3D Object Mesh Surface Reconstruction Based On Depth Image

The 3D object reconstruction, an important issue in machine vision, has been widely used in pattern recognition and robot navigation. We design a RGB-Depth system by fusing TOF camera and color camera for the object surface reconstruction. For this purpose, we study the key issues of the depth image super-resolution reconstruction, multi-view 3D point cloud registration, and 3D point cloud mesh reconstruction in the RGB-Depth 3D reconstruction system.The main contents of this thesis contain the following three aspects:(1) The super-resolution reconstruction methods of fusing color image and depth image is studied. In order to preserve the edge information of depth image, the depth map super-resolution is modeled as an energy optimization problem by modifing a second-order total generalized variation super-resolution method with the color image regularization term and an edge indicator function.The experimental results show that our approach can effectively solve the low depth image resolution.(2) Due to the low accuracy of multi-view point cloud registration, we present an optimization registration method. We calculate the absolute transformation matrix of any point cloud to the reference coordinate system by building a target function and combining adjacent point cloud transformation matrix to minimize the objective function so as to avoid the accumulation of error caused by registration of continuous point cloud.(3) The method of 3D point cloud normal vector estimation is studied. We propose a spherical cover mesh reconstruction method with credibility by considering the relation between signal amplitude of depth camera and the credibility of point cloud information. The proposed method reduces the effect of the noisy 3D point cloud. The experimental results show that the method can effectively reconstruct 3D point cloud mesh surface.