A non-stressful vision-based method for weighing live lambs

Abstract

Abstract Accurate measurement of livestock weight is a primary indicator in the meat industry to increase the economic gain. In lambs, the weight of a live animal is still usually estimated manually using traditional scales, resulting in a tedious process for the experienced assessor and stressful for the animal. In this paper, we propose a solution to this problem using computer vision techniques; thus, the proposed procedure estimates the weight of a lamb by analysing its zenithal image without interacting with the animal, which speeds up the process and reduces weighing costs. It is based on a data-driven decision support system that uses RGB-D machine vision techniques and regression models. Unlike existing methods, it does not require walk-over-weighing platforms or special and expensive infrastructures. The proposed method includes a decision support system that automatically rejects those images that are not appropriate to estimate the lamb weight. After determining the body contour of the lamb, we compute several features that feed different regression models. Best results were achieved with Extra Tree Regression ($R^{2}$=91.94%), outperforming the existing techniques. Using only an image, the proposed approach can identify with a minimum error the optimal weight of a lamb to be slaughtered, so as to maximise the economic profit.