Review: Multiobject tracking in livestock − from farm animal management to state-of-the-art methods

Abstract

Multi-object tracking (MOT) methods have the potential to significantly improve precision livestock farming (PLF) by enabling simultaneous tracking of multiple animals in complex environments. However, research on MOT applications in livestock monitoring is limited, with state-of-the-art (SOTA) models primarily tested on benchmark datasets of pedestrians or vehicles. This systematic review was performed according to PRISMA guidelines. We identified 111 recent papers published from January 2019 to January 2025 using a keyword search for MOT and livestock from three scientific databases. The use-cases, datasets, and algorithms of MOT applied to livestock were thoroughly examined. This review addresses the limitations in existing systems to consistently preserve individual animal identities in long sequences. Key challenges that need to be addressed include frequent occlusions and complex, non-linear motion patterns that are characteristic of livestock behaviour. We identified 21 recent open-source SOTA models currently used in other disciplines (beyond livestock) that offer solutions to these challenges. Our analysis revealed research gaps and opportunities for developing tailored MOT techniques to overcome the challenges of real-world livestock monitoring. For MOT to provide valuable data for PLF purposes, it must perform long−term video analysis and address obstacles such as frequent and long-term occlusion, similar appearances between livestock as well as their non-linear motion. Investigating SOTA models showed that while tracking-by-detection is still the most widely used paradigm, tracking-by-attention, transformer−based end-to-end tracking architecture, provides a novel approach. Improvements in detection association strategies and motion models, as well as innovations in multi-camera tracking, can lead to improved animal health, productivity, and welfare in the livestock industry. This review highlights the importance of adapting and refining MOT methods for livestock monitoring.