Research On Robust Detection And Stable Tracking Method Based On Single Star Target

In the photoelectric tracking system of the star tracker,the star tracker can identify,track and position the individual stellar targets to achieve positioning.It can be widely used in various fields of military and civilian use in aerospace,aviation,nautical and other scenarios,All have high requirements for tracking accuracy,and the lag of the image sensor in the device relative to the control system is one of the main factors that affect the tracking accuracy.With the continuous development,progress and maturity of computer technology,target detection technology and target tracking technology based on computer vision have received more and more attention and attention.At the same time,the research and application of these technologies have become more and more extensive.The main work of this paper is to use the target detection technology based on deep learning and the target tracking technology based on correlation filtering to perform robust detection and stable tracking of single star targets from the perspective of computer vision technology,combined with predictive filtering using adaptive Kalman method,Stable tracking of a single star point in the star tracker's field of view.In this research background,firstly,for the single star target in the field of view,this paper uses the target detection algorithm based on deep learning,to determine the specific position of the star point in the field of view.In order to solve the problem of single star targets occupying a relatively small number of pixels in the field of view and requiring accurate and fast positioning,a new convolutional neural network model structure is designed.Through an improved residual network,it is ensured that the amount of calculation is reduced and the detection speed is improved,Still has good detection performance,the use of multi-scale feature fusion to ensure the robustness of tracking.Then,on the basis of target tracking based on correlation filtering,the joint detection mechanism of tracking confidence and target re-detection are added.In the process of tracking the target,the traditional tracking algorithm based on correlation filtering will lose the tracked target due to various types of interference of the target,such as rapid motion,deformation,leaving and entering the field of view,etc.,which will lead to unstable tracking To reduce tracking accuracy.Therefore,through the maximum response of the tracker to the target and APCE,a joint tracking confidence discrimination mechanism is constructed to continue tracking with high tracking confidence.Otherwise,the target will be relocated in conjunction with the above detector to continue tracking.After experimental analysis,the proposed method shows good tracking stability.Finally,in order to improve the tracking accuracy,in this paper,an adaptive Kalman filtering method for compensating the missed distance is proposed based on the control system.In the actual tracking system,because the image acquisition device lags behind the control system,the tracking accuracy is reduced.In order to achieve stable and accurate tracking control,a CSM model of the moving target is established,and the current state prediction matrix and the prediction error variance matrix are calculated;According to the strong tracking filtering method,the prediction error variance matrix and the maneuver frequency are corrected,and finally,the corrected output information is updated using the corrected parameters.Simulation and experiment prove that this method can significantly reduce the tracking error.This paper is devoted to the research of robust detection and stable tracking methods for single star point targets in star trackers.Focusing on the existing technical problems in the field of computer vision and maneuvering target tracking,corresponding solutions and technical solutions are proposed to enrich the relevant research content of target detection and tracking,merged vision-based detection and tracking technology with maneuvering target tracking control technology,and conducted corresponding simulation and experimental verification.The experiment proves that it has certain theoretical value and practical significance.