The Method Research Of Airborne Three Line Image Matching

Digital photogrammetry technology is widely used for deriving geometric, and radiometric information of objects in the3D real world from2D digital images. Along with the more available of more image data and the increasing application demand for the3D spatial information, the performances of high automaticity, high efficiency and high reliability to the digital photogrammetry are more considered nowadays. Image matching, as one of the key technologies in the digital photogrammetry and remote sensing, computer vision, etc, is always the hot research topic.As a representative of the line array sensor, with the technology of GPS/IMU, ADS40can obtain more high resolution digital aerial images for the same area quickly and accurately, These images can offer sufficient image information and texture information for image matching. Different from image’s character of conventional frame image, traditional image matching algorithm is not fully applicable for ADS40image. Therefore, this paper proposed a tie point matching and dense matching method based on geometrical constraint by POS data, the main research content includes the following two parts:(1)An ADS40image matching method on SIFT features constrained by POS data. For traditional tie point extraction method is difficult to adapt to the ADS40image, this thesis proposed a SIFT feature match method constrained by POS data. Firstly, establish a grid in the reference image and find out a central point in each grid; According to the ADS40projection, project these particular points to match images to obtain a certain number of projection points and form qusi-eipipolar line. Then centered on the respective points of the reference image and qusi-eipiploar line, establish corresponding SIFT features and matching area to get the final match results.(2) Multi-images matching based on space and geometric constraint. In the light of the problem that the low efficiency of traditional global dense matching method, the thesis use the local correlation coefficient method to match, which can obtain a large number of corresponding points quickly to rebuild terrain. Firstly, the reference image is divided into evenly grids with a certain size, and then the grid points are converted to object space points of a certain elevation range by geometric relationship and back-project onto all search images to obtain the corresponding qusi-eipipolar line. After that, using the correlation coefficient as the matching measure, match the reference image grid points with the points on eipipolar line one by one. Finally, taking all matching points to obtain object space points by forward intersection, then back project them to the reference image, comparing the projection points with corresponding match points. If the difference value meet the threshold condition, then save it as the final match point. The experiment results show that the proposed method could improve the matching reliability effectively compared to traditional image matching method, and can access a large number of corresponding points fast, meet the needs of the terrain reconstruction.