Aerial Image Stitching System Based On Global Similarity Priority

With the development of drone,the social development and the progress of industry has met a new direction,and it brings a new broad field for researchers on digital images.Drones can better accomplish a part of previous work in a rather new view,and add a rather new dimension for the design and accomplishment of the work,it offers new demand for the drone developers and researchers of computer on digital images.Aerial photographing is one of the commonest application of drone.The stitching among various photographs is studied and a stitching system using an algorithm based on image features is put forward in this thesis.Besides,the analysis and evaluation on its results are also put forward in this thesis as well.The algorithm is chosen due to its effectiveness and ability to handle complex condition together with a better time efficiency,so it can meet the demand to the system in this thesis.The main content in this thesis contains three parts as follows:(1)The methods based on the image features and their generous process are studied in this thesis.Researches are made as its logical sequence,including the definition of image features and how they design,extract,choose,match and stitch with homography matrix.Besides,needed methods for image enhancement and blending are also included.(2)Researches on the details of the stitching algorithms.Researches focus on the problems in traditional algorithms such as deformation and deformation,and introduce some properties such as local similarity and global similarity to limit algorithm in order to access better result.Then global similarity is finally chosen to be mainly applied in the algorithm,supplied with other properties and it brings good results.(3)Researches on the design and implementation of the stitching system.The system is divided into four modules including user interface,matching,stitching and enhancement.Then they are connected as a whole system after independent implementation and development.(4)A capturing plan is designed to evaluate some aspects of the system such as overlap rate,route,height etc.It is concluded that the system can produce a fine result with reasonable routes,height and overlap.It can also handle the situation with some bad inputs.A stitching system uses a group of overlapped images but a specific drone is raised in the thesis.Besides,the natural image stitching algorithm with the global similarity prior is firstly used with a good results in a drone image stitching system.It is shown from the researches and the results that the system announced in this thesis has a fine ability to stitch images captured by drones and it can handle different situation with different route and height.However,just like other methods using image features,it calls for enough features to attach good results.