Research On The Tilt Image Matching Which Is Based On An Improved SIFT Algorithm

At present,aerial photogrammetry has become the main means of topographic information acquisition by virtue of its convenience,high efficiency and high information tolerance of image acquisition.Classic aerial survey,however,with the aid of vertical photography way to get the surface imaging,and populated areas,architectural structures of city block is very serious,more importantly,to collect images usually contains only the structure of the top surface information,difficult to cover the facade of the building,therefore,difficult to play a role in the 3 d data acquisition.Tilted in recent years,the emergence of photogrammetry to effectively overcome the deficiency of the classical aerial survey,based on tilt photogrammetry to collect the image data fusion the orthographic images and different Angle of tilt,can simultaneously obtain coverage at the top of the building and all aspects of the high resolution imaging,and because of that,since their emergence,from surveying and mapping,geographic information,city planning,and other fields of scholars pay close attention to and attention.As we all know,the image matching is one of the key steps of photogrammetric data post-processing,however,due to the tilt of photogrammetry for tilt image has such problems as large geometric distortion,the change of perspective,makes no longer applies the traditional image matching algorithm,an urgent need to research and development can be used to tilt more visual image matching algorithms.Based on the above analysis,based on the existing vertical photographic image matching algorithm and aiming at the characteristics of tilted photographic images,this paper develops an algorithm that can realize the efficient and high-precision matching of tilted images.The main work is as follows:1)POS information-based auxiliary skew image correction.Based on the initial external orientation element data calculated by POS system attitude information,the relative transformation matrix of tilted image and vertical image is calculated,and the tilted image is corrected by the transformation matrix.SIFT algorithm and its improvement algorithm were used to match the images before and after correction,and the matching results were analyzed.2)multi-view tilted image matching.Taking SIFT algorithm as the prototype,and aiming at the problem that SIFT algorithm needs to build gaussian difference pyramid in the feature point extraction stage,the algorithm has a large amount of computation and poor time efficiency,FAST and Harris-Laplace feature detection operators are adopted to replace DOG operators.As the SIFT algorithm does not have complete Affine invariance,the DOG operator is replaced by MSERs and harris-affine algorithms.Aiming at the problem that SIFT feature point descriptors have 128 dimensions and have a large amount of computation,binary descriptors such as DAISY,FREAK,LATCH and BRISK are adopted for replacement,so as to improve the running speed of the algorithm.After the initial feature detection,the improved algorithm adopts the coarse-to-fine registration strategy,USES the nearest neighbor/next neighbor ratio constraint or hamming distance as constraint conditions to carry out rough matching of feature points,and then USES RANSAC algorithm to carry out precise matching of matching point pairs to obtain correct matching point pairs.Studies have shown that tilt images to correct different perspectives can effectively overcome the perspective changes and geometric distortion problems,more apparent dip image matching algorithm to solve the classical SIFT algorithm is used to tilt the image matching to solve larger perspective transformation caused by the problem of image matching,at the same time solve the SIFT algorithm description problem of poor child time efficiency.The speed and precision of image matching can be improved effectively by using the algorithm in this paper.