Body Size Measurement For Type Classification In Holstein Based On 3D Model

The type classification in Holstein is critical for selecting high producing dairy cows and standardize their breeding,which has important application value.Aiming at solving the problem that a dairy cows’ complete 3D model can’t be reconstructed with high precision by a single device,and to improve the precision and automation of type classification for dairy cow,a point cloud data acquisition system with FM810-GI depth camera as the core was designed.The pretreatment method of point cloud data,the reconstruction method of three-dimensional model of dairy cows and the automatic extraction of linear evaluation body size of dairy cows in point cloud were studied.The research content of this paper aims to provide reference for type classification in Holstein.The specific work contents and conclusions are as follows:(1)Dairy cows’ Point Cloud Data Acquisition System was developed.Dairy cows were taken as the research object,,a method for 3D reconstruction and the measurement of the body size parameters of the dairy cow based on multi-view cloud data was proposed.A hardware system of dairy cows’ point cloud data acquisition was developed,which was composed of FM810-GI,trigger hub,fixed mechanism and computer.In the dairy farm experiment,a multi-channel synchronous acquisition software was designed to obtain three views(left,right and upside)point cloud in the path from the activity field to the milking room.The analysis of the obtained dairy cows’ point cloud model showed that the dairy cows’ point cloud data acquisition system could acquire data more stably.(2)In order to remove the noise of dairy cows’ point cloud data,the data preprocessing methods were researched and selected.The through filtering was used to remove the ground of the dairy cows’ top-view point cloud data,and used to remove most of the background of the dairy cows’ left-view and right-view;The ground in the dairy cows’ left-view and right-view point cloud data was removed by plane template matching method;The sparse outliers in the dairy cows’ point cloud data was removed by the Statistical Outlier Removal filter,and finally the dairy cows’ point cloud data with the background point cloud removed was obtained.This series of data preprocessing methods laid the foundation for subsequent3 D reconstruction.(3)To improve the reconstruction accuracy,a dairy cows’ three view cloud data splicing method based on iteration after obtaining coordinate transformation matrix bystereo calibration was proposed.This method improves a stereo calibration method for three depth cameras,the position relationship of depth cameras was obtained by stereo calibration,and the rotation matrix and translation matrix cloud provide the rough registration.After iteration of ICP algorithm,by using K-Means clustering method to remove redundant point clouds in overlapping areas,a more accurate three-dimensional reconstruction model of cow point clouds cloud be finally obtained.The experimental results show that the maximum real error of calibration is 1.860 mm,It could provide rotation matrix and translation matrix for registration so as to avoid the problem of rough registration which was easy to mismatch by finding feature points.Through error analysis on the length,height and width of the real cow and the 3D cow’s model,the maximum relative error was 1.17% with a high Precision,which could meet the requirements of automatic extraction of body size parameters for type classification in Holstein.(4)The parameters of the dairy cows’ body size were automatically extracted.The automatic measurement of 9 parameters including body height,body depth,chest width,rump angle,rump width,hoof angle,hoof depth,hindlimb side view and breast depth were realized.The parameters of the dairy cows’ body size were automatically extracted and compared with the manually extracted parameters for accuracy analysis.The experimental results showed that the average relative errors of hoof angle,body depth,rump width,heel depth,body height,chest width and lateral vision of hind limbs were 3.80%,rump angle’s average absolute error was 14.9 mm,but the errors of breast depth extraction were large.Except for the depth of the breasts,there was a good linear correlation between the body size parameters automatically extracted and the manually measured values.The maximum determination coefficient2R(hoof angle)was 0.9728,and the minimum(hoop angle)was0.8531.The overall results indicated that the dairy cow reconstruction method was precise and the measurement accuracy met the assessment requirements.The method was applicable in the type classification of Holstein and provides a reference for the type classification of other farming animals.