Research On Method Of Docking Ring Spatial Position Intelligent Perception

Space exploration has always been a scientific activity of human beings.With the development of science and technology,space activities near earth are increasing day by day.However,the frequent development of near-earth space activities has also brought about some problems.One of the serious problems facing us is the failure of satellites in orbit,which not only brings huge losses in economy,but also wastes space resources in orbit.In order to repair and maintain the malfunctioning satellite,it is very important to study the space-related technology for in-orbit service.One of the most important tasks in the in-orbit service space mission is target capture.The service spacecraft should approach the target spacecraft slowly and achieve stable and accurate capture.In the whole process of approaching the target spacecraft,the first problem to be solved is the relative pose measurement,that is,the relative navigation task.Relative pose includes two sets of variables: relative position and relative attitude.The accuracy of the estimation of these two sets of variables has a great influence on the following guidance and control tasks.If a stable and accurate relative position and attitude can be obtained,the next step is to give reasonable control feedback based on these two sets of variables,so that the spacecraft can approach the target safely.This paper studies the relative position estimation problem during the approach of the close range stage(the relative distance between two spacecraft is less than 10 meters).In the process of the approaching stage before our stage,two spacecraft relative attitude has been adjusted to the range of allowable error by star sensor,and in this phase,the measurement error of radar and laser scanning laser rangefinder did not meet the requirements of relative navigation precision,so research proposes to use visual perception to realize the relative position of two spacecraft in close approaching stage.The research content includes the following three aspects:1.Relative position perception method based on convolutional neural network.In the situation that key point on the target spacecraft cannot be detected,image contrast is poor,texture is complex,span process dimension is large,texture characteristics of edge detection effect is not stable,this study try to break through the traditional measurement method that the detect point firstly and match characteristics next to calculate the position of relative position,using convolution neural network supervised regression method,directly estimates the relative position between the service spacecraft and the target spacecraft docking ring from original image shot by three cameras on the installation of service spacecraft.2 Relative position perception data smoothing and relative speed estimation.Kalman filter is used to filter and smooth the relative position obtained from the relative position estimation of the convolutional neural network,and the stable relative velocity estimation is obtained at the same time.3.Relative navigation simulation experiment in simulation environment.Firstly,referencing the real situation of the approaching movement process,this study establishes a mathematical control model to calculate the relative acceleration from relative position and relative velocity.Then a combination of the convolution neural network perception model,Kalman filter and control model is developed in the simulation environment to simulate that service spacecraft approaches from far to near to the target spacecraft docking ring.Finally,the accurate of final position of service spacecraft is counted to evaluation the performances of the convolution neural network perception,the relative position smoothing and the relative velocity estimation.The relative position perception method proposed in this paper is the exploration and research of applying digital image processing,computer vision and deep learning methods in the field of aerospace technology.The simulation results show that the relative position estimation model based on convolutional neural network can be used to calculate the relative position estimation,but there are some random errors.Through the processing of Kalman filter,the error of relative position estimation can be suppressed,and the accurate relative velocity estimation can be obtained.The combination with the relative position perception model,the Kalman filter and the approach control model achieves the expected relative navigation success rate.