Description
The following abstract was presented as part of the Hannah Dairy Research Foundation's Next Generation Resilient Dairying Conference 2025.Dairy farming has faced significant challenges over the past two decades, yet productivity has increased through automation and technology enabling management of larger herds. However, these systems rely on wearable or invasive monitoring devices i.e. collars, tail-mounted accelerometers, or rumen boluses. While effective, such devices present welfare risks - including irritation and entanglement - and impose maintenance and labour constraints. In addition, climate change is intensifying environmental and disease pressures, requiring adaptive monitoring of individual animals to sustain productivity while improving welfare outcomes. We propose video analytics as the next generation of welfare monitoring: providing scalable, continuous, non-contact behavioural assessment using standard CCTV infrastructure already available on many farms. The software-driven nature of this approach allows flexible model updates and adaptation to changing management requirements environments, and breeds.
Our research focuses on developing a video analytics framework combining computer vision, deep learning, and geometric calibration for animal tracking and behavioural analysis. Core challenges include persistent identity tracking under high stocking density, occlusions, and appearance variability; generalising behavioural classification across diverse lighting conditions, camera geometries, and breed morphologies; linking observations to biologically meaningful welfare metrics for evidence-based management to enhance welfare and productivity. These challenges are being tackled through our single and multi-camera tracking work where we employ monocular depth estimation to enhance single-camera tracking and multi-view geometric alignment for automated inter-camera calibration.
As an exemplar application, we present automated bed occupancy estimation for continuous assessment of stall use. By integrating posture classification and barn layout, we derive lying time, stall utilisation, and Cow Comfort Index (CCI). Preliminary analyses indicate spatial variations in stall use correlate with barn layout factors, i.e. proximity to water troughs, highlighting the system’s potential for data-driven optimisation of housing environments.
Future work will extend multi-camera tracking across commercial-scale barns, incorporate physiological proxies such as respiration and rumination, and integrate RFID for identity verification and tracking correction. These developments aim to establish a generalised, non-invasive monitoring pipeline for continuous welfare assessment and behavioural phenotyping. Ultimately, this research supports more resilient, adaptive, and welfare-centred dairy management under increasing climate and labour pressures.
| Period | 24 Nov 2025 |
|---|---|
| Held at | Moredun Research Insitute, Edinburgh |
| Degree of Recognition | International |
Keywords
- Tracking
- Welfare Monitoring
- Deep Learning
- Computer Vision
- Precision Farming