Study 3D Reconstruction Of Indoor Scenes Based On Kinect

3D Three dimensional reconstruction is a hot and difficult topic in recent research. It has been widely used in computer vision, virtual reality, reverse engineering and so on. In order to realize the 3D reconstruction of the scene, the depth map of the Kinect camera is studied and the 3D model of the scene surface is reconstructed in this paper.First of all, this paper is to calibrate the Kinect camera to determine the camera's internal and external parameters as well as the relative position of the infrared camera and depth camera. In order to get the imaging model of the camera, the camera is studied based on the principle of pinhole imaging, the model of which is obtained by using the geometric relationship, and the camera calibration board is used to calibrate the camera.Then the third party library of PCL, OpenNI, is used to control Kinect camera, and grab the depth map and color images taken by the Kinect camera. The depth image noise and visual difference generated by partial occlusion of small black area are handled by using Open CV of depth image processing. The scheme of mean filtering, morphologicalfiltering and median filtering are analysised in this paper. Finally, the filtering algorithm is improved and can achieve satisfactory filtering effect.Point cloud registration is a key step in 3D reconstruction, it is to solve the problem of point cloud registration in the same coordinate system under different coordinate system. This paper uses iterative closest(ICP) algorithm, to achieve the reconstruction of the point cloud,ICP algorithm can achieve accurate registration of point cloud, but it needs to be well initialized. Therefore, the point cloud is initially registered based on the feature point histogram in this paper, and then registration is accuratly matched after the initial registration of point cloud by ICP, and the ICP algorithm is improved appropriately, in order to reduce the computation of the ICP algorithm, voxel filtering principle of point cloud is used at the beginning of the sample, and setting threshold filtering in the ICP distance of corresponding points out.Finally, in this paper, the reconstruction of point cloud were triangulation, using Poisson algorithm to achieve the disordered point cloud triangulation and generate triangular facets. Poisson triangulation algorithm based on the normal vector of a point each for neighboring points, generate triangular meshes, this paper on Poisson algorithm improvement, set the distance threshold, for the triangle, the length of a large filtration, eliminate error patches.The data information is obtained based on a low cost rgb-d equipment by using PCL to deal with the point cloud, in the process of part of the algorithm was improved, to achieve the scene in the 3D reconstruction. It is convenient for application in real life.