Recent advances in computer vision for non-contact phenotyping and weight estimation in livestock:A systematic reviewOA

With the advancement of Precision Livestock Farming(PLF),traditional manual measurement methods for livestock phenotypic trait are seriously limited in meeting modern demands in terms of accuracy,efficiency and animal welfare.In response,computer vision(CV)-based non-contact measurement techniques,especially those integrated with machine learning(ML)and deep learning(DL),have demonstrated significant potential in livestock body size and weight estimation.This study presents a systematic review of recent progress in 2D image-based,3D point cloud-based,and fusion-based CV methods for phenotypic measurement of cattle,pigs,and sheep.The comparative analysis covers core algorithmic frameworks,sensor types,pose normalization techniques,and regression model performance.2D methods offer low-cost deployment with typical measurement error ranging from 5%to 8%,while 3D techniques reduce error to 3%to 5%and achieve R2 values above 0.95.DL-based models can further enhance robustness,especially in dynamic and occluded farm environments.Other than the advances above,key challenges still exist,including low multi-view point cloud registration and model generalization across scenarios,and the scarcity of labeled data.Addressing these disadvantages requires improved multi-modal data fusion,large-scale model transfer strategies,the development of lightweight model design,and real-time deployable architectures.This review provides a structured technical reference for researchers and engineers,especially those new to the field,with insights into end-to-end workflows and expects future innovations that can support intelligent,non-invasive,and large-scale livestock management systems.