Research On 3D Sheep Body Point Cloud Data Registration And Segmentation Algorithm

In the animal husbandry industry,the traditional contact-type sheep body size measurement not only has low measurement accuracy and time-consuming,but also easily causes problems such as stress response to the sheep.Therefore,the accurate measurement of non-contact body size has become a research hotspot in the current animal husbandry industry.Among them,the sheep body point cloud registration and segmentation technology is the key link in the use of machine vision to achieve noncontact accurate measurement of the sheep body size.It can better register the point clouds collected from different perspectives and accurately segment the sheep body point clouds from the complex background,which can effectively improve the accuracy of the non-contact sheep body size measurement,so as to accurately analyze the sheep's physical signs,such as body size,weight,etc.This paper mainly focuses on three aspects of sheep body point cloud initial registration from different perspectives,sheep body point cloud fine registration,and sheep body point cloud segmentation with complex backgrounds.The main research contents of the thesis are as follows:(1)Aiming at the problem of low initial registration accuracy due to the low initial registration accuracy due to the non-key points that are easy to generate key points based on single feature extraction,an initial registration method based on neighborhood and curvature feature parameters is proposed.Using neighborhood feature parameters and curvature feature parameters to sequentially extract key points of the sheep point cloud from different perspectives,according to the similarity of the FPFH descriptors,the initial registration of the sheep point cloud is completed by the sampling consistency algorithm.Compared with the initial registration method based on a single feature,the initial registration of the sheep body point cloud based on the neighborhood and curvature feature parameters has a better effect and higher accuracy.(2)Aiming at the problem that the traditional ICP algorithm finds the nearest point pair point by point and causes the low efficiency of fine registration,an improved ICP method based on k-dtree is proposed.By establishing a k-dtree index structure for the sheep body point cloud,the speed of searching for the closest Euclidean distance in the target point cloud for the points in the source point cloud is accelerated,and the registration of the point cloud is improved on the basis of ensuring the precision of the fine registration.effectiveness.(3)Aiming at the problem of inaccurate segmentation of sheep body point cloud images with complex background using traditional point cloud segmentation methods,an improved K-means sheep body point cloud segmentation method is proposed.By introducing the curvature information,the distance between the point clouds is redefined,and the accurate segmentation of the significant changes in the curvature of the sheep body point cloud data is realized.At the same time,the use of curvature sorting to select the initial clustering center avoids the traditional K-means segmentation algorithm that randomly selects the initial clustering center to cause the segmentation results to be unstable and the segmentation effect is inaccurate.This ensures the uniqueness of the segmentation results and improves the performance of the segmentation results.accuracy.