Research On Key Technologies Of UAV Images Fast Mosaic

Unmanned aerial vehicle (UAV) remote sensing possesses the abilities of instantobservation to the earth, as well as acquiring high temporal and spatial resolution images at lowcost. Therefore, it is widely used in photogrammetry and remote sensing, especially intime-critical event response. Due to the restriction of the image-acquisition conditions, theimages of the large scenes have to be generated by mosaicing technology. The imageregistration method based on scale invariant feature is not only applied to the registrationbetween images with less overlapping, but also used in the registration between imagescontaining motive scene and some defiladed objects. This thesis took the image registrationalgorithm based on scale invariant feature as its thread, and researched the image mosaicingtechnology based on scale invariant feature matching. The main work done in the thesis is listedas follows.1. The significance and development status of UAV image mosaicing technology issummarized and analyzed, and the technical process is listed, the main researched contents areidentified.2. UAV image matching based on features is explored. SIFT and SURF are introduced andanalysed. Randomized kd-tree search algorithm and best bin first (BBF) algorithm areevaluated. The principle of RANSAC algorithm is illustrated. Experimental results show thatSURF is faster in extracting point features than SIFT, and SURF with brightness contrastidentifier is much faster in calculating. An image matching method is designed, with thebrightness contrast identifier SURF feature extraction operator, the BBF search algorithm andthe RANSAC algorithm assembled, to realize image matching.3. The imaging principle and transformation model of UAV image is summarized. Theconcept of homography is illustrated and its solution is worked out. RANSAC algorithm is usedto estimate the original homography, L-M algorithm is used to calculate homography with highaccuracy. Several methods of image interpolation are introduced. The global registration ofUAV image mosaicing is studied. The accumulated errors emerges during the mosaicingprocess may influence the quality of the achieved image. To solve this problem, an overalloptimization adjustment method based on feature point is presented.4. Several methods of image blending are summarized and compared through experiments.A UAV image mosaic method is designed, with the image matching method based on SURFand multi-band blending method applied to the UAV image mosaic. With this method, themosaic of a group of UAV images with three strips is carried out. 5. The main content of the thesis and the technical process of UAV image swift mosaicingare summarized. A method of UAV image swift mosaicing based on SURF is summarized andexperimented, and good results are generated.