Research On Multi-view Matching Method Of Multiple Aerial Image

With the continuous development of surveying and mapping technology, a variety of new equipment is emerging in this field, a large number of digital sensors have been widely used; especially in the field of aerial photogrammetry, the oblique digital sensor is considered to be typical equipment in this field recent years. The oblique digital sensor compared with the traditional film, digital sensor, not only gets the image that can acquire the top information of the buildings, but also obtains the image that can acquire side texture information covering the buildings; because of this superiority, the oblique digital sensor expands its application fields. Although the future of the sensor is very attractive, but its own large rotation angles and high overlap images are a huge challenge to the digital photogrammetric system; aerial triangulation is a core technology in the system, we need to solve the technical problems caused by the oblique images. In this background, the paper will introduce multi-view matching algorithm into the aerial triangulation field, trying to solve the problem that gets the high quality tie points caused by oblique image, the research work and innovation of the article is mainly reflected in the following aspects.1) The oblique image and multi-view matching, that are the research objects in this paper, firstly the paper introduces their basic knowledge, then focus on the advantages compared to the traditional images and the definite reason that the oblique image is not suitable for the old photogrammetric system; This paper also gives the advantage of the multi-view matching compared with traditional single stereo matching, the reason that multi-view matching can be used for oblique image matching. According to the above superiority, paper gives the matching method that suitable for oblique aerial triangulation and its specific implementation process.2) Depending on the process given in the article, the paper gives the relevant knowledge of image pre-processing and feature extraction. The main part of the image pre-processing include generating gray image, pyramids image, and wallis filtering; after analyzing the characteristics of the existing methods, paper gives the method used in this article and the reasons that why paper choose them. For feature extraction, paper introduces its basic knowledge, the principles of SIFT algorithm, the reason that paper choose the SIFT, and a method that calculates the repeatability of feature points. At last, paper gives a series of tests to confirm the reasonableness of method used in the article.3) In the aspects of multi-view image matching and coarse error elimination, paper gives the basic principles of feature restricting GC3that the multi-view matching model used in the article, the improvements compared to the original model and the process of the improved GC3. In the part of coarse error elimination, paper combines the selecting weight iteration with the multi-chip forward intersection, the relative orientation of stereo pair. The article combine one tie points in multiple images with two match points in tow images to complete coarse error elimination to get the higher quality matches results.