Computer vision is moving pig welfare monitoring from pens to individuals

Pig-welfare monitoring is often constrained by a practical contradiction: large groups are easy to film, but individual animals are difficult to follow reliably. A study published on August 24 in *Scientific Reports* introduces an open-source computer-vision pipeline intended to address that problem. The system combines animal detection, trajectory association and re-identification to analyse the behaviour of group-housed pigs at individual level.

The significance is not simply that a camera can “recognise” pigs, but that it can preserve individual identity over time. To generate useful data across hours or days, a system must cope with crossings, occlusion, posture changes, uneven light and dirt. The authors designed a modular architecture so components can be replaced and adapted to different barn conditions instead of relying on one closed model.

The study demonstrates the approach with video from groups of pigs and describes measures such as presence, position, movement, travelled distance, animal proximity and zone use. Such variables may provide early signals of activity, social interaction or deviations from routine. They are not, on their own, diagnoses of welfare, disease or abnormal behaviour. Their value depends on validated thresholds, veterinary observation and housing context.