Research On Automatic Stitching Of Aerial Image And Correcting Algorithm

In recent years, image stitch is a hot research topic in computer vision. Image mosaic is widely used in the UAV reconnaissance, aerial photography and other fields. It has very important significance on the research of aerial image.In the image acquisition process by unmanned air vehicle (UAV). lens distortion and attitude change can produce geometric distortion. This paper first analyzes the distortion of the types and causes of distortion, summarizes the geometric correction process and puts forward the basic steps of distortion correction.Aerial images have large amount of information. Using current image stitch algorithm to stitch aerial images is slow and accuracy is not high. These problems can be the great obstacle for emergency use of high accuracy. Aiming at this problem, this paper puts forward an improved image stitch algorithm based on analyzing the effect of aerial image stitch steps on the aerial image stitching speed and accuracy. At the process of feature extraction, firstly, we build multi-scale image space and select a key scale. Then, we extract feature points at the key scale, and use region removal algorithm to get the key feature of the small region and remove redundant information. At the process of image matching, we propose a hierarchical matching algorithm. First the images are divided into layers according to the scale space feature extraction, and then classify feature points into different layers. Image match begins from the top of the image layers, and a coarse to fine matching process stops until it meets the end condition. At the process of the bundle adjustment, we analyze the relationship between the error convergence process and image stitch, and propose a new iterative threshold for image stitch. Finally, we use the multi-scale fusion algorithm to fuse stitch result and obtain better effect.At the last part of this paper, a large number of experiments conduct to verify the aerial data of different types, and discussed the different situation on the several types of aerial images. Experiments show that, this algorithm can fast stitch aerial image at high stitching precision.