Abstract
Accurate and repeatable assessments of dairy cattle body conformation are essential for productivity, welfare monitoring, and selective breeding. While 3D imaging technologies offer promising alternatives to manual measurements, the performance of low-cost, open-source solutions, such as Kinect 2.0 and OAK-D, remains insufficiently validated under farm conditions. This study assessed the accuracy and agreement of Kinect 2.0 and OAK-D cameras compared with manual caliper-based measurements in 41 dairy cows. Six key exterior traits—three vertical (withers, back, and stature) and three horizontal (chest depth, hook width, and loin width) traits—were measured. Linear regression and Bland–Altman analyses were used to evaluate agreement. OAK-D outperformed Kinect 2.0 in horizontal traits, with regression slopes of 0.908–0.976 and R² values of 0.58–0.61, and showed minimal systematic bias (–0.11 cm; LoA ±3.3 cm). Kinect 2.0 performed better in the vertical dimension (R² up to 0.85) but failed in terms of body depth (R² = 0.016), with a wide LoA (±15.3 cm). The results support the application of the OAK-D for automated exterior scoring in dairy cattle. Its affordability, open architecture, and compatibility with AI underscore its strong potential for scalable, farm-adapted, multiangle systems in precision livestock management.