The Research And Application Of The Unmanned Aerial Vehicle Image Matching Based On Feature Point

Image matching is a fundamental aspect of many problems in computer vision and digital image processing, including object or scene recognition, solving for 3D structure from multiple images, stereo correspondence, and motion tracking. There are troubles in UAV remote sensing images matching since UAV remote sensing system plat is not very stable and the pose of cameras is variable. Therefore, choosing suitable feature auto-matching algorithm for UAV RS images, which have irregular overlap and larger rotation, is critical for applying low-altitude remote sensing images in practice. We do further research in object matching for UAV RS images, especially in feature point matching, including feature point extraction, feature point descriptor and feature point matching strategy. The work and innovations in the thesis are listed as follows.1. Significance and application states of UAV RS image are analyzed. Detailed conclusions and summaries are given on images matching classifications and current research states. The main problems of UAV RS image matching are analyzed.2. The research is given on local feature, which is based on gray information of the image directly, detected from the rotation invariant, the scale invariant and affine invariant. Comparisons and evaluations of these local invariant feature detectors are done. In addition, the feature descriptors are researched and analyzed.3. There are three methods on removing error correspondence, RANSAC method based on linear shaped transformation model, correlation coefficient method and the main orientation method. The effectiveness of three methods is compared in experiment.4. The new feature matching algorithm is proposed, which is colour SIFT based on color image information and SIFT algorithm. A series experimental comparisons of SIFT, SURF and colour-SIFT in UAV RS images are done under effectiveness and adaptability.5. The SIFT operator can not process the large format aerial image and doesn't generate keypoints because of out of memory. The Large_SIFT algorithm based on dividing the large-format aerial image into image blocks is proposed and applyed UAV RS image matching.6. UAV automatic aerial triangulation system, GodWork, is designed, including UAV remote sensing image automatic matching with Large_SIFT algorithm,epipolar constraint and pyramid match, etc.