Research On 3D Reconstruction Of Monocular Vision Based On Improved Sift Algorithm

3D reconstruction can reflect the complete scene information in the image,which is widely used in industrial automation,virtual reality applications,visual navigation and other fields.Therefore,3D reconstruction technology has gradually become a new research hotspot and demand.At present,there are some problems in 3D reconstruction based on stereo vision,such as poor mobility,high price,low accuracy of feature matching and low accuracy of reconstruction.Therefore,this dissertation proposes a convenient and low price 3D reconstruction system based on monocular vision,which studies 3D reconstruction technology through monocular camera calibration,feature extraction and matching,space point cloud reconstruction and other steps.The purpose is to improve the accuracy of feature matching and 3D reconstruction while ensuring the real-time performance of the algorithm.Firstly,to ensure the quality of image acquisition and the effect of reconstruction,the three-dimensional reconstruction system is built.In order to determine the relationship between the world coordinate system and the image pixel coordinate system,the principle of camera imaging is studied to draw out the internal and external parameters of the camera.Aiming at the problem of the accuracy of parameter solution and the distortion of image,the calibration of monocular camera is analyzed theoretically.Through the comparison of several camera calibration methods,Zhang Zhengyou calibration method is selected to get the camera internal and external parameters and distortion parameters.Then,the Sift algorithm based on local features is studied to solve the problems of pose transformation and illumination influence.In order to improve the accuracy of feature extraction and matching,the high frequency points in the Fourier transform domain of image are determined by phase consistency to eliminate the unstable feature points,and Kendall coefficient constraint is introduced to carry out secondary matching to remove the false match points.Finally,on the basis of camera calibration and feature extraction matching,the algorithm of space point cloud reconstruction is studied.The projection matrix is obtained by camera parameters,and the three-dimensional sparse point cloud is obtained by projection matrix.Aiming at the problem of missing image details,the algorithm of 3D reconstruction of dense point cloud is studied.On this basis,the texture mapping method is used to get a more real reconstruction model through triangulation and texture mapping.Through the experiments of camera calibration,feature extraction and matching,and point cloud reconstruction,the comprehensive performance of the algorithm and the measuring results of the reconstructed model are compared and analyzed.The results show that the camera calibration accuracy is 0.004 mm,and the internal and external parameters with high accuracy can be obtained.Compared with the original algorithm,the improved algorithm runs for 0.011 s longer,and the correct matching rate is increased by 4.35%.It is verified that the application of the improved algorithm can improve the accuracy of feature extraction and matching while ensuring the real-time performance.The effect of 3D reconstruction is improved,and the experimental results show that the reconstruction model is more real and effective.