Research Of 3D Reconstruction Method Based On Kinect

Three-dimensional(3D)reconstruction is to determine the spatial location or stereo geometrical shape of objects based on multiple two-dimensional(2D)images or equipment like depth camera.The former is based on stereo matching algorithms,which is always complex and time-consuming,thus greatly limits the application in real projects.However,the latter is based on relative devices,such as TOF or Microsoft Kinect,and is widely used due to its low price and compact shape.With its camera,Kinect can directly obtain the 3D information of the objects or scene,and then reconstruct a 3D model.This paper mainly focuses on the study of 3D reconstruction method based on Kinect.First,this thesis introduces the working principle and error characteristics of Kinect,describes the correction and optimization method of initial depth images obtained by Kinect.For the shortcoming of traditional bilateral filtering,it is improved in this thesis to fulfill the lost data.The effectiveness of the proposed method is proved by the experimental results.And then,we propose an improved method,it can correct and align the depth and RGB image,grow the boundary of depth image to that of color image using region growing algorithm,and then exclude the wrong values.This thesis focuses on the improved depth image restoration method based on fast marching algorithm,which can improve the weight parameters of depth values and image restoration priority.Finally,the depth images are processed by the improved bilateral filtering algorithm.The simulation results show that the proposed method can obtain better results in the boundary regions and more accuracy.The thesis finally discusses the surface reconstruction method using optimized depth map.We first introduce the foreground objects extraction method,as well as the point cloud reconstruction method from 2D to 3D.In order to unify the point clouds from different coordinate to one coordinate,multiple point clouds are registered and fused with ICP algorithm in this thesis.Moreover,in order to reconstruct a smooth 3D surface model,the point clouds are triangulated using the greedy algorithm.The simulation results show that the proposed method can accurately extract foreground objects,complete the registration and fusion of multiple point clouds,and reconstruct a smooth 3D surface model.