Depth Image Based Close Range Non-Cooperative Target Relative Pose Measurement

The ever deepening space exploration boosts the demands for high performance satellite on-orbit service unceasingly.Among the key technologies in satellite on-orbit service,space non-cooperative target close range relative measurement provides essential navigation data for server satellite to carry out rendezvous,and target capture,as well as the subsequent repair and maintenance task.In this thesis,the non-cooperative target relative measurement is considered from the whole satellite on-orbit service task perspective.The main research are as follows:(1)In this research,a solution to the space non-cooperative target close range relative measurement is addressed,which integrates 3D reconstruction,3D recognition and pose estimation as a whole by extensively utilizing depth and color information and combining state of the art methods in 3D reconstruction and 3D recognition.Our approach works closely with the two kinds of formation flying states during rendezvous,and performs target geometry reconstruction,target recognition and pose estimation in an orderly manner.During the phase of flying around,the Kinect Fusion base 3D reconstruction thread collects observation data and fusion them with variating weights into a fine target model.After,the 3D recognition thread firstly coarsely recognizes the instance target pose with the reconstructed model.Then the final iterative closest point based pose estimation pipeline initializes with the coarse pose and performs pose estimation continuously.The feasibility of this solution is verified through ground simulating experiments.(2)A color image-guided depth image up-sampling algorithm,which is important to the data preprocessing in relative measurement,is proposed to tackle the problem of the unsatisfactory lateral resolution of the depth sensor output.Based on the classical ROF model in image restoration,the color image gradient is combined with the total generalized variation regularization term which is refined with an anisotropic diffusion tensor,to guide the depth image detail restoration.And an adaptive PDHG algorithm is used to optimize the up-sampling model to get a good result quickly and accurately.Experiments show that the proposed algorithm has good performance on noise contaminated image up-sampling.