The Study Of3D Pose Estimation For Space Obiect Based On Images

With the improvement of ground-based optical observation system performance, the features of space objects, such as appearance and pose, can provide more critical information to the space situational awareness tasks, which own great values to the applications of spatial object detection and identification.3D pose estimation of space object is an important method to realize the space object monitoring and tracking. This thesis integrated space object characteristics analysis, the database structure, model indexing as well as a variety of image processing methods, proposed a complete3D space object pose estimation scheme based on ground-based telescope images. The scheme utilized image deconvolution method for the restoration of observed images; the space object was then divided into a main component and sub-components, from which relevant geometric features were extracted, then a multi-angle pose model database was created, a feature indexing method was proposed to encode geometric characteristics and similarity matching was performed in the database. Finally the3D pose of space object was estimated.The thesis mainly researched theories and methods of3D pose estimation of space object based on images, contents and innovations are as follows:1. Ground-based telescope imaging simulation was performed and post processing technology was developed. The factors affecting ground optical imaging system were analyzed. Simulation space object image was obtained using proper atmospheric coherence length parameter and noise level, and then recovered by the iterative RL algorithm, which can overcome negative influence of the observation conditions and the post-processing image suitable for target identification and segmentation was obtained.2. Object recognition and image segmentation were studied and realized. Different image segmentation methods were utilized to split space object and background as well as the main and sub components. In addition, opening and closing operations of mathematical morphology is processed. Finally an ideal space object body and sub-components were successfully segmented.3. The multi-view pose model database was constructed and feature indexing was designed. The observation ball was divided by hexagonal grid to get an evenly distributed2D view set of the space object pose. An effective and efficient indexing method was designed to change the2D view of the models as translation invariant, scale invariant and rotation invariant feature vectors.4. The geometric feature extraction and similarity matching were performed and the pose of spatial object was estimated. After the image segmentation, the feature extraction was performed to obtain geometrical parameters of main component and sub-components to generate a feature vector index. By similar matching between the space object vector and model index of the database, the3D pose of space object can be estimated.The3D space object pose estimation method proposed by this thesis owns a complete system structure and related experiment proved its validation and accuracy. The research results can provide technical support to the corresponding data processing software system.