Research On 3D Reconstruction And Target Tracking Of Non-Cooperative Objects In Space

With the development of space technology,human activities to explore space have increased,resulting in a large number of debris and invalid satellites.In order to maintain a sustainable space environment,we need to take the initiative to clean up space debris.In order to capture space debris,the tracking spacecraft must be able to track it stably in a short distance.However,these non cooperative targets are not equipped with docking devices or markers in advance,so they can not interact with the tracking spacecraft to provide the prior information needed for capture.Therefore,in order to achieve close range tracking of spatial targets,the first step is to solve the problem of 3D model reconstruction of non cooperative targets,which provides prior information such as target structure details for spatial target tracking and capturing.In the real space environment,because of the change of illumination environment and the material of object surface,the illumination of optical observation image will be uneven or the information will be lost,which will lead to the error of feature extraction and matching,and then affect the 3D reconstruction effect and target tracking quality.Moreover,the symmetry structure and repetitive texture of the satellite will make the reconstruction results overlap,or even fail.At the same time,the attitude jitter and background interference of the target in space will also improve the false detection rate of the target tracking algorithm and affect the tracking effect.In this paper,the key problems of feature extraction and matching,three-dimensional reconstruction of satelite model and close range tracking of space target are studied.First,in order to reduce the time illumination nonuniformity caused by the change of spatial illumination conditions and the different reflectivity of the external structures of the target,and improve the efficiency of feature extraction and matching accuracy,this paper estimates the illumination component of the image with the minimum perceptible error algorithm,and then corrects the nonuniform illumination of the image with the nonlinear function.Experiments show that the corrected image can provide more detailed information and extract more feature points.Second,in feature matching,due to the structural symmetry and texture repeatability of spatial objects,even in the case of wrong matching,the symmetry viewpoints will meet the geometric conditions suitable for the actual imaging structure,and will not be found by existing algorithms.In this paper,Kalman filter is used to predict the location of features to be matched,limit the search range,and actively search for matching features,effectively reducing the error matching.Then the camera pose is estimated by slam framework,and the sparse model of the target is reconstructed.After the camera pose is optimized by the beam adjustment method,the depth of the image is estimated and the dense model of the model is reconstructed.Third,for space target tracking,in order to take into account the detection efficiency,low SNR threshold is often selected,which will cause a certain degree of false alarm.In the process of tracking,it is difficult to distinguish clutter,false alarm and target mixture,which leads to the unsatisfactory effect of automatic tracking in practice.Combined with Kalman filter and probability data association,considering all measurements within the threshold,data association based on posterior probability information weighting can greatly improve the stability of target tracking in multi false alarm and multi clutter environment.In the process of space target tracking,in order to suppress the false measurement,improve the measurement correlation probability and reduce the calculation.