Automatic Matching Method Of Aerial Image Based On Local Invariant Features

Because the historical aerial images captured the surface morphology, vegetation cover, natural landscape and environmental conditions in the shooting area during the shooting time, the historical aerial images now become valuable information to study the evolution of the surface morphology and the environment change. But most of the historical aerial images lack of photogrammetry parameters due to the limitation of technology measures and years of storing, which can not use traditional orientation methods which based on detailed photogrammetry parameters. The huge number of historical aerial image data makes the orientation by manually selecting control points also seem difficult. The relative orientation of aerial images is the basic step of strip aerial triangulation, and its result directly affects the quality and reliability of the application of aerial images. Therefore, it is particularly critical to obtain the corresponding image points between aerial images then match and orient them automatically.The research of image matching uses the same features of two or multiple images to find a spatial transformation to make them consistent in one space. The study has been successful applied in areas such as cruise missiles, image fusion and three-dimensional street sight making. By using image matching method based on image local invariant feature, the historical aerial images which lack of photogrammetry parameters can be automatically orientated and restore the correct position relationship between aerial images. Therefore, this research focus on using the image matching algorithm to oriented the historical aerial images witch lack of photogrammetry parameters automatically. This article is divided into the following sections:(1) The detection of local invariant features. This research discussed current image orientation methods without control points and introduced the MSER (Maximally Stable Extremal Region) method applied in remote sensing image registration. Ranked-order based adaptive median filter (RAMF) is used to conquer the noise effect of historical aerial images and to achieve a better result.(2) The description and matching algorithm in local invariant features. After extracting features from aerial images using MSER feature detector, the polygon feature regions will be converted to point features that can be described by SIFT described. The BBF matching search algorithm is used to search the matching points, then taking RANSAC algorithm to optimize the matching points. The results showed that, after doing RAMF and RANSAC algorithm optimization can achieve a better matching result of aerial images.(3) The automatic orientation of aerial images. Fiducial marks of the aerial images were extracted accurately by using canny edge detector and Hough Transform to achieve automatic interior orientation of the aerial images, then use the matching points to achieve aerial remote sensing image relative orientation automatically, and verify the accuracy.