| With the development of electronic information transmission and precise sensors,many new image data acquisition modes have emerged in the field of surveying and mapping,such as satellite multi-angle remote sensing,oblique aerial photography,and moving wide baseline photography.This has increased the number of ways to obtain target scene images.However,due to the image acquisition process The influence of some external factors in the image causes large geometric and radiation distortions between the same-named regions of the image,or there are problems such as single ground objects,lack of texture or repetition in special texture images,which makes the use of traditional single matching algorithms,The small number of acquired features of the same name,low accuracy,uneven distribution,and poor stability result in deformation and even failure in the matching process,making it difficult to apply to the stitching of different types of images.Therefore,the realization of automatic splicing of remote sensing images based on general computer algorithms is always a bottleneck in the joint research of digital photogrammetry and computer vision.Based on this,this thesis conducts the research on the general algorithm for automatic splicing of remote sensing images.The main research results are as follows:(1)Introduce the concept,classification and image stitching process of image matching,and further study of the feature extraction and matching principles of the three classic algorithms of SIFT(Scale Invariant Feature Transform),SURF(Speeded Up Robust Features)and MSER(Maximally Stable Extrernal Regions),After a comprehensive comparison experiment of feature extraction and feature matching,the advantages and disadvantages of each algorithm and its application are summarized.The final result shows that there is no one matching algorithm can adapt to all different types of images or images with complex geometric distortion.(2)Propose a remote sensing image stitching algorithm that integrates multiple types of features.First,SIFT algorithm is used for initial matching,and NCC(Normalized Cross Correlation)algorithm is used for positioning refinement to obtain SIFT precise matching points,and then the homography matrix between images is estimated;Then use the Forstner operator to extract evenly distributed corner features of the left image,and use the homography matrix combined with the NCC strategy to match to obtain Forstner precise matches;finally,integrate multiple types of fine matching points,and calculate the splicing model after eliminating mismatching.The experimental results show that the algorithm proposed in this thesis has obvious advantages in terms of the number of points with the same name,spatial distribution and splicing accuracy.(3)Propose a remote sensing image stitching algorithm that integrates point and line features.First,use the improved SIFT algorithm combined with the NCC matching strategy to obtain high-precision points with the same name as the seed point,and then use the improved LSD(Line Segment Detector)operator to extract the stability For the complete line features with higher performance,the candidate matching line segments are screened through homography constraints and slope approximation,and LBD(Line Band Discriptor)descriptors are constructed and matched,and then the line segments with the same name are obtained.Finally,the complementary fusion of dots and lines is realized,and the stitching model between images is calculated to realize the stitching of images.Experimental results show that the algorithm proposed in this thesis integrates the advantages of point and line features,and effectively improves the accuracy of matching and image stitching accuracy.(4)Design and build a digital remote sensing image automatic splicing system,in which the algorithm proposed in this thesis is actually applied and the splicing process is optimized to realize high-precision automatic splicing of remote sensing images. |
PDF Full Text Request