Research On Image Mosaic Algorithm Based On UAV Images

Unmanned aerial vehicle (UAV) has the advantages of small size, flexible, fastresponse etc., to use UAV to collect information can make up of the defects of thesatellite remote sensing and general while collecting information. Therefore, UAV ispaid close attention to extensively. However, limited by UAV’s flight height, flightangle and the shooting equipments, leaflet UAV image can not get the completeinformation of the target area, In order to get more information of the target area andextend the field of view, we can stitch the images got by different angles smoothlyand seamlessly into a panorama image.At present, there are many methods to stitch images. The algorithm based on thecharacteristics only needs to search feature points instead of to search the wholeimage, this advantage makes the algorithm high efficiency. At the same time, thealgorithm based on the characteristics is invariant for the image deformation, graychanges, and occlusion problems. So this type of algorithm is suitable for the UAVimages with high resolution, serious image distortion and big difference of light.Based on the above analysis, the main research contents in this paper are as follows:Firstly this paper analyzes the background and current status of research onimage mosaic. Then it summers up the existing image mosaic algorithms andclassifies them. After that it analyzes the advantages, inadequacy and application fieldof every class of algorithms.Secondly, for registration, this paper firstly uses the bilinear transformationmethod and the nearest neighbor method for geometric correction, and then uses SIFT(Scale Invariant Feature Transform) algorithm to extract the image features. SIFTalgorithm will have a slow speed to search the extreme points. In this paper, a strategyof monolayer extremum search is put forward to solve the problem. After featurematching, the random sample consensus algorithm will be used to remove the falsematching points.Thirdly, the quantity of UAV images is huge, if errors accumulation during the process of the stitching, apparent seam and distorted phenomenon will appear on thepieced panoramic image. In order to successfully stitching the UAV images smoothlyand seamlessly without loss of precision, this paper optimizes the bundle adjustmentand uses it on the panorama generation process. This paper optimizes the selected ruleof the best reference plane about the bundle adjustment. The bundle adjustment isused to optimize global transformation matrix. The transformation matrix will be usedto mosaic the panorama images.The results of the tests show that, the strategy of monolayer extremum searchcan search the extreme points faster when extracting image features while losing acertain degree of accuracy. At the same time, the accumulated error can be reducedwhile using the optimization bundle adjustment method.