Space-based optical space target detection method fusing trajectory information

Abstract

A space-based optical space target detection method fusing track information comprises the following steps: acquiring an image spot position of a point target in a star map of a current position by using a space-based optical camera, and determining a candidate target set according to different motion characteristics of a fixed star and a space target; determining the center-of-mass coordinate and pixel range of the candidate target under the image plane coordinate system according to the orientation of the optical camera, converting the center-of-mass coordinate and pixel range into the sight angle and range under the satellite body coordinate system, carrying out inter-frame trajectory association on the obtained star map sequence, updating the space target library according to trajectory information, and outputting a space target detection result. And repeating the steps to traverse all star maps.

Description

Technical Field

[0001] This invention relates to a space-based optical space target detection method that integrates trajectory information, belonging to the field of space situational awareness. Background Technology

[0002] With the normalization of space exploration activities, the number of spacecraft in Earth's space is increasing daily, and the number of space targets brought by these spacecraft has reached millions, seriously threatening the normal operation of spacecraft in orbit. Reliable detection of space targets is of great significance for maintaining the sustainable development of space activities. Space-based optics is one of the main means of space target detection, with advantages such as long operating range, low power consumption, and low cost.

[0003] Capturing faint space targets typically requires highly sensitive detectors. The presence of numerous stars within the field of view makes space targets easily obscured. Currently, the main methods for target detection based on space-based optical imaging are star catalog projection and inter-frame difference methods. The sheer number of stars makes traditional methods computationally inefficient and slow. Furthermore, the adhesion between stars and targets makes threshold setting difficult, resulting in high false alarm rates or high false negative rates. During space-based optical camera detection, the camera's observations of the same target are discontinuous. Currently, space-based optical target detection focuses on extracting targets from a dense stellar background, without distinguishing between different targets. Distinguishing between different targets relies on orbital information, which is computationally inefficient and difficult to use in high real-time scenarios such as target tracking. Summary of the Invention

[0004] The technical problem solved by this invention is to address the shortcomings of existing space target detection methods, such as low precision and recall and discontinuous target observations. This invention proposes a space-based optical space target detection method that fuses trajectory information.

[0005] The present invention solves the above-mentioned technical problem through the following technical solution: A space-based optical target detection method that integrates trajectory information includes: The star map of the current location is detected by a space-based optical camera, and star map preprocessing and star map recognition are performed to obtain the image spot positions of all point targets; The obtained image spots are subjected to coarse stellar removal to obtain a candidate target set containing space targets and residual stars; Based on the orientation of the space-based optical camera, the centroid coordinates and pixel range of the candidate target in the image plane coordinate system are converted into the line-of-sight angle and range of the candidate target in the satellite body coordinate system. Perform inter-frame trajectory correlation on the star map sequence; Based on the inter-frame trajectory correlation, the line-of-sight angle and range of candidate targets in the satellite body coordinate system, the candidate target set is updated, and missed targets are added and residual stars are removed based on the inter-frame trajectory correlation. The updated space target library is used as the space target detection result. Repeat all judgment steps to detect space targets in the next frame of the star map.

[0006] The space-based optical camera detects targets through either stationary detection or continuous detection. Stationary detection is as follows: Once the space-based optical camera moves to the preset position, it stops moving until it stabilizes before starting to take pictures; Continuous detection is as follows: The space-based optical camera moves at a low speed range and takes pictures simultaneously. During the detection process, there is an overlap area between two consecutive frames of images, and the overlap area meets the preset requirements.

[0007] The methods for star map preprocessing and star map recognition are as follows: The obtained star map is filtered and denoised using Gaussian filtering, and motion blur is restored using Wiener filtering. Set a background threshold to perform threshold segmentation on the processed star map, and obtain a binary image in which the image spots of stars and space objects are separated from the background noise. By scanning binary images using the centroid method and connected component analysis, the centroid positions and circumscribed rectangles of each image spot are obtained.

[0008] The binary image, in which the image spots of the stars and space objects are separated from the background noise, is as follows:

[0009] In the formula, This is an adaptive threshold calculated by taking the mean and variance of pixel values; The centroid method is used to scan the image and then calculate the first moment of the grayscale values ​​of the image after thresholding. The calculation method is as follows:

[0010] In the formula, the parameter appearing for the first time must be explained; Connectivity analysis is used to separate a single point target from other point targets. The separation process includes: The image pixels are scanned one by one from left to right and from top to bottom. According to the set threshold, when a foreground pixel is scanned, the connected component is determined and the image is marked according to the selected 8 connectivity criteria. The outer rectangle is a rectangle that can completely cover the target image patch.

[0011] Based on the different motion characteristics of stars and space targets, the following method is used to coarsely eliminate stars and obtain the candidate target set using dilatational calculus: Based on the sum of squared errors of the relevant sub-images, registration is performed between the previous frame and the current frame according to the characteristics of stellar motion. The registration method is as follows:

[0012] In the formula, and The displacement between the previous frame image and the current frame image is used; the registration method creates a sub-graph by analyzing the position of the set of spots obtained from the connected component analysis in the previous frame, and uses multiple sub-graphs to calculate registration information with high confidence based on clustering information to simplify the calculation; The previous frame image was horizontally shifted. Pixels, vertical translation The pixels are registered with the current frame image with a registration accuracy of 1 pixel.

[0013] Dilation difference is performed on the two registered images. When the difference between pixels in the two image frames is greater than a threshold, the current pixel is considered to be moving. The method of coarse star removal using dilation difference is as follows:

[0014] In the formula, As an intensity enhancement factor, For expansion operation, The structuring element is used; the difference result is the image at the pixel level. The intensity at a certain point is determined; if the intensity is 1, it indicates the presence of a target. Based on connected component analysis, a set of candidate targets, including space targets and remnant stars, is obtained. The information of each candidate target includes the centroid position, gray value, number of pixels, and circumscribed rectangle. The circumscribed rectangle can completely cover the target image patch.

[0015] The method for obtaining the line-of-sight angle and range of candidate targets in the satellite body coordinate system is as follows: Obtain the centroid coordinates of the target in the image plane coordinate system The observation vector in the camera coordinate system is determined to be... ; Determine the azimuth angle of the space-based optical camera relative to the satellite body. and pitch angle The transformation matrix from the camera coordinate system to the satellite body coordinate system is obtained through coordinate rotation and camera mounting matrix. And calculate the target observation vector; Establish a rotating coordinate system, define the pitch and azimuth angles, calculate the relative azimuth and pitch angles of the spatial target using the observation vectors as the line-of-sight angles, and calculate the corresponding angular positions of the four sides of the circumscribed rectangle to define the line-of-sight range.

[0016] The method for calculating the target observation vector is as follows:

[0017]

[0018] The rotating coordinate system uses a right-handed coordinate system, with counter-clockwise rotation being positive. Specifically:

[0019]

[0020] Define pitch angle Define the angle between the XOY plane and the positive direction, and the azimuth angle. Define the angle between the projection of the line of sight onto the XOY plane and the OX axis, and define the positive direction; The method for calculating the relative azimuth and elevation angles of a space target using observed vectors is as follows:

[0021]

[0022] In the formula, the calculation of the line of sight range is the same as that of the line of sight angle, and the angular positions corresponding to the top, bottom, left and right sides of the circumscribed rectangle are calculated respectively. The field of view is defined as .

[0023] The method for inter-frame trajectory correlation of star map sequences is as follows: Using the space target database and the current frame's star map trajectory observations as matching objects, Kalman filtering is applied to the space target database to obtain the predicted position set of each target in the candidate target set in the current frame. The predicted position refers to the relative line-of-sight angle of the target in the current frame; Calculate the predicted location set The candidate target set obtained by star map identification in the current frame Based on the similarity, construct a similarity association matrix; Inertia matching is performed using a similarity correlation matrix. The optimization objective is to minimize the target difference, and positional constraints are set to associate the observation value of the current frame with the trajectory of the spatial target library.

[0024] The similarity matrix is ​​used to calculate the difference between the overlap and distance between two target locations. The calculation method is as follows: In the formula, For intersection, union, and comparison, For distance, Indicates Euclidean distance. To find the diagonal length of the minimum bounding box that covers the two target closed rectangles, the calculation is performed sequentially. and For each element in the matrix, a similarity matrix is ​​obtained, with row variables as follows: The column variable is .

[0025] The method for setting minimizing the target difference as the optimization objective and using the similarity correlation matrix for inertia matching is as follows:

[0026] In the formula, It is a predicted location. and candidate target location similarity, For binary decision variables; The positional constraints are set as follows:

[0027] The observations of the current frame are correlated with the trajectory of the space target library based on the position constraints.

[0028] After completing the association matching between the observations of the current frame and the trajectories of the space target database, the specific type of the target is determined based on the association matching results. The association matching results include: When the correlation matching is that a trajectory matches an observation value, the trajectory information is updated based on the observation value and the space target detection result is output; When the correlation matching situation is that a trajectory does not match any of the observations, determine the reason for the mismatch, select an update scheme according to the type of mismatch reason, and output the space target detection results; When the correlation matching is that an observation does not match any of the trajectories, the space target corresponding to the current observation is determined to be either a space target detected for the first time or a false alarm star. After assigning a new identity, the subsequent five frames of trajectory information are used to determine whether it is a space target or a star. Based on the determination result, the space target detection result is output.

[0029] Reasons for mismatch include: a) The target moved out of the optical camera's field of view; b) The target in dim space was missed; c) The trajectory was a false alarm star, and this frame has been removed. When the mismatch reason is a), the predicted position is outside the current field of view. The current predicted position is calculated based on the camera pointing and the camera field of view. If the mismatch reason is b), mark it as a lost target and store it. The data is maintained within the dataset, and the judgment is made based on the trajectory information of the next five frames. If the mismatch reason is c), mark it as a lost target and store it. The data is maintained within the dataset, and the judgment is made based on the trajectory information of the next five frames.

[0030] The advantages of this invention compared to the prior art are: (1) The present invention provides a space-based optical space target detection method that integrates trajectory information. It obtains the relative line-of-sight angle of the target from the original image without the need for target orbit information, which can reduce the computational complexity. By integrating trajectory information, it can filter out noise interference such as stars and supplement the information of missed targets, thus realizing reliable identification of faint space targets against a dense starry sky background. (2) This invention enables continuous detection of multiple targets in space and assigns a unique identifier to each target. By maintaining the target identifiers during space-based optical camera detection, target information is provided for tasks such as revisiting and tracking; (3) This invention provides a space-based optical space target detection method that integrates trajectory information. Based on trajectory information, it realizes continuous detection of space targets and can detect space targets against a dense starry sky background. It overcomes the problems of low precision and recall and discontinuous target observation values ​​in existing space target detection methods. The overall process is complete and reliable. Attached Figure Description

[0031] Figure 1 A flowchart of the space-based optical space target detection method provided by the present invention; Figure 2 This is a result image of coarse stellar removal using dilatational expansion provided by the present invention; Figure 3 This is a schematic diagram of the target relative line-of-sight angle provided by the present invention; Figure 4 A flowchart of the trajectory association method provided by the present invention; Figure 5 This is a diagram showing the correlation results of space targets during the optical camera detection process provided by the present invention. Detailed Implementation

[0032] A space-based optical space target detection method that integrates trajectory information is proposed. The method uses a space-based optical camera to acquire the image spot positions of point targets in the star map at the current location. Based on the different motion characteristics of stars and space targets, a candidate target set is determined. The centroid coordinates and pixel range of the candidate targets in the image plane coordinate system are determined according to the direction of the optical camera and converted into the line-of-sight angle and range in the satellite body coordinate system. The acquired star map sequence is correlated with the trajectory of each frame. The space target library is updated according to the trajectory information. The space target detection results are output, and the above steps are repeated to traverse all star maps.

[0033] The space-based optical target detection method that integrates trajectory information includes the following steps: The star map of the current location is detected by a space-based optical camera, and star map preprocessing and star map recognition are performed to obtain the image spot positions of all point targets; The obtained image spots are subjected to coarse stellar removal to obtain a candidate target set containing space targets and residual stars; Based on the orientation of the space-based optical camera, the centroid coordinates and pixel range of the candidate target in the image plane coordinate system are converted into the line-of-sight angle and range of the candidate target in the satellite body coordinate system. Perform inter-frame trajectory correlation on the star map sequence; Based on the inter-frame trajectory correlation, the line-of-sight angle and range of candidate targets in the satellite body coordinate system, the candidate target set is updated, and missed targets are added and residual stars are removed based on the inter-frame trajectory correlation. The updated space target library is used as the space target detection result. Repeat all judgment steps to detect space targets in the next frame of the star map.

[0034] Space-based optical cameras detect targets through either stationary or continuous detection. Stationary detection is as follows: Once the space-based optical camera moves to the preset position, it stops moving until it stabilizes before starting to take pictures; Continuous detection is as follows: The space-based optical camera moves at a low speed range and takes pictures simultaneously. During the detection process, there is an overlap area between two consecutive frames of images, and the overlap area meets the preset requirements.

[0035] The methods for star map preprocessing and star map recognition are as follows: The obtained star map is filtered and denoised using Gaussian filtering, and motion blur is restored using Wiener filtering. Set a background threshold to perform threshold segmentation on the processed star map, and obtain a binary image in which the image spots of stars and space objects are separated from the background noise. By scanning binary images using the centroid method and connected component analysis, the centroid positions and circumscribed rectangles of each image spot are obtained.

[0036] Connectivity analysis is used to separate a single point target from other point targets. The separation process includes: The image pixels are scanned one by one from left to right and from top to bottom. According to the set threshold, when a foreground pixel is scanned, the connected component is determined and the image is marked according to the selected 8 connectivity criteria. The outer rectangle is a rectangle that can completely cover the target image patch.

[0037] Based on the different motion characteristics of stars and space targets, the following method is used to coarsely eliminate stars and obtain the candidate target set using dilatational calculus: Based on the sum of squared errors of the relevant sub-images, the previous frame image and the current frame image are registered according to the characteristics of stellar motion.

[0038] Dilation difference is performed on the two registered images. When the difference between the pixels in the two image frames is greater than the threshold, the current pixel is considered to be moving, and star coarse removal is performed using dilation difference. The method for obtaining the line-of-sight angle and range of candidate targets in the satellite body coordinate system is as follows: Obtain the centroid coordinates of the target in the image plane coordinate system The observation vector in the camera coordinate system is determined to be... ; Determine the azimuth angle of the space-based optical camera relative to the satellite body. and pitch angle The transformation matrix from the camera coordinate system to the satellite body coordinate system is obtained through coordinate rotation and camera mounting matrix. And calculate the target observation vector; Establish a rotating coordinate system, define the pitch and azimuth angles, calculate the relative azimuth and pitch angles of the spatial target using the observation vectors as the line-of-sight angles, and calculate the corresponding angular positions of the four sides of the circumscribed rectangle to define the line-of-sight range.

[0039] The method for inter-frame trajectory correlation of star map sequences is as follows: Using the space target database and the current frame's star map trajectory observations as matching objects, Kalman filtering is applied to the space target database to obtain the predicted position set of each target in the candidate target set in the current frame. The predicted position refers to the relative line-of-sight angle of the target in the current frame; Calculate the predicted location set The candidate target set obtained by star map identification in the current frame Based on the similarity, construct a similarity association matrix; Inertia matching is performed using a similarity correlation matrix. The optimization objective is to minimize the target difference, and positional constraints are set to associate the observation value of the current frame with the trajectory of the spatial target library.

[0040] The similarity matrix is ​​used to calculate the difference between the overlap and distance between two target locations; Minimize the target difference as the optimization objective and use the similarity correlation matrix for inertia matching; After completing the association matching between the observations of the current frame and the trajectories of the space target database, the specific type of the target is determined based on the association matching results. The association matching results include: When the correlation matching is that a trajectory matches an observation value, the trajectory information is updated based on the observation value and the space target detection result is output; When the correlation matching situation is that a trajectory does not match any of the observations, determine the reason for the mismatch, select an update scheme according to the type of mismatch reason, and output the space target detection results; When the correlation matching is that an observation does not match any of the trajectories, the space target corresponding to the current observation is determined to be either a space target detected for the first time or a false alarm star. After assigning a new identity, the subsequent five frames of trajectory information are used to determine whether it is a space target or a star. Based on the determination result, the space target detection result is output.

[0041] Reasons for mismatch include: a) The target moved out of the optical camera's field of view; b) The target in dim space was missed; c) The trajectory was a false alarm star, and this frame has been removed. When the mismatch reason is a), the predicted position is outside the current field of view. The current predicted position is calculated based on the camera pointing and the camera field of view. If the mismatch reason is b), mark it as a lost target and store it. The data is maintained within the dataset, and the judgment is made based on the trajectory information of the next five frames. If the mismatch reason is c), mark it as a lost target and store it. The data is maintained within the dataset, and the judgment is made based on the trajectory information of the next five frames.

[0042] The following description, in conjunction with the accompanying drawings and preferred embodiments, provides further details: In the current embodiment, such as Figure 1 The diagram shown is an overall flowchart of a space-based optical target detection method that integrates trajectory information. This method includes the following steps: S1. Use a space-based optical camera to acquire a star map of the current location, perform star map preprocessing and star map recognition, and obtain the image spot positions of all point targets in the image.

[0043] Specifically, the target detection method for space-based optical cameras can be either stationary detection, where the camera moves to the set position and stops, waiting for the camera to stabilize before taking pictures; or continuous detection, where the camera moves slowly while taking pictures. During the detection process, it is necessary to ensure sufficient overlap between two frames.

[0044] The original star image undergoes preprocessing, including filtering and noise reduction, and motion blur restoration. Gaussian filtering is used for noise reduction, and Wiener filtering is used for motion blur restoration. A background threshold is set to perform threshold segmentation on the star image, resulting in a binary image where the image patches of stars and space objects are separated from the background noise.

[0045] The adaptive threshold is calculated by taking the mean and variance of the pixel values.

[0046] By using the centroid method and connected component analysis of the scanned image, the centroid position and circumscribed rectangle of each image spot are obtained. The centroid method calculates the first moment of the grayscale values ​​of the image after threshold segmentation.

[0047] The purpose of the connected component analysis is to separate a single point target from other point targets. The basic process involves scanning image pixels one by one from left to right and top to bottom. Based on a set threshold, when a foreground pixel is scanned, connected component discrimination and image labeling are performed according to the selected 8-connectivity criterion. The circumscribed rectangle is a rectangle that can completely cover the target image patch.

[0048] S2. Based on the different motion characteristics of stars and space targets, dilatational difference is used to perform coarse star removal, resulting in a candidate target set that includes space targets and residual stars.

[0049] Specifically, the motion characteristics of space targets differ from those of stars. First, registration is performed between the previous and current frame images based on the motion characteristics of stars, using the sum of squared errors of the relevant sub-images as the basis for registration.

[0050] In the formula, and This represents the displacement between the previous and current frame images. A significant characteristic of star maps in low-level processing is the large amount of data; to achieve real-time processing, the low-level processing algorithm for star maps cannot be overly complex. The registration method described here creates a sub-map by analyzing the positions of the patch sets obtained from the previous frame's connected component analysis. Multiple sub-maps are then used to calculate registration information with high confidence based on clustering information to simplify the calculation. The previous frame image is horizontally translated... Pixels, vertical translation The pixels are registered with the current frame image with a registration accuracy of 1 pixel.

[0051] Dilation difference is performed on the two registered images. If the difference between pixels in the two image frames exceeds a threshold, that pixel is considered to have motion.

[0052] In the formula, As an intensity enhancement factor, For expansion operation, The structuring element is used. The difference result is the image at the pixel level. The intensity at a given location is used; a value of 1 indicates the presence of a target at that location. Combined with connected component analysis, a set of candidate targets, including space targets and remnant stars, can be obtained. Figure 2 This paper presents the results of coarse stellar removal using dilatational expansion (DIE) on star maps obtained from stationary observations with an optical camera. Information for each target includes its centroid location, grayscale value, number of pixels, and circumscribed rectangle. The circumscribed rectangle completely covers the target image patch.

[0053] S3. Based on the direction of the optical camera, convert the centroid coordinates and pixel range of the candidate target in the image plane coordinate system to the line-of-sight angle and range in the satellite body coordinate system.

[0054] Specifically, the target pixel is At that time, the observed vector in the camera coordinate system is , This refers to the camera's focal length.

[0055] When the optical camera rotates relative to the satellite body at azimuth and elevation angles respectively and At that time, the transformation matrix from the camera coordinate system to the satellite body coordinate system can be obtained through coordinate rotation and camera mounting matrix. Then the target observation vector is:

[0056]

[0057] Rotation uses a right-handed coordinate system, with counter-clockwise rotation being positive.

[0058]

[0059] like Figure 3 As shown, the pitch angle is defined. The azimuth angle is the angle between the XOY plane and the plane, with the upward tilt being positive. The line-of-sight angle is the angle between the projection of the line of sight onto the XOY plane and the OX axis, with positive to the right. For real-time calculation considerations, the line-of-sight angle is evaluated directly in the single-machine coordinate system, without the need to transform to the orbital coordinate system or the Earth's inertial frame. The relative azimuth and pitch angles of the space target are calculated using the observation vector:

[0060]

[0061] The calculation of the line of sight range is the same as that of the line of sight angle; the angle positions corresponding to the top, bottom, left, and right sides of the circumscribed rectangle are calculated respectively. Then the field of view is .

[0062] S4. Perform inter-frame trajectory correlation on the acquired star map sequence.

[0063] Specifically, the matching object is the spatial target database and the observations in the current frame. The spatial target database consists of targets maintained during the detection process. Kalman filtering is applied to the spatial target database to obtain the predicted location set of each target in the current frame. The predicted position refers to the relative line-of-sight angle of the target in the current frame.

[0064] Calculate the predicted location set based on Distance-IoU. With the current frame candidate target set The similarity is used to obtain a similarity correlation matrix. The similarity matrix is ​​calculated as the difference between the overlap and distance between the two target locations:

[0065] In the formula, the first term is the intersection-union ratio, reflecting location information. The vertices of the intersection region are determined by comparing the vertex coordinates of the two rectangles, and the intersection area is calculated. The union area is the area of ​​the two closed rectangles minus the intersection area. The second term is the distance. Indicates Euclidean distance. It is the diagonal length of the smallest rectangle that can cover two target closed rectangles. Calculate sequentially. and For each element in the matrix, a similarity matrix is ​​obtained, with row variables as follows: The column variable is .

[0066] The Jonker-Volgenants algorithm is used for association matching. The shortest augmenting path is found by dynamically maintaining and updating the similarity matrix, with the goal of minimizing the target differences.

[0067] In the formula, It is a predicted location. and candidate target location similarity, The variables are binary decision variables (1 indicates a match, 0 indicates a mismatch), and each predicted position and candidate position is matched only once:

[0068] Complete the association between the observations of the current frame and the trajectories of the space target library. The assignment results include associatable trajectories, unassociated trajectories, and unassociated targets.

[0069] S5. Update the candidate target set, supplement missed targets and remove residual stars based on trajectory information, update the space target library, and output the space target detection results.

[0070] Specifically, the process of trajectory association is as follows: Figure 4 As shown. After the association matching, three situations will occur: 1) a trajectory matches one observation; 2) a trajectory does not match any observations; 3) an observation does not match any trajectory.

[0071] Case 1 indicates that the trajectory and its corresponding target were detected again in the current frame, and the trajectory information was updated based on the observations.

[0072] Case 2 indicates that the trajectory and its corresponding target were not detected in the current frame. This could be caused by the following reasons: a) the target moved outside the optical camera's field of view; b) a target in dim space was missed; c) the trajectory was caused by a star (false alarm), and the frame has been discarded. For reason a, the predicted position is outside the current field of view. The current field of view can be calculated from the camera's pointing direction and the camera's field of view. For reasons b and c, mark them as "lost" targets and store them. The data will continue to be maintained within the dataset. The determination will be made based on the trajectory information from the next five frames.

[0073] Scenario 3 indicates that this observation is the first of its kind, possibly the first detected space target or a false alarm star. A new identifier is assigned to it, and its nature (space target or star) is determined using the trajectory information from the subsequent five frames.

[0074] The subsequent trajectory information of stars and space objects differs. A star's trajectory exhibits a small-amplitude random motion. A space object, relatively faster, exhibits an approximately uniform linear motion for a short period. Therefore, the distinction between a space object and a residual star is made by observing whether the trajectory information in the subsequent five frames maintains approximately uniform linear motion.

[0075] For targets identified as remnant stars, remove them from the maintained target database; for targets that are missed, add their information to the target database.

[0076] S6. Repeat S1-S5 to obtain the next frame of the star map, continuously detect space targets and maintain the continuity of their identification.

[0077] The space-based optical camera continues to capture images, while the algorithm performs space target detection and maintains the space target database trajectory information step by step. Figure 5 The results of the correlation of space targets are shown. Simulation software was used to generate star charts captured by an optical camera during the detection process. The camera's field of view is represented by a red box, and the targets within the box represent observable targets in that frame. Each space target has a unique identifier; white targets indicate targets that were detected but were still being maintained even after they were out of the field of view.

[0078] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solutions of the present invention by utilizing the methods and techniques disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.

[0079] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A space-based optical target detection method that fuses trajectory information, characterized in that... include: The star map of the current location is detected by a space-based optical camera, and star map preprocessing and star map recognition are performed to obtain the image spot positions of all point targets; The obtained image spots are subjected to coarse stellar removal to obtain a candidate target set containing space targets and residual stars; Based on the orientation of the space-based optical camera, the centroid coordinates and pixel range of the candidate target in the image plane coordinate system are converted into the line-of-sight angle and range of the candidate target in the satellite body coordinate system. Perform inter-frame trajectory correlation on the star map sequence; Based on the inter-frame trajectory correlation, the line-of-sight angle and range of candidate targets in the satellite body coordinate system, the candidate target set is updated, and missed targets are added and residual stars are removed based on the inter-frame trajectory correlation. The updated space target library is used as the space target detection result. Repeat all the judgment steps to detect space targets in the next frame of the star map.

2. The space-based optical target detection method according to claim 1, characterized in that: The space-based optical camera detects targets through either stationary detection or continuous detection. Stationary detection is as follows: Once the space-based optical camera moves to the preset position, it stops moving until it stabilizes before starting to take pictures; Continuous detection is as follows: The space-based optical camera moves at a low speed range and takes pictures simultaneously. During the detection process, there is an overlap area between two consecutive frames of images, and the overlap area meets the preset requirements.

3. The space-based optical target detection method according to claim 2, characterized in that: The methods for star map preprocessing and star map recognition are as follows: The obtained star map is filtered and denoised using Gaussian filtering, and motion blur is restored using Wiener filtering. Set a background threshold to perform threshold segmentation on the processed star map, and obtain a binary image in which the image spots of stars and space objects are separated from the background noise. By scanning binary images using the centroid method and connected component analysis, the centroid positions and circumscribed rectangles of each image spot are obtained.

4. The space-based optical target detection method according to claim 3, characterized in that: The binary image, in which the image spots of the stars and space objects are separated from the background noise, is as follows: In the formula, This is an adaptive threshold calculated by taking the mean and variance of pixel values; The centroid method is used to scan the image and then calculate the first moment of the grayscale values ​​of the image after thresholding. The calculation method is as follows: In the formula, the parameter appearing for the first time must be explained; Connectivity analysis is used to separate a single point target from other point targets. The separation process includes: The image pixels are scanned one by one from left to right and from top to bottom. According to the set threshold, when a foreground pixel is scanned, the connected component is determined and the image is marked according to the selected 8 connectivity criteria. The outer rectangle is a rectangle that can completely cover the target image patch.

5. The space-based optical target detection method according to claim 3, characterized in that: Based on the different motion characteristics of stars and space targets, the following method is used to coarsely eliminate stars and obtain the candidate target set using dilatational calculus: Based on the sum of squared errors of the relevant sub-images, registration is performed between the previous frame and the current frame according to the characteristics of stellar motion. The registration method is as follows: In the formula, and The displacement between the previous frame image and the current frame image is used; the registration method creates a sub-graph by analyzing the position of the set of spots obtained from the connected component analysis in the previous frame, and uses multiple sub-graphs to calculate registration information with high confidence based on clustering information to simplify the calculation; The previous frame image was horizontally shifted. Pixels, vertical translation The pixels are registered with the current frame image with a registration accuracy of 1 pixel.

6. The space-based optical target detection method according to claim 5, characterized in that: Dilation difference is performed on the two registered images. When the difference between pixels in the two image frames is greater than a threshold, the current pixel is considered to be moving. The method of coarse star removal using dilation difference is as follows: In the formula, As an intensity enhancement factor, For expansion operation, The structuring element is used; the difference result is the image at the pixel level. The intensity at a certain point is determined; if the intensity is 1, it indicates the presence of a target. Based on connected component analysis, a set of candidate targets, including space targets and remnant stars, is obtained. The information of each candidate target includes the centroid position, gray value, number of pixels, and circumscribed rectangle. The circumscribed rectangle can completely cover the target image patch.

7. The space-based optical target detection method according to claim 5, characterized in that: The method for obtaining the line-of-sight angle and range of candidate targets in the satellite body coordinate system is as follows: Obtain the centroid coordinates of the target in the image plane coordinate system The observation vector in the camera coordinate system is determined to be... ; Determine the azimuth angle of the space-based optical camera relative to the satellite body. and pitch angle The transformation matrix from the camera coordinate system to the satellite body coordinate system is obtained through coordinate rotation and camera mounting matrix. And calculate the target observation vector; Establish a rotating coordinate system, define the pitch and azimuth angles, calculate the relative azimuth and pitch angles of the spatial target using the observation vectors as the line-of-sight angles, and calculate the corresponding angular positions of the four sides of the circumscribed rectangle to define the line-of-sight range.

8. The space-based optical target detection method according to claim 7, characterized in that: The method for calculating the target observation vector is as follows: The rotating coordinate system uses a right-handed coordinate system, with counter-clockwise rotation being positive. Specifically: Define pitch angle Define the angle between the XOY plane and the positive direction, and the azimuth angle. Define the angle between the projection of the line of sight onto the XOY plane and the OX axis, and define the positive direction; The method for calculating the relative azimuth and elevation angles of a space target using observed vectors is as follows: In the formula, the calculation of the line of sight range is the same as that of the line of sight angle, and the angular positions corresponding to the top, bottom, left and right sides of the circumscribed rectangle are calculated respectively. The field of view is defined as .

9. A space-based optical target detection method based on fused trajectory information according to claim 7, characterized in that: The method for inter-frame trajectory correlation of star map sequences is as follows: Using the space target database and the current frame's star map trajectory observations as matching objects, Kalman filtering is applied to the space target database to obtain the predicted position set of each target in the candidate target set in the current frame. The predicted position refers to the target's predicted relative line-of-sight angle in the current frame; Calculate the predicted location set The candidate target set obtained by star map identification in the current frame Based on the similarity, construct a similarity association matrix; Inertia matching is performed using a similarity correlation matrix. The optimization objective is to minimize the target difference, and positional constraints are set to associate the observation value of the current frame with the trajectory of the spatial target library.

10. A space-based optical target detection method based on fused trajectory information according to claim 9, characterized in that: The similarity matrix is ​​used to calculate the difference between the overlap and distance between two target locations. The calculation method is as follows: In the formula, For intersection, union, and comparison, For distance, Indicates Euclidean distance, To find the diagonal length of the minimum bounding box that covers the two target closed rectangles, the calculation is performed sequentially. and For each element in the matrix, a similarity matrix is ​​obtained, with row variables as follows: The column variable is .

11. A space-based optical target detection method based on fused trajectory information according to claim 10, characterized in that: The method for setting minimizing the target difference as the optimization objective and using the similarity correlation matrix for inertia matching is as follows: In the formula, It is a predicted location. and candidate target location similarity, For binary decision variables; The positional constraints are set as follows: The observations of the current frame are correlated with the trajectory of the space target library based on the position constraints.

12. The space-based optical target detection method according to claim 9, characterized in that: After completing the association matching between the observations of the current frame and the trajectories of the space target database, the specific type of the target is determined based on the association matching results. The association matching results include: When the correlation matching is that a trajectory matches an observation value, the trajectory information is updated based on the observation value and the space target detection result is output; When the correlation matching situation is that a trajectory does not match any of the observations, determine the reason for the mismatch, select an update scheme according to the type of mismatch reason, and output the space target detection results; When the correlation matching is that an observation does not match any of the trajectories, the space target corresponding to the current observation is determined to be either a space target detected for the first time or a false alarm star. After assigning a new identity, the subsequent five frames of trajectory information are used to determine whether it is a space target or a star. Based on the determination result, the space target detection result is output.

13. A space-based optical target detection method fusion trajectory information according to claim 12, characterized in that: Reasons for mismatch include: a) The target moved out of the optical camera's field of view; b) The target in dim space was missed; c) The trajectory was a false alarm star, and this frame has been removed. When the mismatch reason is a), the predicted position is outside the current field of view. The current predicted position is calculated based on the camera pointing and the camera field of view. If the mismatch reason is b), mark it as a lost target and store it. The data is maintained within the dataset, and the judgment is made based on the trajectory information of the next five frames. If the mismatch reason is c), mark it as a lost target and store it. The data is maintained within the dataset, and the judgment is made based on the trajectory information of the next five frames.

Images