Research On Cylindrical Panorama Stitching Based On SIFT Algorithms In Photogrammetry System

With the advancement of science and technology, people become increasingly demanding on the outside information. In many cases,panoramic images are needed to show more visual information because single images cannot meet people’s needs. Panorama technology is essential to particular use in specific areas, such as photogrammetry, medical imaging, military security and virtual reality technology. Panoramic images can display full-scene features within the field of view and provide comprehensive information for image viewer. Therefore, it is meaningful to study panoramic image stitching.The purpose of this image mosaic is to montage a set of images with overlapping area taken by photogrammetric system into a 360 °panorama. The key technologies of the image mosaic image are the image registration and the image fusion. For image registration, this thesis selected the better image matching method which has less computing, high efficiency and good robustness based on the feature information. In practice, due to the movement of the camera, the image sequence is not in the same coordinate system.This thesis compared the characteristics of cubic panorama, spherical panorama and cylindrical panorama and then selected a simple way to obtain the original image and the image-rich information-based cylindrical panorama. At first, it need to use cylinder conversion formula to convert the original image into a cylindrical map and then extract their SIFT (Scale-invariant Feature Transform, that is Scale-invariant feature transform) feature points to match their feature points. This thesis describes the major characteristics of SIFT like the formation of its location, scale and orientation invariant feature point in detail. Only through the correct matching feature points can determine the correspondence between the two images. This thesis selects the K-D tree and BBF algorithm to filter their correct matching pair in the matching stage and then removes part of the incorrect matching pair. After that, RANSAC algorithm is adopted to select out the incorrect match of feature points further and generate the transformation matrix. Images will realize image registration under the action in the transformation matrix. In this way, there will be generated directly mosaic obvious seams in the overlap region and the brightness of different images taken under different illumination and there will be different and significant chromatic aberration likely to be caused. This thesis focuses on different ways of using the weighted average method to achieve a smooth transition of the fusion between images, eliminating seams, generate better panoramic mosaic.The way of the usage of outdoor panoramic image stitching algorithm in this thesis can be tested in various scenes and have got an excellent mosaic effect. Experimental results show that the panoramic image stitching algorithm used in this thesis can work quickly and accurately and show robustness to the change in brightness and noise conditions. To Sum up, the panoramic image stitching algorithm has practical value.