Rapid and stress-free measurement of sheep body size is important for precision production and breeding. Most reported methods require sheep to stand in a fixed position with a standard posture, leading to stress and inefficiency. This study proposes a rapid and stress-free method: firstly, sheep's body keypoints are detected through the YOLOv11-pose model and corrected by a self-built correction algorithm during sheep's natural walk; subsequently, five walking body sizes (WBS) and eight posture angles (WPA) are calculated based on the corrected keypoints, and are used to estimate sheep's standing body sizes (SBS) through a regression model. The results show that the keypoint correction algorithm reduces the 2.54% mean percentage error (MAPE) on 17 unfamiliar sheep compared to the uncorrected case when the search radius is 5 pixels. The combined input feature set WPA+WBS outperforms either the standalone WPA or the standalone WBS feature set. The image resolution factor exhibited significance, while shooting distance and lighting intensity did not, and the high-resolution group (848×480) is highly recommended. The Ridge model performs best among four regression models, and its MAPE for body height, hip height, body slope length, abdominal depth, and abdominal circumference are 3.95%, 4.69%, 6.46%, 7.48%, and 6.19%, respectively, with a mean of 5.75% on 17 unfamiliar sheep. Compared to the original 11.54% based keypoint-based method, the accuracy has significantly improved by 50.17%. This method provides a solid theoretical foundation and technical support for rapid and stress-free estimation of sheep body size in production and breeding. Science4Impact Statement: This study proposes an automated method for measuring sheep body size that addresses key challenges in livestock production, specifically in relation to food security and animal welfare. For livestock breeders and farm operations , the method enables rapid, non-invasive measurements during natural walking, thereby complying with internationally recognized animal welfare standards while eliminating the need for physical restraint and minimizing stress-related productivity losses. For agricultural technology companies , the method offers empirically validated performance: the custom-developed correction algorithm improves keypoint localization accuracy, reducing the mean absolute percentage error (MAPE) by 2.54%, while the Ridge regression model demonstrates robust predictive capability (mean MAPE of 6.04% for familiar individuals and 5.75% for unfamiliar sheep), supporting its integration into commercial measurement systems. For agricultural advisory and regulatory bodies, the technique contributes to quality assurance in precision breeding by providing quantified assessments of walking posture and regression-based estimates of standing body size—information that can support the development of breeding efficiency benchmarks. By delivering practical benefits to stakeholders—enhancing productivity in support of food security while safeguarding animal welfare—this work addresses an important gap in non-invasive digital body measurement technologies for livestock.
Shen et al. (Sun,) studied this question.