A cow structural model for video analytics of cow health

Abstract: In livestock farming, animal health directly influences productivity. For dairy cows, many health conditions can be evaluated by trained observers based on visual appearance and movement. However, to manually evaluate every cow in a commercial farm is expensive and impractical. This paper introduces a video-analytic system which automatically detects the cow structure from captured video sequences. A side-view cow structural model is designed to describe the spatial positions of the joints (keypoints) of the cow, and we develop a system using deep learning to automatically extract the structural model from videos. The proposed detection system can detect multiple cows in the same frame and provide robust performance under practical challenges like obstacles (fences) and poor illumination. Compared to other object detection methods, this system provides better detection results and successfully isolates the keypoints of each cow even when they are close to each other.