Fusion of CREStereo and MobileViT-Pose for rapid measurement of cattle body size

Abstract

Accurate measurement of cattle body size is crucial for assessing growth status and making breeding decisions. Existing automated methods either lack precision or suffer from long processing times. In this study, a rapid and non-contact cattle body size measurement method based on stereo vision was carried out. Lateral images of cattle were initially captured using a stereo camera, and depth information was derived from these images using the CREStereo algorithm. The MobileViT-Pose algorithm was then applied to predict body size keypoints, including head, body, front limbs, and hind limbs. The final body size measurements were obtained by integrating depth data with these keypoints. To minimize measurement errors, the Isolation Forest algorithm was used to detect and remove outliers, with the final measurement computed as the average of multiple results. Compared to traditional stereo matching algorithms, CREStereo provided more detailed disparity information and demonstrated greater robustness across varying resolutions. Pose estimation accuracy of the MobileViT-Pose algorithm reached 92.4 %, while improving efficiency and reducing both the number of parameters and FLOPs. Additionally, a lightweight version, LiteMobileViT-Pose, was introduced, featuring only 1.735 M parameters and 0.272 G FLOPs. In practical evaluations, the maximum measurement deviations for body length, body height, hip height, and rump length were 4.55 %, 4.87 %, 4.99 %, and 6.76 %, respectively, when compared to manual measurements. Additionally, the MobileViT-Pose model was deployed, achieving an average body size measurement error of only 2.85 % and a measurement speed of 18.8 fps. The proposed method provides a practical solution for the rapid and accurate measurement of body size.