Research On Key Technology Of Vedio Image Stitching Of UAV

As an emerging aviation remote sensing platform,unmanned aerial vehicle platform has the characteristics of fast starting,low cost and low environmental requirements.High-resolution video camera equipped on the unmanned aerial vehicle can capture high-resolution images.Through the stitching of images,it can be applied in disaster relief,traffic supervision,smart city and other fields.At present,the video image stitching technique of unmanned aerial vehicle is still immature.How to achieve fast and high-precision stitching is still a major challenge.This thesis chooses video image stitching technology of unmanned aerial vehicle as the research content,the relevant technical data at home and abroad are consulted.The characteristics of UAV video images are understood.And the key points are researched.Three research focuses are selected: adaptive research on key frame extraction;optimization of cumulative error in splicing process;image registration algorithm based on feature points.Pointed at these three priorities,the main achievements of this thesis are:(1)Aiming at the change regulation of the video image overlay degree of unmanned aerial vehicle,an adaptive key frame extraction method is proposed.The method is a repeated iterative algorithm.The frame interval extracted by the above key frame is the initial value of the iteration,and the ratio of the actual overlap between the current frame and the previous key frame and the given overlap is continuously calculated and adjusted.Frame spacing of the frame until the overlap requirement is met,the frame image satisfying the overlap requirement is used as the key frame of the current time,and the required frame spacing is satisfied as the initial value of the next extracted key frame,so that the repetition is repeated until all key frame extracted.(2)In order to reduce the cumulative error generated during the splicing process,the sparse beam method adjustment is introduced to minimize the projection error and optimize the absolute homography matrix.The experiment proves that the method solves the problem of splicing misalignment caused by cumulative error to a certain extent.(3)The three feature extraction methods of SIFT,SURF and ORB are studied.Through the experimental comparison of the key frames of UAV video,it is concluded that the SURF algorithm is a better choice from the real-time and quality consideration of UAV image stitching.(4)In the process of feature point matching,a RANSAC matching algorithm combining K-means clustering is proposed.The algorithm classifies the matching point pairs by K-means algorithm,and selects the one with the most data points as the matching pair.By eliminating the remaining matching point pairs,it provides excellent data samples for RANSAC fine matching,thereby improving the accuracy and speed of matching.