A target recognition method applicable to space rendezvous and docking

Abstract

The present invention discloses a target recognition method applicable to space rendezvous and docking, which solves the problems of the existing space rendezvous and docking target recognition methods, namely, slow recognition speed and inaccurate recognition results. The present invention utilizes the data-driven ability of a lightweight neural network to obtain an image with an elliptical cross configuration target as the main body. After removing noise from the image using Gaussian filtering, the LSD algorithm is used for line segment detection. The line segments are grouped based on the angular differences between the line segments, and the lightweight neural network is used to effectively eliminate the ellipses with excessive errors, reducing the computational amount. The initial ellipse set is obtained by adopting the elliptical arc support algorithm. After clustering and fitting the initial ellipse set, an optimal candidate ellipse set is generated. It has the advantages of simplicity, high efficiency and good repeatability. By integrating the lightweight neural network based on data driving, the computational complexity is significantly reduced, the accuracy of ellipse detection is improved, and the speed and accuracy of target recognition are correspondingly improved.

Description

Technical Field

[0001] The present invention relates to a method for identifying a target for space rendezvous and docking, and more particularly to a target identification method applicable to space rendezvous and docking. Background Art

[0002] Identifying space targets through a monocular camera is very common in rendezvous and docking missions. Fast and accurate target identification plays a crucial role in the success or failure of rendezvous and docking. In existing space rendezvous and docking target identification methods, the cooperative target is usually designed as an elliptical cross target. Taking the Hough transform ellipse detection algorithm as an example, this algorithm can effectively detect an ellipse by checking all pairs of feature points on the major axis and using the Hough transform to obtain the minor axis.

[0003] However, since this type of algorithm needs to use feature points as calculation units and calculate the distances between individual feature point pairs separately, it consumes a huge amount of memory, resulting in a slow speed of space rendezvous and docking target identification and an inaccurate identification result. Summary of the Invention

[0004] The object of the present invention is to solve the problem that the existing space rendezvous and docking target identification method has a slow identification speed and an inaccurate identification result, and to provide a target identification method applicable to space rendezvous and docking.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] The present invention provides a target identification method applicable to space rendezvous and docking, which is characterized in that it includes the following steps:

[0007] Step 1: Design an elliptical cross configuration target and install it at the relative position of the docking mechanism. Image the elliptical cross configuration target through a camera to obtain an image of the elliptical cross configuration target;

[0008] Step 2: Use a trained lightweight neural network to detect the initial position box of the elliptical cross configuration target in the image of the elliptical cross configuration target, and obtain a to-be-detected image with the main body being the elliptical cross configuration target and the size being twice that of the initial position box;

[0009] Step 3: Extract edge information from the to-be-detected image to obtain an edge image composed of multiple line segments;

[0010] Step 4: Use the LSD algorithm to detect line segments in the edge image to obtain a line segment set, group the line segments based on the angular difference of the line segments, and output the line segment sets of each group;

[0011] Step 5: Traverse each line segment in the line segment sets of each group, and extract an initial ellipse set from the line segment sets of each group in combination with the ellipse arc support algorithm;

[0012] Step 6: Use the trained lightweight neural network to detect the image to be detected, obtain the real-time position box of the elliptical cross configuration target, obtain the inscribed ellipse according to the real-time position box, and remove the ellipses with large deviations from the inscribed ellipse in the initial ellipse set to obtain the updated initial ellipse set;

[0013] Step 7: Perform an average calculation on the parameters of the ellipses in the real-time position box and the updated initial ellipse set to obtain the final candidate ellipse, and use the final candidate ellipse to identify the elliptical cross configuration target of the docking mechanism to complete the target recognition of space rendezvous and docking.

[0014] Further, the shape of the elliptical cross configuration target described in step 1 is elliptical, the main color is black, and a plurality of white square grids are arranged at intervals on the outer edge, which can make the detection effect more significant, and the middle part is in the shape of a white cross.

[0015] Further, the lightweight neural network described in step 2 is the YOLO_X or SSD lightweight neural network.

[0016] Further, the trained lightweight neural network described in step 2 is: using no less than 500 images of the elliptical cross configuration target, taking the initial position box in each elliptical cross configuration target image as a label, using the elliptical cross configuration target image and the corresponding label as training samples to make a sample training set, and training the lightweight neural network to obtain the lightweight neural network.

[0017] Further, step 3 is specifically: using Gaussian filtering to filter out noise from the image to be detected, performing gradient calculation and non-maximum suppression on the image, and then using the Sobel operator or Canny operator to extract edge information to obtain an edge image composed of multiple line segments.

[0018] Further, the specific process of step 4 is:

[0019] Use the LSD algorithm to detect the line segments in the edge image to obtain the set of endpoint coordinates of the line segments, calculate the direction angle of each line segment relative to the horizontal line, and normalize it to between 0 and 180 degrees. According to the normalized angle, use the complete-link method of hierarchical clustering to group with X degrees as the threshold, and the angle difference between any two line segments in the same group does not exceed X degrees, and output the set of line segments and the quantity of each group.

[0020] Further, in step 4, X takes a value of 14 - 18, and the specific process of using the complete-link method of hierarchical clustering to group with X degrees as the threshold is:

[0021] Step a: The direction angle of each line segment in the edge image relative to the horizontal line is , if is the regular value and remains unchanged, if If it is negative, add 180 degrees, with the direction from the starting point to the ending point of the line segment as the standard, and number each line segment 1, 2, 3,... N , N The value ranges from 7 to 10;

[0022] Step b: Calculate the angular differences between every two of all the line segments , where represents the direction angle of the i th line segment relative to the horizontal line, represents the direction angle of the j th line segment relative to the horizontal line; i and j both range from 1, 2, 3,... N ;

[0023] Step c: Take each line segment as an independent cluster C k = {line segment k}, k is the number of the line segment, k = 1, 2, 3,... N , forming the initial cluster set ;

[0024] Step d: Perform complete-linkage hierarchical merging on the independent clusters in the initial cluster set C with X degrees as the threshold to obtain a new cluster set;

[0025] Step e: Output each new cluster in the new cluster set as the line segment set of each group.

[0026] Furthermore, the specific process of step d is as follows:

[0027] ① Traverse the current initial cluster set C, and take any two clusters C a and C b ;

[0028] ② Define the inter-cluster distance as the maximum angular difference between the line segment pairs formed by the line segments in the two clusters, and calculate the inter-cluster distance which is expressed as:

[0029] ;

[0030] ③ If is satisfied, merge and into a new cluster , update the initial cluster set , and return to ①;

[0031] If , select a cluster from the current initial cluster set C C c Replace the cluster C a and C b any one of them; return to ② until there is no cluster pair that satisfies and can be merged to obtain a new cluster set.

[0032] Furthermore, the specific process of step 5 is as follows:

[0033] Step 5.1: For each line segment in the group, extract the endpoints and sample a number of intermediate points at equal intervals along the line segment to form the sampling point coordinate set P of this group, and perform robust fitting on the sampling point coordinate set P using the RANSAC strategy;

[0034] Step 5.2: Randomly select 5 points from the fitting result in step 5.1 to generate a candidate ellipse; calculate the number of points among the 5 points whose distance to the candidate ellipse is less than the threshold τ respectively;

[0035] Step 5.3: Return to step 5.2, randomly select 5 points again to generate a candidate ellipse and calculate the number of points among the 5 points whose distance to the candidate ellipse is less than the threshold τ. After 5 - 10 calculations, retain the candidate ellipse with the largest number of points less than the threshold τ;

[0036] Step 5.4: Calculate the geometric distance from each line segment sampling point to the candidate ellipse retained in step 5.3 respectively. If the average geometric distance is less than the threshold τ, it is regarded as supported, otherwise it is not supported; at the same time, calculate the included angle between each line segment and the tangent of the corresponding point of the candidate ellipse retained in step 5.2. If it is less than the angle threshold θ, it is regarded as supported, otherwise it is not supported. The line segments that satisfy both of the above supports are used as the enhanced support line segments in this group;

[0037] Step 5.5: Count the number of enhanced support line segments. If it exceeds the threshold N_min, it is regarded as valid and execute step 5.6, otherwise it is regarded as invalid and return to step 5.2;

[0038] Step 5.6: Use the candidate ellipses retained in each group in step 5.3 as the optimal candidate ellipse set;

[0039] Step 5.7: Use the optimal candidate ellipse set as the initial ellipse set.

[0040] Furthermore, the specific method of removing the ellipses with large deviation from the inscribed ellipse in the initial ellipse set in step 6 is as follows: Calculate the distance between the center point of the inscribed ellipse and the center points of the ellipses in the initial ellipse set respectively, and the angle formed by the connection line between the center point of the inscribed ellipse and the center points of the ellipses in the initial ellipse set and the horizontal direction. Remove the ellipses whose center point distance is greater than 20% of the major axis of the inscribed circle ellipse and the ellipses with an angle not less than 5 degrees, and update the initial ellipse set.

[0041] Advantages of the present invention:

[0042] In a target recognition method applicable to space rendezvous and docking provided by the present invention, after designing, installing, and imaging an elliptical cross-shaped target, the target is effectively recognized by using a fast and accurate ellipse detection algorithm, providing support for the successful realization of rendezvous and docking.

[0043] In a target recognition method applicable to space rendezvous and docking provided by the present invention, an image with an elliptical cross-shaped target as the main body is obtained. After removing noise from the image using Gaussian filtering, the edge image can be effectively extracted; after detecting line segments using the LSD algorithm, the line segments are grouped based on the angular differences between the line segments, and a lightweight neural network is used to effectively eliminate ellipses with excessive errors, reducing the computational complexity of subsequent processing. The initial ellipse set is obtained using the ellipse arc support algorithm, and the initial ellipse set is clustered and fitted to generate an optimal candidate ellipse set, and the docking mechanism is recognized using the final candidate ellipses. The method proposed by the present invention is simple, efficient, and has good repeatability, integrating a data-driven lightweight neural network. Compared with the traditional ellipse detection algorithm based on the Hough transform, the computational complexity is significantly reduced, the accuracy of ellipse detection is improved, and the speed and accuracy of target recognition are correspondingly improved. Description of the drawings

[0044] Figure 1 is a flowchart of an embodiment of a target recognition method applicable to space rendezvous and docking of the present invention;

[0045] Figure 2 is a schematic diagram of a simulated elliptical cross-shaped target in an embodiment of a target recognition method applicable to space rendezvous and docking of the present invention;

[0046] Figure 3 is an effect diagram of detecting a simulated elliptical cross-shaped target in an embodiment of a target recognition method applicable to space rendezvous and docking of the present invention;

[0047] Figure 4 is a real elliptical cross-shaped target image in an embodiment of a target recognition method applicable to space rendezvous and docking of the present invention;

[0048] Figure 5 is an effect diagram of detecting a real elliptical cross-shaped target in an embodiment of a target recognition method applicable to space rendezvous and docking of the present invention;

[0049] Figure 6 is a schematic diagram of the results of detecting multiple ellipses in an embodiment of a target recognition method applicable to space rendezvous and docking of the present invention. Detailed implementation manners

[0050] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work belong to the scope of protection of the present invention.

[0051] This embodiment provides a target recognition method applicable to space rendezvous and docking, as Figure 1 shown, including the following steps:

[0052] Step 1: Design an elliptical cross-shaped target. The shape of the elliptical cross-shaped target is elliptical, the main color is black, multiple white square grids are arranged at intervals on the outer edge, and the middle is a white cross shape; and install the elliptical cross-shaped target at the relative position of the docking mechanism, image the elliptical cross-shaped target through a camera, and obtain an image of the elliptical cross-shaped target.

[0053] Step 2: Use no less than 500 images of the elliptical cross-shaped target, and use the initial position box in each image of the elliptical cross-shaped target as a label. Take the image of the elliptical cross-shaped target and the corresponding label as training samples to make a sample training set, and train the YOLO_X or SSD lightweight neural network to obtain a trained YOLO_X or SSD lightweight neural network; use the trained YOLO_X or SSD lightweight neural network to detect the initial position box of the elliptical cross-shaped target in the image of the elliptical cross-shaped target, and obtain a to-be-detected image with the main body being the elliptical cross-shaped target and the size being 2 times that of the initial position box.

[0054] Step 3: Use Gaussian filtering on the to-be-detected image to filter out noise, perform gradient calculation and non-maximum suppression on the image, and then use the Sobel operator or Canny operator to extract edge information to obtain an edge image composed of multiple line segments.

[0055] Step 4: Use the LSD algorithm (Line Segment Detection algorithm) to detect the line segments in the edge image, obtain a set of endpoint coordinates of the line segments, calculate the direction angle of each line segment relative to the horizontal line, and normalize it to between 0 and 180 degrees. According to the normalized angle, use the complete-link method of hierarchical clustering, group with 15 degrees as the threshold, and the angle difference between any two line segments in the same group does not exceed 15 degrees, and output the set of line segments and the quantity of each group.

[0056] In this embodiment, using the complete-link method of hierarchical clustering and grouping with 15 degrees as the threshold is specifically:

[0057] Step a: The direction angle of each line segment in the edge image relative to the horizontal line is , if is regular, the value remains unchanged. If If it is negative, add 180 degrees, with the direction from the starting point to the ending point of the line segment as the reference, and number each line segment as 1, 2, 3,..., 10;

[0058] Step b: Calculate the angular differences between every two of all the line segments , where represents the direction angle of the i -th line segment relative to the horizontal line, represents the direction angle of the j -th line segment relative to the horizontal line; i and j both range from 1, 2, 3,..., 10;

[0059] Step c: Take each line segment as an independent cluster C k = {line segment k}, k is the number of the line segment, k = 1, 2, 3,..., 10, to form the initial cluster set ;

[0060] Step d: Perform complete-linkage hierarchical merging on the independent clusters in the initial cluster set C with a threshold of 15 degrees to obtain a new cluster set; the specific process is as follows:

[0061] ① Traverse the current initial cluster set C, and take any two clusters C a and C b ;

[0062] ② Define the inter-cluster distance as the maximum angular difference between the line segment pairs formed by the line segments within the two clusters, and calculate the inter-cluster distance is expressed as:

[0063] ;

[0064] ③ If is satisfied, merge and into a new cluster , update the initial cluster set , and return to ①;

[0065] If is not satisfied, select a cluster C c in the current initial cluster set C to replace either C a and C b ; return to ② until there are no cluster pairs that satisfy to be merged, and a new cluster set is obtained.

[0066] Step e: Output each new cluster in the new cluster set as the line segment set of each group, and the number of line segments in the new cluster as the number of line segments of each group.

[0067] Step 5. Traverse the line segments of each group, and extract the initial ellipse set from the line segments of the group in combination with the ellipse arc support algorithm; the specific process is as follows:

[0068] Step 5.1. For each line segment in the group, extract the endpoints and sample a number of intermediate points at equal intervals along the line segment to form the sampling point coordinate set P of the group, and perform robust fitting on the sampling point coordinate set P using the RANSAC strategy.

[0069] Step 5.2. Randomly select 5 points from the fitting result in Step 5.1 to generate a candidate ellipse; calculate the number of points whose distances from the candidate ellipse are less than the threshold τ for the 5 points respectively.

[0070] Step 5.3. Return to Step 5.2, randomly select 5 points again to generate a candidate ellipse and calculate the number of points whose distances from the candidate ellipse are less than the threshold τ. After 5 - 10 calculations, retain the candidate ellipse with the largest number of points whose distances are less than the threshold τ.

[0071] Step 5.4. Calculate the geometric distances from the sampling points of each line segment to the candidate ellipse retained in Step 5.3 respectively. If the average geometric distance is less than the threshold τ, it is regarded as supported, otherwise it is not supported; at the same time, calculate the included angle between each line segment and the tangent of the corresponding point of the candidate ellipse retained in Step 5.2. If it is less than the angle threshold θ, it is regarded as supported, otherwise it is not supported. The line segments that satisfy both of the above supports are used as the enhanced support line segments in the group.

[0072] Step 5.5. Count the number of enhanced support line segments. If it exceeds the threshold N_min, it is regarded as valid and execute Step 5.6, otherwise it is regarded as invalid and return to Step 5.2.

[0073] Step 5.6. Use the candidate ellipses retained in each group in Step 5.3 as the optimal candidate ellipse set.

[0074] Step 5.7. Use the optimal candidate ellipse set as the initial ellipse set.

[0075] Step 6. Use the trained lightweight neural network to detect the image to be detected, obtain the real-time position box of the elliptical cross configuration target and the inscribed ellipse of the real-time position box, calculate the distance between the center point of the inscribed ellipse and the center point of the ellipse in the initial ellipse set respectively, and the angle formed by the connection line between the center point of the inscribed ellipse and the center point of the ellipse in the initial ellipse set and the horizontal direction. Eliminate the ellipses whose center point distances are greater than 20% of the major axis of the inscribed circle ellipse and the ellipses whose angles are not less than 5 degrees to obtain the updated initial ellipse set.

[0076] Step 7: Calculate the average of the real-time position box and the parameters of the ellipses in the updated initial ellipse set to obtain the final candidate ellipses, and use the final candidate ellipses to identify the elliptical cross configuration target (target) of the docking mechanism, thus completing the target recognition for space rendezvous and docking.

[0077] Figure 2 is the simulated elliptical cross configuration target in this embodiment. Figure 3 is the effect diagram of identifying the simulated elliptical cross configuration target using the target recognition method of this embodiment. In the dynamic video of the simulated elliptical cross configuration target containing 500 frames, the recognition recall rate reaches 100%, the recognition accuracy reaches 98%, and the frames per second reach 20. Figure 4 is the publicly available real elliptical cross configuration target. Figure 5 is the effect of identifying the real target using the method of the present invention. In the dynamic video of the real elliptical cross configuration target containing 1513 frames, the recognition recall rate reaches 99%, the recognition accuracy reaches 97%, and the frames per second reach 26. Based on the traditional Hough transform ellipse detection algorithm, the recognition recall rate is 95%, the recognition accuracy is 87%, and the frames per second is 0.7.

[0078] Figure 6 is the schematic diagram of the result of detecting multiple ellipses in the embodiment of the target recognition method for space rendezvous and docking according to the present invention. Figure 6 The red color represents the detected ellipses, from which it can be seen that the target recognition method of the present invention is applicable to target recognition in multi-ellipse scenarios.

[0079] Through the verification implementation of the method of the present invention using the dynamic video of the simulated target and the dynamic video of the real target, the target recognition method proposed by the present invention can, compared with the traditional Hough transform ellipse detection algorithm, achieve fast and accurate recognition of the space rendezvous and docking target, which has a supporting effect on the completion of the rendezvous and docking task.

[0080] The above is only the specific implementation manner of the present invention and the comparison of the effects of the relevant specific implementation manners. However, the protection scope of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A target recognition method suitable for space rendezvous and docking, characterized in that: The following steps are involved: Step 1, designing an elliptical cross configuration target and installing it at a relative position of the docking mechanism, imaging the elliptical cross configuration target through a camera, and obtaining an image of the elliptical cross configuration target; Step 2: Using a trained lightweight neural network to detect the initial position frame of the elliptical cross configuration target in the elliptical cross configuration target image, and obtaining an image to be detected whose main body is the elliptical cross configuration target and whose size is twice that of the initial position frame; Step 3: extract edge information from the image to be detected to obtain an edge image composed of multiple line segments; Step 4: Use the LSD algorithm to perform line segment detection on the edge image to obtain a line segment set, group the line segments based on their angle differences, and output the line segment sets of each group; the specific process is as follows: Use the LSD algorithm to detect line segments in the edge image, obtain the endpoint coordinate set of the line segments, calculate the direction angle of each line segment relative to the horizontal line, and normalize it to between 0 and 180 degrees. According to the normalized angle, use the full-link method of hierarchical clustering to group with X degrees as the threshold, and the angle difference between any two line segments in the same group does not exceed X degrees, and output the line segment set and number of each group; the value range of X is 14-18; Step 5: traverse each line segment in each grouped line segment set, and extract the initial ellipse set from each grouped line segment set in combination with the ellipse arc support algorithm; the specific process is as follows: Step 5.1, for each line segment in the group, extract the endpoints and sample several intermediate points at equal intervals along the line segment to form a sampling point coordinate set P of the group, and use the RANSAC strategy to perform robust fitting on the sampling point coordinate set P; Step 5.2: Randomly select 5 points from the fitting results of step 5.1 to generate candidate ellipses; calculate the number of points whose distances from the 5 points to the candidate ellipses are less than the threshold value τ; Step 5.3, return to step 5.2, randomly select 5 points to generate candidate ellipses again, and calculate the number of points whose distances from the 5 points to the candidate ellipse are less than the threshold τ. After 5-10 calculations, retain the candidate ellipse with the largest number of points less than the threshold τ; Step 5.4: Calculate the geometric distances from each line segment sampling point to the candidate ellipse retained in step 5.

3. If the average geometric distance is less than the threshold value τ, it is considered supported, otherwise it is not supported. At the same time, calculate the angle between each line segment and the tangent line of the corresponding point of the candidate ellipse retained in step 5.

2. If it is less than the angle threshold θ, it is considered supported, otherwise it is not supported. The line segment that satisfies both of the above supports is regarded as the enhanced support line segment in the group. Step 5.5: Count the number of enhanced support segments. If it exceeds the threshold N_min, it is considered valid and executes step 5.

6. Otherwise, it is considered invalid and returns to step 5.

2. Step 5.6, taking the candidate ellipses retained by each group in step 5.3 as the optimal candidate ellipse set; Step 5.7, taking the optimal candidate ellipse set as the initial ellipse set; Step 6: Use the trained lightweight neural network to detect the image to be detected, obtain the real-time position frame of the elliptical cross configuration target, obtain the inscribed ellipse according to the real-time position frame, remove the ellipse with a large deviation from the inscribed ellipse in the initial ellipse set, and obtain the updated initial ellipse set; Step 7: average the parameters of the real-time position frame and the ellipse in the updated initial ellipse set to obtain the final candidate ellipse, and use the final candidate ellipse to identify the elliptical cross configuration target of the docking mechanism to complete the target identification of the space rendezvous and docking.

2. According to claim 1, a target recognition method suitable for space rendezvous and docking is characterized in that: The shape of the elliptical cross configuration target described in step 1 is elliptical, the main body color is black, a plurality of white square grids are arranged at intervals on the outer edge, and the middle part is a white cross shape.

3. According to claim 1, a target recognition method suitable for space rendezvous and docking is characterized in that: The lightweight neural network described in step 2 is a YOLO_X or SSD lightweight neural network.

4. A target recognition method suitable for space rendezvous and docking according to claim 3, characterized in that: The trained lightweight neural network described in step 2 is: using an elliptical cross configuration target image containing no less than 500 images, and using the initial position box in each elliptical cross configuration target image as a label, using the elliptical cross configuration target image and the corresponding label as training samples, making a sample training set, and training the lightweight neural network to obtain a lightweight neural network.

5. The target recognition method suitable for space rendezvous and docking according to claim 1, characterized in that: Step 3 is as follows: use Gaussian filtering to filter out noise from the image to be detected, perform gradient calculation and non-maximum suppression on the image, and then use the Sobel operator or Canny operator to extract edge information to obtain an edge image composed of multiple line segments.

6. The target recognition method suitable for space rendezvous and docking according to claim 1, characterized in that: In step 4, the full-link method using hierarchical clustering is used to group with X degree as the threshold, specifically: Step a: The direction angle of each line segment in the edge image relative to the horizontal line is θ. If θ is positive, the value remains unchanged. If θ is negative, it is increased by 180 degrees. The direction from the starting point to the end point of the line segment is used as the standard, and each line segment is numbered 1, 2, 3, ..., N, where N is 7-10; Step b: Calculate the angle difference Δθ between all line segments ij =min(|θ i -θ j ∣,180°-∣θ i -θ j ∣), where θ i Represents the direction angle of the i-th line segment relative to the horizontal line, θ j Indicates the direction angle of the jth line segment relative to the horizontal line; the value range of i and j is 1, 2, 3, ..., N; Step c: Treat each line segment as an independent cluster C k = {segment k}, k is the number of the segment, k = 1, 2, 3, ..., N, forming an initial cluster set C = {C1, C2, ..., C N }; Step d: Perform full-link hierarchical merging of independent clusters in the initial cluster set C with degree X as the threshold to obtain a new cluster set; Step e: Output each new cluster in the new cluster set as a set of line segments for each group.

7. A target recognition method suitable for space rendezvous and docking according to claim 6, characterized in that: The specific process of step d is: ① Traverse the current initial cluster set C and take any two clusters C in the initial cluster set C a and C b ; ② Define the inter-cluster distance as the maximum angle difference between the line segments formed by the line segments in the two clusters, and calculate the inter-cluster distance d(C a ,C b ) is expressed as: d(C a ,C b )=maxΔθ ij (i∈C a ,j∈C b ); ③If d(C a ,C b )≤15°, combined C a and C b The new cluster C a∪b , update the initial cluster set C, and return to ①; If d(C a ,C b )≤15°, select cluster C from the current initial cluster set C c Replace Cluster C a and C b Any of; Return to ② until no d(C a ,C b )≤15° can be merged to obtain a new cluster set.

8. The target recognition method suitable for space rendezvous and docking according to claim 1, characterized in that: The specific method of eliminating ellipses with large deviations from the inscribed ellipse in the initial ellipse set in step 6 is as follows: respectively calculate the distance between the center point of the inscribed ellipse and the center point of the ellipse in the initial ellipse set, and the angle formed by the line connecting the center point of the inscribed ellipse and the center point of the ellipse in the initial ellipse set and the horizontal direction, eliminate ellipses whose center point distance is greater than 20% of the major axis of the inscribed ellipse and ellipses whose angle is not less than 5 degrees, and update the initial ellipse set.

Images