Real-time Measurement and Estimation of the 3D Geometry and Motion Parameters for Spatially Unknown Moving Targets

Because the prior knowledge of spatially unknown moving targets is not available, it is a challenge to conduct on-orbit services, including tracking, approaching and arresting. This paper draws on the idea of the simultaneous localization and mapping (SLAM) of the traditional mobile robot and proposes a two-threaded algorithm framework combining a front-end tracking algorithm and a back-end optimization algorithm. The front-end tracking algorithm is composed of a bundle adjustment (BA) and an adaptive Kalman filter (AKF) for local optimization. The back-end optimization algorithm is used for the global optimization based on the pose-graph to finally achieve the real-time and accurate measurement and estimation of the 3D geometry and motion parameters for the spatially unknown moving targets. First, the data collected by the camera are pre-processed to remove the system noise and unwanted data points. Second, the data association is completed from the processed data based on the feature point method, including multiple descriptors, in preparation for the subsequent measurement and estimation. Then, in the front-end tracking algorithm, the target's rotation information is estimated by solving the Perspective-n-Point (PnP) problem. Based on this estimation, the target's rotation centre and translation information are obtained by the least squares method (LSM). The target's motion parameters are locally optimized by combining the BA and AKF to achieve the point cloud modelling of the target 3D geometric model. Finally, based on the front-end tracking results, a pose-graph for the back-end optimization is initially constructed. Through loop detection, the pose-graph is improved, and global optimization of the target's motion parameters is achieved based on the pose-graph, which ensures the accuracy and real-time performance of the front-end tracking. The experimental results prove that the proposed method can effectively measure and estimate the 3D geometry and motion parameters of unknown moving targets in real time.