An improved PointNet++ point cloud segmentation model applied to automatic measurement method of pig body size

Abstract

Currently, non-contact and automatic measurement of livestock body size is the research orientation of computer-aided livestock breeding and intelligent farming industry. However, directly locating the body size measurement key points on the dense overall point clouds of livestock often results in positioning deviation and affects the measurement accuracy. Additionally, various postures of livestock and the interference of different parts of livestock body are also prone to key points positioning deviations. For more accurate measurement of pig body size, we propose in our paper an improved PointNet++ point cloud segmentation model to subdivide the overall pig point clouds into various parts, such as the pig’s head, ears, trunk, limbs, and tail to localize the body measurement key points in the segmented local parts. A new body size measurement method based on segmentation results, which is integrated with the least squares, point cloud slicing, edge extraction, and polynomial fitting, is also presented in our study so that the pig body size parameters can be more accurately calculated. In our experiment, 25 live pigs and 207 groups of pig body point clouds were used for point cloud segmentation and body size measurements. Compared with manual measurements, the relative errors in our experimental results are listed as follows: body length: 2.57 %, body height (front): 2.18 %, body height (back): 2.28 %, body width (front): 4.56 %, body width (middle): 5.26 %, body width (back): 5.19 %, thoracic circumference: 2.50 %, abdominal circumference: 3.14 %, rump circumference: 2.85 %. To conclude, the new automatic measurement method based on the improved PointNet++ point cloud segmentation model with higher accuracy has a more promising application prospect thanks to its novel features, precise measurement results and stable robustness.