| Digital photogrammetry technology is now widely used for the derivation of geometric, radiometric and semantic information of objects in the 3D real world from 2D digital images. Along with the easy availability of more image data and the increasing application demand for the 3D spatial information, the performances of high automaticity, high efficiency and high reliability to the digital photogrammetry are more considered nowdays. Image matching, as one of the core technologies in the digital photogrammetry, is always the hot research topic.In recent years, with the increasing application of aerial digital sensor, the availiability of large overlap images is easier and easier. As one object probably appears in 15 or more images simultaneously, it has now become an issue of aerial photogrammetry that how to match these corresponding points reliably and accurately. In addition, with the development of low-altitude photogrammetry and increasing use of aerial digital cameras in conventional aerial photography, the situation that adjacent images exist large rotation angle is frequently encountered, so it is extremely urgent to carry on the study on matching images with large rotation angles. Furthermore, high accuracy airbone Positioning and Orientation System (POS system) is used, during the image acquisition the exterior orientation parameters are directly determined at the same time. How to make use of POS data to support image matching in difficult terrain areas (such as images of repetitive texture areas) is also a universal concern.In the light of new problem of image match brought by the current development in photogrammetry, the matching methods of image matching for special images is mainly studied in this thesis. The emphasis are mainly on image matching for large overlap images, image matching for large rotation images as well as POS-supported image matching for repatative texture images, the key issues and technical details of which are studied systematically and in-depth. The research work and innovation of the thesis is mainly reflected in the following areas:1) Research on accurate and reliable matching of large overlap tie points.Large overlap tie point, can not only increase the intersection angle of corresponding light and improve the elevation accuracy, but also provide more redundant observation for bundle adjustment and impove its geometric strength. In this paper, aiming at the shortcoming of traditional aerotriangulation methods, a simple and practical strategy of matching in image space-free network adjustment-matching in object space is proposed. This method can make full use of both advantages of image matching in image- and object-space. Firstly, more accurate exterior orientation elements of images are obtained by image matching in image-space and free network adjustment, which provide approximate values for the image matching in object-space. Secondly, the collinearity geometric constraints are introduced to image matching in object-space, which enables the multi-photo geometrically constrained least squares matching to consider the errors in the image exterior orientation elements flexibly. Finally, the multiple images are simultaneously matched and the reliability and success rate of image are improved. Experiments have been made on one set of large overlap aerial digital frame imagery of some area in China, and the effectiveness of the method is verified. The results have shown that the proposed method can acquire the image tie points successfully for the images with large overlap, which can meet the demand for automatic point transferring in aerial triangulation, and improve the height accuracy of the photogrammetric points significantly.2) Research on strategy of multi-image matching by synthesizing image and space informationFor the stereo matching, because there is no redundant information, errorous matching points cannot be completely removed only by epipolar constraint. As for the large ovelap images, one object probably appear on multiple images simultaneously, and a great deal of redundant information is provided for photogrammeric data process. A strategy of multi-image matching by synthesizing image and space information is put up with in this thesis. Firstly, space information is refined by using image information and then space information in turn guides and restricts the matching on all images, through the cross-use of image and space information, the success rate and reliability of image matching is improved eventually. Experiments have been made on one set of large overlap aerial digital frame imagery of some area in China, and the effectiveness of the method is verified. When applied to dense matching for automatic DSM generation, the quality of DSM can apparently improved.3) Research on matching for images with large rotation angle.When there is larger rotation angle between images, the pyramid image matching methods based on normalized cross correlation often comes to a failure and is unable to obtain satisfactory resuts. In view of this, an image matching method based on SIFT feature matching rotation compensation is developed. Through SIFT feature matching, the rotation angle between images is estimated, which is used for the compensation and improvement in traditional pyramid image matching, and this can achieve the automatic matching of images with large rotation angle. Experiments have been made on one set of large rotation aerial digital frame imagery, and the effectiveness of the method is validated.4) Research on strategy of POS-supported image matchingCurrently, the exterior orientation parameters of images derived by direct georeferencing are not accurate, for high precison target positioning, it's still necessary to perform aerotriangulation on entire block images to. For this actual situation, POS-supported aerotriangulation for repatative texture images is studied. In this thesis, POS data is not only used to predict the initial position of the matching points, but also to derive epipolar constraint to restrict matching. Due to the reduction of the search range and the probability of multi-peak phonominon, the matching success rate of matching repatative textures images is improved dramatically. Furthermore, by introducing POS data in the aerotriangulation, cumbersome steps in the traditional method such as free strip network construction, articial selection of offset point between strips, establishment of approximate relationship between neighboring strips and so on can be eliminated. Experiments have been made on one set of aerial frame imagery, and the effectiveness of the method is validated, which can significantly improve the efficiency and matching success rate of point transferring between strips.Based on the theory and method of image matching proposed in this dissertation, which are applied to the aerotriangulation and automatic DSM generation, the corresponding automatic image measurement experimental system is developed in the Windows environment using Visual C++ language. This system has now applied in the data processing of several national baseic geographic mapping projects successfully.
|
PDF Full Text Request