Computational Vision Use for Evaluation of Confined Dairy Cows Behavior

Changes on dairy cows behavior can be observed as a result of microclimate variances. However, conventional steps that had been used until now are still evasive and they can even lead to some uncertainties in interpretation of results. The experiment described in this paper was conducted on a commercial farm in Sao Pedro (Sao Paulo, Brazil), where dairy cows were monitored through micro-cameras that were installed in strategic places to assess their behavior. The objective of this work was to development software to evaluate the behavior of confined dairy cows in freestalls through digital imaging. It is also important to point out that in this case, software validation was conducted through real-time testing. Obtained images were observed by three different methodologies (indirect way of instantaneous (VII), indirect spaced (VIE), and computer vision (VC)), in order to quantify the percentage of animals that presented behavioral activities, and also to analyze and compare these three distinct registration methods. Based on this study, it was possible to conclude that although computer vision methodology developed for behavioral evaluation presented some constrains, it demonstrated to be an excellent tool to evaluate dairy cows’ behavior.