Research On Key Technologies Of Depth Image Repair And 3d Reconstruction In RGBD Visual Scene

Three-dimensional reconstruction is a process of restoring three-dimensional information of objects in visual scene through two-dimensional image data or three-dimensional point cloud data,and it’s widely used in the fields of three-dimensional digital city,augmented reality,cultural relics protection and so on.With the improvement of depth image acquisition quality of RGBD camera and the demand for three-dimensional refined model in various fields,three-dimensional reconstruction technology based on RGBD data has gradually become a research hotspot.Taking the RGBD data in visual scenes as the experimental data,This paper expands the research and innovation of several key technologies in the process of holes repair and three-dimensional reconstruction of depth images.This paper contains the following sections:(1)An adaptive hole repair algorithm considering edge blending features is presented to solve the hole repair problem in the feature area of deep image edge blending.The algorithm combines the edge information of RGB and depth images,and adds the edge information to the weight value decision in the filter window.Then,the adaptive joint bilateral filter algorithm is used to repair the edge.This algorithm can effectively repair the holes in the edge blending feature area.It ensures that the depth values of the holes in the edge blending area of the depth image are accurate after the repair.(2)Aiming at improving the registration rate and accuracy of point cloud,a point cloud registration algorithm combining NDT and trimmed ICP is proposed.The algorithm combines the characteristics of fast registration rate of NDT algorithm,and the problem that there is only partial overlap in the registration between point clouds.Through the rough registration of point cloud by NDT algorithm and the accurate registration of trimmed ICP algorithm,the algorithm can not only quickly register point cloud data,but also ensure the accuracy of point cloud data registration.(3)Aiming at the problem of three-dimensional reconstruction of point cloud,several common three-dimensional reconstruction algorithms based on mesh triangulation and surface fitting and the construction of color texture of three-dimensional model are studied.The RGBD data of visual scene in the registered is used as the experimental data,and different threedimensional reconstruction algorithms are used for three-dimensional reconstruction experiments.Through experimental analysis and comparison,it is concluded that the greedy projection triangulation algorithm is more suitable for establishing three-dimensional models in visual scenes.