Automatic heart girth measurement for cattle based on deep learning

Body measurement is an important way to assess the health status of livestock. As an indicator of body condition, heart girth is relevant to the weight and health of livestock. The emerging of consumer level sensors provides an efficient and low cost way to monitor the status of livestock. Meanwhile, many studies proposed non-contact body measurement methods utilizing sensors. However, constraints still exist in the pipeline of many data acquisition and measurement methods. To improve flexibility and efficiency, this paper presents an automatic heart girth measurement method based on deep learning. Unlike methods that solely compute on point clouds, we combine the RGB image and 3D point cloud captured by depth sensors to implement the detection and measurement. We train a detector to find the girth of cattle RGB images and project the girth points onto the surface of the livestock. Then we use ellipse to fit the girth curve, and compute its circumstance as the girth length. We test our method on 103 cattle data, the mean error of our results is 6.47%. In comparison with the previous method, our method is more robust and accurate