Correcting animal posture for better high-throughput phenotyping of dairy cow morphology

Abstract

Phenotyping devices based on three-dimensional (3D) imaging have been developed in recent years but many require animal's immobilization during acquisition. For high-throughput purpose and limiting animal's stress, constraint-free system are of interest but measurements may vary according to animal's posture. Here, using a constraint-free system, we determined the influence of head and front-legs positions on morphological measurements, aiming to correct them in the future. Dataset included 130 3D images from 14 dairy cows with different head and legs position. Height at withers (HW), at the sacrum (HS) and at the hips (HH) were determined on 3D images. The results confirmed that the positions of the head and/or legs had a significant influence on HW measurements, but had no effect on HS and HH values. When cows lowered their head, HW were under-estimated up to 3% on average, with a maximal effect when cow head reached ground level (4.8%). HW was then corrected according to head position, based 1)-on a prediction from HH measurement; 2)-a fixed percentage per head position classes or 3)-a continuous percentage of correction. The last two approaches were the most relevant to correct HW measurements (error lower than 1%). Similar results were observed for legs position corrections. These resulted confirmed that the posture of cows in a phenotyping system based on 3D images might lead to important variations in morphological measurements, which can, however, be easily determined and corrected.