Automatic Stitching Of Aerial Sequences Image

Automatic splicing technology of UAV aerial image sequences is a hot research field of remote sensing information processing, computer vision etc.. Its main task is to extract the aerial image sequence, covering the photographing image subset complete information area, and the images are transformed into the same image coordinate, forming a large field of view and high resolution image. At present, the main challenge is how to choose the key areas related to the image frame according to the flight characteristics of the UAV, and panoramic image of the phantom of the opera overlap etc.. Therefore, aiming at these problems, to carry out the key frame extraction and optimization of fusion algorithms, the main work is as follows:1、 For aerial video images containing excessive redundancy problem, affect splicing efficiency, this paper presented an algorithm of key frame extraction based on fuzzy clustering and motion estimation. The algorithm firstly based on the information entropy of image as the feature expression of image content, use of video shot segmentation fuzzy C-means algorithm; then introduces calculated parameters change between the adjacent images phase correlation method, the two principle to extract key frame as the basis, image extraction suitable for panoramic mosaic mission. Implementation results show that, compared with the method of artificial key frame extraction, the use of automatic key frame extraction algorithm not only consume less time, and the smoke image subset removed more in line with the set the image overlap.2、 In the process of panoramic image mosaics, image alignment part will have the problem of phantom of the opera and the mosaic phenomenon, presents a panoramic image mosaic algorithm of constraint local as projective as possible warps. Since the approximation projection transformation algorithm is to divide the image into rectangular blocks number equal, different regions of the independent use of the homography transformation, can effectively improve the problem of different elevation region of the image fusion. But this kind of algorithm for the rectangular block away from the reference image will appear larger distortion and seam surrounding will appear a large number of phenomena such as deformation, the phantom of the opera. And the method of bundle adjustment can transfer problem of error constraint panorama stitching, thereby avoiding the deformation phenomenon in image mosaic. This thesis combines the advantages of the above two kinds of algorithms, to approximate the projection transform based image stitching, the bundle adjustment strategy combined with a perspective transformation, to reduce the distortion phenomenon through the constraint transformation total error. UAV remote sensing image sequence of the experimental results with different elevation information, relative to the as projective as possible warps, with the constraint local as projective as possible warps reduces the irregular splicing results in distortion; relative to the bundle adjustment, higher accuracy, reduces the blurring and ghosting phenomenon.3、 For the translational transform only single direction in aerial video sequences, the traditional graph cut algorithm is easy to cause the problem of segmentation is not balanced, puts forward an improved fusion algorithm of graph cut energy equation. The algorithm uses the image boundary is perpendicular to the direction of motion of the center of the camera as a graph cut endpoint, with negative gradient image point coordinate values and Gauss distance as punished functions of data items. The experimental results show that, compared with the traditional graph cut algorithms, an energy equation is solved effectively the seam is inclined towards the diagonal direction, which makes the result more smooth panoramic image stitching.