Estimating the Diameter at Breast Height of Trees in a Forest With a Single 360 Camera

TL;DR

This work tackles the challenge of obtaining accurate DBH measurements in forests without expensive LiDAR by leveraging a single consumer 360° camera and SfM photogrammetry. The authors integrate a semi-automatic pipeline comprising dense 3D reconstruction via Agisoft Metashape, trunk segmentation through Grounded SAM projections, and robust cross-section fitting with a RANSAC-based Fourier series model to estimate DBH from breast-height cross-sections. They validate on 61 acquisitions of 43 trees, achieving median absolute relative errors in the $5\%$–$9\%$ range, which is only 2–4% worse than LiDAR-based methods but at orders of magnitude lower cost and setup effort. A custom visualization tool supports inspection and validation, and the framework shows promise for scalable, low-cost forest inventories and carbon accounting, with future work focused on automating scaling and expanding validation to broader stands and species.

Abstract

Forest inventories rely on accurate measurements of the diameter at breast height (DBH) for ecological monitoring, resource management, and carbon accounting. While LiDAR-based techniques can achieve centimeter-level precision, they are cost-prohibitive and operationally complex. We present a low-cost alternative that only needs a consumer-grade 360 video camera. Our semi-automated pipeline comprises of (i) a dense point cloud reconstruction using Structure from Motion (SfM) photogrammetry software called Agisoft Metashape, (ii) semantic trunk segmentation by projecting Grounded Segment Anything (SAM) masks onto the 3D cloud, and (iii) a robust RANSAC-based technique to estimate cross section shape and DBH. We introduce an interactive visualization tool for inspecting segmented trees and their estimated DBH. On 61 acquisitions of 43 trees under a variety of conditions, our method attains median absolute relative errors of 5-9% with respect to "ground-truth" manual measurements. This is only 2-4% higher than LiDAR-based estimates, while employing a single 360 camera that costs orders of magnitude less, requires minimal setup, and is widely available.

Figures (5)

• Figure 1: Overview of the proposed 360° camera-based DBH estimation pipeline. 360° video data is collected using a Falcon UAV and a backpack-mounted sensor tower. Agisoft Metashape is used to extract and align images, generate point cloud and mesh, and scale the model. By projecting Grounded SAM masks onto the point cloud, we can extract individual tree trunks. We estimate DBH using Fourier-based fitting with a RANSAC framework on cross sections at breast height. Using an interactive interface, we can visualize and inspect reconstructions and tree trunks with corresponding DBH estimates.
• Figure 2: Examples of short and long trajectory sequences used for data collection. Blue spheres indicate camera positions over time. Scale bars are shown for reference. The short sequence captures a single tree at close range, while the long sequence covers a larger 60m$\times$60m area for multi-tree reconstruction.
• Figure 3: Screenshots of the custom visualization interface. Top: Photogrammetry point cloud rendered with original colors. Bottom left: Segmentation view highlighting identified tree trunks (in red). Bottom right: Sidebar panel showing tree ID and DBH estimate when a user clicks the tree trunk in the visualizer.
• Figure 4: Comparison of geometric fitting methods on cross-sections from both image-based (left column) and LiDAR-based (right column) point clouds. Top row: Partially observed trees. Bottom row: Fully observed trees. The truncated Fourier series (orange) provides more robust fits than ellipse (cyan) and B-spline (red) methods. Absolute relative errors for the Fourier method are: 6.18% and 3.94% (image-based), and 2.08% and 10.69% (LiDAR-based).
• Figure 5: Comparison of DBH estimation performance across observability, platform, sequence length, and sensing modality. Each block contains top-row Estimated vs. reference DBH scatter plots and bottom-row box plots of relative error (%), comparing three fitting methods.