Motion Estimation Of Space Target Based On 2D/3D Image Data Fusion

To reduce space debris, extend satellite life and recycle some parts of obsolete satellites, orbit servicing for non-cooperative space target becomes a research priority in many countries. Orbit servicing for non-cooperative is based on target closing and capture. And target motion parameters acquisition is a prerequisite to achieve these operations. Since the target structure and motion parameter are unknown, taking into account the sensor volume, accuracy and sampling frequency, this article adopts visual multi-sensor fusion method to achieve the target motion parameters estimation. Therefore, this article studies non-cooperative space target motion parameters estimation based on 2D/3D vision cameras facing to orbit servicing for non-cooperative space target. The main contents are as follows:Firstly, aiming at the 3D camera calibration in target reconstruction, an optimization function of attitude is established based on orthogonal projection model for the using of depth measurement data. In the view of the big measurement error, a convex optimization method is given to improve algorithm robustness and ensure the achievement of the global solution. This method translates the original optimization to a linear optimization problem constrained by a linear inequality. And then interior point method can solve this problem. Global convergence proof is given. This algorithm is numerically simulated. The results show that this method can rapidly converge with its strong robustness.Secondly, on account of 2D/3D data fusion in target reconstruction, joint bilateral filtering method based on confidence is designed. This method which combines with the high-resolution 2D gray image scene can enhance the 3D image resolution and keep the original edge information. As the scene with 2D grayscale images and 3D depth images are often have the similar discontinuities, at first the joint bilateral filtering function is established adding gray image into the filter function. At the same time in order to ensure accuracy of the depth image after resolution increasing, building confidence map based on depth image gradient because of the big edge region measurement error in the image. Adding the confidence map into joint bilateral filtering function, the joint bilateral filtering method based on confidence in this article is done. The simulation shows this method has excellent edge preservation and high precision. And the method can effectively complete scene reconstruction with some indoor scenes shooting by a Kinect.Finally, For estimation non-cooperative space target motion parameters, three-dimensional scene flow calculation is built. Under the assumption of same lighting, rigid body and small motion, using of two 3D reconstruction scenes frames, the three-dimensional scene flow calculation is built based on optical flow. Adding equations of motion in depth direction, this article deduces global scene flow algorithm and local scene flow algorithm for different scenarios. Scenes flow optimization function is established and solved by Jacobi iteration method. Then the target motio n parameters can be estimated. The Simulation result shows this method can effectively calculate target motion with rapid measurement.