Research And Implementation Of UAV Image Mosaic Methods

In recent years,with the development of China's economy,the UAV industry has made vigorous development.UAV has been widely used in the military investigation,environmental monitoring,emergency command,agricultural production,and other fields.In these applications,UAV is needed to take images or videos.When using an UAV to take photos,due to the limitation of flight height and camera focal length,a single image can not cover the whole target area.Therefore,it is necessary to mosaic multiple aerial images taken by UAV to obtain the image covering the whole target area.Image mosaic has been a topic in domestic and foreign research.The complete image mosaic includes image acquisition,image preprocessing,image registration,and image fusion.The accuracy and speed of image mosaic are affected by every step of image mosaic.If the quality of the acquired images is not high,then the corresponding preprocessing is necessary.In the image registration stage,one or more homography matrices are used to align the overlapping areas of images.The image fusion stage combines the matched images to get an image with good visual effect and can reflect the information of the images to be spliced.To achieve high-quality image mosaic in a short time,the thesis studies the UAV image mosaic methods,and the main work is as follows:(1)In view of the disadvantages of large computation and long time-consuming in traditional Scale-Invariant Feature Transform(SIFT)algorithm,the thesis firstly marks the Region of Interest(ROI)in the images to be stitched and optimizes the feature detection area.In the phase of feature extraction,the parallel acceleration of the multithreaded is used,which greatly improves the speed of detecting feature points and generating descriptors.Then,the descriptor generated by SIFT algorithm is reduced to 64 dimensions.Finally,the improved bidirectional matching method is used to match the feature points.(2)To improve the accuracy of image mosaic,firstly,the coordinates of feature points are normalized.Then,the matching points are constrained twice by using H and S components in the HSI color model based on the method of Random Sample Consensus(RANSAC).At the same time,the projection error of matching is eliminated by using the Thin-plate Spline(TPS)method,the alignment ability of the images is improved.Finally,use the fade-in and fade-out weighted average method for image fusion.(3)The thesis evaluates the results of image mosaic from subjective and objective aspects.Compared with the image results of SIFT method and As-Projective-As-Possible(APAP)warps method,which verifies the effectiveness of the improved SIFT method.(4)The thesis designs an aerial image mosaic system of UAV.The system integrates three kinds of image mosaic methods,which have good practicability.Users can choose the images to be spliced and the splicing method,which increases the interactivity and operability of image mosaic.In conclusion,the improved SIFT method proposed in the thesis is significant in UAV aerial image mosaic.At the same time,the UAV aerial image mosaic system brings great convenience to image mosaic.