A high dynamic star tracking method based on random sampling consensus

Abstract

The application discloses a high-dynamic star tracking method based on random sampling consensus, and solves the technical problem that the existing star tracking method is difficult to overcome the great change of star point positions between frames, and thus cannot perform fast matching of star point positions between frames. The application uses an optimized random sampling consensus algorithm to overcome the great change of star point positions between frames caused by the fast maneuver of a satellite platform, quickly enumerates matched star points, realizes the matching of star points between frames, and thus realizes the fast identification of star points of a current frame by using the star points identified in a previous frame, can adapt to high-dynamic star tracking of 4-degree high-speed maneuver between two continuous frames of a carrier, can correctly and quickly match between two star maps within 360 degrees, and can improve dynamic capability. The method can be used in satellite obstacle avoidance, navigation and other aspects of carrier navigation which have demands for high-speed maneuver.

Description

Technical Field

[0001] This invention relates to star sensor tracking methods, specifically to a high dynamic star tracking method based on random sampling consistency. Background Technology

[0002] With the development of aerospace technology, more and more spacecraft are entering space, making the limited space orbital resources increasingly scarce. The risk of collisions between spacecraft is increasing, and the competition among nations for space dominance is intensifying. This new situation places higher demands on the maneuverability of satellite platforms.

[0003] When a spacecraft moves at high speed, especially rotating, the star points undergo affine transformations between frames, involving both displacement and rotation. This makes matching star points between frames difficult. For example... Figure 1 As shown, Figure 1 (a) A star chart is given for right ascension and declination values ​​of [1, -89]. Figure 1 (b) A star map is given when the right ascension and declination are [1, -85]. By comparison, it can be seen that in addition to the change in the position of the star points, there is also a large-angle rotation of 25 degrees superimposed between the two. This complex relationship of star point movement poses a huge challenge to star tracking methods. However, existing star tracking methods are unable to overcome the huge changes in the position of star points between frames, which makes it impossible to quickly match the position of star points between frames. Summary of the Invention

[0004] The purpose of this invention is to provide a high-dynamic star tracking method based on random sampling consistency, so as to solve the technical problem that existing star tracking methods are unable to overcome the huge changes in the positions of star points between frames, which makes it impossible to quickly match the positions of star points between frames.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A high-dynamic star tracking method based on random sampling consistency is characterized by the following steps:

[0007] Step 1: Acquire star map in real time using star sensor; Statistically analyze the position and grayscale information of each star point in the previous frame star map and the current frame star map, and select a portion of star points in each of the two maps as a dataset of star points to be matched based on the star point position and grayscale information, and construct dataset matrices respectively.

[0008] Step 2: Construct a fast star point scanning and matching mechanism, and use the fast scanning and matching mechanism to generate column numbers for the star point datasets to be matched in the previous frame star map and the current frame star map respectively;

[0009] Step 3, based on the column sequence number generated in step 2, the corresponding column vectors are extracted from the two data set matrices constructed in step 1, and the corresponding star point matrices are constructed respectively;

[0010] Step 4, the transformation matrix H corresponding to the two star point matrices constructed in step 3 is calculated;

[0011] Step 5, the data set matrix of the current frame star map is mapped back to the data set matrix of the previous frame star map through the transformation matrix H calculated in step 4, so as to eliminate the translation error and rotation error between the previous frame star map and the current frame star map; the star point coordinates of the mapped star point i in the current frame star map are associated and matched with the star point coordinates of the corresponding star point j in the previous frame star map by using the nearest neighbor association strategy, so as to calculate the position matching error θ ij of each star point;

[0012] Step 6, the position matching error θ ij is compared with the preset error threshold σ; when θ ij <σ, the corresponding star point coordinates in the previous frame star map and the current frame star map are extracted to form a corresponding matching point vector, and all the matching point vectors are arranged in columns to form an inner point set The number of corresponding star points T in the inner point set is counted, and the star number δ i of the matched star point is obtained to form a star number vector;

[0013] Step 7, the number of counted star points T is compared with the preset star point number threshold μ;

[0014] If T≥μ, step 8 is executed;

[0015] If T<μ, step 2 is returned, and the column sequence number is regenerated by using the star point fast scanning matching mechanism until the preset iteration number τ is reached, or until T≥μ, step 8 is executed;

[0016] Step 8, the matching star point coordinates in the inner point set and the corresponding star number vector are obtained to construct a target surface matrix and a celestial sphere matrix respectively;

[0017] Step 9, based on the target surface matrix and the celestial sphere matrix constructed in step 8, the three-axis pointing of the current frame star map is calculated by using the Quest method;

[0018] Step 10, according to the star table, the star numbers of all the star points in the current frame star map are obtained by using the nearest neighbor association strategy according to the three-axis pointing obtained in step 9;

[0019] Step 11, steps 1-10 are repeated to perform high dynamic star tracking.

[0020] Further, step 1 is specifically:

[0021] Real-time star map is acquired by star sensor; position of each star point and gray information of each star point in previous frame star map and current frame star map are counted, each star point in previous frame star map and current frame star map is sorted in descending order based on the gray information, and M and N star points in two frame star maps are respectively selected as a set of star point data to be matched, 5≤N≤25 and is an integer, 5≤M≤25 and is an integer, and then a set of matrixes is respectively constructed:

[0022]

[0023] Wherein, x M is the X-axis coordinate of the Mth star point in the previous frame star map in the image target rectangular coordinate system; y M is the Y-axis coordinate of the Mth star point in the previous frame star map in the image target rectangular coordinate system; is the X-axis coordinate of the Nth star point in the current frame star map in the image target rectangular coordinate system; is the Y-axis coordinate of the Nth star point in the current frame star map in the image target rectangular coordinate system.

[0024] Further, step 2 is specifically:

[0025] A star point fast scanning matching mechanism is constructed, and column sequence numbers [a, b, c] and [m, n, p] are generated for the set of star point data to be matched in the previous frame star map and the current frame star map by using the fast scanning matching mechanism;

[0026] Step 3 is specifically:

[0027] Based on the column sequence numbers generated in step 2, three column vectors with column sequence numbers [a, b, c] are extracted from the set of matrixes pre_vect to form a star point matrix Three column vectors with column sequence numbers [m, n, p] are extracted from the set of matrixes post_vect to form a star point matrix

[0028] Further, step 4 is specifically:

[0029] The transformation matrix H corresponding to the two star point matrices formed in step 3 is calculated:

[0030]

[0031] Further, step 5 is specifically:

[0032] The transformation matrix H calculated by step 4 maps the data set matrix post_vect back to the data set matrix pre_vect to eliminate the translation error and rotation error between the previous frame star map and the current frame star map; by using the nearest neighbor association strategy, the star point coordinates (x i ,y i ) of the star point i in the current frame star map after being mapped by the transformation matrix H are associated and matched with the star point coordinates of the corresponding star point j in the previous frame star map, so as to calculate the position matching error θ ij of each star point:

[0033]

[0034] Wherein, Δx ij and Δy ij are calculated by the following formula:

[0035]

[0036] Further, step 6 is specifically:

[0037] The position matching error θ ij is compared with the preset error threshold σ; when θ ij <σ, the star point coordinates of the corresponding star points in the previous frame star map and the current frame star map are extracted to form a corresponding matching point vector All the matching point vectors are arranged in columns to form an inner point set The number T of the corresponding star points in the inner point set is counted; and the star number δ i of the matched star points is obtained to form a star number vector i∈[1,…,T].

[0038] Further, step 7 is specifically:

[0039] The counted number T of star points is compared with the preset star point number threshold μ;

[0040] If T≥μ, step 8 is executed;

[0041] If T<μ, step 2 is returned to use the star point fast scanning matching mechanism to generate the column sequence numbers [a, b, c] and [m, n, p] again until the preset iteration number τ is reached, or until T≥μ, step 8 is executed;

[0042] Wherein, μ is 30% to 60% of the smaller value of M and N.

[0043] Further, step 8 is specifically:

[0044] The matching star point coordinates in the inner point set and the corresponding star number vector are obtained Constructing target surface matrix respectively and celestial sphere matrix

[0045] Wherein (α i1 ,β i1 )…(α iT ,β iT ) respectively represent the corresponding right ascension and declination values of star δ i1 …δ T in the star catalog, and f is the focal length of the camera.

[0046] Further, in step 7, μ is 40% of the smaller value of M and N.

[0047] Advantages of the present application:

[0048] The high dynamic star tracking method based on random sampling consensus provided by the present application uses an optimized random sampling consensus algorithm, overcomes the huge changes in the positions of star points between frames caused by the rapid maneuvering of a satellite platform, quickly enumerates matched star points, realizes the matching of star points between frames, thereby realizes the rapid identification of star points in the current frame using the star points identified in the previous frame, can adapt to high dynamic star tracking of 4 degrees of high-speed maneuvering between two consecutive frames of a carrier, can correctly and quickly match between two frames of star maps within 360°, and can improve the dynamic capability. The method can be used in satellite obstacle avoidance, navigation and other aspects of carrier navigation that require high-speed maneuvering. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 is a schematic diagram of the positions of star points on two frames of star maps, wherein (a) is a star map when the right ascension and declination values are [1, -89], and (b) is a star map when the right ascension and declination values are [1, -85];

[0050] Figure 2 is a result diagram of matching the current frame of star maps with the previous frame of star maps using an existing method; wherein (a) is a star map in which all star points have been identified and the star numbers are known in the previous frame, and (b) is a star map photographed in the current frame, in which the star numbers of all star points are unknown;

[0051] Figure 3 is a schematic diagram of star points in the current frame of star maps and star points in the previous frame of star maps that are successfully matched and obtain corresponding star numbers in the embodiment of the present application; wherein (a) is a star map in which all star points have been identified and the star numbers are known in the previous frame, and (b) is a current frame of star maps in which some star numbers are obtained through association matching, and the star numbers of the remaining star points are star numbers obtained through back calculation of the pointing matrix;

[0052] Figure 4The current frame star map and the previous frame star map in the embodiment of the application are correctly matched with a large rotation angle (160°); wherein (a) is a star map in which all star points have been identified in the previous frame, and (b) is a current frame star map with a large rotation angle, and the pointing direction and star point information of the current frame star map are successfully solved by using the method of the embodiment. DETAILED DESCRIPTION

[0053] The technical solutions of the application will be described clearly and completely in combination with the drawings and embodiments. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0054] The principle of implementing the high dynamic star sensor tracking method based on random sampling consensus is to associate the identified star points in the previous frame with the unidentified star points in the current frame by using the identified star points and star numbers in the previous frame, and the successfully associated star points obtain the corresponding star numbers, and the pointing direction of the camera optical axis of the current frame can be solved by combining the target coordinates of the star points on the detector.

[0055] The application is based on an optimized random sampling consensus algorithm, and a star point fast scanning matching mechanism is proposed in the application, as shown in Figure 2 Figure 2 All star numbers of the star points in (a) have been successfully identified; Figure 2 (b) is a current frame star map to be identified, and a well-known algorithm in the image registration field, random sampling consensus (RANSAC) algorithm, is used herein. The algorithm can effectively eliminate the unmatched points in feature matching. In order to meet the high dynamic star sensor tracking requirement, the sample selection strategy of the algorithm is changed importantly, and some star points in the current frame star map will be successfully matched with the star points in the previous frame star map.

[0056] The successfully matched star points in the current frame star map can obtain the star numbers by using the star numbers in the previous frame star map, and the effect is as shown in Figure 3 The high dynamic star tracking method based on random sampling consensus of the embodiment will be described in detail.

[0057] The embodiment includes the following steps:

[0058] ​Step 1, real-time acquisition of star map by star sensor; the star sensor works in full-sky recognition mode when it is started, first completes the calculation of the current optical axis pointing according to the star map recognition algorithm, and completes the recognition of all star points in the field of view; after completing the initial attitude determination, it is switched to tracking mode, then the positions of each star point and the gray information of each star point in the previous frame star map and the current frame star map are counted, based on the gray information, each star point in the previous frame star map and the current frame star map is sorted in descending order respectively, and the first M and N star points in the two frames of star maps are selected as the matching star point data set respectively, 5≤N≤25 and is an integer, 5≤M≤25 and is an integer, and then the data set matrix is constructed respectively:

[0059]

[0060] Wherein, x M is the X-axis coordinate of the Mth star point in the previous frame star map in the image target rectangular coordinate system; y M is the Y-axis coordinate of the Mth star point in the previous frame star map in the image target rectangular coordinate system; is the X-axis coordinate of the Nth star point in the current frame star map in the image target rectangular coordinate system; is the Y-axis coordinate of the Nth star point in the current frame star map in the image target rectangular coordinate system.

[0061] Step 2, construct a star point fast scanning matching mechanism, and use the fast scanning matching mechanism to generate column sequence numbers [a, b, c] and [m, n, p] for the matching star point data set in the previous frame star map and the current frame star map respectively;

[0062] Step 3, based on the column sequence numbers generated in step 2, extract the three-column vector with column sequence number [a, b, c] from the data set matrix pre_vect to form a star point matrix extract the three-column vector with column sequence number [m, n, p] from the data set matrix post_vect to form a star point matrix

[0063] Step 4, calculate the transformation matrix H corresponding to the two star point matrices formed in step 3:

[0064]

[0065] Step 5, map the data set matrix post_vect back to the data set matrix pre_vect through the transformation matrix H calculated in step 4, to eliminate the translation error and rotation error between the previous frame star map and the current frame star map; using the nearest neighbor association strategy, the star point coordinates (x i , y i ) of the star point i in the current frame star map after mapping by the transformation matrix H are associated with the star point coordinates The correlation matching is performed, so as to calculate the position matching error θ of each star point ij :

[0066]

[0067] wherein, Δx ij and Δy ij are calculated by the following formula:

[0068]

[0069] Step 6, comparing the position matching error θ ij with a preset error threshold σ; extracting the star point coordinates corresponding to the star point coordinates in the previous frame star map and the current frame star map when θ ij <σ, to form a corresponding matching point vector All the matching point vectors are arranged in columns to form an inlier set The inlier set is counted, and the number of corresponding star points T is obtained; and the star number δ of the matched star point is obtained to form a star number vector i . i∈[1,…,T].

[0070] Step 7, comparing the number of counted star points T with a preset star point number threshold μ;

[0071] If T≥μ, step 8 is executed;

[0072] If T<μ, step 2 is returned, and the column sequence numbers [a, b, c] and [m, n, p] are regenerated by using the star point fast scanning matching mechanism until a preset iteration number τ is reached, or until T≥μ, step 8 is executed.

[0073] Wherein, μ is 30% to 60% of the smaller value of M and N, and is preferably 40%.

[0074] Step 8, obtaining the matching star point coordinates in the inlier set and the corresponding star number vector to construct a target surface matrix and a celestial sphere matrix f is the focal length of the camera.

[0075] Wherein (α i1 ,β i1 )…(a iT ,β iT ) respectively represent the corresponding right ascension and declination values of the star number δ i1 …δ T in the star catalog, and f is the focal length of the camera.

[0076] Step 9, based on the target surface matrix and the celestial sphere matrix constructed in step 8, the three-axis pointing of the current frame star map is calculated by using the Quest method.

[0077] Step 10, using the three-axis pointing obtained in step 9, the coordinates of all possible star points under the three-axis pointing are calculated according to the star catalog, and the star numbers of all star points in the current frame star map are obtained by using the nearest neighbor association strategy; the star map recognition effect is as shown in Figure 4

[0078] Step 11, repeat steps 1-10 to perform high dynamic star tracking.

[0079] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any change or replacement within the technical scope disclosed in the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.​

Claims

1. A high dynamic star tracking method based on random sampling consensus, characterized in that, The method comprises the following steps: Step 1, acquiring a star map in real time through a star sensor; Step 2, constructing a star point fast scanning matching mechanism, and generating column serial numbers for the to-be-matched star point data sets in the previous frame star map and the current frame star map respectively by using the fast scanning matching mechanism; Step 3, based on the column serial numbers generated in step 2, extracting corresponding column vectors from the two data set matrices constructed in step 1 respectively, and respectively forming corresponding star point matrices; Step 4, calculating the corresponding transformation matrix H according to the two star point matrices formed in step 3; Step 7, comparing the number T of the counted star points with a preset star point number threshold μ; Step 5, mapping the data set matrix of the current frame star map back to the data set matrix of the previous frame star map through the transformation matrix H calculated in step 4 to eliminate the translation error and the rotation error between the previous frame star map and the current frame star map; using the nearest neighbor association strategy to associate and match the star point coordinate of the mapped star point i in the current frame star map with the star point coordinate of the corresponding star point j in the previous frame star map, so as to calculate the position matching error θ of each star point ij ; Step 6, match the position error θ ij with the preset error threshold σ; extract θ ij <σ, the corresponding star coordinates in the previous frame star map and the current frame star map form the corresponding matching point vector, and all the matching point vectors are arranged in columns to form the inner point set Statistics of inner point set The corresponding star point number T, and the star number δ of the matched star point is obtained i , to form the star number vector; If T ≥ μ, step 8 is executed; If T < μ, returning to step 2 to generate column serial numbers again by using the star point fast scanning matching mechanism, until a preset iteration number τ is reached, or until T ≥ μ, and step 8 is executed; Step 9, based on the target surface matrix and the celestial sphere matrix constructed in step 8, calculating the three-axis pointing of the current frame star map by using the Quest method; Step 8, obtaining the inlier set The target surface matrix and the celestial sphere matrix are respectively constructed according to the coordinates of each matched star point and the corresponding star vector in the star image. Step 10, calculating all possible star point position coordinates under the three-axis pointing according to a star catalog by using the three-axis pointing obtained in step 9, and obtaining the star numbers of all star points in the current frame star map by using a nearest neighbor association strategy; Step 11, repeating steps 1-10 to perform high dynamic star tracking. Step 1 is specifically:

2. The random sample consensus based high dynamic star tracking method of claim 1, wherein, acquiring a star map in real time through a star sensor; counting the positions of star points and the gray scale information of the star points in the previous frame star map and the current frame star map, and respectively sorting the star points in the previous frame star map and the current frame star map in descending order based on the gray scale information, and respectively selecting the first M and n star points in the two frame star maps as to-be-matched star point data sets, 5 ≤ n ≤ 25 and n is an integer, 5 ≤ M ≤ 25 and M is an integer, and then respectively constructing data set matrices: Step 2 is specifically: wherein x M is the X-axis coordinate of the Mth star point in the previous frame of star map under the X-axis coordinate of the image target surface rectangular coordinate system; y M is the Y-axis coordinate of the Mth star point in the previous frame of star map under the Y-axis coordinate of the image target surface rectangular coordinate system; is the X-axis coordinate of the Nth star point in the current frame of star map under the X-axis coordinate of the image target surface rectangular coordinate system; is the Y-axis coordinate of the Nth star point in the current frame of star map under the Y-axis coordinate of the image target surface rectangular coordinate system.

3. The random sample consensus based high dynamic star tracking method of claim 2, wherein, constructing a star point fast scanning matching mechanism, and generating column serial numbers [a, b, c] and [m, n, p] for the to-be-matched star point data sets in the previous frame star map and the current frame star map respectively by using the fast scanning matching mechanism; Step 3 is specifically: Step 4 is specifically: Based on the column sequence number generated in step 2, a three-column vector with column sequence number [a, b, c] is extracted from the data set matrix pre_vect to form a star point matrix A three-column vector with column sequence number [m, n, p] is extracted from the data set matrix post_vect to form a star point matrix 4. The random sample consensus based high dynamic star tracking method of claim 3, wherein, calculating the corresponding transformation matrix H according to the two star point matrices formed in step 3; Step 5 is specifically:

5. The random sample consensus based high dynamic star tracking method of claim 4, wherein, Step 6 is specifically: The transformation matrix H calculated by step 4 maps the data set matrix post_vect back to the data set matrix pre_vect to eliminate the translation error and rotation error between the previous frame star map and the current frame star map; using the nearest neighbor association strategy, the star point coordinates (x i ,y i ) of the star point i in the current frame star map after mapping by the transformation matrix H are associated and matched with the star point coordinates of the corresponding star point j in the previous frame star map, so as to calculate the position matching error θ ij of each star point. where Δx ij and Δy ij are calculated by the following equations:

6. The random sample consensus based high dynamic star tracking method of claim 5, wherein, Step 7 is specifically: The position matching error θ ij is compared with a preset error threshold σ; θ ij <σ, the coordinates of the corresponding star points in the previous frame star map and the current frame star map form a corresponding matching point vector All the matching point vectors are arranged in columns to form an inner point set The inner point set is counted The corresponding number of star points T; and the star number δ of the matched star points are obtained i , to form a star number vector 7. The random sample consensus based high dynamic star tracking method of claim 6, wherein, comparing the number T of the counted star points with a preset star point number threshold μ; If T ≥ μ, step 8 is executed; If T < μ, returning to step 2 to generate column serial numbers [a, b, c] and [m, n, p] again by using the star point fast scanning matching mechanism, until a preset iteration number τ is reached, or until T ≥ μ, and step 8 is executed; Wherein, μ is 30% to 60% of the smaller value of M and N. Step 8 is specifically:

8. The random sample consensus based high dynamic star tracking method of claim 7, wherein, 9. The high dynamic star tracking method based on random sampling consensus according to claim 8, characterized in that: acquiring an inlier set coordinates of each matching star point and the corresponding star vector constructing a target surface matrix and a celestial sphere matrix where (α i1 ,β i1 )…(α iT ,β iT ) represent the right ascension and declination values, respectively, of the stars δ i1 …δ T corresponding in the star catalogue, and f is the focal length of the camera. In step 7, μ is 40% of the smaller value of M and N. ​

Images