Reparation with moving least squares sampling and extraction of body sizes of beef cattle from unilateral point clouds

Body measurements of beef cattle at various growth stages are crucial for assessing breeding quality. To decrease the stress responses of live cattle during measurement, noncontact body size measurement using a deep sensor is conducted. After directly obtaining the point cloud data (PCD) of a unilateral surface, novel point cloud repair and body measurement are proposed. The PCD of beef cattle are first preprocessed using Euclidean clustering, conditional filtering, and Random Sample Consensus. To merge the unilateral PCD into the entire contour, a Euclidean transformation matrix and moving least squares are applied. To address holes caused by stitching due to camera angle restrictions, various upsampling techniques based on least squares deformation are investigated to repair holes in different surfaces with varying growth stages. Finally, polynomial fitting is chosen to be in accordance with the feature point distribution of the body height (BH), chest girth (CG), back height (AH), waist height (WH), ischial tuberosity, and shoulder end of beef cattle. Using the formulas of Euclidean distance and cubic B -spline curve fitting, the values of BH, AH, WH, CG, and body length are obtained and then compared with manual measurement data. The predicted values of the five body sizes of 100 beef cattle had a mean absolute error and mean absolute percentage error of 7.77 cm and 5.42 %, respectively. The largest absolute error was 12.05 cm, and the smallest absolute error was 4.12 cm. A series of PCD processing and repair methods provides a new method for the noncontact automated measurement of beef cattle. This approach promotes more effective practices in livestock welfare and trait assessment compared with traditional management.