Pose Estimation and Simulation of Non-Cooperative Spacecraft Based on Feature Points Detection

• Published in: 2024 IEEE 25th China Conference on System Simulation Technology and its Application (CCSSTA) pp. 12 - 16
• Main Authors: Gao, Xiang, Liao, Ying, Zhou, Huanhuan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 21.07.2024
• Subjects: attitude estimation; Background noise; Convolutional neural networks; Feature extraction; feature point matching; neural network; Neural networks; non-cooperative spacecraft; Pose estimation; Shape; Space vehicles; Systems simulation; Training
• DOI: 10.1109/CCSSTA62096.2024.10691843

This paper presents an innovative framework for pose estimation based on feature points matching. First, a series of spatial points are selected as feature points on the shape of the spacecraft, and a descriptor is generated for each feature point. Then the projection point coordinates of the spatial point in the image are obtained according to different pose labels. In the offline training stage, the projection points and corresponding descriptors are used as labels of the input images for training the neural network. In the running stage, the convolutional neural network takes spacecraft image as input to obtain the descriptors and the position of keypoints in the image. Based on the descriptor, the matching relationship between the space points and the image feature points is established. Then the matching relationship is used to solve the spacecraft attitude using Random Sample Consensus (RANSAC) and Efficient Perspective-n-Point (EPnP). Extensive experiments have shown that this attitude estimation method is highly robust to image background noise.