| The gullies are deep in the Karst area,with numerous Fenglin and Fengcong and complex landscapes.It is difficult and costly to obtain topographic information.With the rapid development of remote sensing platforms,the application of UAV technology has become more and more extensive,and multi-view oblique aerial photogrammetry has become the mainstream development direction.Compared with the traditional orthophoto,the multi-view oblique aerial image has a complete texture,but the data volume is large and the image geometric distortion is large.In the multi-view oblique aerial photogrammetry in the Karst area,due to the mutual occlusion and shadow between the Fenglin and the Fengcong,it is more difficult to match and stitch the multi-view oblique aerial image in the area.When matching the oblique aerial images of Karst landforms,the traditional feature detection algorithm does not have complete affine invariance,and the correct matching point pairs are few.Aiming at these shortcomings,this paper constructs an improved feature detection algorithm based on complete affine invariant framework.The feature detection algorithm before and after the improvement is used to study the feature point rough matching and fine matching,and compared with the traditional classical feature detection algorithm.At the end of the paper,the fusion algorithm of multi-view oblique aerial image is discussed in this area.The main research contents and conclusions of this paper are as follows:(1)When multi-view oblique aerial photogrammetry photogrammetry,the acquired images are not adjacent and non-serialized,and the effects of different routes have different brightness and local shadows.In the Karst area,the difference in brightness and local shadows of multi-view oblique aerial image is more pronounced,resulting in a change in pixel values at the same location,which in turn affects image matching.For the homogenization and color pretreatment of images,three kinds of preprocessing methods based on histogram equalization,Retinex enhancement algorithm and Wallis filter algorithm are compared and analyzed.Studies have shown that for the Karst topography,the image homogenization method based on the Wallis filter algorithm works best.(2)When matching multi-view oblique aerial images of Karst landform,the classic feature detection algorithm does not have complete affine invariance,and the correct matching point pairs are few,sometimes not even.In view of these shortcomings,this paper improves the SURF,AKAZE and SIFT algorithms based on the complete affine invariant framework,and then increases the number of feature points of the multi-view oblique aerial image of Karst landform,and then accelerates the FLANN algorithm by KD-TREE to complete image feature rough matching.(3)Aiming at the problem that RANSAC algorithm takes a long time in feature matching,the paper constructs a PROSAC algorithm to purify the rough matching point pairs,so that the correct matching point pairs are more evenly distributed on the image.Compared with the RANSAC algorithm,the algorithm has a faster search speed,more accurate culling of incorrectly matched point pairs,and can estimate a better homography matrix,which makes the stitching more accurate.(4)The traditional stitching fusion algorithm has poor effect on image blurring and stitching seam processing.This paper constructs a graph cut algorithm to search for optimal stitching lines,and then uses multi-resolution fusion algorithm to fuse multi-view oblique aerial images.In order to verify the stitching fusion effect of the proposed algorithm,the stitching experiment was carried out by using four group images of Karst landform of multiview oblique aerial image.The stitching fusion effect of the algorithm and AANAP algorithm,OcvStitch algorithm,AutoStitch software and ICE software was compared.The experimental results show that the s stitching fusion effect of the proposed algorithm is better for the multi-view oblique aerial image of Karst landform. |
PDF Full Text Request