UAV Remote Sensing Image Fast Mosaic Based On Improved SIFT Algorithm

Remote sensing UAV has characteristics of high efficiency,flexibility,low cost,and no casualties when observing the ground.They are widely used in military reconnaissance,agricultural perception,rescue and other fields.In order to obtain more data information,remote sensing UAV fly at low altitude.As a result,the area covered by a single remote sensing image is generally small.However,in the application of reality remote sensing,it is required to perceive the environment and goals at a wider region scale.Therefore,multiple consecutive multi-frame remote sensing images with smaller coverage areas need to be spliced together.Thus,a single frame panoramic remote sensing image with a wide coverage area is obtained.In this paper,the fast splicing method of UAV remote sensing images was studied.The improved SIFT algorithm was used to quickly splice a large number of continuous multi-frame remote sensing images,and the real-time splicing was greatly improved under the premise of reducing the splicing accuracy.The main research results are as follows:(1)There are a large number of feature points in the classical SIFT algorithm for high-resolution UAV remote sensing image splicing.Generally,the time cost of classical SIFT algorithm is larger as a result of huge feature points.In the paper,the classical SIFT algorithm was improved by the the monolayer extremum search and the minimum point suppressing.In detail,the monolayer extremum search was adopted to reduce the time cost of extreme point searching.The minimum point suppressing was applied in reducing the number of feature points.Experiment results show that the proposed algorithm improved the real-time performance of remote image registration.(2)There will be a “ghost shadow” in overlapping region of registration image due to the influence of light intensity and registration accuracy.In the paper,triangular-weighted Poisson fusion algorithm was adopted to eliminate “ghost shadow”.In detail,the weighted coefficient was determined by trigonometric function,and the image is initially fused by weighted average fusion algorithm.Poisson fusion algorithm was used to further optimize the fusion image.Experiment results show that triangular-weighted Poisson fusion algorithm effectively eliminated the “ghost shadow” and the texture transition of the overlapping area was smoother.(3)The effect of accumulated errors can not be ignored,when the number ofstitched images is larger.In the paper,the pyramid splicing strategy and the LM algorithm were adopted to reduce the accumulation of errors.From the second layer of the pyramid,the LM algorithm was applied in optimizing the transformation matrix so that the image could be accurately aligned.Experiment results show that the pyramid splicing strategy and the LM algorithm reduced the cumulative error and improved the splicing quality of the sequence image.