Kinect-based Measurement Of Animal Signs Under Complex Conditions

Since the advent of the new era of socialism with Chinese characteristics,the living standard and quality of life of the Chinese people have gradually improved.The need to obtain animal signs without causing stress reactions is increasing.Moreover,animal protection personnel want to measure the body characteristics of animals without disturbing them,and the requirements are gradually increasing.Therefore,the method in this paper makes it possible for breeders and animal protection personnel to realize the hope of obtaining animal sign values without contact measurement.Contactless measurement has always been one of the main research directions of machine vision.In this paper,point cloud reconstruction is realized by using Kinect depth vision camera to obtain the point cloud information of pigs.Then,the volume,surface area,body length,body width and body height of pigs were obtained through point cloud reconstruction to predict the weight of pigs.This paper introduces in detail the classical algorithms of point cloud reconstruction: point cloud filtering,point cloud segmentation,point cloud reconstruction,and introduces the classical SVR algorithm in machine learning.After several experiments,this paper proposes a combination of improved voxel filtering algorithm and Gaussian filtering algorithm in point cloud filtering.In point cloud segmentation,a segmentation algorithm based on random sampling consensus algorithm(RANSAC)combined with conditional European clustering is proposed.A surface reconstruction algorithm based on upsampling fusion α-shape is proposed for point cloud reconstruction.For weight prediction,the α-shape algorithm combined with SVR algorithm was used to predict the weight of pigs.The body state features of pigs obtained by the point cloud reconstruction technology combined with machine learning algorithm in this paper are compared with the actual measurement of pig body state feature values.Finally,it is found that the relative error between the pig’s body state eigenvalues obtained by the algorithm in this paper and the actual measurement is within 5%.