Research On 3D Laser Scanning Point Cloud Data Registration Algorithm

With the rapid development of 3D scanning technology and the computer data processing technology,the three-dimensional point cloud data,with its unique advantages,is playing an important role in more and more areas,such as reverse engineering?defect detection?heritage conservation?3D printing and so on.However,being limited to the sight range of the coordinate measurement device,the size of the object and the influence of the environment,sometimes it is unable to get the full three-dimensional point cloud data of the object with single perspective stitching.Thus,It is needed to put the three-dimensional scan point clouds together that obtained under different perspectives stitching,which is known as the registration of two piece of three-dimensional point clouds.The target of registration of two piece of three-dimensional point clouds is to finding the rotation matrix and the translation matrix between the two piece of point cloud.In order to realize the purpose of registration of two piece of point clouds,However,the process of present registration of two piece of three-dimensional point clouds has following problems such as lowly registration precision and slowly compute speed.To improve the problem of above problems,the following study and experiment have been carried out in this study.The basic mathematical principle about the registration of point cloud date has been summarized.The two piece of point cloud acquired from different perspectives are set as the sources point cloud and the target point cloud respectively.Based on the FPFH descriptor and a sample consensus method,the rough registration of two piece of point cloud has been completed.By extracting the key point of points of point cloud date and calculating the surface normal of key points,then using the normal feature to calculate FPFH feature descriptor,and using the sample consensus method to finish the rough registration.The experimental results show that using this method can effectively optimize the initial matching position of the point cloud date.Making use of the initial value obtained from the rough registration and combining with the recent iteration point(ICP)algorithm to achieve accurate registration of point cloud data.In order to reduce the amount of date of point cloud data and to improve the calculation efficiency,a voxel grid method is introduced to streamline the point cloud data.Reuse RANSAC algorithm to remove mismatch dotted pair to improve the accuracy of registration.After the reconstruction of precis matching dotted pair iteration is repeated until with the initial iterative value until a certain condition is satisfied.Compared with the traditional iterative closest point(ICP)algorithm,the improved iterative closest point(ICP)algorithm is greatly improved in matching accuracy and computing speed.Finally,with PCL(point cloud library),the data of dragon Stand and bunny from Stanford university point cloud database,an experiment has been done,in which the ICP method has been compared with the method suggested in this study.The experiment result has shown that the later is superior to the former in such aspects as time saving and matching degree.