Registration And Integration Methods Of Range Images

Recently, with the development of range sensor, register and integration of multiple range images have gained more and more concerns and have been applied in many domains such as CAD reverse engineering, ill product detection, and non- contact measurement of 3D human body. In this paper, the problem of hierarchical triangulation, registration and integration of multiple range images is addressed. Multiple range images can be obtained by turning the 3D object different angles around sensor. The range images of different viewpoints are measured by phase measuring profilometry, then they are registered to estimate the relative transformations among the viewpoints, and are integrated to obtain a complete 3D geometric description of object. Our work mainly includes the following three aspects: Hierarchical triangulation representation of 3D object surface The triangular meshes are often used to represent object surfaces in computer graphics and geometry modeling, because they can represent objects of arbitary shape and are easily constructed from sensed 3D data. The multiple range images in our exprement are 3D distant data sets. A large quantity of closest points and closest distances need to be calculated in our register algorithm, so we should construct triangular meshes of 3D distance data, then construct a k-d tree based on triangular meshes in order to find closest points and closest distance more quickly. Therefore the triangulation construction of multiple range images is important in our register process. We pick up 3D data set according to sampling based on equal intervals. Neighbouring 3D data are connected in order to construct original triangular meshes. In this method, because there are thousands of triangles in triangular meshes, it is not convenient to find and calculate. So we should simplify this kind of orginal triangles. Our mesh simplification algorithm can get hierarchical triangulation of range images by deciding the desired length, acceptable deviation and the maximum allowable shape change measure. Registration of multiple range images In order to integrate range images of different viewpoints, the motion transformations among viewpoints have to be obtained. Our registration algorithm is an integration of the iterative closest point (ICP) algorithm with random sampling and least median of squares (LMS) estimator. The registered range images are represented by 3D data sets while the registering range images are represented by triangular meshes model. Our LMS registration algorithm need largely computations of the closest points and closest distances from a 3D point to triangular meshes model which will consume much time. In order to accelerate computation, we put forward two methods to shorten the compute process: one method is to construct k-d tree based on triangular meshes. the other method is to adopt surface-based nearest point search method along with the orthogonal projection of data point and triangular meshes onto same projection plane. Finally we can obtain two motion parameters by using the two method, we select the more precise one. In addition, we don register range images sequentially in order to reduce and avoid the error accumulation, we construct the star-shaped topology among all view-points, after we select one view-point as the center of star-shaped topology, the remain view-points range images register with it. If star-shaped topology can抰 be constructed, we transform the similar star-shaped topology.