Biological Science: Research Developments and Innovations Vol. 2

Contactless sensing is becoming a central component of precision dairy management because cameras can observe animals repeatedly without attaching devices, interrupting routine behaviour or requiring repeated handling. This critical narrative review evaluates computer vision, depth imaging, infrared thermography and artificial intelligence as tools for monitoring health and welfare in dairy cattle. The literature was selected from open scholarly sources using a transparent, topic-specific search covering 1 January 2000 to 4 July 2026, with emphasis on peer-reviewed studies that provided biologically interpretable outcomes or meaningful validation under farm conditions. Evidence is strongest for automated body condition assessment, locomotion and lameness phenotyping, individual identification, classification of major behaviours, and extraction of thermal signatures related to udder inflammation, heat load and acute stress. Yet apparently high algorithmic performance often reflects narrow datasets, repeated images from the same animals, controlled camera geometry or labels derived from subjective human scores. Thermal imaging is particularly sensitive to ambient conditions, camera angle, emissivity assumptions, contamination and the anatomical region selected, while visible-light systems remain vulnerable to occlusion, lighting variation and domain shift between farms. Across modalities, external validation, longitudinal evaluation and decision-level clinical utility lag behind proof-of-concept accuracy. The most defensible near-term role of contactless monitoring is therefore continuous screening and risk stratification rather than autonomous diagnosis. Progress depends on animal-level data partitioning, multi-farm benchmarking, uncertainty-aware models, standardised reference phenotypes, multimodal fusion, edge deployment and welfare-centred evaluation of false alarms and missed cases. Contactless precision monitoring can strengthen dairy cattle health and welfare surveillance, but its value will be determined less by model novelty than by biological validity, generalisability and integration into actionable farm and veterinary workflows.