| The body size parameters and body weight of pigs are important indicators of pig breeding and production,which have important application value for large-scale and standardized breeding of pigs.Traditional body size parameter acquisition and body weight measurement use direct contact measurement method,which requires more staff and larger site requirements,and is easy to cause stress response of pigs.Aiming at the defects of traditional methods,this paper adopts a method of pig body size parameter acquisition and weight estimation based on point cloud,which provides a fast and convenient method for pig body size parameter acquisition and weight estimation.The main work and conclusions of this paper are as follows :(1)Construction of measurement platform and collection of pig point cloud data.Two Kinect cameras are calibrated by Zhang Dingyou calibration method to obtain the internal parameters,external parameters,distortion parameters and the relative position between the cameras.(2)The original scene cloud data prepossessing.Firstly,the filter and the outlier detection based on multivariate Gaussian distribution are used for the coarse denoising of the point cloud to remove the noise of the scene point cloud and the obvious outliers in the geometric sense.At the same time,the plane removal algorithm based on RANSAC is used to remove the background initially.Secondly,a point cloud segmentation method based on color region growing is proposed.The clustering index of point cloud is judged by HSV threshold to complete the segmentation of point cloud.Finally,a point cloud reduction method based on fuzzy C-Medoids optimized cuckoo algorithm is proposed.The research shows that when the reduction rate is 25 %,the algorithm retains the edge and other feature information well,and the degree of sparseness is consistent with the initial point cloud.(3)Pig point cloud registration.In this paper,a key point extraction method based on unsupervised pig point cloud is proposed at first.By comparing the initial key points obtained in the original pig point cloud with the key points extracted in the second time after random matrix transformation,the two key points are regressed by the feature scale transformation loss function and the point-to-point loss function,and the high-value pig point cloud key points are obtained,and the feature descriptors of these key points are obtained.SAC-IA coarse registration is carried out through these key points,and ICP based on Octree fine registration iteration is carried out on the basis of coarse registration,so as to obtain the complete point cloud of pigs.Research shows that the registration speed of this algorithm is about 40.2 % higher than the traditional ICP algorithm.(4)Collection of pig body size parameters and estimation of body weight.In this paper,the coordinate system is corrected by ground normal vector,and the pig body size parameters are obtained by geometric relationship.By comparing the multiple linear regression model,stepwise regression analysis model and Robust regression model,a better method was selected to fit the weight of pigs.Among the four indicators of absolute error,relative error,root mean square error and relative root mean square error,the estimation results of the fitting model based on stepwise regression are better than those of other fitting models.The absolute error of weight estimation is-3.801～3.258,the root mean square error is 1.846,and the relative root mean square error is 2.59 %,which has high estimation accuracy. |
PDF Full Text Request