Research On Matching And Mosaicing Methods For UAV Image Sequences

In recent years, UAV remote sensing is widely used in emergency rescue, power line patrol, construction of digital city, construction of new countryside, but the development of UAV image processing software is still lagging behind, such as the image matching is time-consuming, low quality, as well as the stitching of image sequences has a large accumulated error and the effect is poor, thus affect the application of UAV remote sensing. Therefore, research and improve the existing methods for UAV image sequences matching and mosaicing is very significant.Based on the analysis of the basic principle of existing methods for UAV image sequences matching and stitching, according to the UAV image with high resolution,large rotation angles, large scale changes, high overlapping rate as well as other features, We proposed a few new methods that which are adapted to the UAV image sequences processing, This research work mainly include the following aspects.(1) It analyzed the research status of UAV image sequences matching and stitching, mainly introduces advantages and disadvantages of existing methods for UAV image matching and stitching at home and abroad, and puts forward a kind of thought which can not rely on single matching method and fixed threshold to deal with the image sequences matching.(2) For the problems that UAV images in rapid stitching exist serious cumulative error, the deficiencies of RANSAC algorithm, and hard to determine the relationship of image position, we proposed image classification matching and stitching, RANSAC algorithm in combination with relative orientation to remove mismatched points and POS data using,etc. making the fast stitching for better result.(3)Proceed from the principle of feature matching, due to the current method is time-consuming, and uneven distribution of feature, after a detailed analysis of the relationship between the threshold and the feature extraction,this paper proposed the method of the grid feature extraction of threshold of automatic adjustment, as well as introducing the ORB characteristic descriptors, an effective solution to the uneven distribution of feature and the time-consuming match problems.(4) Because the existing image sequences matching method has a low level of automation, bad quality of matching, the lack of effective quality assessment and other issues, making the match with a certain blindness. In this paper, we according to the result of accuracy assessment of the relative orientation, uniformity assessment and three degrees of overlap detection, proposed a effective assessment method for UAV image sequences matching, making program could automatically change match Strategies, and thus on the basis of quality assurance to improve the automated level of processing.