Study On 3D Reconstruction Of Monocular Sequential Images

Vision is the main way for human to perceive the external world.The purpose of computer vision is to enable the computer to have the function of human vision,to be able to perceive and understand images,to restore the description of three-dimensional scenes,and to complete special tasks in a specific environment.3D reconstruction has always been one of the most popular research directions in the field of computer vision.It is a technology that studies how to obtain 3D information of objects in space through 2D information and establish mathematical models of 3D objects that suitable for computer representation and processing.Therefore,3D reconstruction is a common scientific problem and core technology in computer-aided geometric design,computer graphics,computer animation,computer vision,medical image processing,scientific computing and virtual reality,digital media creation and so on.There are many methods for 3D reconstruction,and the most widely used method is structure from motion based on multiple images.This method simulates the visual model of human eyes.It has the advantage of intuitive to use and great expansibility.The main work of this thesis is as follows:This thesis discusses the research status of 3D reconstruction at home and abroad,introduces the related theoretical basis of 3D reconstruction based on image sequence matching,including coordinate system,camera model,camera calibration,polar geometry,basic matrix estimation,camera attitude estimation,triangulation reconstruction,etc.A basic matrix estimation method based on matching distance is proposed,which is combined with polar distance and residual difference to optimize the estimation of basic matrix.Due to the existence of camera distortion,after removing the data with relatively large distortion on the edge,the proposed algorithm is used to estimate the fundamental matrix,and the optimal fundamental matrix is obtained when the matching distance is the minimum.Experiments show that the proposed method has advantages in average residual and average polar distance while implementing robust basic matrix estimation.In order to solve the problem of point cloud sparsity in the reconstruction model of the structure from motion algorithm,an improved reconstruction method is proposed.First,the matching data is generated by contrast context histogram and estimate the image basic matrix.Then,the basic matrix is decomposed to obtain the translation and rotation matrix.Then,the projection matrix was calculated combine with the camera internal parameters.Finally,generate the 3D point cloud using triangulation method.The algorithm has high matching precision and robustness.Realizes feature point matching through displacement,which makes up for the disadvantages of matching data insufficient in the low-frequency region of image.Experimental results show that the point cloud generated by this algorithm is denser than that of the existing algorithm.In real environment,the algorithm can be used to reconstruct 3D objects.