Research On Automatic Image Stitching And 3D Reconstruction

Computer vision has a wide range of applications in mobile robot navigation,industrial detection,military reconnaissance,etc.,and attempts to create artificial intelligence systems that can obtain "information" from images or multidimensional data.Image detection,image index creation,medical image analysis And topology modeling also has newer applications.In this paper,a new three-dimensional modeling method is proposed.The three methods of camera calibration,image matching,image fusion and three-dimensional reconstruction are studied by theoretical analysis,simulation analysis and experimental verification.The accuracy of the three-dimensional reconstruction.(1)The basic theory of camera imaging model,coordinate system transformation,matrix operation and camera calibration is introduced.The conversion between camera coordinate system and world coordinate system and the mathematical derivation of camera parameters are analyzed in detail.The current mainstream Zhang Zhengyou plane calibration Method for simulation experiments and practical experiments,with high calibration accuracy and good robustness.(2)This paper introduces a new multi-view matching framework based on the new invariant feature.The framework is mainly for SIFT corners,scattered scale spaces,and the use of fuzzy local gradients,but also the introduction of a novel outlier refusal process,the pattern matching probability model validation,the process based on the incorrect feature matching background distribution Verify pairwise feature matching.Use Bundle Adjustment to resolve the consistency of all camera parameters.(3)In order to maximize the retrieval of image in the image feature extraction and matching algorithm,a multi-band fusion algorithm is proposed and elaborated in detail.In this paper,the introduction of gain compensation and automatic correction steps,a single picture of multi-band information to be integrated to eliminate the multi-picture stitching between the possible redundancy and contradictory information to improve the uneven light caused by the matching vignetting phenomenon,Accuracy,and misalignment caused by the parallax effect caused by the undesired movement of the optical center and radial distortion.(4)This paper deals with the problem of three-dimensional image reconstruction.Although the 2D problem has been fully studied,the actual effect of 2D or multidimensional image splicing needs to be improved and the 3D reconstruction based on 2D image splicing is still difficult.The previous method is to manually enter the image sequence in order to establish a matching image,this method is subject to manual restrictions.In this paper,we use the multi-image matching problem as the starting point,and use the invariant local feature multi-view matching framework to find the matching between all the images,combined with the mark point depth information,through the affine transformation to match the different surface To achieve the effect of three-dimensional reconstruction,the formation of the target clear,complete and accurate information description.