Research On Key Technologies Of 3D Reconstruction Based On Depth Image

Three-dimensional reconstruction is one of the research hotspots in computer vision and this technology has been widely used in medical research,the protection of cultural relics,game design,and other fields.3D reconstruction include depth data acquisition,point cloud registration and fusion,and generating surface.This paper focus on depth image inpainting and point cloud registration,and put forward a novel method.Kinect is developed by the Microsoft company for a somatosensory game,but it aroused people’s attention greatly.Then it is used to acquire depth image.Because of the limited of Kinect sensor’s distance measurement,there are several artifacts that one encounters in the depth map like holes,noise and so on.In this paper,a novel edge information to guide the filtering based strategy to improve the quality of the depth map.First of all,both color image and raw depth data are used to extract initial edges,then fuse these edges together to get the final detection result which as the guidance.Secondly,combining the edge information and the iterative median filtering algorithm to fill the black holes.Finally,the adaptive median filter can be used to smooth noise.Experimental results show that the proposed approach is able to generate high quality depth maps and the edges of the objects are sharpened.Point cloud registration is the first key operation for 3D reconstruction,which will greatly affect the final results of the model.Up till now,ICP algorithm has been the most widely used in the point cloud registration.But ICP algorithm requires a good initial position and lack of stability and the ICP algorithm may not converge to a global minimum,which can not produce the reliable results.For the above problems,this paper proposed a novel point cloud registration method that is a Hausdroff distance based on D4PCS point cloud registration algorithm.Firstly,based on Hausdorff distance of principal curvature was used to keep and extract the feature points,and exclude the non-criticle points for point cloud simplification.Then,D4PCS algorithm is used for coarse registration.Finally,ICP algorithm is used for refinement.The result show that our algorithm can align point clouds with lower overlapping ratio compared with ICP and more general shapes.