Research Of Registration Algorithm Of 3D Reconstruction Based On Range Image

With the rapid development of science and technology, computer software and hardware technology are rapidly growing. It requires high quality of the objects'reproduce and services to meet the needs of different levels. Rely on the range images obtained from multiple and different perspectives from the acquisition, adjust them under the unified coordinate system, and then operate data fusion, grid process, texture mapping and et al. We can finally get the actual three-dimensional geometric objects model. So the registration of range image is the first key operation, which will greatly affect the final results of the model.This paper shows the main process of three-dimensional reconstruction of objects'model and the basic principles of image registration briefly, and then describes the proposed registration algorithm.Up till now, ICP algorithm has been the most widely used in the domain of range image registration methods, whereas the genetic algorithm has also been used for its simplicity and good global search capability. ICP algorithm requires a good initial position to avoid the uncertainty of the algorithm's convergence, which can not produce the reliable results. So based on the principles of ICP algorithm and the characteristics of the algorithms derived from, this paper presents a method combined a rough registration from the genetic algorithm with the fine registration from ICP algorithm to obtain the fine registration result. In order to avoid premature mature of traditional genetic algorithm, we propose a way of using simulated annealing algorithm to improve the traditional genetic algorithm. The new algorithm accept the vicious solutions conditionally and retains some non-optimal gene to approach the local optimal solution according to Metropolis criterion, which can ensure effective genetic diversity and out of the trap of local optimal solution to make the algorithm converge to the global optimal solution mature eventually.We take the coarse registration from genetic algorithm as the initial position's estimation of ICP. Then we find common areas, select the control points and solve the corresponding point with LTS to refine the rough registration from the improved genetic algorithms and complete the fine registration between the two range images. Experimental results show that the algorithm has no requires to initial position and improves the accuracy of the original ICP algorithm. It has a strong robustness and complete automation without manual manipulation.According to summarize the main research topics, we give the direction of future research of range image registration in the end of this article.