Yolo11-Based Pose Estimation and Geometric Locomotion Trait Extraction in Beef Cattle: A Proof-Of-Concept Study
Year: 2026
Type: preprint
Abstract: Objective assessment of locomotion-related traits could improve repeatability in beef cattle structural phenotyping. This proof-of-concept study evaluated a YOLO11-based workflow for cattle detection, localization of five limb keypoints, and extraction of forelimb step length, hock angle, and hind-limb angle from controlled side-view RGB images. Detection and pose datasets contained 1,000 images from 138 steers and 1,200 images from 120 steers, respectively, with animal-level training, validation, and test partitions. On held-out animals, detection mAP@50–95 was 0.9678 and pose mAP@50–95 was 0.9950. Mean keypoint localization error was 9.86 ± 5.76 pixels, or 2.46% of the larger cattle bounding-box dimension, with percentage of correct keypoints (PCK) values of 0.943, 0.998, and 1.000 at thresholds of 0.05, 0.10, and 0.20. Forelimb step length showed good agreement between manual and model-derived measurements (MAE 4.68 cm; Lin’s CCC 0.903). Hock and hind-limb angles showed weak individual-animal concordance (CCC 0.146 and 0.317).
Source: Research Square
Institution: University of Illinois Urbana-Champaign
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7225812580
Cites:
Cited by:
Related to: