Digital Close-range Aerial Triangulation Based On SIFT Matching

In recent years, with the development of computer technology, imageprocessing and computer vision, the demand of photogrammetric automation isgetting higher and higher, in the digital photogrammetry, image matching is the keyto realize automatic measurement, and the foundation to get the3-D information ofthe object, it is also the emphasis and difficulty of image processing direction thatwe research in recent decades. Close range image matching technique is nowwidely used in many aspects, such as terrain reconstruction, pattern recognitionand so on, but because there are many repeated textures and a lot of shielding casein the close-range image, which lead to the result that the matching of close rangeimage is more difficult than conventional aerial image matching. In this context,the close-range image matching algorithms are studied in this paper, the purpose isto solve the automatic acquisition of connection points in the aero-triangulation.The research content mainly includes the following aspects:(1)Through the understanding of close-range image matching at home andabroad, some existing algorithms are studied, the matching strategies of pyramidstructure are used and on the basis of this a bidirectional matching,RANSACconstraint, disparity constraint and epipolar constraint to remove the wrongmatching is introduced, the least squares exacting matching on the bottom of thePyramid is used to improve matching accuracy.According to the research above,the paper proposed the matching strategy and process. The error of the matchingstrategy is reduced to the minimum, so that it can meet the accuracy requirementsof the relative orientation and apply to the close-range image matching.(2) According to the matching strategy which is formulated, the content of theimage preprocessing is determined, mainly including Wallis filtering, distortioncorrection and the formation of the pyramid images. (3) The construction and adjustment of the regional network.According to theclassic aerial triangulation theory and the characteristics of the close-range image,the automation processing of close-range data is realized, mainly including therelative orientation and model connection of close range images, absoluteorientation of free nets.(4)On the basis of theoretical analysis, close-range image pre-processing﹑image matching, construction and adjustment of free network are realized, takingVC++6.0as the development platform, using Canon5D Mark Two filming for theexperimental data of close range images.