Automated Methodologies to Assess Welfare of Rhesus Macaques (Macaca mulatta) in a Breeding Colony

Abstract

This thesis endeavours to explore the application of machine learning-based methodologies to gain valuable insights into the behaviour of captive Rhesus macaques. The primary objectives are to enhance their welfare and streamline management practices. Rhesus macaques hold significant importance in biomedical research, making their welfare a paramount concern in such environments. The subjects of this study reside within a breeding colony in the UK, serving as a source of individuals for neuroscientific laboratories. These macaques are grouped into breeding and juvenile cohorts, with continuous surveillance via a comprehensive CCTV system at the Centre of Macaques, comprising cameras within each enclosure. Efficient macaque management is crucial not only for ensuring their well-being but also for facilitating successful breeding programs and the subsequent supply of animals to research laboratories. However, the process of guaranteeing both their welfare and the collection of informative behaviour data necessitates specialized personnel, incurring significant time and financial costs. This study presents three distinct implementations of computer vision-based pipelines that leverage video footage from the existing CCTV system and researcher-recorded videos to autonomously collect and analyse behaviour of interest. This innovative approach not only saves time in data collection and analysis but also extracts previously unattainable information for the facility's staff. The first project focuses on temperament assessment in macaques, employing three models based on object detection and pose estimation. These models analyse pre-existing researcher-recorded videos designed for the same purpose. The second project investigates diverse foraging patterns based on varying food mixes provided. This automated methodology utilizes object detection to calculate the percentage of macaques engaged in foraging activities from CCTV footage. The third project aims to evaluate two enrichment items, one food-based and one non food-based. The automated methodology relies on object detection algorithms to extract data concerning the number of individuals interacting with these objects over specified time intervals. These projects harness existing camera infrastructure and furnish invaluable insights into neophobia, temperament, preferred food choices, and enrichment planning, all of which inform crucial management decisions. Consequently, this thesis underscores the advantages of employing such methodologies and illustrates their potential for broader application, with the capacity to enhance the welfare and management of non-human primates in similar facilities.