Learning end-to-end respiratory rate prediction of dairy cows from RGB videos

M. Wang, S. Li, R. Peng, S.E. Räisänen, A.M. Serviento, X. Sun, K. Wang, F. Yu, M. Niu

doi: external page 10.3168/jds.2023-24601

An end-to-end computer vision method has been developed to monitor the respiratory rate (RR) of dairy cows, crucial for assessing their health and welfare. The new approach utilizes the advanced Transformer model, VideoMAE, to analyze video data and automatically select relevant regions like the cow's abdomen for RR prediction. Unlike traditional methods that require multiple processing steps, this method simplifies the process and reduces errors. Tested on video data from 6 cows, the model demonstrated strong performance with a mean absolute error of 2.58 breaths per minute and a Pearson correlation of 0.86. This innovative method shows promise for automated RR monitoring in dairy farms and could potentially be expanded to include other behavioral and identification monitoring.