Study On Binocular Vision Measurement For Satellite Pose Based On Model Matching

A space robot has to gain the precise pose information of the target while it performsin-orbit services. Currently, most pose measurement works by searching the pre-installedsigns or salient features on the target. However, when the pre-installed signs fail or thedistance between space robot and the target is large, it is difficult to obtain the featureinformation on target surface. In this paper, the method of satellite pose measurementwas researched based on model matching technology, which was supported by the projectof the National High-Tech Research and Development Program (“863” Program) ofChina in the field of aerospace called “an intelligent space robot for in-orbit services”.And the effectiveness was verified throughout simulation. This thesis’s work included thefollowing aspects:Firstly, the theoretical database of a model was created and the virtual image of asatellite was generated. The model was created by the ProE software, and then modelboundary data points were extracted after the model was meshed by the ANSYS software.The binocular images were created based on OSG virtualization technology. In order tobe more realistic, noise was added to the virtual images.Secondly, the three-dimensional data of edges was obtained. After the noise wasfiltered, the image edges were extracted. Corresponding points were found by binocularmatching; the sub-pixel position of edges was located throughout fitting method. Andthen the three-dimensional data of edges was reconstructed based on trigonometricparallax principle.Thirdly, iterative closest point (ICP) algorithm was selected to calculate the poserelationship between the theoretical data and the three-dimensional data according to thecharacteristics of the measured object in this paper. Considering the occurrence ofsingular value and local minimum, singular value exclusion method based on thethreshold and initial registration method were given for solving these problems.And thek-D tree searching method was used to accelerate this algorithm.Finally, simulation experiments were designed to evaluate these methods whichwere mentioned above. In order to get close to the real circumstance of measurement, themeasured data was gained in different perspectives. The results of these experimentshave showed that the method in this paper can meet the requirement of real application.