The Improvement Of ICP Algorithm And Research On Large Scale Point Cloud Registration Method

Three dimensional point cloud registration is an important part of 3D reconstruction,and it has a wide range of applications in various fields.For example,in the industrial field,it can be used to detect whether there are defects in the body parts;In the medical industry,it can be used to simulate human organs and identify the patient's lesions,etc..In recent years,with the improvement of the precision of 3D scanning equipment,it is very easy to get the accurate 3D model of the object.Therefore,the research of 3D point cloud registration algorithm has gradually become the focus of research.The process of registration of point cloud data is the different points of view,time is measured in different coordinate systems of two or more point cloud data through a rotation and translation transformation,they are unified into the same coordinate system,so as to obtain the complete information of the objects and perform the visual operations.The existing point cloud registration algorithm has two major aspects: on the one hand,the most commonly used ICP(Iterative Closest Points)algorithm can meet the requirements of people on the experiment in a certain extent,but when the corresponding point selection is made,the nearest point is simply used as the corresponding point,this will cause the mismatch points of the algorithm,thus affecting the accuracy of registration;on the other hand,when the point cloud data of large scale,the registration process will take a lot of time,causing problems of poor registration algorithm in real time.To solve these problems,this paper mainly from the following points:(1)This paper insight into the registration process and some existing problems of the traditional ICP algorithm and its improved algorithm,and puts forward an improved ICP algorithm based on the feature descriptor of spin-image.In this algorithm,the matchingpoint cloud is processed before the registration,and the basic shape feature of point cloud is reduced while the amount of data is reduced.Then find out the two key points of the point cloud,calculates the matching image feature descriptor rotation point cloud point,and according to the characteristics of the two feature descriptor similarity to determine the nearest point and then complete the registration of ICP,has better convergence effect.(2)In order to effectively solve the problem of poor real-time performance of large data point cloud registration,this paper deeply studies the parallel registration algorithm of point cloud based on GPU(Graphics Processing Unit).Details of the registration process of the EM-ICP algorithm and Softassign algorithm,combined with GPU,realize the parallel algorithm of GPU based on EM-ICP and Softassign,greatly improves the efficiency of point cloud registration,improve the real-time performance of the algorithm.(3)The design and implementation of the registration system of cloud points based on improved ICP algorithm based on the improved algorithm proposed in this paper,by using the object oriented programming method for each module of the system are analyzed and designed in detail.The system is mainly divided into point cloud display,point cloud filtering module and point cloud registration module,and the point cloud registration module uses the improved ICP algorithm proposed in this paper.