Research On Algorithms For UAV Aerial Image Stitching

Because of these advantages as low cost, no casualties risk, strong survival ability, good mobility and easy to use in low-altitude shot, UAV has obvious advantage over ordinary reconnaissance and satellite systems. The UAV aerial images stitching technology is widely used in dangerous areas investigating, real- time detection of battlefield, marine environmental monitoring, natural disaster monitoring and evaluation, and attracted so much attention of home and aboard that grow to be a hot in image processing area.In the UAV aerial images Stitching, image has high resolution and quantity of images is very large. If global matching strategy is used, the registration error is small but the speed of matching is rather slow. If local matching strategy is used, the speed of matching is fast, but the registration error is very large.Combining the advantages of both, according to the characteristics of UAV aerial image sequence, a new image registration algorithm is proposed based on second matching. Firstly, this algorithm searches last image of second consecutive match of the current image and mark it, and then for the following images of the same strip, searches the first matching by order and the second matching by reverse order from the marked image until the match failed. The current strip match ends when there is overlap between the first and second consecutive match. This algorithm is effective in reducing the number of matches between images, improving the speed of image registration. Finally, this paper verifies the effectiveness of the algor ithm by testing three groups of multi strip aerial image data.When the image is globally aligned, this paper used the number of matched feature points as weights to establish the maximum spanning tree, and used the total error of image feature points as weights to search the error center. Then, we computer all other images projection model to error center. Finally, this paper verifies that it has the same projection error as the method that use distance as weight to search physical center, sometimes event better.Image fusion has the problem of ghosting and uneven color. This paper adapter the idea of graph cuts to image segmentation. We use the method that Boykov improved based on augmenting paths algorithms and used to solve minimum cut / maximum flow problem to search the best line in the overlap areas of two images so that each pitch is just from the only image. This method can solve ghosting problem. Then we use multi-band image fusion algorithm to composite the images so that the result image transitio n look softer beside edges. Through experiment verification, the boykov’s improved maximum flow algorithm indeed Accelerate speed of searching the best line and the multi-band fusion algorithm met our expectations.