Measurement And Estimation Of Three-Dimensional Geometry And Motion Parameters Of Spatially Unknown Moving Targets

As human activities for the universe increase,the technical requirements for space-based services are increasing.Due to the fact that human activities in space exploration in the past failed to fully consider the impact on the space environment,there is now a large amount of space debris near the Earth’s orbit,which seriously affects the development of space exploration.Therefore,the management of the space environment and the cleanup of space junk are urgently needed.Most space junk are unknown moving targets,and their unknowns and uncertainties pose great challenges to the management of the space environment.To remove space debris,it is necessary to accurately identify the target motion state and three-dimensional structure information in advance.In the existing literature,the vision-based measurement method can effectively solve the problem,which is a hot topic in current research.This thesis uses the ToF camera Kinect V2 as the data acquisition sensor.The depth imaging principle of the Kinect V2 camera is introduced,and then the Kinect V2 camera is calibrated to correct the systematic error caused by the distortion.Based on the calibration,the error analysis and correction of depth measurement for the Kinect V2 camera are performed to avoid the subsequent measurement and estimation result of unknown moving targets with excessive system error.For excessive noise in the collected data,pre-processing of the measured data is performed to eliminate the noise.The target is segmented from the collected data to achieve frame selection tracking of the target.The existing feature extraction algorithms are compared by experiments.The characteristics of each algorithm are analyzed to determine the appropriate feature extraction algorithm for data association.Aiming at the problem that the measurement and estimation is easily interrupted for the data association between two frames,a map point structure method with double descriptors is proposed for data association,which prepares for the measurement and estimation of 3D geometry and motion parameters of unknown moving targets.By analyzing the difference between the traditional SLAM problem and the unknown motion target state estimation problem,and comparing the advantages and disadvantages of the existing filter method and the graph optimization method,a two-threaded algorithm framework combining front-end tracking algorithm and back-end optimization algorithm is proposed to measure and estimate the target motion state and 3D geometry.The front-end tracking algorithm decouples the estimation of target rotation information and translational information,and combines bundle adjustment(BA)and adaptive Kalman filter(AKF)for local optimization.The back-end optimization algorithm performs global optimization based on the pose graph to eliminate the cumulative error of the target motion.The word-of-bag model is created based on the target collection data,and then the loop detection is completed to improve the pose graph.The physical experiment platform and the software platform are built to verify the proposed algorithm.The experimental setting is that the target is fixed at the end of the UR arm,and the Kinect V2 camera is fixed at a position outside the arm.Through the hand-eye calibration,the pose relationship between the target and the camera is determined,and the ground-truth information is provided for the analysis of the experimental results.Experiments on three different motion forms of the target verify the validity and accuracy of the proposed measurement and estimation algorithm.