RGB-D video-based individual identification of dairy cows using gait and texture analyses

Abstract

The growth of computer vision technology can enable the automatic assessment of dairy cow health, for instance, the detection of lameness. To monitor the health condition of each cow, it is necessary to identify individual cows automatically. Tags using microchips, which are attached to the cow’s body, have been employed for the automatic identification of cows. However, tagging requires a substantial amount of effort from dairy farmers as well as induces stress on the cows because of the body-mounted devices. A method for cow identification based on three-dimensional video analysis using RGB-D cameras, which capture images with RGB color information as well subject distance from the camera, is proposed. Cameras are mostly maintenance-free, do not contact the cow’s body, and have high compatibility with existing vision-based health monitoring systems. Using RGB-D videos of walking cows, a unified approach using two complementary features for identification, gait (i.e., walking style) and texture (i.e., markings), is developed.