Key Techniques Of Image Information Processing For Non-cooperative Target-directed Autonomous Spacecraft

With the development of space technology, proximity operations have become more and more important in many space-related applications, such as Rendezvous and Docking, Satellite Servicing, Satellite Capture, etc. The tendency is to develop autonomous spacecrafts that are independent of navigation equipments outside and targeting at non-cooperative spacecrafts. So it is necessary to study the technology of acquiring the motion and structure of non-cooperative spacecraft. Under the background, we research a method of information processing based on image features, which is targeting at non-cooperative spacecraft and has two parts for different tasks, namely far-field image information processing and near-field image information processing. Key technologies in the method are discussed and the solutions are proposed. The logic circuits of some solutions are designed and implemented.The task of image information processing in far field is to detect, recognize and track target spacecraft. The key technologies studied in this phase are dim-small object detection and block object extraction. A detector of dim-small object is designed, which is based on energy accumulation and track association. Performance of the detector is improved through non-maximal suppression after energy accumulating. A block object detector is proposed which is based on gray entropy of image and can extract the area of the target integrally and accurately.The task of image information processing in near field is to estimate the motion and structure of non-cooperative target spacecraft. The key technologies in this phase are features extraction, features match and structure from motion (SFM). Firstly, the logic circuits for Harris detector are demonstrated and a fast method for Hough Transform based on pre-storage weighted matrix is established. The modules of gray histography statistics and neighborhood extraction have the potential to find wide applications in various algorithms of image processing. Secondly, the algorithms of features matching based on prediction by optical flow are proposed. The efficiency of initialing correspondences is higher through motion estimation and the reliable correspondences are determined by the measures of similarity which are constructed using properties of features. Thirdly, the algorithms based on lines for SFM are proposed systematically, including directly linear transformation, robust estimation and bundle adjustment. Therefore, rich and stable line features in the target are used to estimate structure and motion with high precision and robustness while the chase spacecraft circles the non-cooperative target.The algorithms presented in this dissertation not only have applications in proximity operations, but also show great potential in many other fields such as model reconstruction of buildings, ego-motion of robots, virtual reality, etc.