Research On Non-cooperative Target Reconstruction And Posture Measurement Based On Active And Passive Fusion

As an important perception mode for spacecraft or space robots,visual measure-ment has experienced the development of teleoperation monitoring,cooperative and semi-cooperative target measurement.As the supporting technology of the next gen-eration space mission,the requirement of the measurement system in robustness,ap-plicability and computational ability is obviously higher than that of the existing mea-surement technology.Therefore,guided by the requirement of on-orbit service task,this topic has carried out in-depth research on the related technology of visual measure-ment,and realized the non-cooperative target measurement method based on the fusion of 1D laser ranging and monocular vision.The main work is as follows.Firstly,the mathematic model of the measurement system is established,and a non-iterative calibration method of camera internal parameters and a fast and robust calibration method of sensor external parameters are proposed.The two methods to-gether achieve the fast and robust calibration of the parameters required by the mea-surement system model.Compared with the existing classical methods,the efficiency of our proposed method is improved by about 20+ times under the same calibration accuracy.Secondly,we have improved the related technology,and proposed an improved robust estimation method of fundamental matrix and a non-iterative pose estimation method based on Grbner basis solution.The proposed basic matrix estimation method suppresses the interference of mismatched points and improves the success rate of ini-tialization.At the same time,the improvement of pose estimation algorithm ensures that the system can measure steadily with fewer tracking points,and overcomes the problem of non-cooperative targets lacking significant features to a certain extent.The above improvements effectively improve the robustness and measurement accuracy of the system.In addition,for the key problem of laser-vision fusion,we creatively put forward the fusion scheme of 1D laser range information and monocular vision information in general scene,completed the absolute scale estimation,and realized the reconstruction of target in Euclidean space and pose measurement.We also introduce the concept of key frame cluster,which effectively improves the measurement accuracy and compu-tational eff-iciency of the system.The validity of data fusion and the accuracy of the estimated absolute scale are verified by experiments,and the validity of the proposed method is proved.Finally,aiming at the problem of error accumulation in long-time measurement,we propose a solution to real-time monitoring system error based on fused scale information and restraining scale drift by adjusting local results when necessary.Experiments show that this method can effectively guarantee the accuracy and stability of the system in the long-term measurement process.At the same time,a global pose map optimization problem based on loop is constructed,which guarantees the accuracy and applicability of the system on a large scale.On the basis of the above research results,we analyze the characteristics of the pro-posed method through simulation experiments,high-precision turntable experiments,open data set experiments,mobile platform experiments and other experimental scenar-ios.The accuracy and applicability of the proposed method are evaluated from different dimensions such as target characteristics,motion states and environmental conditions.The feasibility of non-cooperative target reconstruction and pose measurement method based on laser-vision fusion is also verified.