Research On Small Body Vision Based Terrain Relative Navigation

Nowadays,small body exploration has become an important project in deep space exploration.For the development of space economy,it is important to verify deep space exploration technology.In the detection of small bodies,the problem of accurate landing can not be achieved by only relying on inertial navigation.Therefore,it is especially important to study the relative navigation method based on computer vision.The terrain relative navigation method is used on the selected series of navigation points to solve the pose of the detector,which is used to correct the error of the inertial navigation,which can effectively assist the detector to achieve accurate landing.This paper first introduces the recent visual navigation methods,and analyzes their own advantages and disadvantages.Then,according to the problem that the small body has less real image data and difficulty in obtaining,a visual navigation simulation platform based on CINEMA 4D is proposed,which is used to generate simulated small body images and true values of camera position for experiments.This paper introduces and implements the terrain relative navigation method based on the observation image and the planning image in the case of small body 3D reconstruction data obtained before navigation.Terrain relative navigation is mainly divided into two parts: image feature matching and external parameter solving.In the first part,for the selected navigation points and navigation planning images,the SIFT features are extracted and matched with the navigation observation images,and the method of using image geometric constraints,including the polar geometry constraint and homography constraint,is proposed for filtering the existing error matching.Aiming at the fact that there is a large scale difference between the planning image and the observation image and may cause the matching result is not good,the image scaling based on bilinear interpolation is proposed to reduce the scale difference between the two images.For the close-range imaging,when the matching points are too concentrated after the homography filtering,a multi-homography filtering method is proposed to keep more matching points,so that the distribution range of the matching points is expanded.In the second part,for the features that have been matched and the corresponding three-dimensional data from the 3D reconstruction model,the single-image spatial resection method is used to solve the external parameters of the camera to achieve relative navigation of the terrain.Experimental analysis was carried out on parameters,data,etc.that may affect navigation accuracy.The experimental results show that the method of this paper can effectively realize the terrain.relative navigation function of small body.