UAV Image Mosaic Technology Based On Matching Of Point And Line Features

The UAV (Unmanned Aerial Vehicle) remote sensing has advantages of obvious flexibility, maneuvering, high efficiency, low cost and the ability to improve the status of the lack of high resolution remote sensing data in cloudy and misty areas effectively, and thus make it be broadly applied in the field of management of land and mineral resources, geological environment evaluation and disaster prevention, rapid acquisition of space information and topographic map update. Therefore, the low altitude photogrammetry based on UAV remote sensing platform and ordinary digital camera has become a hot research pot in the field of remote sensing and photogrammetry both domestic and overseas. However, because the flying height of UAV is low in general, single image obtained has less coverage. In order to solve the contradiction between image coverage and resolution, and to reflect the situation over the study area accurately and in time, it is needed to match the obtained images and mosaic them together to form a full scenes.This dissertation first introduces the composition of UAV remote sensing system, analyzes the planning parameter setting of UVA flight route, and carries out distortion correction and dodging for the obtained images. Then SIFT algorithm is introduced into the UAV image matching, and the thorough research to the algorithm is carried on. According to the characteristics and patterns of the UAV images, this dissertation estimates the overlap area of the images before using the SIFT algorithm in image matching, which reduces the computation burden of the matching effectively. For UAV images in difficult areas, in terms of the image feature is not obvious or relatively scarce, we adopt a method combines point and line features for the matching. When doing UAV image stitching, splicing error accumulations is a problem that cannot be ignored. In order to control error accumulations, three strategies are proposed:the best overlapping degree strategy, the best datum strategy and the best stitching path strategy. Specifically, the research work and innovations in this dissertation are mainly the following several aspects:1. According to the parameters of UAV (Unmanned Aerial Vehicle), estimates the overlap of adjacent images firstly, calculates the overlap areas of adjacent images by using phase correlation method, take the average of the overlap areas of final images. This method not only avoids the expanding, but also controls the excessive the narrowing. Through the calculation of overlap area of images, reduces the scope of image which produced the feature points, shorts the time cost of production of feature points. Obtaining the optimal Gaussian nuclear size that applies to the images of UAV, the experiment shows that the feature points’ number and precision are optimal which are produced under this size, and this size relative to the fixed nuclear size, its time efficiency to increase by almost20%.2. Aiming to the UAV images in some difficult areas, that is to say the object features of images are scarce or not obvious, this dissertation proposes a match method which combined point features and line features, on the basis of completing preliminary matches by using the point features, carrying out the further fine matching by using line features as a supplement. The method makes full use of complementary characteristic of point features and line features, and makes the robustness of image matching is improved by about10%.3. Aiming the error accumulation problem in the process of UAV images stitching, there are three strategies is proposed:the best overlapping degree, the best datum and best stitching path. Through the experiment testing, the range of the best overlap degree of UAV image is25%～37.5%, according to this number, applying rarefies to the UAV images before joining together. Remote sensing image information entropy as a basis to choose the best datum, experiments prove that using information entropy to choose the reference images are able to produce rich feature points, and it is easy to be with adjacent image matching. Searching the path of image splicing by using Ant Colony Algorithm, experiments show that this algorithm can quickly and accurately search the best stitching path.4. Through the analysis of the various parameters in UAV route planning, finds out the relationship between each parameter, based on the relationship between them, and proposes the method to best route design. The quality of UAV images which are obtained by using the route is designed by this method is improved effectively, provides a good data source for image matching. The distortion of UAV images is corrected by the system distortion model which is adjusted by the beam method of additional parameters, corrected images eliminates the impact of distortion to matching. Carrying on gray-scale adjustment to images by using the method of histogram adjustment, through gray-scale adjustment, eliminates the impact of uneven exposure to matching.