| Advances in science and technology continue to facilitate mankind's exploration of the vastness of space,with spaceflight activities becoming more frequent and more and more satellites being sent into space.With the increase of spacecraft and debris,the on-orbit service technology of non-cooperative spacecraft,which is applied to space debris removal,recovery of failed spacecraft,fuel refuelling,and on-orbit maintenance and upgrade,is receiving more and more attention from researchers in various countries and has become a hot spot for research in the space field.The key to the on-orbit servicing technology is to solve the problem of high precision real-time measurement of the relative position of non-cooperative targets during the close range phase of two spacecraft.In this paper,we study the trinocular stereo visual position measurement technology for non-cooperative targets in space,including the convergent trinocular stereo visual measurement mathematical model,fast identification of non-cooperative target docking rings,significant feature extraction,trinocular cross-polar constraint matching of feature points,stereo visual position measurement,algorithm porting optimization and DSP multi-core parallel processing and other key technologies.For the mathematical model of stereo vision measurement,the imaging mechanism of the vision camera is analysed,and the one-to-one mapping relationship between 2D image coordinates of spatial points and 3D spatial coordinates is derived;the theoretical model of binocular stereo vision,the stereo vision polar line constraint criterion and the trinocular stereo vision measurement model are analysed,and the feasibility of trinocular stereo vision measurement is demonstrated.The relationship between the structural parameters of the measurement model and the measurement accuracy,as well as the relationship between the placement structure of the stereo vision camera and the effective detection field of view,are investigated to provide a theoretical basis for the system configuration and optimal configuration of the convergent trinocular stereo vision model.Through the monocular camera calibration and stereo vision calibration of the trinocular measurement system,as well as the design of the total station external measurement scheme between the camera coordinate system and the world coordinate system,the internal and external parameters of the trinocular measurement system are obtained,laying the foundation for the stereo vision highprecision positional measurement of non-cooperative targets.The improved arc support line segment fitting algorithm and significant feature point detection algorithm for multi-elliptical features are investigated for the significant geometric features(docking ring elliptical features and significant feature points)of the non-cooperative target surface.Firstly,the improved arc support line segment ellipse fitting algorithm is proposed for large size,multi-elliptical feature non-cooperative target images in complex scenes: the image is pre-processed with bilateral filtering algorithm for edge-preserving noise reduction according to the noise characteristics of the image;in the process of ellipse parameter fitting according to the geometric structure of both inner and outer rings of the docking ring and engine nozzle are concentric circles,the arc segment grouping mechanism is optimised to use more elliptical edge arc segments to fit accurate elliptical parameters,which improves the conversion rate from the initial elliptical set to the final elliptical parameters and further improves the elliptical detection efficiency.A perfect ellipse screening mechanism is established using the stereo vision polar line constraint criterion and the docking ring radius constraint to eliminate the interference of invalid ellipse information and achieve the extraction of correct circles.Afterwards,the connected areas are quickly marked by the twice traversal method and area and curvature constraints are added to extract the salient feature points of the target surface.Finally,the docking ring plane and salient feature point data of the non-cooperative target are reconstructed using stereo vision principles to establish the target coordinate system and calculate the relative positional relationship between the target and the visual measurement system.The problem that the docking ring and salient feature points of a non-cooperative target are not clearly and stably imaged in the unstable state of the target is considered as a non-cooperative target with unknown geometry.The ORB algorithm based on the combination of FAST operator and Harris operator is used to quickly detect the feature point information of the trinocular image,and the method of using trinocular crosspolar line constraint is proposed to match the same-name feature points of the trinocular.Through the additional polar line constraints brought by the third camera,the uniqueness of the geometric position of the target feature point on the trinocular image is ensured,effectively avoiding the problem of mis-matching feature point localization brought about by image similarity matching in complex scenes.This results in denser matching point data and higher matching accuracy,which improves the applicability and robustness of the positional measurement method.In view of the current situation that the on-orbit service visual measurement method is difficult to be used for autonomous real-time attitude measurement on board,an embedded platform for trinocular stereo visual attitude measurement is built.In order to improve the computational efficiency of the embedded platform's attitude measurement algorithm,a multi-core processing method based on the data flow model is designed,which completes the process of acquiring trinocular images from the camera,performing key feature extraction of the non-cooperative target in the image processing unit,and then performing key feature extraction to the attitude processing unit.The data flow processing model is completed from the acquisition of trinocular images by the camera,to the extraction of key features by the image processing unit,and then to the stereo visual pose solution by the pose processing unit.The output update rate of the final pose measurement data is 1Hz.In order to validate the effectiveness of the proposed method,the non-cooperative target motion simulation experiments from 15.0m to 1.8m and the non-cooperative target ground test experiments were completed.The results of the target motion simulation experiments show that the trinocular vision measurement system incorporates three sets of binocular vision measurements in two combinations,and has higher accuracy and safety reliability in positional resolution than the single binocular vision measurement system.In addition,the ground test results show that the embedded platform of trinocular stereo vision position measurement in complex scenes has a test range of 12.0m-1.7m,and the overall relative position measurement accuracy is better than 1.3cm;the relative attitude measurement accuracy is better than 1.7°.This paper experimentally verifies that the method has high positional solution accuracy and high robustness,and can provide real-time continuous high-precision positional measurement results of non-cooperative targets. |
PDF Full Text Request