Automatic Measurement of Multi-Posture Beef Cattle Body Size Based on Depth Image

YE Wenshuai,KANG Xi,HE Zhijiang,LI Mengfei,LIU Gang
DOI: https://doi.org/10.12133/j.smartag.SA202210001
2022-01-01
Abstract:Beef cattle in the farm are active, which leads the collection of posture of the beef cattle changeable, so it is difficult to automatically measure the body size of the beef cattle. Aiming at the above problems, an automatic measurement method for beef cattle's body size under multi-pose was proposed by analyzing the skeleton features of beef cattle head and the edge contour features of beef cattle images. Firstly, the consumer-grade depth camera Azure Kinect DK was used to collect the top-view depth video data directly above the beef cattle and the video data were divided into frames to obtain the original depth image. Secondly, the original depth image was processed by shadow interpolation, normalization, image segmentation and connected domain to remove the complex background and obtain the target image containing only beef cattle. Thirdly, the Zhang-Suen algorithm was used to extract the beef cattle skeleton of the target image, and calculated the intersection points and endpoints of the skeleton, so as to analyze the characteristics of the beef cattle head to determine the head removal point , and to remove the beef cattle head information from the image. Finally, the curvature curve of the beef cattle profile was obtained by the improved U-chord curvature method. The body measurement points were determined according to the curvature value and converted into three-dimensional spaces to calculate the body size parameters. In this paper, the postures of beef cattle, which were analyzed by a large amount of depth image data, were divided into left crooked, right crooked, correct posture, head down and head up, respectively. The test results showed that the head removal method proposed based on the skeleton in multiple postures hads head removel success rate higher than 92% in the five postures. Using the body measurement point extraction method based on the improved U-chord curvature proposed, the average absolute error of body length measurement was 2.73 cm, the average absolute error of body height measurement was 2.07 cm, and the average absolute error of belly width measurement was 1.47 cm. The method provides a better way to achieve the automatic measurement of beef cattle body size in multiple poses.
What problem does this paper attempt to address?