Research On Fast Stitching Method Of Middle-low Altitude Remote Sensing Images

Image stitching is the process of combining multiple photographic images withoverlapping fields to produce a segmented panorama with a high resolution. Imagestitching is widely used in many fields nowadays such as medical imaging, videostitching, digital maps, UAV/satellite remote sensing image stitching and so on. Theimage stitching process can be divided into three main steps: image matching, imageregistration and image fusion.With the increasingly widespread application of unmanned aerial vehicle(UAV) in today’s world, how to handle a large number of remote sensing imagesacquired by UAVs has become a very important research content. One of themost concerns is the fast stitching of large amount of UAV remote sensing images,the reason is that the remote sensing images acquired by UAVs play an importantrole in many fields such as emergency rescue, environment monitoring, forestprotection, power line inspection, cartography and so on. Therefore, how to merge alarge number of UAV remote sensing images and get the whole scene after it is quitenecessary, it has higher application requirements in every respect. However, in theprocess of image stitching, because of the influence of the shooting angle andillumination variations, when two images are mosaicked, there inevitably exists aseam-line on the stitching boundary. Therefore, the image fusion algorithm ofoverlapping area is especially important.In this paper, we first introduce the main steps of image stitching and summarizethe technologies referenced by this paper. After that we implement the method ofremote sensing images stitching. Moreover, the existing problems of the existed remote sensing image fusion algorithm are analyzed and improved. A new weightedfusion algorithm is presented, which can make the stitching image smoothness andseamless, eliminated the influence of illumination variations and improve the stitchingeffect obviously.Finally, due to the high-resolution of remote sensing image, inversetransformation of pixel coordinate points for fusion cost too much time, therefore, thenew algorithm takes advantage of GPU to accelerate the rate of remote sensing imagefusion and then indicates the efficiency of the algorithm through experiments.