Research On Sparse Reconstruction And Dense Reconstruction Based On Monocular Vision

At present,the development of computer vision is more and more rapid,and the three-dimensional reconstruction of the environment in computer vision research is a major research hot spot.In the visual environment reconstruction,the camera need to acquire image information,and then the target image is processed,calculated to obtain the three-dimensional information of the environment and complete the reconstruction of the environment.Moreover,with the widespread use of electronic devices such as mobile phones and pads,people are paying more and more attention to photographing images and completing reconstruction using only monocular cameras.Therefore,this article studies the environment reconstruction based on monocular vision for the monocular camera has the characteristics of simple structure,low cost and flexible operation,it can easily obtain the image information of the environment and implement reconstruction.1.The basic theory of three coordinate systems,camera model,polar geometry,internal parameter matrix and external parameter matrix are introduced first.The process of solving essential matrix and basic matrix is analyzed.The camera calibration method is used to calibrate the camera.2.Experiments are performed on the images obtained by the camera to verify the advantages and disadvantages of preprocessing such as graying,filtering and sharpening.And SIFT algorithm is carefully studied to complete the feature extraction of the image,and the final experiment is performed.3.The basic steps of the structure from motion(SFM)in sparse reconstruction are described,and the bundle adjustment method in the step is analyzed.On this basis,a bundle adjustment method based on the Levenberg-Marquardt algorithm is adopted as the key step of the SFM algorithm,and the sparse reconstruction of the environment is realized.4.The procedure of Clustering Views for Multi-view Stereo(CMVS)and Patch-based Multi-view Stereo(PMVS)is studied,and the optimization of SFM input with CMVS is completed,and the dense reconstruction of the target object is achieved through the PMVS step.