial image reconstruction method that leverages 3D Gaussian Splatting as a continuous, densifiable scene representation for trunk measurement. After SfM-MVS initialization and Gaussian optimization, we extract a dense point set from the Gaussian field using RaDe-GS's depth-aware cumulative-opacity integration and associate each sample with a multi-view opacity reliability score. We then estimate DBH from trunk-isolated points using opacity-weighted solid-circle fitting. Evaluated on 10 plots with field-measured DBH, TreeDGS reaches 4.79,cm RMSE (about 2.6 pixels at this GSD) and outperforms a state-of-the-art LiDAR baseline (7.91,cm RMSE), demonstrating that densified splat-based geometry can enable accurate, low-cost aerial DBH measurement.

TreeDGS: Aerial Gaussian Splatting for Distant DBH Measurement

TL;DR

TreeDGS tackles the problem of estimating tree diameter at breast height (DBH) from stand-off UAV RGB imagery in pixel- and view-limited forests. It combines a SfM/OpenMVS initialization with 3D Gaussian Splatting (RaDe-GS) to produce a dense, opacity-informed surface representation, which is then densified into trunk-focused samples via depth-aware sampling and multi-view reliability cues. A trunk-segmentation step isolates stem geometry, and an opacity-weighted, slice-wise solid-circle RANSAC with height-wise taper regression yields accurate DBH estimates at h_BH, outperforming a UAV LiDAR baseline in field tests across 10 plots (RMSE 4.79 cm, MAE 3.70 cm). The approach demonstrates that dense, opacity-guided Gaussian splat geometry can enable accurate, low-cost aerial DBH measurement from RGB imagery, with potential extensions to joint trunk primitives and end-to-end supervision.

Abstract

Aerial remote sensing enables efficient large-area surveying, but accurate direct object-level measurement remains difficult in complex natural scenes. Recent advancements in 3D vision, particularly learned radiance-field representations such as NeRF and 3D Gaussian Splatting, have begun to raise the ceiling on reconstruction fidelity and densifiable geometry from posed imagery. Nevertheless, direct aerial measurement of important natural attributes such as tree diameter at breast height (DBH) remains challenging. Trunks in aerial forest scans are distant and sparsely observed in image views: at typical operating altitudes, stems may span only a few pixels. With these constraints, conventional reconstruction methods leave breast-height trunk geometry weakly constrained. We present TreeDGS, an aerial image reconstruction method that leverages 3D Gaussian Splatting as a continuous, densifiable scene representation for trunk measurement. After SfM-MVS initialization and Gaussian optimization, we extract a dense point set from the Gaussian field using RaDe-GS's depth-aware cumulative-opacity integration and associate each sample with a multi-view opacity reliability score. We then estimate DBH from trunk-isolated points using opacity-weighted solid-circle fitting. Evaluated on 10 plots with field-measured DBH, TreeDGS reaches 4.79,cm RMSE (about 2.6 pixels at this GSD) and outperforms a state-of-the-art LiDAR baseline (7.91,cm RMSE), demonstrating that densified splat-based geometry can enable accurate, low-cost aerial DBH measurement.
Paper Structure (27 sections, 16 equations, 10 figures, 2 tables)

This paper contains 27 sections, 16 equations, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials
  3. UAV RGB imagery.
  4. UAV LiDAR
  5. Manual field-to-reconstruction association.
  6. Proposed Pipeline
  7. Problem Statement
  8. Structure-from-Motion and Multi-View Stereo
  9. Coarse-to-fine matching.
  10. SfM reconstruction.
  11. Coarse densification with OpenMVS openmvs_github.
  12. Reconstruction with Gaussian Splats
  13. Gaussian parameterization.
  14. RaDe-GS depth-plane formulation.
  15. Adaptive training with RLGS.
  16. ...and 12 more sections

Figures (10)

  • Figure 1: TreeDGS: RGB-only for DBH estimation with 3D Gaussian Splatting. (a) High-fidelity TreeDGS reconstruction from distant UAV RGB imagery as an optimized set of 3D Gaussians. (b) Surface points extracted via opacity-consistent sampling (built on RaDe-GS zhang2024rade) and segmented into stem vs. vegetation to isolate trunk geometry for DBH fitting. (c) DBH errors against field measurements, shown as the distribution of DBH estimates relative to the field DBH distribution; TreeDGS + opacity-weighted circle fitting reduces error vs. UAV LiDAR + cylinder fitting malladi2025digiforests (RMSE/MAE: 4.79/3.67 cm vs. 7.91/5.04 cm) at ground sample distance (GSD) of approximately 1.84 cm.
  • Figure 2: Pixel-limited trunk observations from distant aerial imagery. At $\sim$70 m altitude ($\mathrm{GSD}\approx 1.84$ cm/px), a typical pine stem can occupy only $\sim$13 pixels across in a single RGB image, making per-image diameter cues highly quantized, sensitive to occlusion, and difficult to measure precisely.
  • Figure 3: Region of interest (ROI) and plot layout (10 subplots). Each subplot corresponds to one 0.2-acre circular plot (radius 16.05 m). The circle indicates the field plot boundary used for tree inclusion, while the dashed polygon outlines the ROI used to clip and organize aerial products for per-plot processing.
  • Figure 4: RGB and LiDAR acquisition patterns. (a) RGB imagery was collected with high overlap and mixed viewing angles to support SfM/MVS and improve trunk visibility. (b) LiDAR was flown with dense grid (lawnmower) flight lines to obtain uniform coverage across the plot network.
  • Figure 5: TreeDGS pipeline. RGB images are reconstructed with SfM/MVS and represented as 3D Gaussian splats $G$. We surface-sample dense points using opacity $\alpha$ and weights $w$, segment trunks $T$, and estimate DBH by opacity-weighted circle fitting on multiple trunk slices.
  • ...and 5 more figures