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

As a new way of remote sensing,unmanned aerial vehicle(UAV)remote sensing technology has become a research hotspot in recent years because of its low cost,high efficiency,flexibility and convenience.UAV remote sensing technology is widely used in military fields such as battlefield reconnaissance,bomb dropping and surveillance,as well as mapping,disaster monitoring and traffic control.However,UAV as a remote sensing platform,its flight height is limited,airborne camera focal length is limited,leading to the smaller single remote sensing image information,to obtain the complete information of the target area,a large number of remote sensing image mosaic into a large panorama.The traditional image stitching algorithm can not meet the requirements of real-time,robustness and accuracy of UAV remote sensing.In view of these situations,this paper studies and studies the image stitching technology at home and abroad,compares and analyzes the popular image stitching method,and proposes an improved algorithm of UAV remote sensing image mosaic based on feature.Image preprocessing is a preparatory work before image stitching,which is directly related to the accuracy of stitching.When UAV gets the remote sensing image,it is easy to produce the change of shooting angle and illumination condition.The UAV is small in body shape,light in quality and easy to shake in the fuselage.These may result in obvious or geometric distortion of the image luminance drop.In this paper,the image is corrected by image denoising and image interpolation,and the noise and distortion are eliminated effectively.Image registration is the core of image stitching,which is directly related to the quality of stitching.In this paper,a comprehensive and in-depth study of feature based image registration is carried out,and an improved algorithm is proposed.First of all,the introduction of Hessian detection operator feature detection SURF algorithm,algorithm of invariance to scale change;secondly,select the feature description of BRIEF operator in ORB algorithm,realize the algorithm of invariant to rotation changes;then,the similarity calculation with Hamming distance,use the BF algorithm to achieve the characteristics of coarse matching;finally,using the the distance threshold removal large distance matching,and then RANSAC algorithm to complete purification.The experiment shows that the algorithm has good real-time performance,strong robustness and high accuracy.Image fusion is a fusion of images that complete the registration into an image,which requires the smooth transition between the brightness and the color of the image,and there is no crack at the connection.However,even two continuous UAV remote sensing images,the brightness and chroma of them are not the same.If directly stitching after registration,there will be seam and ghost in the image overlap area.In view of these conditions,this paper uses the Poisson fusion algorithm to complete the seamless fusion and stitching,and realizes the smooth transition of pixels.