Precision Bovine Phenotyping: 3D Point-Cloud Sensing, Processing, and On-Farm Applications

Cattle-body dimensions support growth monitoring, body-conformation assessment, health management, genetic evaluation, and breeding decisions. Traditional contact measurement is labour-intensive, affected by operator and posture, and may induce handling stress. Three-dimensional (3D) point clouds enable contactless and repeatable estimation of linear dimensions, girths, volume, and related phenotypes. We searched Web of Science Core Collection, IEEE Xplore, and CNKI on 30 September 2026 for publications from January 2010 to September 2026 and synthesised 93 studies. Across cattle studies, reported relative errors were commonly in the approximately 2–6% range under controlled or semi-controlled conditions, but the estimates were trait-, breed-, sensor-, posture-, and protocol-dependent and therefore are not pooled effect sizes. Multi-view reconstruction improved surface completeness; deep segmentation and keypoint models improved automation; and 2D–3D fusion often reduced dependence on dense 3D annotation. Evidence remains weaker for repeatability, systematic bias, cross-breed transfer, and long-term commercial-farm performance than for single-site accuracy. The main barriers to deployment are dynamic occlusion, non-rigid deformation, environmental variability, annotation cost, computation, maintenance, and integration with farm workflows. The review therefore recommends unified anatomical landmarks and reference protocols, reporting of bias, repeatability, reproducibility, uncertainty, and Bland–Altman agreement, and external validation across breeds and farms. Cattle-tested methods should be separated from transferable computer-vision methods when assessing readiness for breeding and production applications. Emerging 4D temporal point clouds, self-supervised learning, multimodal fusion, foundation models, and digital twins remain promising but require direct cattle-measurement validation.