Intelligent Video Analysis-Based Pose Recognition and Localization for Improving Dairy Cow Welfare

Ensuring the welfare of dairy cows is essential for sustainable and ethical livestock farming. This study presents an integrated system that combines pose recognition and spatial localization to monitor cow behavior in real-time. Using a convolutional neural network (YOLOv11n) trained on a newly collected and augmented dataset, the framework accurately classifies cow postures (standing or lying) and determines their location within predefined sub-zones of the barn. The system achieved a pose detection accuracy of 91.20% and 85.12 on daytime and nighttime test sets, respectively. Validation on the external MMCows dataset confirmed its robustness and generalizability across different barn layouts (overall accuracy of 93.84%). Regarding localization accuracy, the proposed method achieved 93.28% and 95.74 on daytime and nighttime test sets, respectively. By relying solely on RGB cameras, the method ensures low cost, scalability, and ease of deployment. This vision-based approach enables continuous monitoring, early anomaly detection, and informed decision-making in precision livestock management. The dataset and pretrained models are available at the following URL (after paper acceptance).