Home cage monitoring of group-housed animals in highly enriched enclosures

Home cage monitoring is a key goal for animal welfare as it allows behaviour assessment in a familiar setting without disturbance. There are challenges in adapting existing methods to species such as primates that are housed in groups in large, highly enriched enclosures. Segmentation models such as Yolo provide a reliable mechanism for breaking down complex environments into different components. Together with the use of the Segment-Anything-Model from Meta it is possible to create an annotated dataset for a complex environment and train a model in a reasonably short time (<1 month). For this project a Yolo segmentation model was trained to detect rhesus macaques (group size 2-20 monkeys) in highly enriched indoor environments (enclosures measuring 8.4 m x 2.7 m x 2.8 m with multiple levels of enrichment). The current setup allows the monkeys to be monitored 24 hours a day, 7 days a week (for 22 different groups of monkeys) with output measures including foraging and enclosure use. A system of checks allows model performance to be tracked over time and for rapid retraining of the model if something changes in the environment (such as lighting or novel enrichment items). The effectiveness of the model is demonstrated by showing increased foraging when a scatter feed is given.