Fixed star target identification and positioning method under star map background

By combining star map recognition and Kalman filtering technology, the problem of insufficient space debris monitoring accuracy in the existing technology is solved, high-precision monitoring and positioning of space debris is achieved, and the monitoring capabilities of the space-based platform are improved.

CN118936458APending Publication Date: 2024-11-12XIAN INST OF OPTICS & PRECISION MECHANICS CHINESE ACAD OF SCI

Patent Information

Application Number
CN202411001274.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing technology is unable to effectively monitor and locate space debris, especially weak targets, which leads to insufficient monitoring accuracy of space-based platforms, affecting the stability of the space environment and the safety of aerospace activities.

Method used

Using a combination of star map recognition and Kalman filtering, the camera posture is constructed through the all-sky star map autonomous recognition algorithm and QUEST algorithm, distinguishing stars from moving targets, obtaining Ascension Declination Data, and using Kalman filtering for position prediction and window search to achieve continuous tracking of space debris.

Benefits of technology

It realizes high-precision monitoring and positioning of targets in the entire celestial range, effectively distinguishes between star and non-stair targets, and improves the monitoring accuracy and real-time performance of the space-based platform.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118936458A_ABST
    Figure CN118936458A_ABST
Patent Text Reader

Abstract

The invention provides a fixed star target identification and positioning method under a star map background. The method comprises the following steps: identifying a part of fixed stars in a picture shot by a camera based on an all-day star map autonomous identification algorithm; the method comprises the following steps: constructing an initial camera attitude based on a QUEST algorithm, and obtaining initial camera attitude information; identifying a local sky area shot star map based on the initial camera attitude information, and distinguishing a fixed star and a moving target in the shot image; obtaining right ascension and declination data of the detection target based on the initial attitude of the camera, and constructing an initial state of the detection target; and based on the initial state of the detection target, performing position prediction on the detection target by adopting Kalman filtering, and completing continuous tracking on the detection target by using a window search mode. According to the method, a technical strategy of combining star map recognition and Kalman filtering is adopted, so that target monitoring in a whole celestial sphere range and effective distinguishing between a fixed star target and a non-fixed star target are realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of star sensors, and in particular to a method for identifying and locating star targets under a star map background. Background Art

[0002] Space-based platform space debris monitoring is an important technology in the field of space monitoring. Its main purpose is to detect, track, identify and confirm space debris. Space-based platforms mainly use space-based remote sensing detection, space-based collision perception detection and spacecraft surface sampling and analysis to measure and monitor space debris. It is of great significance to ensure the safety of space activities and maintain the stability of the space environment.

[0003] There are two major problems in the space debris monitoring mission of space-based platforms: (1) It is impossible to detect space point targets using conventional space point target detection methods (image difference method, trajectory detection method and matching correlation method, etc.), which affects the effective and real-time monitoring performance of space-based monitoring platforms for small and weak space targets; (2) The monitoring mission requires the precise positioning of the detected space debris, but the positioning accuracy of existing space-based monitoring platforms is limited, which affects the high-precision measurement capability of the position of small and weak space targets. Summary of the invention

[0004] In view of this, the present invention proposes a method for identifying and locating star targets under a star map background to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention proposes a method for identifying and locating star targets under a star map background, comprising:

[0006] Identify some stars in the pictures taken by the camera based on the autonomous recognition algorithm of the all-sky star map;

[0007] Construct the initial camera posture based on the QUEST algorithm to obtain the initial camera posture information;

[0008] Recognize the local sky area star map based on the initial camera posture information, and distinguish the stars and moving targets in the captured picture;

[0009] Acquiring right ascension and declination data of the detection target based on the initial attitude of the camera, and constructing an initial state of the detection target;

[0010] Based on the initial state of the detection target, the Kalman filter is used to predict the position of the detection target, and the window search method is used to complete the continuous tracking of the detection target.

[0011] Optionally, the process of identifying some stars in the picture taken by the camera based on the autonomous recognition algorithm of the all-sky star map includes:

[0012] The shooting camera posture is determined based on the all-sky star map recognition technology, the sky area position pointed by the field of view axis is determined based on the shooting camera posture, all observation targets in the camera field of view are obtained, and a number of matching star pairs are output.

[0013] Optionally, the process of establishing the initial camera posture based on the QUEST algorithm includes:

[0014] Based on the matching star pair, an observation star direction vector in a star sensor body coordinate system and a corresponding navigation star direction vector in a celestial coordinate system are obtained, a Wabba problem is defined, and the Wabba problem is solved based on a q-method and a QUEST method to obtain an optimal attitude estimation, and the initial camera attitude information is obtained based on the optimal attitude and converted into a pitch angle, a yaw angle, and a roll angle.

[0015] Optionally, the Wabba problem is as follows:

[0016]

[0017] Among them, J is the objective function, a k is the weight coefficient. Assuming that all the weight coefficients are consistent, if the vectors in the sets w and v are accurate data, the corresponding minimum result of the loss function is 0; if there is a deviation between the vectors in the sets w and v, then the corresponding loss function result is greater than zero. At the same time, the larger the corresponding value of the loss function, the greater the deviation between the algorithm's estimated value R of the posture matrix and the ideal value A.

[0018] Optionally, the process of solving the Wabba problem based on the q-method and the QUEST method respectively includes:

[0019] Characterizing the objective function of the Wabba problem in the form of quaternions;

[0020] Based on the Lagrange multiplier, the objective function is maximized and a derivative operation is performed;

[0021] The matrix eigenvector in the derivation operation process is solved based on the QUEST method to obtain the optimal posture estimation.

[0022] Optionally, the process of solving the matrix eigenvector in the derivation operation process based on the QUEST method includes:

[0023] The matrix eigenvectors are represented as Rodriguez parameters, and the solution of the Rodriguez parameters is converted into a linear equation system, and the linear equation system is solved based on Gaussian elimination method.

[0024] Optionally, the process of identifying a star map shot in a local sky area based on the initial camera posture information includes:

[0025] Based on the initial camera attitude information, a reference star map is generated in combination with a navigation database;

[0026] Sort all the star points in the photographed star map and the reference star map according to the size of the coordinates;

[0027] Based on the nearest neighbor matching principle, the star points in the reference star map are divided into star targets and non-star targets;

[0028] Set the limiting magnitude based on the magnitude data error of the shooting camera and select a reference star.

[0029] Optionally, the process of using Kalman filtering to predict the position of the detected target includes:

[0030] Based on the initial state of the detected target, the target prior state estimation and prior state error covariance matrix prediction are performed, and the right ascension and declination states of the detected target in the k+1 frame are estimated a priori based on the optimal position estimation result, and the position coordinates of the CCD plane corresponding to the shooting camera where the right ascension and declination of the target in the k+1 frame are located are calculated according to the attitude matrix. This output is used as the window center position of the subsequent window search part;

[0031] Optionally, the process of completing continuous tracking of the detection target using the window search method includes:

[0032] The position coordinates of the CCD plane of the camera corresponding to the target right ascension and declination in the k+1th frame are taken as the window center position, and the window center radius is obtained;

[0033] Using the window center radius as the search radius, the detection target is identified and located, and the observation value of the target in the k+1th frame is obtained;

[0034] The optimal position estimation of the detected target in the k+1th frame and the update of the error covariance matrix are performed based on the observed values.

[0035] Optionally, the window center radius is defined as follows:

[0036] R w =4×E p +4 of which R w is the window center radius,

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] Aiming at the monitoring task of space debris targets of space-based platforms, the present invention adopts a technical strategy of combining star map recognition with Kalman filtering to realize target monitoring within the entire celestial sphere; adopts a reference star map nearest neighbor matching method to realize the recognition of observed star points in the captured star map, and after all star targets in the captured star map are identified, the remaining unidentified detection targets are identified as non-star targets, thereby completing the effective distinction between star targets and non-star targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0040] Figure 1 It is a schematic diagram of the flow of the method for identifying and locating star targets under the background of a star map of the present invention;

[0041] Figure 2 A schematic diagram of target identification and target positioning in an embodiment of the present invention;

[0042] Figure 3 A schematic diagram for describing the Wabba problem in an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of the target recognition algorithm framework in an embodiment of the present invention;

[0044] Figure 5 Schematic diagram of an observed star and a reference neighboring star in an embodiment of the present invention;

[0045] Figure 6 It is a schematic diagram of element sorting in an embodiment of the present invention;

[0046] Figure 7 A schematic diagram of element search and matching in an embodiment of the present invention;

[0047] Figure 8 is an ideal star map corresponding to a shooting camera in a certain posture in an embodiment of the present invention, wherein (a) is an ideal star map without brightness noise, and (b) is an ideal star map with brightness noise;

[0048] Fig. 9 Graph showing the influence of position noise on recognition rate in an embodiment of the present invention;

[0049] Fig.10 is an algorithm flow chart of a star target identification and positioning method in an embodiment of the present invention;

[0050] Fig.11Schematic diagram of the whole celestial sphere autonomous star map algorithm in an embodiment of the present invention;

[0051] Fig.12 Schematic diagram of a local sky area star map algorithm in an embodiment of the present invention;

[0052] Fig.13 This is a target recognition result diagram in an embodiment of the present invention;

[0053] Fig.14 A schematic diagram of Kalman predicted position in an embodiment of the present invention;

[0054] Fig.15 Schematic diagram of window search in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to be able to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0056] In order to monitor space debris targets on space-based platforms, this embodiment adopts a technical strategy combining star map recognition and Kalman filtering to achieve target monitoring of space debris in the entire celestial sphere. The algorithm flow is as follows: Figure 1 As shown in Figure 1, the algorithm mainly includes the following parts, namely, full-sky star map recognition, target recognition based on local sky star map recognition, target search and positioning based on Kalman filtering, etc. Here we mainly explain the target recognition and target positioning steps. The basic idea is as follows Figure 2 shown.

[0057] In the target recognition module, this embodiment adopts a target recognition method based on local sky star map matching, determines the shooting camera posture through the full celestial star map recognition technology, and then determines the celestial position of the field of view axis, and then uses the local sky star map recognition algorithm to match and identify all the observation targets in the field of view, so as to complete the distinction between star targets and non-star targets, realize the target recognition task in the star background of the observation star map, and determine the right ascension and declination position information of the target based on the coordinate conversion relationship; in terms of target tracking, this embodiment adopts a target tracking algorithm based on Kalman filtering, first estimates the target state, and then predicts the approximate position of the imaging plane where the target is located in the next frame, uses a window scanning method to determine the corresponding actual observation data, and uses the Kalman filter method to optimize its position data, so as to complete the rapid recognition and tracking of the target. The detailed steps of the recognition and tracking stage are as follows:

[0058] Identification Phase

[0059] Step 1: Use the autonomous recognition algorithm of the all-sky star map to complete the recognition task of some stars in the captured image, and based on this, use the QUEST algorithm to complete the establishment of the initial attitude of the camera;

[0060] Step 2: Based on the initial attitude information, the algorithm switches to the recognition of the star map taken in the local sky area, and completes the distinction between stars and moving targets in the captured pictures, completing the recognition task of high-orbit weak point targets;

[0061] Figure 2 In the figure: “○” indicates the observation target that the algorithm determines to be a star, and “□” indicates the observation target that the algorithm determines to be a detection target.

[0062] Tracking Phase

[0063] Step 1: Based on the current camera attitude, complete the calculation of the right ascension and declination data of the detected target and establish the initial state of the moving target;

[0064] Step 2: Based on the initial state, use Kalman filtering to complete the prediction, and use the window search method to complete the continuous tracking of the weak target;

[0065] In the figure, the dotted line “□” represents the predicted position of the detected target by the Kalman filter algorithm.

[0066] The above is the basic idea of ​​the algorithm of this embodiment. The target recognition method based on local sky area star map matching and the target tracking method based on Kalman filtering in the two stages of the algorithm are described in detail later.

[0067] This embodiment first completes the posture information solution of the shooting camera, and then completes the matching of all star targets in the local sky area under the premise that the posture information is known. The specific contents related to the algorithm will be described in detail below.

[0068] Attitude calculation: After the full celestial star map recognition is completed, the algorithm outputs at least 3 pairs of matching star pairs, and uses them to obtain the direction vector of the observation star in the star sensor body coordinate system and the corresponding navigation star direction vector in the celestial coordinate system. Combined with the relevant information obtained, the attitude data of the star sensor is calculated (the rotation matrix of the star sensor body coordinate system relative to the celestial coordinate system), and converted into pitch angle, yaw angle, and roll angle. This section first introduces the wabba problem ( Figure 3 ), and explains the mathematical model of posture solution, and then introduces the QUEST algorithm used in this embodiment to achieve the initial posture solution task.

[0069] Wabba problem: After completing the sub-pixel positioning of the star points in the captured star map and the recognition of the all-sky star map, assuming that N star points are successfully identified, where N ≥ 3, then their corresponding vector set W in the star sensor body coordinate system can be expressed as w = [w1, w2, w3, ..., w N ], and the vector set V of the observed star corresponding to the navigation star in the celestial coordinate system can be expressed as v = [v1, v2, v3, ...v N ], the relationship between the two is shown below:

[0070] w=Av (1)

[0071] Where: A represents the posture transformation matrix.

[0072] When there are two non-parallel observation vectors in space, the vector-based attitude solution model shows that this information can be used to uniquely determine the rotation matrix of the object, that is, the corresponding three-axis position data of the star sensor. When the number of successfully matched star points in the star map recognition link is greater than 3, there are multiple pairs of observation star vectors and navigation star vectors involved in the attitude solution. At this time, there is no rotation matrix that makes all N equations in formula (2) valid. At this time, we need to find a solution to satisfy the statistical error of the N observations. Based on the above discussion, formula 2 can be used to define the objective function of the mathematical model corresponding to this problem, that is, to find a matrix R in the solution space to ensure that the objective function can obtain the minimum value, which is the wabba problem.

[0073]

[0074] Where: J is the objective function, a kis the weight coefficient, and it is assumed that all weight coefficients are consistent. Theoretically, if the vectors in the sets w and v are accurate data, then the minimum result of the corresponding loss function is 0; if there is a deviation between the two, then the corresponding loss function result is greater than zero. At the same time, the larger the corresponding value of the loss function, the greater the deviation between the estimated value R of the attitude matrix and the ideal value A of the algorithm. There are three main solutions to the wabba problem: q-method, Newton iteration method and QUEST method.

[0075] Select the pose solving method:

[0076] For embedded systems, the numerical calculation method based on Newton iteration requires more iterative operations in the actual solution process. The whole process is relatively complicated, so it is rarely used in practice. In this section, only the q-method and QUEST method are explained, and the corresponding attitude solution method is selected based on this.

[0077] q-method:

[0078] Expanding the corresponding expression of the loss function J gives the following formula:

[0079]

[0080] From the analysis, we can see that the equivalent problem of solving the problem of minimizing the loss function is to solve The objective function maximization problem. If the objective function J2(q) is represented in the form of quaternion, the form conversion can be performed as shown below

[0081] J2(q)=q T Kq (4)

[0082] Among them: K is a square matrix of dimension 4×4, and the expressions of its corresponding specific elements are as follows:

[0083]

[0084] σ=trace(B)

[0085] Z=[B 23 -B 32 B 31 -B 13 B 12 -B 21 ] T

[0086] Because the quaternion expression has constraints, it will be limited. Therefore, the objective function J2(q) can be maximized by introducing the Lagrange multiplier method. Therefore, the above mathematical problem can be transformed into the following expression:

[0087] J2(q)=q T Kq-λq T q (5)

[0088] The above formula is used to differentiate the variable q, and the result of the corresponding expression is set to 0, then

[0089] Kq=λq (6)

[0090] From the analysis, we can see that the optimal posture is the eigenvector of matrix K. Substituting it into formula (4) and formula (6), we get

[0091] J2(q)=q T Kq=q T λq=λq T q=λ (7)

[0092] Then the eigenvalue result of matrix k is the value of the objective function, so the eigenvector corresponding to the maximum value of the eigenvalue of matrix k is the best estimate of q.

[0093] QUEST method:

[0094] For the wabba problem, when choosing to apply the q-method to solve it, it is necessary to obtain the reference direction vector of the corresponding direction in the celestial coordinate system that matches the observation vector corresponding to the observed star in the body coordinate system of the star sensor. The q method takes minimizing the mean square error as the objective function and gives the best attitude estimation in the corresponding sense. However, it should be noted that when solving the problem through this method, it involves the operation of solving the matrix eigenvectors and eigenvalues, which requires more embedded calculations, but in actual application, the requirements for system real-time performance are usually relatively high. Based on the basic framework of the q-method, the steps of solving the eigenvector by applying the method QUEST are further optimized to reduce the complexity of the operation, thereby meeting the high requirements for real-time performance of the system. The corresponding contents of the algorithm include:

[0095] According to formula (4)(5)(6), we can know

[0096]

[0097] Since the value corresponding to the loss function J will tend to zero, the optimal estimate of the maximum eigenvalue λ opt It can be approximately written as the following formula:

[0098]

[0099] But the approximate estimate of λ opt The numerical accuracy is sufficient to meet the needs of actual engineering. Therefore, after the best eigenvalue is estimated, the corresponding best eigenvector can be calculated. The result is the best posture for estimation. The specific steps for solving the eigenvector are as follows:

[0100] The eigenvector problem is expressed as

[0101] p=[(λ opt +σ)IS] -1 Z (10)

[0102] where p is defined as the Rodriguez parameter, which has the form

[0103]

[0104] The definitions of the symbols in the formula refer to formula (4). From this analysis, we can see that the eigenvector value can be obtained by solving the inverse of the matrix, but the whole process requires a lot of calculations. Therefore, we write formula 10 as formula 12, and then use Gaussian elimination method to solve this linear equation system

[0105] [(λ opt +σ)IS]p=Z (12)

[0106] After the calculation of the eigenvector is completed, the solution of the attitude quaternion can be completed (Formula 13), that is, the optimal attitude estimation is completed.

[0107]

[0108] Target recognition method

[0109] The algorithm flow of this embodiment is as follows Figure 4 As shown, this section mainly explains the key reference star map generation, fast search and matching, and target recognition technology.

[0110] Reference star map generation: The digital star map simulation algorithm process mainly includes coordinate transformation and grayscale transformation. In the process of generating the reference star map of the local sky area in this embodiment, there is no need to perform the grayscale transformation step. Only the position coordinates of the star points in the reference star map on the imaging plane need to be used. The detailed process is as follows:

[0111] After the recognition of the full celestial star map is completed, the attitude matrix M value is determined. If the right ascension and declination data of a star in the navigation star database in the celestial coordinate system are (α i ,δ i ), if the magnitude of the star is lower than the limiting magnitude of the camera, and αi , δ i If the requirements in the following formula are met, the corresponding star can be imaged in the detector plane.

[0112]

[0113] In the formula, if the camera has a circular field of view, then R is the field of view radius (the corresponding value of R is half of the FOV diagonal), and (α0, δ0) represents the right ascension and declination information corresponding to the camera's visual axis. Then the direction vector of the star in the star sensor coordinate system is:

[0114]

[0115] Where: [X i , Y i , Z i ] is the direction vector of the star in the celestial coordinate system, is the direction vector of the star in the star sensor coordinate system, and M is the attitude matrix.

[0116] In addition to the above differences from the digital star map simulation in the simulation stage, the generation of the reference star map is required to meet a certain real-time performance in the local sky area star map recognition process. If we use the judgment standard of formula 14 to traverse all the star points in the navigation database, due to the large number of stars, reaching 534484, the loop operation takes a long time and it is difficult to meet the application requirements. Therefore, we need to design a fast search method to achieve fast search of star data in a certain area in the direction of the field of view. At present, a commonly used idea is to divide the sky area into several sub-blocks, such as the declination belt method, the cone method, the spherical rectangle method, the inscribed square table division, etc. The above methods have improved the efficiency of the sky area search to a certain extent, but with the increase of the number of limiting magnitudes, the retrieved sub-blocks still contain a large amount of redundant star information, which affects the retrieval efficiency. With the development of on-chip storage technology, capacity indicators are no longer the main factor restricting algorithm design. Therefore, this embodiment uses local neighbor star storage to complete the search process.

[0117] Fast matching search algorithm

[0118] After the reference star map is generated, assume that the ideal position of the star in the reference star map numbered #index is marked as (x c ,y c ), the corresponding observed star coordinates in the captured star map are (x im ,y im ), defining the distance d between the two on the detector plane mc Formula 16. If d mcIf the value is less than a certain threshold ε, the observed star is judged to be the #index navigation star with the number of the neighbor ( Figure 5 ).

[0119]

[0120] Here we need to note that due to the close positions of some navigation stars in the celestial coordinate system, coupled with position noise interference, multiple reference stars appear in the neighborhood of the corresponding observation star, such as Figure 6 As shown in the figure, it is impossible to determine whether the observed star corresponds to a specific reference star, but for target recognition applications, we only need to pay attention to whether the imaging target is a star target. Therefore, when there are multiple reference stars in a certain neighborhood of the observed star, it is still considered to be a star target. According to the above-mentioned nearest neighbor matching principle, the discrimination between star targets and non-star targets is completed.

[0121] (2) Sorting and matching

[0122] For the camera in this embodiment, since the limit detection magnitude is +10.5Mv, there are many stars in the field of view. Therefore, if the solution is performed one by one in a traversal manner, the calculation complexity is high. Therefore, this embodiment adopts a fast matching search algorithm to complete the above matching task. The basic idea is as follows:

[0123] Because the reference and observed star maps match, the distance between stars that are far apart does not need to be calculated by formula. Based on the above ideas, we first sort the coordinates and then match and identify them. That is, before identifying the local sky map, all the star points in the observed star map and the reference star map are sorted from small to large according to their corresponding X coordinates ( Figure 6 ), and then matching and identification are performed based on the nearest neighbor principle.

[0124] Sequence A and sequence B represent the sorted reference star map and observed star map respectively. This can be regarded as the matching process of sequence A and sequence B. Taking the matching of star No. 2 in observed sequence A and the stars contained in reference sequence B as an example, the specific matching and identification method is as follows: Figure 7 shown.

[0125] Step 1: To reduce the multiplication and division operations in the distance calculation and improve the calculation speed of the algorithm in the embedded platform, we avoid directly using formula (17) to calculate the distance d between the two star points. Instead, we use subtraction operations to calculate the coordinate difference dx (X coordinate difference) and dy (y coordinate difference) between the two. The formula is as follows:

[0126]

[0127] Step 2: The calculation of the No. 7 star in the A sequence and the star in the B sequence starts with the No. 9' star. In the comparison process between the No. 7 star in the A sequence and the star in the B sequence, the No. 9' star is the first star point in the B sequence that satisfies dx<r. Analysis shows that if the dx value of the No. 9 star and the No. 1' and No. 2' stars is greater than r, since the sequence is a queue comparison after the increase order, the dx value of the No. 7 star after the No. 9 star and the No. 1' and No. 2' stars must be greater than r, so there is no need to compare the No. 7 star with them.

[0128] Step 3: During the alignment and matching process of sequences A and B, if the dx value between star No. 7 and a star point in sequence B is less than r, the comparison will continue with the star point in sequence B at the next position. For example, star No. 7 in A will be compared with stars No. 9', 4', 7', and 8' respectively.

[0129] Step 4: When the dx calculation and comparison between star No. 7 and star No. 5' is performed, it is found that the dx value is greater than r, then the matching process between star No. 7 and the B sequence star is terminated. Since the A and B sequences are compared in increasing order, the dx values ​​between star No. 7 and the 3', 11', and 10' stars following star No. 5' must be greater than r, so it can be skipped directly.

[0130] Step 5: When the No. 7 star is matched with the B-sequence star, the dx and dy values ​​of the No. 7' star meet the requirement of being less than r, so there is a unique reference navigation star within the neighborhood radius r of the No. 7 star, so the No. 7 navigation star in the A sequence and the No. 7' observation star in the B sequence match each other. By analogy, the matching relationship between the navigation star and the observation star can also be obtained.

[0131] Step 6: Similarly, the 1st, 5th, 6th, 8th and 10th stars in the A sequence are compared with the elements in the B sequence respectively, thereby completing the matching and identification of the observed star map and the reference star map.

[0132] From the above analysis, it can be seen that the search and matching method of sorting first and then comparing can effectively reduce the number of comparisons between sequences, thereby improving the running speed of the algorithm.

[0133] (3) Reference star selection

[0134] Due to the influence of many factors such as stellar brightness changes, spectral characteristics, and detection systems, the current shooting camera, i.e., the star sensor camera, has some constraints on the accuracy of measuring the star brightness data. That is, the star target detected by the shooting camera is different from the magnitude of the corresponding star in the Ty2 star catalog, that is, there is a certain error, which leads to some dark stars in the star catalog with brightness lower than the detection limit magnitude appearing in the observed star map, while some stars in the star catalog with brightness higher than the detection limit magnitude do not appear in the photographed star map.

[0135] like Figure 8(a) is the ideal star map corresponding to the camera in a certain posture. The star map after adding brightness noise in the same posture is shown in Figure (b). Through analysis, it can be found that if there is no noise interference, Figure 8 In (a), stars a, b, and c that are brighter than the camera's limiting magnitude appear in the star map. However, when noise appears, the bright stars a, b, and c that should have appeared in the captured star map are Figure 8 (b) are all gone, that is, they become missing stars, and for Figure 8 The dark stars d and e that do not exist in (a) appear Figure 8 (b), that is, adding star points. For the target recognition process, the existence of missing stars will not affect it, but adding stars may cause some troubles to the target recognition process. Since the information of the added stars has not been stored in the navigation database, the algorithm will identify it as a non-star target in the local sky area star map recognition process, and then mistakenly determine the added star as a target. If the limiting magnitude of the shooting camera is +10.5Mv, and the star with a magnitude brightness record of +11.0Mv in a certain star catalog may appear in the shooting star map due to brightness noise. At this time, if the +10.5Mv star catalog is used to generate a reference star map, the star cannot be identified and is mistakenly determined by the algorithm as a non-star target. To avoid the above problems, the limiting magnitude corresponding to the star catalog database generated by the reference star map should be higher than the camera limiting magnitude. Generally speaking, since the error of the limiting magnitude data accurately calibrated by the shooting camera is generally within ±0.3Mv, the deviation of ±0.5Mv we use here is sufficient to meet the needs, so this embodiment uses +11.0Mv as the limiting magnitude when generating the reference star map.

[0136] Algorithm Testing

[0137] The recognition success rate of the target recognition algorithm based on reference star map matching is tested to evaluate the performance of the algorithm. In the simulation experiment, there are interference factors that affect the recognition rate of the algorithm, including brightness and star point position noise. In this embodiment, the Monte Carlo method is used to perform 1000 random tests during the test process, and the position of the detection target is randomly added to the randomly generated star map, and the corresponding star map recognition success rate is counted. In this experiment, a star map containing the detection target is defined to be successfully recognized by the algorithm when the following conditions are met:

[0138] (1) The star target was not mistakenly identified as the detection target; (2) The detection target was identified by the algorithm. The detection camera parameters are as follows: FOV is a 2° circular field of view, the coordinates of the center position of the imaging plane are (512,512), the detector resolution is 1024×1024, the pixel size is 13μm×13μm, the bit depth is 16bit, the focal length of the optical system is 381mm, and the detection limit magnitude is +10.5Mv. MATLAB programming was used in the experiment, and the PC was configured with Windows7 operating system, 4GB memory, and 3.4Ghz Intel 7 processor.

[0139] like Fig. 9 As shown in Figure 1, the target recognition success rate change curve of the proposed algorithm is when the standard deviation of the star point position noise increases from 0 to 2.0 pixel at the beginning and the brightness noise is 0.3Mv. Fig. 9 It can be seen that this method is highly robust to position noise. When the standard deviation of position noise increases from 0 pixel to 2.0 pixel, the recognition rate of the algorithm remains relatively stable, and has been maintained at more than 97.6%. The reason for the failure of target recognition is that the background stars in some sky areas are too dense, and the distance between the simulated detection target and the star target is too close, making it difficult to distinguish them through the neighboring position relationship between the reference star and the observed star, but the algorithm recognition rate remains at a high level, and the recognition result is relatively reliable.

[0140] The local star map recognition technology can effectively complete the recognition in each frame of the captured image and mark it in the form of a rectangular frame. However, the recognition calculation amount of each frame of the image is large and cannot meet the real-time requirements at this stage. Therefore, this embodiment uses Kalman filtering technology in the subsequent tracking stage to realize the positioning of the captured target.

[0141] Kalman filter basics

[0142] The essence of the Kalman filter model is to complete the optimal estimation of the system state vector by the state transition information and the observed value. The model mainly includes the iterative steps of "prediction-measurement-correction", and then reduces the interference of random noise according to the measurement state of the system, thereby completing the task of estimating the system state. In other words, the Kalman filter model mainly uses the measurement state of the system to complete the optimal estimation of the original state of the system in a certain statistical sense from the conditions interfered by noise. At present, the Kalman filter model is widely used in control, guidance, navigation and other engineering. The main steps of the iterative process of the Kalman filter model are summarized as follows:

[0143] Step 1: According to formula (18) and formula (19), the target prior state estimation and prior state error covariance matrix prediction are performed.

[0144]

[0145] Where x represents the system state matrix, A is the state transfer matrix, P is the covariance matrix, and Q is the measurement noise matrix.

[0146] Step 2: Update the Kalman filter gain according to formula (20).

[0147]

[0148] Where H represents the system state observation matrix, R is the measurement covariance matrix, and K is the Kalman gain coefficient.

[0149] Step 3: Update the error covariance matrix of the posterior state according to formula (21).

[0150] P(k+1)=[IK(k+1)H(k+1)]{(k+1|k) (21)

[0151] Step 4: Update the posterior state estimate according to formula (22).

[0152]

[0153] Where Z represents the system observation state matrix.

[0154] Step 5: For the next frame k=k+1, return to step 1.

[0155] Model Application

[0156] Based on the basic principle of the Kalman filter model, this section mainly describes in detail the application of the model to the positioning process of small targets. Since the time window T between adjacent frames of the captured image is small, we model the movement state of the target within T time as uniform motion, that is, assuming that the angular velocity of the target in the right ascension and declination directions is and constant, so the right ascension and declination positions α(k+T) and β(k+T) of the target at time k+T are equal to the right ascension and declination positions α(k) and β(k) at time k plus the target's motion angle in their respective directions, that is, and Based on the above uniform motion model, the target motion equation can be approximately written as formula (23):

[0157] x(k+1)=A(k+1|k)x(k)+B(k+1|k)w(k) (23)

[0158] in:

[0159]

[0160]

[0161] In the formula, α and β are the right ascension and declination values ​​of the target, respectively. and are the angular velocities of the target in the right ascension and declination directions respectively, k is the frame number of the sequence (k=0, 1, 2...), T is the time interval between adjacent frames, and w(k) is the noise.

[0162] Generally speaking, the elements in w(k) can be regarded as uncorrelated white noise with a mean of 0, that is, the expected value of the noise satisfies E{W(k)}=0. Therefore, the covariance matrix of the noise can be written as E{W(k)W T (j)}=Q(k)δ kj In the form of kj is the Dirac equation, if and only if k = j, δ kj =1; and the expression of Q(k) is as follows, w1(k) and w2(k) represent the noise of the target in the right ascension and declination directions respectively

[0163]

[0164] In the formula, and are the noise variance of the target in right ascension and declination directions respectively;

[0165] In practical applications, the coordinates of the target on the CCD plane are usually used as observation values. To improve the calculation speed, the detection target measurement equation expression is constructed in this embodiment as follows:

[0166]

[0167] In the formula,

[0168] v(k) is the target measurement error noise, so the variance matrix of the target measurement error can be written as follows:

[0169]

[0170] In the formula, and are the measurement noise variances of the target in right ascension and declination directions, respectively.

[0171] Since we use right ascension and declination data in the measurement equation, and the CCD plane coordinate information is obtained by taking pictures through the sub-pixel positioning algorithm, it is necessary to convert the plane coordinate data into right ascension and declination data to obtain the measurement value of the measurement equation. The corresponding calculation method is as follows:

[0172] Assuming that b is the direction vector of the target in the celestial coordinate system, and r is the direction vector of the target in the camera body coordinate system, the attitude matrix A satisfies the following conditions, namely

[0173] r=Ab (31)

[0174] in:

[0175]

[0176] Where (x, y) is the coordinate of the target on the CCD plane, (x0, y0) is the center point of the CCD detection plane, f is the focal length of the camera optical system, and matrix A is the attitude matrix, which satisfies the orthogonal matrix definition, that is, A-1 = AT, so the relationship between b and r can be written as follows:

[0177]

[0178] Therefore, the right ascension and declination values ​​α and β of the target can be calculated by the formula:

[0179]

[0180] Here, the right ascension and declination values ​​of the target are α∈[0, 2π] and β∈[-π / 2, π / 2]. Through the above formula, we can obtain the measurement value of the target.

[0181] Specific process of the algorithm

[0182] In practical applications, when we track the detection target in the entire captured star map, a lot of computing time is required. Therefore, in order to reduce the algorithm running time and improve real-time performance, this embodiment combines the Kalman filter model with the windowing technology. The overall algorithm process framework diagram is as follows: Fig.10 As shown in the figure, it mainly includes the steps of target recognition, initial state establishment, Kalman filter prediction and window tracking. The target recognition stage is mainly based on the full-day recognition and local sky area star map recognition technology to complete the recognition task of the target under the star background, and then establish the initial state of the Kalman filter model. The subsequent observation star map sequence uses the Kalman filter method to predict the position of the observed target, and searches for the target in the search window, and outputs the optimized target position to complete the positioning task. The corresponding specific process is described as follows:

[0183] initialization

[0184] The purpose of initialization is twofold: target recognition and target motion state establishment, and the results are used as the input of the Kalman filter model. In the process of target recognition and positioning, the target in the captured image must be recognized first. This process is necessary to further estimate the initial state of the target and run the subsequent Kalman filter prediction. For the first two frames of captured images, we use the full celestial autonomous star map recognition algorithm ( Fig.11 , where the green circles represent the stars identified by the algorithm in the entire celestial sphere) and the local sky star map recognition algorithm ( Fig.12 , where the blue circles represent the stars identified by the algorithm in the local sky area), completing the distinction between stellar targets and non-stellar targets ( Fig.13 , where the red box represents the target identified by the algorithm), and then the target is identified, thereby obtaining the coordinate information of the target on the detector plane, which is represented by (x -1 ,y -1 ) and (x0, y0).

[0185] Combined with the attitude matrix of the first two frames calculated by QUEST attitude solution, the right ascension and declination information (α -1 , β -1 ) and (α0, β0) (Formula (32) to Formula (33)).

[0186] On this basis, the angular velocity of the target in the right ascension and declination directions is calculated respectively to obtain the target initial state estimation, as shown in formula (34):

[0187]

[0188] Therefore, the target initial state is:

[0189]

[0190] Kalman prediction

[0191] Predict the target's prior position coordinates in the next frame to determine the window center. After the initial state is determined, according to formula (18), the optimal position estimation result of the detected target in the kth frame is The prior estimation of the right ascension and declination state of the target detected in the k+1th frame can be completed. And according to the attitude matrix A, the position coordinates of the target right ascension and declination in the k+1th frame corresponding to the CCD plane of the shooting camera are calculated. This output is used as the window center position for the subsequent window search part ( Fig.14 , where the red dot represents the position of the target predicted by the algorithm in the next frame).

[0192] Window Tracking

[0193] After the window center prior position is estimated by Kalman filtering, the target needs to be identified in the search window and the optimal estimation result of the target is finally output. The specific process is as follows:

[0194] Step 1: Take the predicted value of the k+1th frame As the center position coordinates of the window tracking;

[0195] Step 2: Use the window center radius Rw as the search radius to complete the recognition and positioning of the target, thereby obtaining the observation value Z(k+1) of the target in the k+1th frame, where the search window radius Rw is defined as follows:

[0196] R w =4×E p +4 (36)

[0197] In the formula

[0198]

[0199] Step 3: According to formula (22), the update of the optimal position estimate of the target detected in the k+1th frame is completed, that is,

[0200] Step 4: According to formula (21), complete the update of the error covariance matrix of the target detected in the k+1th frame, that is, P(k+1).

[0201] Through the above self-window search method ( Fig.15 , where the red dotted circle represents the target search range of the algorithm within the window), avoiding the time-consuming star map recognition and target recognition process, reducing the computing load of the embedded platform, and thus shortening the running time of the algorithm to a certain extent.

[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying and locating star targets under a star map background, characterized in that: include: Identify some stars in the pictures taken by the camera based on the autonomous recognition algorithm of the all-sky star map; The initial camera posture is established based on the QUEST algorithm to obtain the initial camera posture information; Recognize the local sky area star map based on the initial camera posture information, and distinguish the stars and moving targets in the captured picture; Acquiring right ascension and declination data of the detection target based on the initial attitude of the camera, and constructing an initial state of the detection target; Based on the initial state of the detection target, the Kalman filter is used to predict the position of the detection target, and the window search method is used to complete the continuous tracking of the detection target.

2. The method for identifying and locating stellar targets under a star map background according to claim 1, characterized in that: The process of identifying some stars in the pictures taken by the camera based on the autonomous recognition algorithm of the all-sky star map includes: The shooting camera posture is determined based on the all-sky star map recognition technology, the sky area position pointed by the field of view axis is determined based on the shooting camera posture, all observation targets in the camera field of view are obtained, and a number of matching star pairs are output.

3. The method for identifying and locating stellar targets under a star map background according to claim 1, characterized in that: The process of establishing the initial camera posture based on the QUEST algorithm includes: The observation star direction vector in the star sensor body coordinate system and the corresponding navigation star direction vector in the celestial coordinate system are obtained based on the matching star pair, and the Wabba problem is defined. The Wabba problem is solved based on the q-method and the QUEST method to obtain the optimal attitude estimation. The initial camera attitude information is obtained based on the optimal attitude and converted into pitch angle, yaw angle and roll angle.

4. The method for identifying and locating stellar targets under a star map background according to claim 3, characterized in that: The Wabba problem is as follows: Among them, J is the objective function, a k is the weight coefficient. Assuming that all the weight coefficients are consistent, if the vectors in the sets w and v are accurate data, the corresponding minimum result of the loss function is 0; if there is a deviation between the vectors in the sets w and v, then the corresponding loss function result is greater than zero. At the same time, the larger the corresponding value of the loss function, the greater the deviation between the algorithm's estimated value R of the posture matrix and the ideal value A.

5. The method for identifying and locating stellar targets under a star map background according to claim 3, characterized in that: The process of solving the Wabba problem based on the q-method and the QUEST method respectively includes: Characterizing the objective function of the Wabba problem in the form of quaternions; Based on the Lagrange multiplier, the objective function is maximized and a derivative operation is performed; The matrix eigenvector in the derivation operation process is solved based on the QUEST method to obtain the optimal posture estimation.

6. The method for identifying and locating stellar targets under a star map background according to claim 5, characterized in that: The process of solving the matrix eigenvector in the derivative operation process based on the QUEST method includes: The matrix eigenvectors are represented as Rodriguez parameters, and the solution of the Rodriguez parameters is converted into a linear equation system, and the linear equation system is solved based on Gaussian elimination method.

7. The method for identifying and locating stellar targets under a star map background according to claim 1, characterized in that: The process of identifying the local sky area captured star map based on the initial camera posture information includes: Based on the initial camera attitude information, a reference star map is generated in combination with a navigation database; Sort all the star points in the photographed star map and the reference star map according to the size of the coordinates; Based on the nearest neighbor matching principle, the star points in the reference star map are divided into star targets and non-star targets; Set the limiting magnitude based on the magnitude data error of the shooting camera and select a reference star.

8. The method for identifying and locating stellar targets under a star map background according to claim 1, characterized in that: The process of using Kalman filtering to predict the position of the detected target includes: Based on the initial state of the detection target, the target prior state estimation and the prior state error covariance matrix prediction are performed, and based on the optimal position estimation result, the right ascension and declination states of the detection target in the k+1th frame are estimated a priori, and the position coordinates of the CCD plane of the shooting camera corresponding to the right ascension and declination of the target in the k+1th frame are calculated according to the attitude matrix, and the output is used as the window center position of the subsequent window search part.

9. The method for identifying and locating stellar targets under a star map background according to claim 1, characterized in that: The process of using window search to continuously track the detected target includes: The position coordinates of the CCD plane of the camera corresponding to the target right ascension and declination in the k+1th frame are taken as the window center position, and the window center radius is obtained; Using the window center radius as the search radius, the detection target is identified and located, and the observation value of the target in the k+1th frame is obtained; The optimal position estimation of the detected target in the k+1th frame and the update of the error covariance matrix are performed based on the observed values.

10. The method for identifying and locating stellar targets under a star map background according to claim 9, characterized in that: The window center radius is defined as follows: R w =4×E p +4 Where R w is the window center radius,

Citation Information

Patent Citations

  • Star map identification method based on radial and dynamic circumferential modes

    CN111174776A

  • Spacecraft autonomous navigation method based on space target azimuth information

    CN115979276A

  • Spatial target detection method based on starlight attitude measurement

    CN116952206A

Cited By

  • Calibration method based on double shooting and imaging equipment

    CN119573772A

  • Method and device for measuring and calculating sight line of space target of strapdown platform

    CN121521136A