Research On Algorithm Of Relative Pose Estimation For Non-Cooperative Spacecrafts Based On Binocular-Vision

The applications of non-cooperative measurement range from aerospace to military, intelligent transportation, medicine and multimedia. It is the key of some aerospace activities, for example on-orbit satellite maintenance, satellite components upgrade, building space station and capture enemy satellite. The technology of non-cooperative measurement will be a mjolnir for human space exploration. Recently, the research of measuring system based on CCD has been a hot topic in around the world of rendezvous and docking's researchers for its low cost and easy availability.The visual measuring method which is presented by this paper takes the background of a project of 863. It is designed for the performance of 3 DOF satellite rendezvous and docking to 5 DOF satellite in the air-floated platform. The paper presents a model-based binocular visual measuring method which is based on the achievements of visual measurement of non-cooperative satellite's rendezvous and docking. This method effectively overcomes the failure which is reasoned by lost the predesignated target's features or inaccurate measurement of features compared to the other algorithms that build on the recognition of target's features. It includes rough estimate module and refine estimate module. The advantage of the algorithm is not constrained the physical structure of the spacecraft.It recognizes the non-cooperative objective and gets its position information in the rough estimate module. Considering the huge information is operated in the binocular system, the module takes stereo vision function of OpenCV which was optimized by the parallel programming to reduce the system's burden. It can get the non-cooperative objective's 3D position on CCD frame system by OpenCV's stereo vision function. And then, through the absolute orientation it will be easily get the non-cooperative objective's position information relative to CCD frame system. For improving the robust of the absolute orientation to coarse matching error between the objective and the image, the module mix the process of RANSAC (random sample consensus) in absolute orientation to reject the huge error. From the simulation, it concludes that the absolute orientation combined with RANSAC can effectively get rid of the matching error and enhance its robust.Because of there must be some error of calculation in the process of recognition to real objective. It was designed the refine estimate module to reduce the error of calculation. However the refine estimate module need the coarse result from rough estimate module to start. In the refine estimate module, it selects 3D position of some points which are on the model's surface on non-cooperative frame system, and then projects these points to image. Minimize the distance between the projection and the real non-cooperative image and will get renew objective information. From the simulation, it concludes that the refine estimate module will be able to minimize the error to realize the success of rendezvous and docking on condition that the attitude error is under the 20 degree and the location error is under the 1 centimeter respectively.This paper will be a good reference to the performance of non-cooperative rendezvous and docking both in theory and practice.