The Research Of 3D Model Construction Based On Kinect

The technology of 3D model construction has realized the digital representation of the real space information,and it could reproduce the 3D model of objective reality in real in the computer.At present,because the 3D reconstruction is increasingly used in various industries and fields,the validity and accuracy of the 3D reconstruction has attracted high expectations.This thesis studies 3D construction based on Kinect by the method of volume reconstruction.We arrive at the method of 3D construction using the global data cube which is obtained by voxelization of model.The volume reconstruction method is decomposited into two processes: voxelization of model and loop closure detection.The voxelization consists of both the calculation of the SDF value and the fusion of the color texture.The 3D motion estimation is needed to get the global data.The research content mainlyincludes:First of all,a 3D reconstruction method based on voxel has beenproposed.This method is carried out for 3D reconstruction according to the properties value of the voxel,so it needs to solve the problem of the initialization calculation and update of the values in the voxel.In this thesis,According to the depth and color information collected by Kinect,the distance between the point and the surface which is called the SDF value is calculated and the color texture of coincidence images are fused.As the large amount of SDF value calculation,this thesis uses the distance of the projection point to the local surface to replace the SDF value,and updates the image information of different frames to the corresponding voxel.The calculation and fusion results are optimized using different weight functions.Secondly,a 3D motion estimation method based on optical flow method has been proposed.According to the method,the image roration and translation variables are calculated to estimate the camera pose.The position of 3D model in the body is calculated by the camera pose.Based on the motion of the rigid body and the Lucas-Kanade optical flow method,this paper presents the calculation formula of the motion parameters of the 3D model.And based on the theory of image Pyramid,this method calculates the 6 degrees of freedom movement parameters of 3D model from the image of Pyramid,and calculates the rotation matrix and translation vectoraccording to the motion parameters,then utilizes the last frame image's camera pose to calculate current frame's camera pose.Finally,a method of loop closure detection which is based on visual dictionary has been proposed.The method extracts features from every single frame to generate visual words,then it builds visual vocabulary.It relies on visual word matching between images to estimate image similarity.It is by means of posterior probability distribution of closed loop hypothesis to predict whether the next frame image is closed.The experiment indicates that this method can accurately detect the closed loop and it can prevent continuous introduction of redundant image which could lead to decline of reconstruction accuracy.This method improves the efficiency and accuracy of 3D reconstruction.In this thesis,theoretical analysis and experimental results show thatThe research of this paper can obtain a good result of 3D reconstruction.Our researchis helpful for the development of 3D reconstruction technology.