Research On Robust Algorithms For Non-cooperative Spacecraft Target Position Estimation

With the further exploration of space,the types of space missions have gradually achieved more diversified development,such as autonomous space rendezvous and docking,cleaning of space debris such as meteorites or failed spacecraft,on orbit maintenance and service of spacecraft,etc.In these space activities,the acquisition of non-cooperative target pose information is the key to complete the task.As an effective method of non-cooperative target pose estimation,stereo vision pose measurement technology has been widely concerned by many countries.According to the requirement of relative pose measurement of non-cooperative targets in space,this paper discusses the robust algorithm and error analysis of pose estimation of non-cooperative targets.the following studies have been carried out:Firstly,by analyzing the research algorithms and measurement equipment application status of non-cooperative target relative pose measurement at home and abroad,the measurement equipment and the existing mainstream algorithms supporting it are evaluated with indicators such as power consumption,calculation volume,and accuracy.The relative position and attitude measurement scheme based on feature for geometric calculation of the camera has high research and application value.Secondly,through the analysis of the pinhole imaging model and the definition of the related coordinate system,a parallel binocular imaging model was established.A distortion model of the vision system was established,and camera calibration experiments based on the two major platforms of OpenCV and MATLAB were completed.Evaluation of the experimental results was performed with reprojection error,efficiency,and interface visual effects as indicators.At the same time,the calibration parameters of the binocular vision system were obtained,which provided an experimental basis for the non-cooperative target pose solution.Then,in view of the difficulty of satellite attitude recognition algorithm classification,a non-cooperative target pose measurement algorithm for satellite solar panel recognition with widely representative meaning and stable feature performance is proposed,and it is designed by applying special data normalization The three-layer classifier cascade structure and the cascade rule combining classic and flexible non-maximum suppression algorithms are introduced.At the same time,in order to solve the problem of low efficiency of traditional feature extraction methods,color integration is proposed through time-consuming analysis and modular segmentation Map and supporting feature extraction methods.The validity and practicability of the proposed method are verified by experiments.Finally,the above algorithm is applied to the relative non-cooperative target relative pose measurement system platform.The platform mainly includes three subsystems of binocular stereo vision,non-cooperative target detection and relative pose and truth values.By designing targeted experiments,the rationality of the proposed method is verified.Combining the relative pose measurement results and sub-experimental results of non-cooperative targets,the error sources are analyzed from three aspects: binocular imaging system error,solar panel detection algorithm error,and feature point matching error.