Research On The Optimization Of UAV Aerial Image Stitching Of High-rise Buildings

With the continuous development of science and technology,people's methods of image acquisition are changing rapidly,from the original measurement and designing to the advanced means such as UAV aerial photography,satellite remote sensing and so on,but even though,they still can not meet the constantly updated practical needs.Nowadays,UAV has become a mainstream means of acquiring images for its small size,light weight and convenient access to high-definition images.But due to the height limitation of UAV aerial flight to obtain the images of large buildings with great elevation in the process of image merging,there are often many problems for the results,one is the height limitation of UAV aerial flight,the other is the height effect of buildings themselves with elevation,causing the aerial photography of buildings with elevation to be non-orthographic and to have various degrees of distortion,thus bringing great difficult to the image registration,and how to make a big difference in efficient merging of buildings with elevation is the research problem of this thesis.This thesis summarizes the relevant techniques used in feature-based image merging,whose basic steps include the extraction and matching of feature points,the estimation of model parameters and image transform.Based on the relevant methods in each step of image registration and fusion,this thesis emphasizes the image registration on the basis of SIFT features and the related algorithm about the image fusion based on wavelet transform.1.The extraction of feature points: This thesis firstly applies the SIFT algorithm and the Harris algorithm to extract feature points and compare the characteristics of these two methods with each other,then after extracting the feature points,the random sampling consistent algorithm is introduced to eliminate the grosses and errors of the feature points extracted by the two algorithms,and the data used in the subsequent image fusion are obtained at last.2.Image fusion: This thesis combines the theories and methods of wavelet transform and ALPHA image fusion as the fusion algorithm,which will be introduced detailedly in the forth chapter.At first,the author only uses wavelet fusion as the last fusion method of image,however,with the experiment being carried out,the author finds that the waveletfusion cannot perfectly integrate UAV images of tall buildings and the ghosting and splicing dislocation appear obviously,therefore,ALPHA fusion is applied on the basis of wavelet fusion in this thesis,which can solve the problems of the ghosting and splicing dislocation after wavelet fusion better,but the color of the fused image with ALPHA fusion is not very good,which needs further solution.3.The method in this thesis has been using to process images in the computer field whose internal storage is kilobyte,while it is used to process image data in this thesis and its internal storage is megabyte,so this thesis optimizes the program so that it can eliminate the redundant data while merging in the process of wavelet fusion and it can operate on such a computer with 4G internal storage.Owing to the large amount of image data processed by this program,the image fusion data after rarefying are added to the program in this thesis,and then the splicing quality of image fusion data after rarefying is observed to meet the requirements and to generate the final fusion images,thus saving time and improving efficiency.The fusion rules directly determine the quality of image fusion,so this thesis pays much attention to low frequency and high frequency fusion rules among image fusion rules.In order to evaluate the quality of image merging,this thesis introduces the evaluation index of image fusion quality.This thesis adopts the subjective evaluation index in the fusion quality evaluation index to evaluate the merged images.The experiment proves that the quality of merged images is high with the method mentioned in this thesis.