Research On Technologies Of Recognition And Reconstruction For Dynamic Texture In Augmented Reality

Recently, three-dimensional (3D) reconstruction of natural scenes in Augmented Reality (AR) has become a hotspot in the field of computer vision, which has wide application prospects and important value in practicality in Virtual Reality (VR), natural disasters prediction, games animation, military simulation and other areas.However, reconstruction of dynamic texture in AR is not yet mature, and the key technologies involved need further study and exploration. This paper mainly studies the key technologies of recognition and reconstruction for dynamic texture in AR: Studying a robust recognition method for dynamic texture, we first sdudied the existing recognition method for dynamic texture, then proposed a novel dynamic texture recognition method which combined adaptive optical flow threshold algorithm and variational level set method. Experiments demonstrate that the proposed recognition method can recognized dynamic texture fast and accurately.Camera calibration is a necessary step to recover 3D information from 2D images scene,according to the characteristics of image sequences, we firstly recovered the camera parameters by using ACTS system, and then corrected them by using the proposed correction method of minimizing the mean error to obtain more accrate camera parameters.For dynamic texture reconstruction, first we extracted the corresponding points of dynamic texture region by making use of the results of recognition and adaptive optical flow threshold,then we accurately recovered the three-dimensional coordinates of dynamic texture with camera parameters and corresponding points, and finally we made use of triangulation analysis to complete triangulated 3D points of space and mapped texture into the 3D models.Then we obtained the realistic reconstruction results of dynamic texture.All algorithms proposed have been emulated by programming, and the simulation results demonstrated the validity and feasibility of our methods. Experimental results show that the proposed recognition method is an efficient method with high accuracy and the reconstruction method is fast for dynamic texture reconstruction with strong sense of reality, which laid the foundation for combination technologies of the virtual scene and the reality, and also the interactive technology.In the meanwhile, the research of this paper has greatly promoted the development of reconstruction technology for natural scenes in AR.