Automated measurement of beef cattle body size via key point detection and monocular depth estimation

The measurement of beef cattle's body size is the key problem to be solved in beef cattle breeding. Automatic body size measurement were still influenced due to the light, distance, and other factors. To achieve measurement of beef cattle's oblique length, body height, chest depth, and hoof diameter under different distances and lighting conditions, an automatic measurement method based on key point detection and monocular depth estimation was proposed. Firstly, the cattle were detected by the YOLOv5s network, then, the key points were detected through the Lite-HRNet network and the pixel distance of the body size was calculated. Secondly, the monocular depth information was extracted through the Global-Local Path networks to obtain the relative depth value of the images. The relationship between the actual distance and the relative depth was obtained by using the actual distance calibration. Finally, the ratio between the pixel length and the actual length was obtained through the imaging principle of the RGB camera, and the beef cattle images in different scenarios were measured. It was concluded that the proposed system could complete the measurement work of beef cattle straight side images under no sky or small sky background. Thirty beef cattle images were measured and the average relative error of body height, body oblique length, chest depth and hoof diameter were 6.75 %, 7.55 %, 8.00 %, 8.97 %. The results showed that the proposed method could measure the body size of beef cattle at different distances and light conditions.