Fully Automatic Registration Of Point Clouds In Frequency Domain

In recent years, with the development of science and technoloty and the computer technology, reverse engineering has become an emerging science rapidly development. As a research hot spot of computer technology field, reverse engineering technology which can produce digital model via an object physical model is obtained more and more widely used. This paper makes a detailed introduction of the registration technology of point cloud data in reverse engineering.Via scan the sme object with terrestrial laser scanner in different position and different directions to obain the point cloud data which have repeat regions. Because each scan was carried out in the current coordinate system, so it needs to find a single coordinate system, and establish the corresponding relation of one to one region by the overlapping area, then align and merge the point cloud data in the single coordinate system, in order to obatin the complete data model of the measured object, this process is called point cloud registration.Most classical automatic registration techniques are iterative in nature, or require user intervention in order to specify matching feature points. While the iterative algorithm is easy to fall into local maxima, and depends on the initial registration position, thus easily leads to wrong registration. Therefore, it is necessary to improve and perfect the existing registration algorithms, or find a new registration algorithm to avoid the existing problems in the method of the introduced algorithms. The registration method proposed in this paper is a new kind of point cloud registration method.To the splicing problem of multi station point cloud obtained from terrestrial laser scanning, this paper proposes a new technique for the registration of point cloud, the method in this paper changes point cloud from space domain to frequency domain by the Fourier Transform, and the registration will be executed in frequency domain. The method presented in this thesis does not require an initial estimation of the transformation, and avoids problems of the classical iterative techniques by employing the Fourier transform, which decouples the estimation of rotational parameters from the estimation of the translational parameters. By interpolation, the point cloud will be grid, and the points evenly distributed in the grid, this is prepare for the point cloud change form spatial domain to the frequency domain by the Fourier transform. According to the features of terrestrial laser scanning, the axis of rotate is vertical, only the angle around the z-axis is the rotate angle, the other rotation angles which is rotate around x-axis and y-axis is 0. Fourier-Mellin transform is extended to three-dimensional, deduced the transform theory and rotation and scale theory, and then transformed the data into log-polar transformation, in the log-polar coordinates, rotation and scale parameters can be isolated, after determining the rotation and scale parameters, then calculate the translation parameters.Point cloud registration algorithm has varying needs in terms of registration precision, and the computational complexity and robustness. The advantage of the algorithm is efficient in anti-noise, high robustness, and can greatly save the computational time. In this paper, the application of Fourier transform in different platforms can be quickly realized, with wide availability, and according to the different hardware platforms, the method proposed in this paper can also be highly scalable, and meet the accuracy requirements. Finally, the experimental results also show that, the algorithm in this paper as compared with the traditional algorithm has certain advantages.