Application Research Of SLAM In Video Satellite Processing

With the continuous development of remote sensing,video satellites are gradually developing into a new type of earth observation mode.Video satellites can obtain highresolution dynamic video of the area by continuously shooting a certain area of the surface.Compared with traditional static remote sensing images,video satellites provide the possibility to capture and track ground moving targets in real time.However,due to factors such as hardware and software,there are few related studies and relatively lack of technology.Video satellites do not have the ability to locate and track moving targets in orbit.The application potential and potential commercial value of video satellites have not been fully exploited.In order to make satellite video have in-orbit positioning and tracking capability of certain target,this paper uses SLAM(Simultaneous Localization And Mapping)technology as the starting point,combining SLAM with satellite video processing technology.The application of SLAM in satellite video was studied.The algorithm of in-orbit processing and application of video satellite is preliminarily established.The process of in-orbit target localization and tracking of video satellite is constructed.The prototype SLAM framework suitable for video satellite target detection is proposed.A relatively complete set of video satellite SLAM application theory is formed.The system and related experiments were carried out.The main research contents are as follows:Fast feature extraction and matching algorithm.Based on the characteristics of video satellites,this paper analyzes the problems existing in the application of conventional computer vision algorithms in satellite video.Furthermore,a feature detection and matching algorithm suitable for satellite video is studied.In order to ensure uniform distribution of feature points,Gaussian difference pyramid is used for corner detection.To ensure matching efficiency,binary descriptor and brute force matching algorithm are adopted.Finally,the algorithm is verified in terms of feature point distribution and efficiency.Satellite pose estimation and conversion.From the perspective of SLAM application in video satellites,this paper studies the method of recovering the relative motion of interframe video satellites based on visual image matching and using homography matrix recovery.In the case where the relative pose is determined between frames,a method of converting the relative pose into a true pose under the world coordinate system(such as WGS84)is derived.At the same time,a method for solving the problem of the absolute scale uncertainty of the motion estimated by the visual odometer is proposed by using the in-orbit observation data.Video moving target detection and positioning.Firstly,aiming at the detection and localization of video moving target,the principle and method of conventional video moving target detection are analyzed.Then the algorithm for extracting moving target is studied under the condition that the computing resources are limited on satellite and the moving target of satellite video is small and the texture is weak.At the same time,combined with the relative pose and the attitude and orbit data measured by the visual odometry,the relationship between the moving target pixel coordinates and the realworld coordinates(such as WGS84)is derived by using the triangulation principle and the rigid body coordinate transformation.