Research On Stitching Technology Of Low Altitude Remote Sensing Image Based On Unmanned Aerial Vehicle

Image stitching has always been an important research direction in image processing, plays an important role in the field of image processing. Using image stitching technology can make a series images with overlapping area join together into a complete image after image registration and image fusion.Under the impetus of commercialization, UAV remote sensing booming in recent years. Because of its lightweight, flexible, strong real-time performance, low price relative to the satellite remote sensing and not easily affected by bad weather and other advantages, favored by many countries and regions. In simple terms, UAV remote sensing is to obtain the images of the area which is unmanned aerial vehicle flying over by using the camera aboard. The image obtained by UAV remote sensing has a number of large and high resolution, so the realization of UAV remote sensing image stitching is a hotspot of research in recent years.In this paper, image stitching technology was applied to UAV remote sensing images. For this problem, I did some research on image stitching first. In this paper I studied two important process of image stitching technology: image registration and image fusion. Based on that, I choose using image registration method based on the characteristics of the image in this paper. In the process of image registration, this paper mainly introduces three kinds of Feature points extraction algorithm according to the time sequence of the development of algorithms: SIFT(Scale Invariant Feature Transform) algorithm, SURF(Speeded Up Robust Features) algorithm, the O RB(Oriented FAST and Rotated BRIEF) algorithm. The three methods were applied to the UAV remote sensing images, the speed to detect feature points is the standard to compare the basis of this three algorithms, finally I choose the ORB feature point detection algorithm. For matching feature points I choose KNN algorithm, this paper firstly studied on search strategies kd tree search algorithm, but as a result of kd tree search algorithm under the condition of high dimension the comp lexity of the algorithm is high. so in the search algorithm, I select LSH(local sensitive hashing algorithm), further reduce the time spent in the process of feature points matching. After image registration, the next step is image fusion. In this part of this paper is based on the optimal seam line search via graph cuts and poisson fusion. In the process of the optimal seam line search based on graph cut, I propose a weight value calculation method, in this method combines the pixel color and gradient information, try to avoid find image stitching line which pass the important areas and avoid splicing dislocation, in that cases maybe influence the effect of stitching. I implemented the algorithm and compared this algorithm with the traditional one in this paper. Finally I get the images together through the method that proposed in this paper.