Research On Aerial Image Stitching Technology Based On Improved SURF And APAP

With the development of science and technology,aerial UAV technology is widely used in agriculture,industry,military,and other fields.But because of the limitation of the camera perspective,the content contained in a single image cannot meet the needs of information acquisition.Therefore,it is necessary to stitch the collected aerial images into a large-scale panoramic image with rich content.Aerial images are easily affected by illumination,scale,and rotation changes,and have the characteristics of discontinuous image sequence and parallax.We focuses on the study of feature extraction algorithms and image warping methods that can adapt to the characteristics of aerial images to obtain high-precision and natural aerial images.The main research contents of this paper are as follows:In order to obtain high-precision panoramic aerial images,this paper improves the robust SURF algorithm.Aiming at the problem of inaccurate edge positioning and high error matching rate of the SURF algorithm,an aerial image stitching algorithm based on long-distance features is proposed.First,we propose the Canny-SURF feature detection algorithm,which uses bilateral filtering to improve the Canny algorithm to detect the edge of the scene and combine it with the SURF feature as feature points.Then,the L-FREAK binary descriptor is introduced to replace the original SURF floating-point descriptor to improve matching efficiency.The L-FREAK descriptor only uses the medium and long distance sampling point information to determine the main direction,which further improves the algorithm's tolerance to rotation changes.Finally,the proposed algorithm was used to process the data set provided by Mikolajczykg and the actual aerial image,and compared with SIFT and SURF and other classic algorithms.Experimental results show that the proposed algorithm is robust to images with various feature changes,and its registration speed and accuracy are both suitable for aerial image stitching.Traditional aerial image stitching algorithms cannot overcome the problem of parallax,which causes ghosting and distortion in images after stitching.Therefore,this paper improves the APAP algorithm with high alignment quality,and proposes the Linear-Preserve-Mesh warps method in aerial image stitching.First,we mesh the image and construct an energy function that contains the vertex information of the mesh.Then,the local homography matrix of the APAP algorithm is used to align the local details,the similarity constraint is used to ensure the natural transition and distortion correction of the image.The straight line protection constraint is used to prevent the straight line structure in the scene from bending during the deformation process.Finally,the minimum error solution is solved in the sparse linear system to obtain the optimal mesh vertex set,which guides the warping of the image.The effectiveness of the proposed algorithm is verified by experiments on multiple standard datasets.The results show that compared with Auto Stitch,APAP and other algorithms,the proposed algorithm has stronger comprehensive performance and can reduce ghost and distortion problems.In addition,the proposed image warping method is applied to the actual collected UAV aerial images with a certain parallax.The local details of the stitched image are well aligned.Compared with the original APAP algorithm,the naturalness of the image is significantly improved,and the distortion of the non-overlapping area is also improved.