Three-dimensional reconstruction of the trunk volume of dairy cows using RGB-D images obtained by UAV for body mass prediction

Accurate estimation of body mass (BM) in dairy cattle is essential for nutritional management, productivity monitoring, and animal welfare. In this context, three-dimensional (3D) morphometric descriptors derived from remote sensing technologies have emerged as promising alternatives to conventional weighing methods. This study proposes and evaluates a methodological framework for reconstructing the dorsal trunk volume of dairy cows using RGB-D data acquired by an unmanned aerial vehicle (UAV). Data were collected on a commercial dairy farm under a tie-stall system, with animals moving along a defined pathway after milking, enabling standardized overhead image acquisition. Top-view images were obtained using an Intel RealSense D435i depth camera mounted on a UAV platform (3DR Solo). Point clouds were generated and processed in CloudCompare, including dorsal surface segmentation, normal estimation, and volumetric mesh reconstruction using Poisson surface reconstruction. Only closed and geometrically consistent meshes were retained for analysis. The dataset comprised 25 dairy cows, resulting in 25 observations, with BM ranging from 500 to 750 kg, stratified into 50 kg intervals. Correlation analyses showed strong positive associations between reconstructed dorsal volume and BM, both globally (r = 0.979; ρ = 0.989) and across classes. A simple linear regression model, evaluated using a 70/30 train–test split and leave-one-out cross-validation, demonstrated high predictive performance (R² = 0.967 training; R² = 0.931 testing). These results confirm the technical feasibility and zootechnical relevance of UAV-based RGB-D acquisition combined with 3D reconstruction for non-invasive BM estimation in dairy cattle, highlighting its applicability in precision livestock farming systems.