中低空遥感图像序列快速拼接方法研究

发布时间：2018-11-11 11:38

【摘要】：图像拼接是将多个具有一定重叠度的摄影图像序列合成为全景的高分辨率图像的过程。图像拼接目前被广泛使用在诸多应用领域，如医学成像、视频拼接、数字地图和卫星照片的拼接、遥感图像拼接等。图像拼接一般由图像匹配、图像配准、图像融合几个主要步骤组成。随着国内外对无人机的应用日益广泛，如何快速处理无人机获取的大量遥感图像数据已经成为一项十分重要的研究内容，其中主要以无人机图像的拼接最受关注。其原因在于，遥感图像在多个领域内都有重要的作用。比如应急救援、电力线路巡查、地图制图等，因此，如何将所获取的大量遥感图像进行合并，以获取拼接后整个拍摄场景的图像在各个方面均有较高的应用需求。然而，在遥感图像拼接过程中，由于拍摄角度不同及光照明暗变化等原因，导致拼接后的图像会出现明显的拼接缝现象。又因图像两两间重叠度较大，对重叠区域进行融合的阶段所需时间较长。因此，对遥感图像重叠区域进行融合的算法在整个拼接过程中尤为重要。本文首先对图像拼接基本流程中的主要步骤及其所涉及的相关技术进行了概括总结，然后，综合考虑遥感图像序列的特点实现了本文中所介绍的遥感图像序列拼接方法，并在此基础上总结了原有图像融合算法的一些缺点，从而提出一种新的遥感图像加权融合算法，该算法对原融合算法存在的问题进行了改进。算法能够渐进的对遥感图像重叠区域进行融合，较好的消除了光照明暗的影响，实现了遥感图像的无缝拼接，使拼接效果有较好的提升。最后，在对图像重叠区域进行融合的过程中，，取待拼接原图像中坐标点处的像素值时需在基准图像坐标系中将像素坐标进行逆变换从而确定某一坐标对应原图像中的坐标值。由于遥感图像分辨率较高，因此，按像素坐标对所有像素点进行坐标逆变换所消耗的时间较多，导致融合拼接速率较慢。文中改进的融合算法考虑到对像素坐标逆变换的并行性，利用GPU并行加速技术对遥感图像融合拼接期间的像素坐标逆变换过程进行并行处理从而提高融合拼接效率、优化算法性能，最终通过实验证明了该算法的有效性。
[Abstract]:Image mosaic is the process of synthesizing multiple sequences of photographic images with certain overlap into panoramic high resolution images. Image mosaic is widely used in many applications, such as medical imaging, video mosaic, digital map and satellite image mosaic, remote sensing image mosaic and so on. Image stitching consists of image matching, image registration and image fusion. With the increasing application of UAV at home and abroad, how to quickly process a large number of remote sensing image data obtained by UAV has become a very important research content, in which the most attention is paid to the image mosaic of UAV. The reason is that remote sensing images play an important role in many fields. Such as emergency rescue, power line inspection, map mapping, so how to obtain a large number of remote sensing images are merged to obtain the image of the whole scene after stitching has a higher application demand in all aspects. However, in the process of remote sensing image stitching, because of different shooting angle and the change of illumination and shade, the stitching will appear obviously. Because of the large overlap between the two images, it takes a long time to fuse the overlapped regions. Therefore, the fusion algorithm for overlapped regions of remote sensing images is particularly important in the whole mosaic process. In this paper, the main steps in the basic process of image stitching and the related technologies are summarized. Then, considering the characteristics of remote sensing image sequence, the method of remote sensing image sequence mosaic is realized. On this basis, some shortcomings of the original image fusion algorithm are summarized, and a new weighted remote sensing image fusion algorithm is proposed, which improves the existing problems of the original fusion algorithm. The algorithm can gradually fuse the overlapped region of remote sensing image, eliminate the influence of light and dark, realize the seamless stitching of remote sensing image, and improve the effect of remote sensing image. Finally, in the process of image overlapping region fusion, when the pixel value of coordinate point in the original image is to be stitched, the pixel coordinate should be inversely transformed in the reference image coordinate system to determine the coordinate value corresponding to a certain coordinate in the original image. Because of the high resolution of remote sensing images, it takes more time to invert all pixels according to pixel coordinates, resulting in a slow fusion rate. Considering the parallelism of inverse transformation of pixel coordinates, the improved fusion algorithm uses GPU parallel acceleration technology to parallel process the inverse transformation of pixel coordinates during remote sensing image fusion and stitching, so as to improve the efficiency of fusion and stitching. The performance of the algorithm is optimized and the effectiveness of the algorithm is proved by experiments.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP751

【参考文献】