Automatic Mosaic Technique Of UAV Images Based On Features

The art of mosaic using unmanned aerial vehicle (UAV) remote sensing image has high value in application to battlefield surveillance and time-critical event response. In this thesis, the investigation and discussion of automatic feature-based UAV image mosaic were carried out.The main work and innovations in the thesis are listed as follows.1. Image matching based on features was explored. Three scale invariant feature detectors including SIFT, SURF and CenSurE were analysed. Three search algorithms including linear search algorithm, randomized kd-tree algorithm and hierarchical k-means tree algorithm were evaluated. Experimental results show that CenSurE is the fastest operator extracting point features, hierarchical k-means tree is the quickest searching algorithm, and all the three scale invirant feature detectors can satisfy the precision demand of UAV image registration.2. The model of UAV image registration was studied. The solutions to relative homography and absolute homography were worked over. Homography relationship between original image and the down-sampling image was presented, a fast solution to geometric transform parameters was carried out, and its dependability has been validated in experiment.3. The principle of spare bundle adjustment was studied. The problem of errors accumulation of geometry transform parameters was solved by using spare bundle adjustment. According to the sequence information of UAV images, strategy of bundle adjustment based on sequence information has been designed, and its stability has been validated in experiment.4. The main content of image composite was summarized. The solution to generating the consistency of photometric information of UAV images has been deeply analysed. Several image merge algorithms were compared through experiments. The need of image mosaic was analysed and an idea of interest-oriented mosaic was proposed. A simple implementation of interest-oriented mosaic was carried out.5. A method of automatic feature-based UAV image mosaic has been worked out by summarizing conclusions of the thesis and the whole flow chart was drawed. A automatic mosaic system for UAV images was developed, a set of datas was processed on the system, and good results were generated.