Vision Simulation And Detection Of Circular Target For On-orbit Service

On-orbit service is a key technology to ensure the continuous and stable operation of space vehicles and had received more and more attention in recent years.In on-orbit service missions,the service spacecraft need to gradually approach and finally capture the target spacecraft.During the approaching,the relative pose measurement of the two spacecraft determines whether the mission will be success or not.The target spacecraft in the mission usually has the characteristics of noncooperative targets,whose structures don't contain visaul target.It is difficult to obtain the relative pose between spacecrafts directly.At present,the main measurement method is to combine the visual imaging features of the natural structure of the target spacecraft and design the algorithm.As the ubiquitous structure of most spacecraft,the ring structure can be used to measure the relative position between spacecrafts because of its visual imaging characteristics.This subject studies the method of generating visual simulation samples of spacecraft and several detection methods of ring targets on the ring structure of a specific spacecraft.In this paper,a spatial three-dimensional model of ring structure of a specific target spacecraft is constructed,and Open Inventor,which is a graphics library,is applied to realize the fast visual simulation of the 3D model under different relative poses and different illumination conditions.According to the characteristics of camera parameters and pose parameters of software simulation system,an automatic calculation method for sample images of ring target parameters has been proposed in this paper.Visual simulation images and tags can be used to test the performance of ring detection algorithms.In the part of detection methods of the ring target,this paper abstracts the ring target detection into the detection of two concentric circles and discusses the time complexity of the circle Hough transform and the random circle detection.Considering the problem of selecting too many candidate circles for random circle detection,two optimization algorithms based on random circle detection are proposed in this paper,which greatly reduces the number of candidate circles.The computational efficiency of two optimization algorithms and circle Hough transform is tested on the simulation sample set.When the appropriate parameter is selected,optimization methods can greatly improve the computational efficiency while ensuring the detection accuracy.This paper also points out the detection limit of optimization algorithms.Considering the rapid development of convolutional neural networks in the field of target detection and segmentation,this paper studies the specific process of calculating the ring target parameters based on convolutional neural networks.This paper proposes the use of semantic segmentation network to approximate the ring target for the visual simulation of the ring structure.At the same time,this paper designs neural networks for edge localization,which realizes the precise positioning of the edge points of the ring target.This paper also proposes a complete process for calculating the ring target parameters by using convolutional neural networks,and and the intersection of union for the detection results of ring targets is higher than 98.7% under different conditions.