Registration Of Point Clouds And 3D Reconstruction

The privacy of point cloud goes better and the number of points becomes larger with the development of the optical equipment.The increase of the number leads to a more complexity,which contributes to more toughness of finding an accuracy result and realizing a fast registration.A fast and robust registration method is hard to find in the 3D Object reconstruction.This article concentrates on the aspect of slowness.This paper proposes Bi-ICP,Partition-Based-Trimmed ICP to solve the slowness of registration.About the accuracy and the ratio propability in the Trimmed ICP,the paper puts up the homogeneous search method to registration and get the ratio propability.On the strategy of registration of point clouds,based on the optimization of linear topology network,the paper puts up the min unit of the local registration method.With the methods,this strategy keeps the shortness of running time in sequential registration and higher accuracy of accumulated registration.The registration method between two point clouds is the foundament of multi-view registration.The most widely-used algorithm in registration is ICP.Based on ICP,this paper improves the algorithm to Bi-ICP and Partition-Based-Trimmed ICP,a higher speed in registration is get.The equivalence of rotation matrix and quaternion matrix tells the step width of each iteration in the form of angle.Bi-ICP improves the ICP method by the addition of rotation angles in two directions,which increases the step width and the registration speed.For the Partition-Based-Trimmed ICP method,this paper improves the registration process of point clouds of different types,based on the Trimming ICP method.Since the contribution on the convergence speed of different point-pairs is different,the speed of registration goes up as in the process of pruning ?-ball.Meanwhile,this article introduces the Gaussian Model into the Partition-Based-Trimmed ICP method to search the ?-ball and -branch adaptively.In order to solve the problem of finding homogeneous areas precisely,the entropy is brought in to search the homogeneous point-pairs accurately.The relationship of Entropy and RMS is built up and it leads to the function of searching homogeneous point pairs.The theory of dissipative structure and convex functions help to find the homogeneous point-pairs.Finally,an accurate and robust method for registration is complete.Furthermore,the experiments on Standford Repository and Mian's Datasets prove the excellence of this algorithm.With the expectation of acquisition of fast registration in sequential registration strategy and accurate registration in accumulated registration strategy,this article proposes the smallest-topology-unit strategy.With the complement of local registration and original point cloud as the links of neighbor units,it improves the RMS of registration and the time of reconstruction successfully.Despite of the creations above,this article optimizes the Delaunay Triangulation process with the characteristics of constrcture lighting.The geometric transform and the coding features of Gray code and phase shifting guarantee a fast and robust triangulation method.A merging method based on the Bi-ICP and Partition-Based-Trimmed ICP obtains a faster registration.With the optimization above,the registration of point clouds obtains a better RMS and faster speed and then,the construction of the small scene object is complete.