Reserach Of Three Dimensional Measuring Device Based On Dual Kinect Camera

Pig breeding industry is an important industry sector of animal husbandry in China,providing basic food materials and national energy supply for the country.The size of pig body is the important basis for evaluating the health index of pig breeding.The traditional body size parameters are usually measured manually by workers using measuring tape and other tools.This traditional measurement method not only increases the workload of workers,but also makes the measurement difficult and error large.Because of the direct contact with pig body during measurement,it is easy to cause stress response to pigs,which is not conducive to the healthy growth of pigs.We used Kinect camera to obtain the point cloud data of black pig,and studied the point cloud data preprocessing algorithm,point cloud registration algorithm and body size extraction algorithm to realize the non-contact measurement of pig body size.The main work and conclusion are as follows:(1)Combined with the fact that the manual measurement process of pig body size in the farm is tedious,it is easy to cause stress response to pig body,and the high cost of the existing three-dimensional measurement device is difficult to be widely used in the breeding industry,A three-dimensional measurement device based on double Kinect is built,and a three-dimensional measurement software is developed,which has the characteristics of simple scene layout,high measurement accuracy,fast calculation speed and non-contact measurement.(2)In-depth study of point cloud denoising,point cloud reduction and point cloud segmentation and other preprocessing algorithms.With the continuous improvement of the accuracy of 3D scanning equipment,the number of point clouds collected by the equipment is very large,which increases the time and difficulty of subsequent point cloud registration and other operations.In this paper,a k-means clustering point cloud reduction method based on octree is proposed,which takes the real leaf nodes of octree as the initial clustering center,improves the speed and robustness of clustering,and then uses the combination of RMS curvature and Euclidean distance to eliminate redundant point cloud.The result of reduction can not only retain the detailed features,but also the density distance between the simplified point cloud and the original point cloud consistency.(3)Aiming at the problem that the traditional point cloud registration algorithm has poor registration results for two pieces of point clouds with similar features and insufficient overlapping parts and large distance from each other,a fast registration method based on 3D sphere calibration is proposed.Firstly,three spheres are used to calibrate the camera to obtain the initial pose of the point cloud adjusted by the rigid body transformation matrix.Then,based on the traditional ICP algorithm,K-D tree search is used to find the corresponding point pair,and the Euclidean distance threshold and the angle threshold between the direction vector are used to eliminate the wrong point pair.Finally,the global error iteration of the optimized corresponding point is carried out to obtain the optimal registration result.(4)Automatic extraction of pig body size is realized.Firstly,the point cloud after registration is unified into the standard coordinate system established by principal component analysis,then the coordinate system is corrected by using ground normal vector,and the pig body orientation is determined according to the characteristics that the point cloud in the second half of the pig body is larger than the point cloud in the first half.Finally,the measurement point is extracted based on the characteristics of the pig body shape to complete the measurement,which has the characteristics of high measurement accuracy and no stress response.