Research On Splicing Technology Of Drone Aerial Images

In recent years,with the continuous rise of the drone industry,more and more researchers have been attracted to engage in various research and development related to drones(UAV),which has led to rapid development of drone technology.The UAV have also been widely used in mapping,emergency command,and agricultural irrigation.The great application of the UAV aerial photography technology is especially reflected in coping with emergencies.In case of sudden natural disasters,the aerial remote sensing system carried by unmanned aerial vehicles(UAV)can quickly move to the disaster area and report the dynamic monitoring of post-earthquake disaster,geological landslide and debris flow.In addition,the UAV can provide more powerful help for subsequent reconstruction in the disaster area.Due to the traditional UAV platform is limited by the flying height and the focal length of the camera.However,the images captured by the UAV can not completely be cover the entire target area.In addition,the stability of the drones is usually poor because the size and power of the UAV are limited,which in turn the drones appear to be tilted and shaken during the flight.So the images captured by UAV will blur or even distorted.For the problems in the aerial images,this paper proposes an improved ORB algorithm based on quadtree structure and Gaussian pyramid theory.In our Algorithm,the quadtree structure is used to improve the feature point extraction non-uniformity of the ORB algorithm,and the Gaussian pyramid theory is used to improve the transformation robustness of feature points.In addition,based on the improved ORB algorithm,a real-time splicing of panoramic images algorithm is realized,which enables the output of wide-field,high-resolution and high-quality panoramic images.Our algriotmw improves the splicing rate and quality of the aerial image of the drone,and better satisfies the real-time requirements in image splicing.The research contents of this paper are as follows:(1)The image preprocessing is carried out for the image heterogeneity,image noise and image enhancement caused by the difference in the corresponding time of the pixel of the airborne camera focal plane array in the UAV aerial image.(2)Analyzed and compared the mainstream methods of feature point detection and matching in current UAV aerial image Mosaic technology.The feature calculation principle and feature point descriptor calculation principle of ORB algorithm are emphatically studied.Through experiments,it is found that the original ORB algorithm has too concentrated feature points and weak robustness against scaling.(3)Aiming at the over-concentration of feature points in the output region of the original ORB algorithm and the two deficiencies of the robustness of feature response to scale transformation,improvements were made.First,the input image is iteratively segmented based on the quadtree structure,and the features of each region are calculated separately to retain the feature points with the largest response in the region,so as to solve the problem of feature extraction being too concentrated.Secondly,in view of the weak robustness of the original ORB algorithm's feature points to cope with scale changes,SURF feature points'computational ideas were utilized to select points exceeding the threshold as feature points in the 3 3 pixel neighborhood in the Gaussian Pyramid space to enhance the ability of feature points to cope with scale changes.Finally,violence matching is used to search the corresponding feature points in traditional ORB algorithm.This paper also USES the nearest neighbour matching algorithm to speed up the feature matching process and reduce the time of ORB algorithm matching link.(4)Based on the improved ORB algorithm,UAV aerial image splicing is realized.The experimental comparison shows that the improved algorithm is better than the traditional image splicing algorithm and can better meet the real-time requirements.