Research On Vision Based Satellite Motion Analysis Pose Measurement And Capture Position Detection

With the continuous efforts to the exploration of space resources by human beings,there are a large number of space debris in the near-Earth space that potentially threaten the operation of spacecraft.Besides,the failed satellites in space orbit also need to be repaired or removed.Due to these actual on-orbit service demands,the capture of space non-cooperative targets has important practical implications.Among the space non-cooperative target capture technology,machine vision plays an important role,and the visual information realizes the control and navigation of the space manipulator.During the imaging processing of the target star by the onboard camera,the motion analysis,three-dimensional reconstruction,pose tracking and capture position recognition of the target star are accomplished.Each procedure is carried out step by step according to the distance between the capturing system and the target star,and the tracking and measurement accuracy are also continuously improved.Although manipulator control technology and computer vision measurement technology have made great progress in recent years,how to achieve autonomous and intelligent capture of space non-cooperative targets is still a very challenging research topic.Especially when considering the harsh and varied illumination conditions in the space,the complex electromagnetic environment and the limited computing power of the onboard computer processor,the space vision technology has raised higher requirements than the ground environment.Based on the monocular,binocular vision theory and deep learning based object detection technology,this article proposes a set of space vision measurement and detection strategies based on different distances,and implements satellite ground measurement and detection experiments based on satellite models and existing equipment in the laboratory.Firstly,this article analyzes state of the art and existing research methods of the non-cooperative target capture strategy and vision technology,then determines the basic approach of vision based space non-cooperative target capturing,which determines the basic research idea and research content of this article.On the basis of the relevant theory,we further build a system to carry out related validate experiments.In order to initially determine the rotational speed of the satellite at a long range to determine the feasibility of the capture task,the vision system turns on the monitoring mode.Calculating the rotational speed of the satellite by singular value decomposition and Rodrigues formula by extracting successive satellite images from the monocular camera;In order to accurately measure the relative pose of the satellite at medium range,stereo vision technology is introduced,and the vision system turns on the acquisition mode.First calculate 3D coordinates of the feature points with stereo image pairs and thus obtain the sparse point cloud of the satellite,and then implement the adjacent point cloud registration to obtain accurate relative pose results,finally reconstruct the satellite three-dimensional dense point cloud through multiple view images.At the short range,the detetion and recognition of the mechanical docking device on the satellite are completed to achieve the docking task in the final stage,and the vision system turns on the tracking mode.Through the object detection method based on deep learning,the bracket on the satellite is efficiently and reliably identified.Finally,relevant experiments are carried out to verify the validity and reliability of the above theories and algorithms.According to the experimental data and results,Conclusion can be made that the adopted method effectively solves the above difficulties and problems.