Body Size Measurement And Weight Prediction For Dairy Cows Based On 3D Point Cloud

Dairy industry is an important part of animal husbandry in China,the body size and weight are important indexes of dairy farming and dairy production,they have important application values for large-scale and standard breeding of dairy cows.The problems of heavy workload and stress reaction of dairy cows were focused.On the basis of analyzing and summarizing the domestic and foreign research results on body size measurement and weight prediction,kinect v2 was used to obtain the 3D point cloud of Holstein dairy cows.The preprocessing methods,the repair method of the point cloud missing areas and the automatic extraction method of the body size parameters were studied,the weight prediction models were constructed.Finally,the body size measurement and weight prediction of dairy cows were achieved without contact,which provided quick and easy methods for obtaining cow's body size and weight.The main research work and conclusions are as follows:(1)Point cloud of dairy cows were acquired and pre-processed.Holstein dairy cows were taken as the research object and the initial point cloud data of 45 cows were collected by kinect v2,5 frames of point cloud were gotten from every cow,a total of 225 frames were acquired.The initial point cloud of dairy cows included background point clouds such as ground,horizontal railings,vertical railings,and outliers.The through filtering method was used to remove part of the point cloud.The ground point cloud was removed using the template matching method with normal lines.The horizontal rail point cloud was removed using the line template matching method and the series bounding cylinder method.Euclidean clustering method was used to remove vertical rail point cloud.The average distance estimation method was used to remove outlier data.Finally,point clouds containing only individual cows were obtained.(2)There was a missing area in the point cloud after pretreatment.A method for repairing the missing area of point cloud based on improved cubic B-spline curves was studied.Firstly,the extracted dairy cow's point clouds were sliced and projected along the x-axis direction of the point cloud coordinate system.In each slice points,some points were filled in the adjacent points with larger spacing.Finally,cubic B-spline curves were used for fitting the filled slice points.A total of 225 frames dairy cows' point clouds were repaired using the optimal values of parameters h and L.The results show that the average frame approximation error is reduced by 26.70%,and the large missing area on the point clouds are repaired and the sparseness of point clouds has also been improved.The proposed algorithm has better uniformity and approximation performance than the cubic B-spline method,which provides an effective method for repairing large missing point cloud area.(3)Automatic body size measurement methods for dairy cows based on geometric characteristics were studied and proposed.In order to achieve automatic measurement of 6 body size parameters such as body height,body straight length,body oblique length,belly width,shoulder width,and body depth that were highly related to cow weight,the methods of body size parameters extraction based on geometric characteristics was studied.The results showed that the mean absolute error of body depth was 0.0097 meters;the mean absolute errors of body height,belly width,and shoulder width were both less than 0.0200 meters;the mean absolute errors of body oblique length and body straight length were 0.0765 meters and 0.0771 meters,respectively,which were larger than the other 4 body measurements.The mean relative error of body height was 0.9201%,which was the smallest among 6 body measurements;the mean relative errors of body depth,belly width and shoulder width were all less than 4%,second only to body height;the mean relative errors of body oblique length and body straight length were 4.8623% and 5.0471%,which were larger than the other 4 body measurements.The mean relative error of six body measurements was 3.0687%,which had higher accuracy and could meet the weight prediction requirements.(4)Dairy cows' weight prediction models were established and optimized.The data of weight and body size parameters of 45 groups of cows were divided into 35 training groups and 10 verification groups.Respectively,multiple linear regression model,regression analysis model,partial least squares model,BP neural network and RBF neural network model were used to construct cow weight prediction models.The verification groups samples were used to verify the prediction accuracy.The results showed that partial least squares prediction model had higher prediction accuracy than other prediction models.The absolute error of partial least squares prediction model was between-20.903 kilograms and +27.089 kilograms,and the root mean square error was 17.236 kg.The relative error was between-3.024% and +4.915%,and the relative root mean square error was 2.867%.Finally,the partial least squares model was chosen as the weight prediction model of dairy cows,which had a higher prediction accuracy.