An automatic colour-based computer vision algorithm for tracking the position of piglets

Abstract

Artificial vision is a powerful observation tool for research in the field of livestock production. So, based on the search and recognition of colour spots in images, a digital image processing system which permits the detection of the position of piglets in a farrowing pen, was developed. To this end, 24,000 images were captured over five takes (days), with a five-second interval between every other image. The nine piglets in a litter were marked on their backs and sides with different coloured spray paints each one, placed at a considerable distance on the RGB space. The programme requires the user to introduce the colour patterns to be found, and the output is an ASCII file with the positions (column X, line Y) for each of these marks within the image analysed. This information may be extremely useful for further applications in the study of animal behaviour and welfare parameters (huddling, activity, suckling, etc.). The software programme initially segments the image in the RGB colour space to separate the colour marks from the rest of the image, and then recognises the colour patterns, using another colour space (B/(R+G+B), (G-R), (B-G)) more suitable for this purpose. This additional colour space was obtained testing different colour combinations derived from R, G and B. The statistical evaluation of the programme apos;s performance revealed an overall 72.5% in piglet detection, 89.1% of this total being correctly detected.