Speaker
Description
The accurate measurement of individual, high-quality logs is a critical operational task in the timber trade, particularly during timber auctions where roundwood is traded individually and commands premium prices per cubic meter. In these high-stakes environments, even minor deviations in volume estimation can result in substantial economic losses for either buyers or sellers. Consequently, ensuring accuracy, transparency, traceability, and reproducibility is essential. Currently, log volume estimation predominantly relies on manual measurements using conventional stereometric formulae, such as those of Huber, Smalian, or Newton, in accordance with frameworks like the German Agreement for the raw wood trade. However, these traditional methods are constrained by simplified geometric assumptions, human error, tool calibration issues, and forestry rounding rules. Recent technological advancements in artificial intelligence , image-based methods, and three-dimensional (3D) reconstruction offer a promising pathway to overcome these operational bottlenecks. While small-scale devices equipped with Light Detection and Ranging (LiDAR) or Red Green and Blue (RGB) sensors have demonstrated potential, the integration of these technologies into fully automated, end-to-end workflows for individual log identification and volume estimation remains largely unexplored in operational practice. To address this gap, this study proposes an automated methodological framework that synergizes two-dimensional (2D) instance segmentation based on RGB information with 3D photogrammetric reconstruction, allowing log volumes to be estimated directly from precise geometric models. The proposed workflow comprises four primary stages. First, RGB imagery is acquired using consumer-grade mobile devices (specifically an iPhone 15 Pro and an iPad Pro mounted on a gimbal) providing a flexible, accessible data collection method suitable for auction yards. Second, instance segmentation of individual logs is performed using a YOLOv8-Seg model based on two-dimensional Convolutional Neural Networks (2D CNNs). This strategic use of 2D CNNs robustly isolates logs from complex backgrounds while bypassing the computational heavy lifting and extensive annotation requirements associated with fully 3D deep-learning architectures. Third, the segmented image data undergoes photogrammetric processing to generate detailed 3D virtual reconstructions of each log. Finally, the system automatically calculates the individual log volumes derived directly from the reconstructed 3D geometry. By closely integrating instance segmentation and photogrammetry, this pipeline transitions raw field imagery into highly accurate volume metrics with minimal manual intervention. The practical implications are significant across the forest-wood value chain. For forest owners and auction organizers, it ensures fair valuation and compliance with standardized trading frameworks through transparent, digital documentation. For timber buyers, it mitigates uncertainty and measurement disputes. Ultimately, this methodology reduces operator subjectivity, enhances measurement accuracy, and lays a robust foundation for the future integration of automated digital twins in forest operations and high-value timber trading.
| Keywords | "Photogrammetry; Instance Segmentation; Roundwood; |
|---|
