The Research Of Automatic Relative Height Generation Based On Aerial Stereo Images

Considering the contradiction between the difficulty in obtaining elevation and the application requirements on elevation information, this paper improved essential methods in photogrammetry and image matching, presented the conception of relative height, and studied the approach of automatic relative height calculation in aerial stereo images on the basis of corner detection and image matching. Finally, the experiment was verified and the prospect of application on relative height was suggested.In order to imply the automatic calculation of relative height, this paper has studied image matching based on corner feature. On the basis of analyzing and comparing existing corner detection algorithms, SUSAN algorithm has been adopted and improved in this paper for corner detection and image matching for high-resolution images. Aiming at some problems existed in the traditional SUSAN algorithm, such as over detection and edge sensitive for irregular texture image area, we proposed an improved algorithm. Firstly, the threshold of gray difference is calculated adaptively according to the relationship between local and global image contrast. Secondly, rectangular template is adapted to determine the possible limits of SUSAN region from the edge. Finally, the corners are detected based on above two steps. Experiments show that the corners location is detected more accurately with the improved algorithm, and the results are more reliable and less redundant with much higher speed in image marching.In general, because of there is no enough prior knowledge, the search space of corner matching needs to be extended to the whole image, and it leads to an obvious inefficiency and low accuracy in the course of matching. In order to improve the speed and the accuracy, the method of two-grade image matching was proposed in this paper. Firstly, we split the image into sub-images and determine an image block in every sub-image; then use blocks to proceed a rough matching and confirm the spatial-relationship between two images. Through this way, the searching range of corners decreases availably, and the matching speed rises greatly. Used in multi-scale corner matching based on image pyramid, two-grade image matching can prevent the error in corner matching from occurring in top layer and propagating into other layers.With the homonymy points obtained from the course of image matching based on corners, we transformed original images into epipolar images, and then simplified the mathematical relationship between ground objects and image points. Finally, we proposed the method of automatic relative height calculation in stereo image, and inferred the theoretical precision of the method. Relative height can express the changes in elevation of ground objects accurately, and decrease the difficulty in obtaining elevation information from stereo image.Because the direction of aerial photography approximates to vertical, after stereo image transformed into the attitude parallel to the base-line, the angle between base-line and horizontal can represents the attitude of stereo image. Meanwhile, the homonymy points are in the same row in stereo image, and the relative height can be calculated with the abscissa of homonymy points and the angle between base-line and horizontal. The experiment indicated that relative height can express the changes in elevation of ground objects accurately and precisely, and satisfy the requirement of the elevation information without exterior orientation elements.