Analysing the space-usage-pattern of a cow herd using video surveillance and automated motion detection

Abstract

As the presence of a human observer can influence the behaviour of animals, a camera based system in which the cows cannot leave the surveyed area, and that automatically detects the animals and the areas of the barn used, would be a beneficial tool to analyse dairy cattle herd activity. The aim of this study was to provide such a system and to demonstrate it in an example analysis. A group of thirty-six Holstein Friesian dairy cows housed at the Chamber of Agriculture in Futterkamp in Northern Germany were recorded with eight cameras. The barn was partitioned into rectangular areas and automated motion detection was implemented and applied to detect the areas with cow activity. For the automated motion detection high evaluation metrics were reached. Area under receiver-operating-characteristic (ROC) curve and precision-recall (PR) curve varied among the cameras from 0.85 to 0.94 (0.9 ± 0.03), respectively, 0.68 to 0.95 (0.88 ± 0.09), while accuracies and precision ranged from 0.783 to 0.88 (0.844 ± 0.033) and 0.544 to 0.938 (0.833 ±0.134). As an example, video recordings from five days in April and June 2019 were used each and compared afterwards. Areas visited more in April or June could be specified, and herd preferences in the use of lying cubicles and feeding troughs were revealed. It was found that space usage was most stable in the middle of the period between morning and afternoon milking and it became more unstable towards afternoon milking. The presented tool has the potential to visualise and evaluate dairy cows’ space usage and was developed in Python code. It is available on GitHub.