Automated measurement of livestock body based on pose normalisation using statistical shape model

Livestock body measurement is a key indicator for predicting their live weight, probability of illness, or readiness for market, to help farmers and breeders control their health con-dition and slaughter performance. Livestock body measurement methods develop from manual ways based on conventional tools to non-contact ways based on analysing data obtained from remote sensing instruments, with a new trend to extract body measure-ments from 3D point clouds. But the accuracy and robustness of current methods are affected by noise and point cloud data missing caused by self occlusions of animals. In this paper, a method to solve the problems of animal body measurement is proposed. In this method, the statistical shape model for animals are fitted to livestock point cloud data, and then body measurements are extracted from the reconstructed meshes. With the help of the statistical shape model, the livestock body measurement method solves the problem of missing point cloud data and inconsistency caused by livestock movement in precision animal husbandry. The method is verified on two typical livestock including cattle and pigs. In cattle body measurement, the overall estimation accuracy is 91.95%, while in pig body measurement, the accuracy is 87.63%. Ways to solve the point cloud data missing by parametric reconstruction and the data inconsistency by normalising animal poses are presented with the proposal of the method of this paper, which can be easily adapted to measure objects with non-rigid articulated deformations.(c) 2023 IAgrE. Published by Elsevier Ltd. All rights reserved.