Research And Implementation Of Drone Aerial Image Stitching Technology

The aerial photography of the drone is based on the unmanned aerial vehicle,equipped with remote sensing equipment to obtain information,and the computer image is processed to obtain the final image.Compared with systems such as satellite remote sensing,drone aerial photography is smart,lightweight,low-cost,and superior in performance,and has advantages in the field of aerial photography.Aerial photography of drones is widely used in hazardous area investigation,battlefield investigation,marine environment monitoring,natural disaster monitoring,etc.However,due to the limitations of flight height,camera focal length and other factors during the aerial photography of drones,single images are difficult to comprehensive.The target area is displayed,so it is common practice to take multiple images for automatic stitching.With the increasing number of aerial photography applications for drones,its image stitching technology has attracted the attention of many scientists,pushing image processing technology into a high-speed development stage.In this paper,the current research achievements and application status of image mosaic technology are summarized and analyzed.It is pointed points out that this technology field has been based on feature-based methods in recent years as the research focus and development direction.UAV image mosaic is divided into four processes:image preprocessing,feature detection,image registration and image fusion.Image preprocessing is to perform grayscale and sharpening on the acquired image to simplify the image calculation workload and enhance the image clarity.In addition,geometric correction is needed to reduce the impact of image distortion.In the process of image mosaic,if the feature points are extracted directly by using the SIFT algorithm,the number of feature points will be very large,resulting in a significant increase in the workload of image registration process,and the registration efficiency will be reduced.In order to avoid this problem,SURF algorithm is adopted in this study,and Hessian detector is introduced to enhance the stability of feature points set,remove the unstable edge points in advance,so as to improve the efficiency of the algorithm.Image registration is the process of matching and superimposing images.After image feature extraction,geometric transformation is used to perform feature matching.In this paper,the improved k-d tree BBF algorithm is used to improve the matching efficiency,and then the random sampling consensus algorithm is used to remove the wrong matching points.Image fusion is the process of fusing two or more images into one.In this paper,the level of image fusion and four common fusion methods are analyzed and studied.Wavelet transform fusion method is used for image fusion.The experimental results show that,compared with SIFT algorithm,the introduction of Hessian detector,although the number of feature points extracted by SURF algorithm is greatly reduced,it has higher accuracy and efficiency.The improved BBF algorithm has stronger registration ability,faster matching speed and efficiency,and stronger anti-jamming ability.The system can be processed simply and conveniently,and the final shape is mosaic image,and the effect is excellent.