A fast star map recognition method based on attitude information

Abstract

The present invention relates to a partitioned rapid star map recognition method based on attitude information. The star closest to the image center is used as a reference star. The angle characteristics and distance characteristics of the reference star and its neighboring stars are used to construct an angle feature vector, a distance feature vector, and a distance mapping vector. The discreteness between the reference star and the navigation star is calculated based on the distance mapping vector to obtain a candidate navigation star. The angle and distance information are used to search for the corresponding starting edges of the reference star angle feature vector and the candidate navigation star angle feature vector and perform a cyclic shift. The star with the highest similarity is ultimately the correct navigation star. Real-time prediction and rapid recognition are then performed. The advantages of the present invention over the prior art are that the angle feature vector of the reference star is constructed, and a distance mapping vector containing an anti-interference code is calculated to improve the robustness of the algorithm. The computational complexity is minimized, real-time performance is improved, the recognition accuracy of the algorithm is guaranteed, and the reliability of the algorithm is improved.

Description

Technical Field

[0001] The invention belongs to the field of aerospace and relates to a partitioned fast star map recognition method based on attitude information. Background Art

[0002] Spacecraft must constantly monitor their precise three-axis orientation during space missions. Therefore, determining the spacecraft's attitude with high precision has long been a research hotspot. This attitude information is primarily measured in real time by attitude sensors, including star sensors, gyroscopes, horizon instruments, and magnetometers. Star sensors are among the most accurate attitude sensors currently available. Using stars as a reference, they can accurately measure attitude data down to the arcsecond level with excellent stability. Star sensors operate in two modes: spatial stray mode and tracking mode, with tracking being the primary mode. When in spatial stray mode, the star map must first be preprocessed to extract the centroid coordinates of star points. Next, star map recognition methods are used to determine the corresponding star signs and related information in the navigation star catalog for these extracted star points. Finally, the attitude is determined using the star coordinate vectors in both the star sensor coordinate system and the celestial coordinate system. The star map recognition step is the most critical. In 2006, SILANI et al. proposed the polestar method in "Star Identification Algorithms: A Novel Approach & Comparison Study," published in IEEE Transactions on Aerospace and Electronic Systems, Vol. 42, No. 4. This method uses angular distance as a feature. After selecting a reference star, it selects all neighboring stars within a certain radius and calculates the angular distances between these neighboring stars and the reference star. A binary vector is then generated using a mapping method and matched against a database of navigation stars using a voting approach. In 2008, ZHANG et al. published "Full-sky Autonomous Star Identification Based on Radial and Cyclic Features of Star Patterns," published in Image and Vision Computing, Vol. 26, No. 7. They divided the star pattern recognition process into two steps: first, using the radial distance features between the reference star and its neighbors for initial matching to narrow the search range, and then using circumferential features to match the reference star to its unique corresponding navigation star. In 2007, LEE et al. developed an improved grid method in the article "Star pattern identification technique by modified grid algorithm" published in "IEEE Transactions on Aerospace and Electronic Systems", Volume 43, Issue 3. They replaced the traditional Cartesian coordinate system with a polar coordinate system and constructed a polar coordinate grid, thereby improving the robustness of the existing grid method.In their article "Modified Grid Algorithm for Noisy All-Sky Autonomous Star Identification," published in IEEE Transactions on Aerospace and Electronic Systems, Vol. 45, No. 2, in 2009, NA et al. introduced the concept of a minimum loss function from optimization problems into star map recognition, using the minimum loss function to measure the difference between the reference star pattern and the navigation star pattern. Furthermore, they incorporated new magnitude features into the recognition process and used them as weights in the loss function, making the algorithm more efficient. However, existing algorithms either use the nearest neighbors of the primary star as calibration stars, which can easily lead to algorithm failure due to mismatches, resulting in low recognition accuracy; or suffer from poor real-time performance due to overly large databases and complex recognition steps. Furthermore, some algorithms suffer from low recognition accuracy when the star map contains pseudo-stars, missing stars, and significant star position noise. Therefore, reliably and quickly identifying star points in a star map is a challenging task. Summary of the Invention

[0003] In order to solve the problems existing in the prior art, the present invention proposes a partitioned fast star map recognition method based on attitude information, which solves the problem of star map recognition in harsh environments. It has a wider scope of application while ensuring good recognition accuracy and real-time performance.

[0004] The present invention and technical solutions are described as follows:

[0005] The basic idea of ​​the present invention is that, first, the method of the present invention uses the star closest to the center of the image as the reference star, and uses the angle characteristics and distance characteristics of the reference star and its neighboring stars to construct the angle feature vector, distance feature vector and distance mapping vector. Secondly, based on the distance mapping vector, the discreteness between the reference star and the navigation star is calculated, the navigation star library list is shortened, and the candidate navigation star is obtained. Then, the corresponding starting edges of the reference star angle feature vector and the candidate navigation star angle feature vector are searched and cyclically shifted using the angle and distance information. Then, based on the angle feature vector, the cumulative angle feature vector is constructed, and the similarity between the reference star and the qualified navigation star is calculated. The star with the highest similarity is finally obtained, which is the correct navigation star. Finally, based on the attitude information provided by the inertial navigation system, the position of the star point on the CCD plane is predicted in real time to achieve rapid identification of stars in the subsequent star map.

[0006] The present invention provides a partitioned rapid star map recognition method based on attitude information, which is characterized by comprising five steps: feature extraction, navigation star library construction, candidate navigation star screening, reference star recognition, and rapid star map recognition based on attitude information, as follows:

[0007] Step 1: Feature Extraction

[0008] The present invention uses two rotation and translation invariant features: the Euclidean distance D between the neighboring star and the central navigation star. i , the relative angle Φ between consecutive adjacent stars and the central navigation star i , and construct the star pattern of each navigation star based on these two features, such as Figure 2 As shown. Let each navigation star be at the center of the field of view, and consider all neighboring stars within the radius R to extract the above two features. The neighboring stars are recorded as S={S1, S2, ..., S N The relationship between R and the star sensitive field of view FOV is:

[0009]

[0010] Where f is the focal length, ρ is the pixel size, and radius R is the maximum distance from the center.

[0011] Step 2: Navigation star library construction

[0012] In the navigation star library, each navigation star has an angle feature vector Φ, a distance feature vector D, and a distance mapping vector Λ. Using the position coordinates of each neighboring star and the navigation star in the image coordinate system, combined with Φ i and D i The expression of the angle characteristic vector Φ between the navigation star and the neighboring star can be obtained by Φ=(Φ1, Φ2,…, Φ N ) and the distance feature vector D=(D1,D2,…,D N ). Among them, Φ i and D i They are the adjacent angles of the two neighboring stars and the radial distances between the neighboring stars and the reference star when the reference star is located at the center of the CCD in the ideal star map. The above features are uniformly arranged in a counterclockwise manner. In addition, it is necessary to construct the navigation star distance mapping vector It is used to shorten the navigation star library list and obtain the candidate navigation star partition (i.e. a set of filtered navigation stars). i is the mapping value obtained from the ideal star map, N D is the dimension of the distance mapping vector. With the candidate navigation star partition, the angle feature vector Φ and the distance feature vector D can be used to identify the unique navigation star corresponding to the reference star.

[0013] Step 3: Candidate navigation star partition

[0014] Step 3.1: For the star image taken by the star sensor, after the star point centroid positioning process is completed, the star point closest to the image center is selected as the reference star, and the angle feature vector between the reference star and the neighboring star is extracted. And the distance feature vector d=(d1,d2,…,d N ), then construct N D dimensional reference star distance mapping vector in, and d i are the adjacent angles between the two neighboring stars and the radial distances between the neighboring stars and the reference star when the reference star is near the center of the CCD in the real star map; i This is the mapping value obtained from the actual star map.

[0015] Step 3.2: Next, partition the star list in the navigation star library based on the distance mapping vector λ, that is, screen out the candidate navigation stars that meet the conditions to form a new area (set). In this way, the reference star identification process in step 4 only needs to be performed on the navigation stars in this partition. When using image signals to solve the two-dimensional centroid coordinates of star points, problems such as stray light, instrument noise, and low accuracy of the centroid algorithm will be encountered, resulting in a certain deviation in the centroid coordinates, which in turn leads to the radial distance d between the reference star and the neighboring star. i Deviation, T i The value calculation is inaccurate, T i Represents the mapping position of the i-th neighboring star. In order to improve the anti-noise performance of the method of the present invention, when calculating T i When d i / (R / N D ) satisfies the relevant constraints, then the T i The value is easily affected by noise and needs to be mapped to the distance vector An anti-interference code is added in , and ζ1 is the anti-interference code threshold.

[0016] Step 3.3: Calculate the reference star distance mapping vector Distance mapping vector to navigation star The discreteness X between them. The navigation stars whose discreteness X is less than or equal to the discreteness threshold ζ2 are determined as candidate navigation stars, and the candidate navigation star partition is obtained.

[0017] Step 4: Reference Star Identification

[0018] With the candidate navigation star partition, we can then use the angle feature vector of the reference star to And the distance feature vector d is used to identify the unique corresponding navigation star. The specific steps are as follows.

[0019] Step 4.1: Due to the randomness of the rotation angle when the star sensor takes the star map, the starting edges of the reference star and the navigation star’s constructed feature vectors are inconsistent. In order to obtain the correct reference star and navigation star cumulative angle feature vectors ε and E, and calculate the similarity P between the two (the meaning is described later), it is necessary to ensure that the angle feature vectors Φ and Φ of the navigation star and the reference star are consistent. is constructed from the same starting edge, i.e., vector Φ, The first digit in is equal to the first digit in vector D and d. Therefore, the distance and angle information of a reference star should be used to compare the Based on this, we traverse the angle feature vector Φ and distance feature vector D of all the navigation stars in the candidate navigation star partition in turn, and filter out the ones containing The navigation star is then used to calculate the accumulated angle feature vector and similarity.

[0020] Step 4.2: Characteristic vector of the angle between the reference star and the navigation star After aligning the starting edge (starting angle), Φ calculates the similarity between the two and identifies the correct navigation star corresponding to the reference star.

[0021] Step 5: Rapid star map recognition based on attitude information

[0022] For the star sensor shooting at time k and k+1, the attitude of the two shots obviously changes, and the projection of the star on the CCD imaging plane will also move accordingly. In order to quickly obtain the predicted position of the centroid of the star point in the star map shot at time k+1, avoid the recognition of the star map in the entire sky area and improve the overall processing speed of the system, the short-term high-precision attitude change information provided by the inertial navigation can be used to accurately predict the position of the star point in real time, as follows:

[0023]

[0024] Where A, D, and θ are the optical axis orientation and rotation angle of the star sensor at time k+1, respectively. These can be calculated based on the attitude information provided by the inertial navigation system. Once the predicted coordinates of the star point on the CCD plane are obtained based on the attitude information provided by the inertial navigation system, star point extraction and centroid positioning can be performed within a small neighborhood centered around the predicted coordinates. Rapid verification identification is performed on all extracted star points to prevent incorrect star point extraction. Once the actual centroid coordinates of a known star in the carrier coordinate system are obtained, an attitude determination algorithm such as QUEST can be used to calculate the attitude transformation matrix between the carrier coordinate system and the celestial coordinate system at the current moment.

[0025] The advantage of the present invention over the prior art is that it constructs the angle characteristic vector of the reference star The distance feature vector d is calculated based on the distance feature vector d to obtain the distance mapping vector λ containing the anti-interference code to improve the robustness of the algorithm; the candidate navigation star partition is obtained by calculating the discreteness between the reference star and the navigation star, which can reduce the amount of calculation to the greatest extent and improve the real-time performance; the distance feature vector D and the angle feature vector Φ of the navigation star are searched to see whether the distance and angle information of the reference star are included. To find the starting edge, and through the circular shift operation, get the new angle feature vector between the reference star and the navigation star Φ′, ensures the recognition accuracy of the algorithm; calculates the angle feature vector The algorithm's reliability is further improved by calculating the cumulative angle characteristic vectors ε and E of Φ' and the similarity P between the reference star and the navigation star. After identifying the correct navigation star, the center of mass position of the star's projection in the star map at the current moment is recursively predicted in real time based on the attitude information provided by the inertial navigation system. After performing a rapid verification identification on all extracted star points and eliminating incorrectly extracted star points, an attitude determination algorithm such as QUEST can be used to calculate the attitude transformation matrix between the carrier coordinate system and the celestial coordinate system at the current moment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 : Schematic diagram of the star map recognition process in the present invention

[0027] Figure 2 : Schematic diagram of navigation star mode

[0028] Figure 3 : Reference star pattern diagram

[0029] Figure 4 :Matching instructions Figure 1

[0030] Figure 5 :Matching instructions Figure 2 DETAILED DESCRIPTION

[0031] The present invention will now be described in further detail with reference to the accompanying drawings.

[0032] The present invention proposes a partitioned rapid star map recognition method based on attitude information, which includes five steps: feature extraction, navigation star library construction, candidate navigation star screening, reference star recognition and rapid star map recognition based on attitude information.

[0033] like Figure 1 The overall workflow is: Calculate the angle characteristic vector of the reference star The distance feature vector d is obtained, and based on the distance feature vector d, the distance mapping vector λ containing the anti-interference code is obtained; for the navigation star whose number of neighboring stars meets the threshold range, the discreteness X of the distance mapping vectors λ and Λ between the reference star and the navigation star is calculated to obtain the candidate navigation star partition; the distance and angle information pairs of the reference star are searched in the distance feature vector D and the angle feature vector Φ of the navigation star. And align the starting edge to obtain the new angle characteristic vector between the reference star and the navigation star Φ′; calculate the angle characteristic vector The cumulative angle characteristic vector ε of Φ′=(ε1,ε2,…,εm )、E=(E1,E2,…,E n ), and calculate the similarity P between the reference star and the navigation star to identify the correct navigation star. i and E i The i-th cumulative angle between the reference star and the navigation star is respectively. The center of mass position of the star is predicted in real time based on the attitude information provided by the inertial navigation system to achieve rapid star map recognition based on attitude information.

[0034] Example 1

[0035] Step 1: Feature Extraction

[0036] See also Figure 2 :Specifically include:

[0037] Two rotational and translational invariant features—the Euclidean distance D between neighboring stars and the central navigation star i , the relative angle Φ between consecutive adjacent stars and the central navigation star i The calculation is as follows:

[0038]

[0039] Where x i and y i are the coordinates of the ith neighboring star in the image coordinate system, N is the total number of all neighboring stars in the field of view, x c and y c is the coordinate of the navigation star in the image coordinate system.

[0040] Step 2: Navigation star library construction

[0041] Construct distance feature vector D=(D1,D2,…,D N ), the starting edge D i The selection is different, corresponding to the angle feature vector Φ and the distance feature vector D, only a circular shift is performed, which does not affect the essential characteristics of the vector. min Construct D and Φ for the starting edge, where D min =min{D1,D2,…,D N}, it does not matter which edge is chosen as the starting edge.

[0042] The distance characteristic vector D between the navigation star and the neighboring star has been obtained before. N ), the following is the calculation of the navigation star distance mapping vector Λ through the distance characteristic vector D. The radial distance D between the navigation star and the neighboring star i The maximum is R, and the distance mapping vector is an N D dimensional vector signal, N DThe value is given by the following formula, where the value of k is determined by experimental testing. The value of each bit in the vector is 0 or 1, and the specific value selection method is as follows. First, let N D The dimension distance feature vector is all 0, substitute D1 into the following formula to get T1, then the T1th position of the distance mapping vector Λ is 1. Traverse the D in the distance feature vector D in turn i (i∈1~N), then we get N D dimensional navigation star distance mapping vector

[0043] N D =2 k ; k∈Z (5)

[0044]

[0045] Step 3: Screening and partitioning candidate navigation stars

[0046] When adding anti-interference code, for the radial distance d that satisfies formula (7) i , in the T i and T i The +1 bit is set to 1, satisfying the radial distance d of formula (8) i , in the T i and T i The -1 bit is set to 1. This increases the success rate of matching when the discreteness of the distance mapping vectors λ and Λ between the reference and navigation stars is subsequently calculated, ensuring a higher probability that the candidate navigation stars within the partition contain the correct reference star. The introduction of the anti-interference code gives the method of the present invention better noise immunity (this noise primarily refers to position noise). ζ1 is the anti-interference code threshold.

[0047]

[0048] Calculate the reference star distance mapping vector Distance mapping vector to navigation star To calculate the dispersion X between the reference and reference stars, first perform a bitwise AND operation on λ and Λ and accumulate the sum to obtain the value B. The B value represents the number of neighboring stars that can match the reference star and the reference star. Then, subtract the B value from the sum of the number of neighbors of the reference star to obtain the dispersion X. The specific calculation formula is as follows.

[0049]

[0050] The smaller the discreteness X of a certain navigation star, the more closely it matches the reference star, and the greater the possibility that the navigation star is the navigation star to be identified. The navigation stars with discreteness X less than or equal to the discreteness threshold ζ2 are determined as candidate navigation stars and classified into the candidate navigation star partition to prepare for the reference star identification in step 4, as shown in the following formula:

[0051] X≤ζ2 (10)

[0052] Furthermore, the number of neighbors of the primary star is also a valuable star map feature. Comparing the neighbor counts of the reference and guide stars allows for the rapid elimination of unqualified guide stars. Therefore, only the discreteness calculation is performed on guide stars whose neighbor counts fall within a certain threshold, which helps improve overall execution efficiency.

[0053] Example 2

[0054] Step 4: Reference Star Identification

[0055] See also Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 :

[0056] Step 4.1: Align the starting edge. Figure 2 Taking the star pattern of the navigation star in as an example, if S1 is the starting edge, the distance characteristic vector D and the angle characteristic vector Φ of the navigation star are as follows.

[0057]

[0058] Star sensor real shot star map Figure 3 For example (assuming there are two missing stars and three pseudo stars), then take any side, for example, take S5 as the starting side, then the d, As shown in the following formula.

[0059]

[0060] Based on the distance and angle information of the reference star Based on this, we traverse the angle feature vector Φ and distance feature vector D of all candidate navigation stars in turn, and filter out the ones containing The next step is to calculate the accumulated angle feature vector and similarity to identify the correct navigation star. In addition, when locating the centroid of the star point in the real star map, there are problems such as stray light, instrument noise, and low accuracy of the centroid algorithm, which makes the distance and angle information difficult to be accurately determined. There is a certain error, so in the traversal process, as long as the (D j ,Φ j ) satisfies the following constraints, that is, meets the matching conditions, and subsequent similarity calculation can be performed.

[0061]

[0062] Where, d i is the distance characteristic of the reference star, D j is the j-th distance feature of the navigation star, is the i-th angle characteristic of the reference star, Φj is the j-th angle feature of the navigation star, ζ3 is the distance threshold, and ζ4 is the angle threshold.

[0063] by Figure 2 、 Figure 3 For example, based on the characteristics of the reference star To illustrate this process. At this time, the angle feature vector Φ, All included Φ6 is the 6th position in Φ, in In the third position, Φ is circularly shifted 5 positions to the left. Shift left by 2 bits:

[0064]

[0065] Find Φ′, The cumulative angle feature vectors E and ε are as shown in the following formula, and the similarity P is calculated (the specific method is described later), and P = 10. It can be seen that in the harsh environment where there are two missing stars and three false stars in the real star map, as long as the angle feature vectors Φ and The starting position of the reference star is correctly matched, and the reference star and its corresponding correct guide star can still achieve a very good match, while the similarity of the remaining incorrect guide stars will be very low. The method of the present invention is highly robust to the problem of missing stars and false stars in real star maps.

[0066]

[0067] Step 4.2: The following describes the angle characteristic vector between the reference star and the navigation star. How to calculate the similarity between the two after aligning the starting edges (starting angles) (hereinafter Φ is considered to be aligned with the starting edge). Use the angle characteristic vectors of the reference star and the navigation star respectively Φ constructs the respective cumulative angle feature vector ε=(ε1,ε2,…,ε m ) and E=(E1,E2,…,E n ), where each element is defined as:

[0068]

[0069] Calculate the similarity P between the cumulative angle feature vector ε and E. The initial value of P is set to 0. The calculation method is as follows:

[0070] Step 4.2.1: ε i and E j In the example, the subscripts i and j start counting from 1, i.e., i=j=1.

[0071] Step 4.2.2: ε1 versus E j Traverse (j from 1 to n), if ε1 and Ε k (k∈1~n) is close, that is, it satisfies the cumulative angle constraint (18), then it is considered that ε1 and E k Matching, similarity P = 1, and let i = i + 1 = 2, j = k + 1, ε2 for E j Traverse (j from k+1 to n), ζ5 is the cumulative angle threshold;

[0072] |ε i -Ε j |≤ζ5 (17)

[0073] Step 4.2.3: If ε1 and E1~E n If none of them match, let i=i+1=2, j=1, ε2 is equal to E j Traverse (j from 1 to n), at this time P = 0;

[0074] Step 4.2.4: When i>m or j>n is satisfied, the cumulative angle feature vector matching ends. The similarity P at this time represents the number of successful cumulative angle feature vector matches between the reference star to be identified and a candidate navigation star.

[0075] Example 3

[0076] For the convenience of intuitive understanding, Figure 2 、 3 For example, given the specific angle values ​​(ignoring noise), the angle feature vector Φ of the navigation star S is (23, 10, 32, 25, 50, 40, 25, 30, 35, 10, 30, 50), and the cumulative angle feature vector E is (23, 33, 65, 90, 140, 180, 205, 235, 270, 280, 310, 360); the angle feature vector of the reference star S′ is Cumulative angle feature vector ε = (23, 65, 90, 115, 140, 180, 205, 222, 235, 255, 270, 310, 360). Calculate the similarity P between the cumulative angle feature vector ε and E according to steps 4.2.1 to 4.2.4, and we get P = 10. Figure 5 The matching between the cumulative angle feature vectors ε and E is specifically marked in FIG.

[0077] In addition, if the above method is not used to first set the angle feature grid Φ circular shift alignment, and then calculate the similarity P by finding the cumulative angle feature vector ε and E, and then directly use the angle feature vector Arrange Φ from small to large and calculate The angle feature vector Φ of the sorted navigation star S is (10, 10, 23, 25, 25, 30, 30, 32, 35, 40, 50, 50), and the angle feature vector of the reference star S′ is Calculate according to Step 1 to Step 4 The similarity P between Φ is P = 5. It is easy to see that the recognition ability of the method of the present invention is more advantageous and more robust.

[0078] Step 5: Rapid star map recognition based on attitude information

[0079] For the star sensor shooting at time k and k+1, the attitude of the star sensor during the two shootings has obviously changed, so the projection of the star on the CCD imaging plane will also move accordingly. In order to quickly obtain the centroid of the star point in the star map shot at time k+1 and avoid the recognition of the star map in the entire sky to improve the real-time performance, the short-term high-precision attitude change information provided by the inertial navigation can be used to predict the position of the star point in real time, see formula (2).

[0080] After obtaining the predicted coordinate values ​​of the star points on the CCD plane based on the attitude information provided by the inertial navigation system, star point extraction and centroid positioning operations can be performed within a small neighborhood centered around the predicted coordinate values. Rapid verification identification is performed on all extracted star points to prevent incorrect star point extraction. Once the actual centroid coordinates of the known star in the carrier coordinate system are obtained, attitude determination algorithms such as QUEST can be used to calculate the attitude transformation matrix between the carrier coordinate system and the celestial coordinate system at the current moment.

[0081] Stars with magnitudes less than 6 from the SAO J2000 catalog were selected as guide stars for the star map matching task. Performance was tested using a noise-free simulated star map. Furthermore, position noise, spurious stars, and magnitude noise were added during the simulation to analyze the algorithm's performance. The simulation parameters are shown in Table 1.

[0082] Table 1 Simulation parameters

[0083]

[0084] A simulation experiment was conducted by randomly generating 10,000 ideal simulated images. Performance tests were then conducted by adding varying intensities of position noise, magnitude noise, and numbers of pseudo-stars to these 10,000 ideal simulated star images to verify the reliability and robustness of the proposed method. The test results for the ideal simulated star images are shown in Table 2. A grid method was used for comparison.

[0085] In an ideal situation, there is no star position noise, magnitude noise, or pseudo-stars in the simulated image. Table 2 summarizes the performance of the two methods under ideal conditions.

[0086] Table 2 Algorithm performance under ideal conditions

[0087]

[0088]

[0089] The results in Table 2 show that the proposed method outperforms the grid method in recognition accuracy and maintains a relatively good average time consumption, but is slightly inferior to the grid method. Next, the robustness of the proposed method is tested by adding position noise, magnitude uncertainty, and pseudo-stars to simulated images.

[0090] Table 3 Algorithm performance with position noise

[0091]

[0092] Table 3 shows that the proposed method is less sensitive to position noise. This is primarily because the proposed method maps radial distance features onto a high-dimensional vector to generate a distance mapping vector. This, combined with the aid of an anti-interference code, reduces sensitivity to position noise to a certain extent. Consequently, this technique maintains a recognition rate above 96%.

[0093] Table 4 Algorithm performance with magnitude noise

[0094]

[0095] By matching the corresponding starting edges of the angle feature vectors and performing similarity calculations after cyclic shifting, the method of the present invention can still identify the corresponding navigation stars with high accuracy even when there are multiple missing stars, as shown in Table 4.

[0096] Table 5 Algorithm performance with pseudo-stars

[0097]

[0098]

[0099] When the star map contains pseudo stars, the recognition rate of the method of the present invention changes relatively steadily, and can still maintain a high level under the adverse condition of 4 pseudo stars. The reason why the recognition rate of the method of the present invention is high when pseudo stars are present is the same as the reason why the recognition rate is high when there are missing stars. In simple terms, even when there are multiple missing stars and pseudo stars, the method of the present invention can still ensure that the reference star and the navigation star have a high similarity P. For details, please refer to Figure 5 explanation.

[0100] In summary, the present invention first calculates the angle characteristic vector The distance feature vector d lays a good foundation for the subsequent processing steps; then, based on the distance feature vector d, the distance mapping vector λ containing the anti-interference code is calculated. For the navigation star whose number of neighboring stars meets the threshold range, the discreteness X of the distance mapping vectors λ and Λ between the reference star and the navigation star is calculated, and finally the candidate navigation star partition is obtained, which narrows the search range of the star and improves the processing speed; then, the starting edge of the feature vector of the navigation star and the reference star is found, and the angle feature vector is obtained based on the starting edge. Φ is cyclically shifted to obtain the new angle feature vector Φ′; Then, using the angle feature vector Φ′ calculates the cumulative angle feature vectors ε and E, and finds the similarity P between the reference star and the navigation star. The one with the largest similarity P is the correct navigation star. Finally, once the full-sky star map recognition is completed, the attitude information provided by the inertial navigation system can be used to quickly recognize the star map of subsequent shots, thereby improving the overall calculation speed of the system.

Claims

1. A partitioned fast star map recognition method based on attitude information, characterized by: The process includes five steps: feature extraction, navigation star library construction, candidate navigation star partitioning, reference star identification, and rapid star map recognition based on attitude information. The details are as follows: Step 1: Feature Extraction Using two rotation and translation invariant features - the Euclidean distance D between the neighboring star and the central navigation star i , the relative angle Φ between consecutive adjacent stars and the central navigation star i , and construct the star pattern of each navigation star based on these two features, let each navigation star be at the center of the field of view, and consider all the neighboring stars within the radius R to extract the above two features. The neighboring stars are recorded as S={S1,S2,…,S N }, the relationship between R and the star sensitive field of view FOV is: Where f is the focal length, ρ is the pixel size, and radius R is the maximum distance from the center; Step 2: Navigation star library construction In the navigation star library, each navigation star has an angle feature vector Φ, a distance feature vector D, and a distance mapping vector Λ. The position coordinates of each neighboring star and the navigation star in the image coordinate system are combined with Φ i and D i The expression of the angle characteristic vector Φ between the navigation star and the neighboring star is obtained by Φ=(Φ1, Φ2,…, Φ N ) and the distance feature vector D=(D1,D2,…,D N ), where Φ i and D i They are the adjacent angles of the two neighboring stars and the radial distances between the neighboring stars and the reference star when the reference star is located at the center of the CCD in the ideal star map. The above features are uniformly arranged in a counterclockwise manner. In addition, it is also necessary to construct the navigation star distance mapping vector It is used to shorten the navigation star library list and obtain the candidate navigation star partitions, where Λ i is the mapping value obtained from the ideal star map, N D is the dimension of the distance mapping vector. With the candidate navigation star partition, the angle feature vector Φ and the distance feature vector D are used to identify the unique navigation star corresponding to the reference star. Step 3: Candidate navigation star partition Step 3.1: For the star image taken by the star sensor, after the star point centroid positioning process is completed, the star point closest to the image center is selected as the reference star, and the angle feature vector between the reference star and the neighboring star is extracted. And the distance feature vector d=(d1,d2,…,d N ), then construct N D dimensional reference star distance mapping vector in, and d i are the adjacent angles between the two neighboring stars and the radial distances between the neighboring stars and the reference star when the reference star is near the center of the CCD in the real star map; i The mapping value is obtained from the actual star map; Step 3.2: Partition the star list in the navigation star library based on the distance mapping vector λ, that is, screen out the candidate navigation stars that meet the conditions to form a new area. In this way, the reference star identification process in step 4 only needs to be performed on the navigation stars in this area. When using image signals to solve the two-dimensional centroid coordinates of the star point, there will be problems such as stray light, instrument noise, and low accuracy of the centroid algorithm, which will cause a certain deviation in the centroid coordinates, thereby causing the radial distance d between the reference star and the neighboring star to be larger. i Deviation, T i The value calculation is inaccurate, T i Represents the mapping position of the i-th neighboring star. In order to improve the anti-noise performance, when calculating T i When d i / (R / N D ) satisfies the relevant constraints, then the T i The value is easily affected by noise and needs to be mapped to the distance vector Add anti-interference code, ζ1 is the anti-interference code threshold; Step 3.3: Calculate the reference star distance mapping vector Distance mapping vector to navigation star The discreteness X between them is determined, and the navigation stars with discreteness X less than or equal to the discreteness threshold ζ2 are determined as candidate navigation stars to obtain the candidate navigation star partition; Step 4: Reference Star Identification With the candidate navigation star partition, the angle characteristic vector of the reference star is used to And the distance feature vector d to identify the unique corresponding navigation star. The specific steps are as follows: Step 4.1: Due to the randomness of the rotation angle when the star sensor takes the star map, the starting edges of the reference star and the navigation star constructed feature vectors are inconsistent. In order to obtain the correct reference star cumulative angle feature vector ε and the navigation star cumulative angle feature vector E, and calculate the similarity P between the two, it is necessary to ensure that the vectors Φ, The first digit in the vector D and d are equal; the distance and angle information of the reference star are Based on the above, the angle feature vector Φ and distance feature vector D of the navigation star in the candidate navigation star partition are traversed in turn, and the candidate navigation star is selected. The navigation star is then calculated and the cumulative angle feature vector and similarity are calculated. Step 4.2: Characteristic vector of the angle between the reference star and the navigation star After aligning the starting edge, Φ uses the angle characteristic vector of the reference star and the navigation star respectively Φ constructs the respective cumulative angle feature vector ε=(ε1,ε2,…,ε m ) and E=(E1,E2,…,E n ), calculate the similarity between the cumulative angle feature vector ε and E, and identify the correct navigation star corresponding to the reference star; Step 5: Rapid star map recognition based on attitude information For the star sensor shooting at time k and k+1, the attitude of the two shots obviously changes, and the projection of the star on the CCD imaging plane will also move accordingly. In order to quickly obtain the predicted position of the centroid of the star point in the star map shot at time k+1, avoid full-sky star map recognition and improve the overall processing speed of the system, the short-term high-precision attitude change information provided by the inertial navigation is used to accurately predict the position of the star point in real time, as follows: Where A is the longitude of the star sensor's optical axis at time k+1, Q is the latitude of the star sensor's optical axis at time k+1, and θ is the rotation angle of the star sensor at time k+1. It is calculated based on the attitude information provided by the inertial navigation system. After obtaining the predicted coordinate values ​​of the star points on the CCD plane based on the attitude information provided by the inertial navigation system, star point extraction and centroid positioning operations are performed within a smaller neighborhood around the predicted coordinate values. All extracted star points are quickly verified to prevent incorrect extraction of star points. After obtaining the actual centroid coordinates of the known stars in the carrier coordinate system, the QUEST attitude determination algorithm is used to calculate the attitude conversion matrix between the carrier coordinate system and the celestial coordinate system at the current moment.

2. A method for rapid star map recognition based on attitude information according to claim 1, characterized in that: During step 1 "feature extraction", two rotation- and translation-invariant features - the Euclidean distance D between the neighboring star and the central navigation star i , the relative angle Φ between consecutive adjacent stars and the central navigation star i The calculation is as follows: Where x i and y i are the coordinates of the ith neighboring star in the image coordinate system, N is the total number of all neighboring stars in the field of view, x c and y c is the coordinate of the navigation star in the image coordinate system.

3. A method for rapid star map recognition based on attitude information according to claim 1, characterized in that: In step 2, "building the navigation star library", the distance feature vector D = (D1, D2, ..., D N ), the starting edge D i The selection is different, corresponding to the angle feature vector Φ and the distance feature vector D, only a circular shift is performed, which does not affect the essential characteristics of the vector. min Construct D and Φ for the starting edge, where D min =min{D1,D2,…,D N }, it does not matter which edge is chosen as the starting edge; according to the distance characteristic vector D = (D1, D2, ..., D N ), calculate the navigation star distance mapping vector Λ through the distance characteristic vector D; the radial distance D between the navigation star and the neighboring star i The maximum is R, and the distance mapping vector is an N D dimensional vector signal, N D The value is given by the following formula (5), where the value of k is determined by experimental testing. The value of each bit in the vector is 0 or 1. The specific value selection method is as follows: First, let N D The dimension distance feature vector is all 0, substitute D1 into the following formula (6) to get T1, then the T1th position of the distance mapping vector Λ is 1, and the D in the distance feature vector D is traversed in turn. i , i∈1~N, then we get N D dimensional navigation star distance mapping vector N D =2 k ;k∈Z (5) 4. A method for rapid star map recognition based on attitude information according to claim 1, characterized in that: In step 3, "selection and partitioning of candidate navigation stars", when adding anti-interference code, for the radial distance d that satisfies formula (7), i , in the T i and T i The +1 bit is set to 1, satisfying the radial distance d of formula (8) i , in the T i and T i -1 bit is set to 1; when the discreteness of the distance mapping vectors λ and Λ between the reference star and the navigation star is calculated later, the successful matching rate is guaranteed to be higher and the probability that the candidate navigation star in the partition contains the correct reference star is higher; an anti-interference code is introduced to improve the anti-noise performance; ζ1 is the anti-interference code threshold:

5. A method for rapid star map recognition based on attitude information according to claim 4, characterized in that: When calculating the discreteness of the distance mapping vectors λ and Λ between the reference star and the navigation star, the reference star distance mapping vector is calculated. Distance mapping vector to navigation star When calculating the discreteness X between the reference stars and the reference stars, first perform a bitwise AND operation on λ and Λ and accumulate the sum to obtain the B value. The B value represents the number of neighboring stars that can match the reference star and the reference star. Then, subtract the B value from the sum of the number of neighboring stars of the reference star to obtain the discreteness X. The specific calculation formula is as follows:

6. A method for rapid star map recognition based on attitude information according to claim 4, characterized in that: When calculating the discreteness of the distance mapping vectors λ and Λ between the reference star and the navigation star, the smaller the discreteness X of the navigation star, the more closely it matches the reference star and the navigation star, and the greater the possibility that the navigation star is the navigation star to be identified. The navigation stars with discreteness X less than or equal to the discreteness threshold ζ2 are determined as candidate navigation stars and classified into the candidate navigation star partition to prepare for the reference star identification in step 4, as shown in the following formula: X≤ζ2 (10).

7. A method for rapid star map recognition based on attitude information according to claim 1, characterized in that: In step 4.1, "Align the starting edge," if S1 is the starting edge, the distance characteristic vector D and angle characteristic vector Φ of the navigation star are as follows: Assuming that there are two missing stars and three pseudo stars in the star map taken by the star sensor, take any side and take S5 as the starting side, then the d, As shown in the following formula:

8. A method for rapid star map recognition based on attitude information according to claim 1, characterized in that: In step 4.1 "Align the starting edge", use the distance and angle information of the reference star to align the Based on this, we traverse the angle feature vector Φ and distance feature vector D of all candidate navigation stars in turn, and filter out the ones containing The navigation star is then accumulated and the similarity calculation is performed to identify the correct navigation star. When locating the centroid of the star point in the real star map, there are problems such as stray light, instrument noise, and low accuracy of the centroid algorithm. The distance and angle information are not consistent. There is a certain error. In the traversal process, as long as the (D j ,Φ j ) satisfies the constraints of the following formula (13), that is, meets the matching conditions, and then the subsequent similarity calculation is performed: Where, d i is the distance characteristic of the reference star, D j is the j-th distance feature of the navigation star, is the i-th angle characteristic of the reference star, Φ j is the jth angle feature of the navigation star, ζ3 is the distance threshold, ζ4 is the angle threshold; the angle feature vector Φ, All included Φ6 is the 6th position in Φ, in In the third position, Φ is circularly shifted 5 positions to the left. Shift left by 2 bits: Find Φ′, The cumulative angle feature vectors Ε, ε, calculate the similarity P, as long as the angle feature vectors Φ, The starting position of the reference star is matched correctly, and a very good matching effect can still be obtained between the reference star and its corresponding correct navigation star, which has strong robustness:

9. A method for rapid star map recognition based on attitude information according to claim 1, characterized in that: Step 4.2 "Angle characteristic vector between reference star and navigation star After aligning the starting edge, Φ is used to calculate the angle characteristic vectors of the reference star and the navigation star. Φ constructs the respective cumulative angle feature vector ε=(ε1,ε2,…,ε m ) and E=(E1,E2,…,E n ), where each element is defined as: Calculate the similarity P between the cumulative angle feature vector ε and E. The initial value of P is set to 0. The calculation method is as follows: Step 4.2.1: ε i and E j In the above equation, the subscripts i and j start counting from 1, i.e., i=j=1; Step 4.2.2: ε1 versus E j Traverse, j from 1 to n, if ε1 and Ε k Close, k∈1~n, that is, satisfying the cumulative angle constraint (17), then it is considered that ε1 and E k Matching, similarity P = 1, and let i = i + 1 = 2, j = k + 1, ε2 for E j Traverse, j from k+1 to n, ζ5 is the cumulative angle threshold; |e i -E j |≤ζ5 (17) Step 4.2.3: If ε1 and E1~E n If none of them match, let i=i+1=2, j=1, ε2 is equal to E j Traverse, j from 1 to n, at this time P = 0; Step 4.2.4: When i>m or j>n is satisfied, the cumulative angle feature vector matching ends. The similarity P at this time represents the number of successful cumulative angle feature vector matches between the reference star to be identified and a candidate navigation star.

Images