Research On Feature Extraction And Matching Based On UAV Tilted Images

Oblique photography is a high-tech developed in the field of international surveying and mapping in recent years.Which is a breakthrough in aerial photogrammetry.For traditional rthophoto,which breaks the limitation of shooting only from a vertical angle.Many sensors are mounted on the same flight platform,with one vertical and four vertical.The image is captured at five different angles,and the texture information of the ground object is obtained as much as possible,introducing the user into the visual model world.Faced with so many image maps,fast and efficient image stitching is particularly important.How to quickly and efficiently complete the image matching algorithm is the need of modern development.The quality of image stitching mainly depends on the accuracy of image matching.The main problems with image matching of Oblique photography:(1)The slated image is taken at a certain tilt angle of the camera's main optical axis,and the aerial image is mostly a natural scene.The scene is complex and susceptible to imaging conditions and environment,resulting in numerous distortions and imitations between images;(2)The commonly used feature-based image matching algorithm is very inefficient,which is inconsistent with the rapid development of today's society.Therefore,for the application of image matching algorithm in practical work and life,the algorithm needs to be improved and optimized.Therefore,the research of image matching algorithm has very important theoretical and practical significance.Based on the above research background,this paper uses the image of a aerial photography as the data source of six groups of experiments,using MSER's good affine invariance,regional stability and regional size variability and SIFT good scalability and uniqueness.Besides,its characteristics of easy access to local regions,and based on SIFT and MSER,propose a matching algorithm based on MSER affine invariant region features--MSER-F(where Maximally Stable Extremal Regions-Features)focusing on the problem of oblique image feature point extraction and image matching.The main ideas of the proposed MSER-F algorithm are as follows: Firstly,the MSER region is detected on the Gaussian multi-scale space,and then the MSER region is fitted by the affine ellipse.Secondly,the fitting region is described and matched based on the SIFT descriptor.A large number of initial matching point sets,followed by the RANSAC method to eliminate the points.According to the proposed MSER-F algorithm,six sets of experiments were designed using VS2010 and OpenCV2.4.9,and the reliability of the MSER-F algorithm was analyzed.The main conclusions drawn are as follows:(1)By analyzing the experimental data of experiments 1,2 and 3,which proves that the SIFT algorithm has strong rotation invariance and anti-scaling,and it has strong robustness;SIFT algorithm in affine deformation and image Perspective distortion is weak,and its ability to resist affine and perspective distortion is very poor.(2)Through the comparative analysis of Experiment 1 and Experiment 4,MSERF has higher robustness,which can prove that the anti-rotation performance of the new algorithm is good;by comparing the results of correct matching logarithm,in terms of anti-rotation deformation,The MSER-F algorithm is still weaker than the SIFT algorithm.(3)With the comparison between experiment 2 and experiment 5,the number of matching of MSER-F algorithm is still good,and the image matching work can also be completed.The ability of MSER-F algorithm to resist scaling transformation is good;the result of correct matching logarithm is compared.In terms of anti-scaling deformation,the MSER-F algorithm is still weaker than the SIFT algorithm.(4)The comparison between experiment 2 and experiment 5,MSER-F has higher robustness,which proves that MSER-F algorithm has excellent ability to resist affine deformation;compares the result of correct matching logarithm,and resists affine In terms of deformation,the MSER-F algorithm is more superior than the SIFT algorithm.(5)Time and efficiency are as important in real work.The number of feature points detected by SIFT algorithm is significantly higher than the number of feature regions detected by MSER-F algorithm.Therefore,it is considered a reason that the final matching number of MSER-F algorithm is lower than SIFT.according to the statistical average operation time,the operation speed of the MSER-F algorithm is only 0.239 times that of the SIFT algorithm,and the operation speed of the MSER-F algorithm is faster.