Authors:

Colton Gomoll

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Traditional methods for assessing tree inventories are time-consuming and as-such prove expensive and inefficient for large-scale applications. To address this issue, The Digital Photogrammetry Research Group (DPRG) is developing an automated tree inventory pipeline to segment individual trees within point cloud datasets and derive each tree's biometrics. In order to analyze pipeline results for quality control, reference maps are essential. However, reference maps based on manual measurements in the field can sometimes be inaccessible or non-existent. This project proposes an alternative method for generating reference maps for quality control based on reconstructed point clouds collected using BackPack systems. Horizontal slices were made from normalized point cloud data and refined through a semi-automated process to obtain stem and outlier (non-stem) clusters. Tree locations and DBH were then extracted from the reference maps using least- squares circle fitting on stem clusters and compared to pipeline results. The method was tested on both natural forest and plantation datasets, collected at the University of Georgia's Whitehall Forest site using commercial (Hovermap) and in-house (KNAP) Backpack systems. For trees with DBH greater than 4 inches, comparison between pipeline and reference map results showed over 80% precision, 88% recall, and 86% F1-score amongst all datasets. Plantations showed higher precision, recall, and F1-scores when compared to natural growth forest. This approach offers a practical solution for quality control in the absence of field-based references, allowing for rapid feedback and iterative improvement of the tree inventory pipeline. Keywords: Reference Data; Quality Control; Tree Inventory; Point Cloud; BackPack LiDAR

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Colton Gomoll