Research On Three-dimensional Image Mosaic Algorithm

In recent years, three-dimensional reconstruction plays a bigger and bigger role invarious fields, so three-dimensional image mosaic technology has also attracted moreattention. Three-dimensional image mosaic technology mainly processes the gotten pointcloud data. Then implements registration of point cloud data those need to be registered inproper order.The article mainly takes three-dimensional point cloud data as research object. Itsobjective is to study how to establish the topology relationship between scattered point clouddata and how to implement registration of point cloud data. To a certain extent, it solves theproblem of improving the efficiency of algorithm. In view of this, in this paper an improvedalgorithm whose direction can be controlled for finding k nearest neighbors of secondaryrasterizing point cloud data is proposed. Also an improved algorithm of iterative closest pointwhich is based on the unit quaternion is proposed. The paper’s main work is as follows:Firstly, it briefly introduces the method of how to get point cloud data. And the methodto establish topology between scattered point cloud data has been introduced in detail,including octree method, k-dtree method, rasterization method and so on. Also the objectiveevaluation about various of method has been given. Then rasterization methodimplementation principle has been studied specifically.Secondly, based on space grids division principle and the controllability of searchingextension direction, an improved algorithm whose direction can be controlled for finding knearest neighbors of secondary rasterizing point cloud data is proposed. In the first place, theminimum bounding box of point cloud data will be secondarily rasterized and to reduce thepoints contained in each grid. Then in the process of searching the closest point of candidatepoint, reducing the number of grid those need to be searched by the means of controlling thesearch range extension direction and reduce the number of points those need to be compared. Eventually make the efficiency of algorithm for searching the nearest point has a certaindegree of improvement. It verifies the correctness and rationality of the algorithm from twoaspects of theory and practical experiment.Thirdly, with the analysis of its principle and the existing problems, this paper brieflyintroduces the process of classical ICP registration algorithm. On the basis of the unitquaternion been used to solve the transformation matrix parameters, taken the distancebetween the corresponding points as the similarity metric standard,, in the front and backiteration, take the distance difference corresponding points for convergence condition, animproved algorithm of iterative closest point which is based on the unit quaternion isproposed. Combined with improved rasterization closest point search algorithm, it shortensthe closet point search time and makes the efficiency of it own improved. And it verifies thecorrectness and rationality of the algorithm from two aspects of theory and practicalexperiment.