Sheep body automatic measurement based on fusion of color and depth image

Visual livestock measurement techniques offer non-contact operation, high efficiency, and reduced animal stress. However, traditional 2D imaging lacks depth data, while 3D reconstruction faces computational and environmental constraints, limiting real-world applicability. This study presents a novel RGB-D fusion method for efficient sheep morphometric analysis. Using Kinect V2 sensors, top-down and lateral RGB-D data were captured from Dorper sheep. An optimized YOLOv8pose_slimneck framework was used to detect keypoints of body dimension in color images, with depth values derived from aligned RGB-D pairs. Six body dimensions—body length, height, rump height, chest depth, chest width, and rump width—were calculated. On-farm experiment demonstrated that the automated body dimension measurements achieved <5 % error, and subsequent liveweight prediction based on these visual measurements yielded a mean error of 5.3 %. This approach demonstrates strong practical feasibility for implementing precision sheep farming.