A robust machine vision system for body measurements of beef calves
David Weales, Medhat Moussa, Cole Tarry
Abstract
Measuring body dimensions is a useful method of assessing the health and growth of young beef cattle. However, performing these measurements in a barn environment can present a number of unique challenges. The objective of this paper is to design an image capture system and image processing algorithm that can meet these challenges. The system uses two RGB-D cameras to collect images from the top-left and top-right of the calf. Images were collected in a barn environment along with ground truth body measurements. Colour image processing was used to remove the background by making use of a deep learning instance segmentation model for each camera. The segmented data from the two cameras was registered to create a 3D image of the calf, which was then used to measure a few key body dimensions. The experimental results showed a mean error of 0.2 cm for heart girth, -0.8 cm for withers height, -0.2 cm for midpiece height, and -2.1 cm for pin height.