Automated Detection and Tracking of Individual Piglets for Real-Time Behavioral Monitoring

Automated monitoring of piglet behavior offers potential benefits for animal welfare assessment in modern pig production. This paper describes a system for individual piglet detection and tracking in farrowing environments using standard RGB cameras. We evaluated several deep learning architectures to address the specific challenges of farrowing pens, where piglets often cluster together, partially occlude one another, and move unpredictably. Comparative testing showed that a YOLOv5l detector combined with a SORT tracking algorithm achieved promising results (98.6% MOTA, 93.9% IDF1) at 38 frames per second under controlled test conditions. The system extracts basic behavioral metrics including total distance traveled, average speed, and idle time, with mean squared errors of approximately 0.01 when compared to manual annotations. These preliminary metrics may contribute to automated activity analysis related to piglet welfare status. Our results suggest that individual tracking in farrowing environments is feasible with conventional hardware, though performance may vary under different farm conditions. The current implementation focuses on fundamental movement parameters, with limitations in tracking stationary objects and distinguishing between tightly clustered piglets. This work represents an incremental step toward precision livestock farming applications, potentially supporting more targeted welfare monitoring in commercial settings when further developed and validated across diverse environments. The datasets used in this study are made publicly available.