Line Extraction ,matching And Three-dimensional Reconstruction In Satellite Images

Straight lines are significant and stable features in digital photos, aerial photographs and satellite images. Line extraction and matching is a primary problem of three-dimensional reconstruction , moreover,it is a key technology and a classical issue in many fields,such as computer vision,feature recognition and image registration.In this paper,we firstly extract line information accurately through improved Hough transformation,match line based on homography constraint and epipolar-line constraint, and reconstruct three-dimensional line based on homonymous line and three-dimensional point cloud data . The following aspects are mainly studied:Edge detection. The basic theory of edge detection are discussed. This paper summarizes and compares different algorithms of edge detection.By analyzing the advantages and disadvantages of every algorithm and thinking over the application purposes of edge image,it selects Canny operator to detect edge.The merit of this operator includes high signal to noise ratio , high precision and single-edge response .Line extraction.The basic theory of Hough transformation is introduced, and against the problems of line extraction algorithmic,the algorithmic of Hough transformation was improved. Adaptive median filter, neighbourhood calculation and grey edge image being changed into bianry image are put forward in this algorithmic.The anti-jamming capability of this algorithmic is stronger and the extraction speed is higher.Matching line.A novel algorithm about matching homonymous line based on homography and epipolar-line constraint is presented. In the first instance,it transforms sets of straight line segments extracted from different images to the same image coordinate system. After that, lighting of geometric properties of lines, the candidate matched straight lines are ascertained. The overlaps of the candidate matched straight lines are confirmed by epipolar-line constraint, The support property region of the overlap line is ascertained by definition and is described.Then, lighting of gray properties of support property region of lines, it ascertains the matched straight lines. In the end, it excluds the matched lines which are error with cross-correlation criterion to acquire the fast and reliable matching line. Reconstruction of three-dimensional line.Against the deficiencies of existing algorithmics of reconstruction of three-dimensional line, the algorithmic of reconstructing three-dimensional line based on homonymous line and three-dimensional point cloud data is put forward. According to basic principles of collinearity equations, it initially sets elevation value,then, verifies and refines elevation value, intersects beam and object plane, refines three-dimensional line,in the end,it completes reconstruction of three-dimensional line.The process of reconstructing three-dimensional line is uncomplicated and easy to understand, and it obtains three-dimensional straight line which has higher accuracy.