Computer vision-based livestock weight measurement: principles, methods, and challenges

Livestock body weight is a critical indicator for optimizing feeding management, enabling precision nutrition, and supporting intelligent breeding. Traditional weighing methods, however, are often labor-intensive, inefficient, and may induce stress responses in animals, limiting their practicality in large-scale farming systems. In recent years, computer vision-based non-contact weight measurement has attracted increasing attention in smart animal husbandry due to its advantages in efficiency, non-invasiveness, and scalability. This review provides a comprehensive synthesis of recent progress in this field, with three primary contributions. First, it systematically summarizes state-of-the-art techniques for non-contact livestock weight measurement, covering data acquisition strategies, image preprocessing methods, and modeling approaches. Second, it presents a critical comparative analysis of existing methodologies, with particular emphasis on the performance characteristics and limitations of machine learning-based vs. deep learning (DL)-based models. Third, it highlights the growing significance of advanced three-dimensional (3D) reconstruction techniques, such as 3D Gaussian Splatting, in enhancing measurement accuracy and robustness. Finally, this review identifies persistent challenges, including variations in animal posture under dynamic conditions, environmental interference, and the scarcity of high-quality labeled datasets. Based on these insights, it outlines future research directions, emphasizing multi-source data fusion, the development of highly generalizable model architectures, and cross-modal DL frameworks, with the goal of building robust, adaptive, and scalable livestock weight measurement systems.