Research On Method Of3D Reconstruction By Image Sequences And Error Analysis

One of the main contents of computer vision is3D reconstructionwhich reconstructs the3D information from2D images. The newreconstruction technology based on image sequences is produced by thecombination of the computer vision and the computer graphics. Given aset of2D images, we can recover the3D model of the objects. Theresearch results have a significant application prospect, and presenttremendous value.The problem of reconstruction is an inverse problem of the3Dinformation reconstructed by2D images, which is determined by thecomplexity of the imaging model and very vulnerable to the influence ofthe noise. Multiple links of reconstruction may introduce errors, so it isvery necessary to analyze how the error affects the reconstruction for improving the reconstruction accuracy.This paper work is mainly divided into three parts:In the first part, the projection model under different models wasstudied. This paper focuses on the algebraic form and the geometricmeaning of the perspective projection model and affine projection basedon establishing camera observation model and introduces the three typicalaffine projection models--Orthographic projection, scaled-orthographicprojection and paraperspective projection. In addition, this paper analyzesthe error source, focusing on the distortion error and its correction model.In the second part, three factorization methods based on imagesequences and recursive perspective projection reconstruction algorithmwere introduced. Face model was reconstructed by two algorithms–scaled-orthographic projection and perspective projection. The processfrom affine reconstruction to Euclidean reconstruction was completed byadding measurement constraints. By using recursive algorithm in theperspective projection reconstruction can improve the recoveryaccuracy. In the third part, this paper studied the influence of the image noiseto the reconstruction of different feature points by using the theory ofmatrix perturbation, and analyzed the distribution of the camera distortionerror and how the image matching error of different sizes affected theaccuracy of the3D reconstruction. In addition, the influence of differentprojection models was studied. In the end, the results were checked by thesimulation.