This review summarizes the latest advancements in computer vision-based techniques for non-contact measurement of livestock body dimensions and weight. Traditional methods for collecting livestock phenotypic data, such as manual measurements, are labor-intensive, time-consuming, and can cause stress to animals, leading to economic losses. In contrast, modern technologies like next-generation AI, visual processing, intelligent sensing, multimodal fusion, and robotics offer faster, more efficient, and non-invasive data acquisition. The review focuses on three main areas: 3D reconstruction of livestock, body dimension acquisition using computer vision, and weight estimation using computer vision. 3D reconstruction technology is crucial for obtaining accurate livestock phenotypic data. It involves capturing 3D point cloud data using RGB cameras, laser scanning, or 3D cameras. RGB-based methods are cost-effective but have limitations due to the lack of depth information and the need for multiple cameras. Laser scanning provides high precision but is expensive and requires precise alignment. 3D cameras offer a balance between cost and accuracy, enabling real-time data acquisition and reducing the impact of livestock movement. For body dimension acquisition, RGB image-based methods are widely used but have limitations in capturing three-dimensional parameters. 3D point cloud methods, on the other hand, provide more detailed spatial information and are more accurate. Segmentation techniques, including geometric and neural network-based methods, are used to extract key body measurements from 3D point clouds. Weight estimation methods include linear regression models based on body measurements and neural network-based approaches. Linear regression models use 3D reconstruction data to estimate weight, while neural networks, particularly convolutional neural networks (CNNs), offer higher accuracy and can process large datasets efficiently. The review also discusses challenges in non-contact phenotypic data acquisition, including the need for accurate 3D reconstruction models, the lack of high-quality public datasets, and the inefficiency of current point cloud acquisition methods. Future research directions include improving neural network models, standardizing data, and developing more efficient data collection methods. Overall, computer vision-based techniques are becoming increasingly important in precision livestock farming, offering a more efficient and non-invasive way to collect livestock phenotypic data. These technologies have the potential to significantly enhance data collection efficiency, accuracy, and animal welfare in livestock management.This review summarizes the latest advancements in computer vision-based techniques for non-contact measurement of livestock body dimensions and weight. Traditional methods for collecting livestock phenotypic data, such as manual measurements, are labor-intensive, time-consuming, and can cause stress to animals, leading to economic losses. In contrast, modern technologies like next-generation AI, visual processing, intelligent sensing, multimodal fusion, and robotics offer faster, more efficient, and non-invasive data acquisition. The review focuses on three main areas: 3D reconstruction of livestock, body dimension acquisition using computer vision, and weight estimation using computer vision. 3D reconstruction technology is crucial for obtaining accurate livestock phenotypic data. It involves capturing 3D point cloud data using RGB cameras, laser scanning, or 3D cameras. RGB-based methods are cost-effective but have limitations due to the lack of depth information and the need for multiple cameras. Laser scanning provides high precision but is expensive and requires precise alignment. 3D cameras offer a balance between cost and accuracy, enabling real-time data acquisition and reducing the impact of livestock movement. For body dimension acquisition, RGB image-based methods are widely used but have limitations in capturing three-dimensional parameters. 3D point cloud methods, on the other hand, provide more detailed spatial information and are more accurate. Segmentation techniques, including geometric and neural network-based methods, are used to extract key body measurements from 3D point clouds. Weight estimation methods include linear regression models based on body measurements and neural network-based approaches. Linear regression models use 3D reconstruction data to estimate weight, while neural networks, particularly convolutional neural networks (CNNs), offer higher accuracy and can process large datasets efficiently. The review also discusses challenges in non-contact phenotypic data acquisition, including the need for accurate 3D reconstruction models, the lack of high-quality public datasets, and the inefficiency of current point cloud acquisition methods. Future research directions include improving neural network models, standardizing data, and developing more efficient data collection methods. Overall, computer vision-based techniques are becoming increasingly important in precision livestock farming, offering a more efficient and non-invasive way to collect livestock phenotypic data. These technologies have the potential to significantly enhance data collection efficiency, accuracy, and animal welfare in livestock management.