Research On Remote Sensing Image Stitching Based On ORB Features By Unmanned Aerial Vehicle

Image stitching is the process of splicing a complete image through multiple reality images acquired in the same scene by specific transformation.In recent years,image stitching technology has been widely used in medical image processing,computer vision,remote sensing image processing,virtual reality technology and other fields.Especially,with the rapid development of unmanned aerial vehicle(UAV)technique,remote sensing image stitching based on UAV system has been the hotspots.First the basic processes of image stitching are introduced,including image acquisition,registration and fusion.Then aim to serve low process speed problems of traditional SIFT or SURF algorithms that unfitness in real-time-application for UAV remote sensing images,an improved fast remote sensing image stitching algorithm based on ORB feature points detecting is proposed.One of the key processes is image registration.Feature points detect is the first stage of this process.Feature points and feature descriptions are quickly extracted according to the ORB features extraction.Then,a two-step matching method based on K-Nearest Neighbor(KNN)and random sample consensus algorithm(RANSAC)is adopted to match these points.A search strategy based on the K-d tree is used to search the feature points pairs in rough matching step,and the RANSAC algorithm is used to remove the fault matching points and realize fine matching step.The intensity of light may change rapidly causing by the UAV continuous shooting,as a result,lead to the color aberration difference between images.If fusion and stitching directly without any process,image blurring and stitching ghosting dislocation in the overlapping area will unavoidable.To solve the problem,a Laplacian Pyramid fusion method is proposed.Our method achieves image smoothing and seamless image stitching,eliminates the stitching ghosting dislocation phenomenon effectively.According to the height of the shooting and the types of the objects,we divided the experiment scenes into three typical types.The stitching results are compared with the SIFT and SURF algorithm respectively.The experiment results show that at the same matching precision,the proposed algorithm is 20 times faster than the classical SURF algorithm,and 41 times faster than the SIFT algorithm.