Group detection and group parameter estimation method based on photoelectric detection unmanned aerial vehicle group target

Abstract

The invention relates to the field of unmanned aerial vehicle groups, and aims to solve the problem that an existing detection means cannot quickly and accurately classify and judge anti-unmanned aerial vehicle cluster low-slow small targets. A grouping detection and group parameter estimation method based on a photoelectric detection unmanned aerial vehicle group target is provided, a single-frame image of the unmanned aerial vehicle group target is collected, and the pixel position coordinate and the pixel gray value of each unmanned aerial vehicle in the single-frame image are extracted; setting the pixel gray scale of each unmanned aerial vehicle in the single-frame image to be 1, setting the gray scale of other background pixels to be 0, outputting pixel position coordinates with the pixel gray scale being 1, and performing normalization processing to obtain target point coordinates; the obtained target point coordinates serve as input of a clustering algorithm for clustering processing, and a final clustering center of the multiple group targets is obtained; and fitting the sample points of the group targets through a principal component analysis method to form an elliptical shape, and obtaining the center position, main axis direction, long and short axis length and eccentricity information of each group target, thereby completing the initial grouping detection of the unmanned aerial vehicle group targets.

Description

Technical Field

[0001] The present invention relates to the field of drone swarms, and in particular to a method for swarm detection and swarm parameter estimation based on photoelectric detection of drone swarm targets. Background Art

[0002] With the development trend of UAV swarm flight control measurement technology towards scale, informatization, intelligence and diversification, swarm target tracking has become an important issue in current research. Actively exploring the detection, tracking and identification of UAV swarms, acquiring the spatial characteristics of swarm targets and perceiving and grasping the situation of swarm targets are important parts of UAV swarm flight control measurement technology.

[0003] Anti-UAV clusters mainly target UAV targets such as UAV clusters. It requires the use of multiple widely distributed detection systems to obtain multi-dimensional, multi-domain, and multi-band characteristic parameters of group targets. Through multi-level collaborative perception processing, the detection capability of anti-UAV cluster targets in actual complex environments is improved; and the data detected by multi-platform reconnaissance is used to determine the target's exact location and real-time motion status, so as to realize the identification and early warning of anti-UAV cluster targets.

[0004] However, due to the low flying altitude of drone swarms and the large low-altitude blind spot of existing radar detection systems, anti-drone swarms are not easy to be detected in low-altitude areas; at the same time, the relative flight speed of anti-drone swarms is slow, and their Doppler effect is not obvious. They are easily interfered by weather and clutter, and are easily confused with bird flocks. The amount of observation data is large, which undoubtedly increases the difficulty of target distinction of anti-drone swarms; therefore, it is difficult to judge the authenticity of low, slow and small targets only through radar detection means, and it is impossible to effectively track the actual targets.

[0005] Therefore, it is imperative to find a method that can quickly and accurately classify and judge anti-UAV cluster targets in complex environments, achieve effective tracking of group targets, and provide monitoring personnel with real-time dynamic parameter information of anti-UAV clusters. Summary of the invention

[0006] The purpose of the present invention is to solve the technical problems that the existing radar detection means are unable to quickly and accurately classify and judge the low, slow and small targets of anti-UAV clusters, as well as effectively track them, and to propose a group detection and group parameter estimation method based on photoelectric detection of UAV group targets.

[0007] The design idea of ​​the present invention is: in the initial stage of drone group target tracking, the drone group targets are grouped and detected through single-frame images, the targets that meet the characteristic distance are divided into the same group, and the spatial shape of each group is fitted according to relevant characteristics to obtain the center, major and minor axis lengths, eccentricity and main axis angle of the shape, so as to provide spatial feature information for group target tracking.

[0008] To achieve the above objectives, the technical solution proposed by the present invention is:

[0009] The method for swarm detection and swarm parameter estimation based on photoelectric detection of drone swarm targets is special in that it includes the following steps:

[0010] S1, collect a single frame image of the drone group target, and extract the pixel position coordinates and pixel grayscale value of each drone in the single frame image;

[0011] S2, set the pixel grayscale of each drone in the single frame image to 1, set the grayscale of other background pixels to 0, and output the pixel position coordinates with pixel grayscale of 1;

[0012] S3, normalizing the pixel position coordinates with a pixel grayscale of 1 output in step S2 to obtain the target point coordinates;

[0013] S4, using the target point coordinates obtained in step S3 as the input of the clustering algorithm to perform clustering processing, obtaining the final clustering centers of multiple group targets, and dividing all target points into clusters corresponding to the final clustering centers to form multiple group targets;

[0014] S5. The sample points of each group target are fitted into an elliptical shape through the principal component analysis method to obtain the center position, main axis direction, major and minor axis lengths, and eccentricity information of each group target, thereby completing the initial group detection and group parameter estimation of the drone group targets.

[0015] Furthermore, step S3 is specifically as follows:

[0016] S3.1. Set the normalized image size to W*H, and obtain the pixel position coordinates (Ti_x, Ti_y) with a pixel grayscale of 1 in a single-frame image, where i=1…m, m is the number of drones in a single-frame image, and m≥1;

[0017] Where W is the vertical resolution of the image, and H is the horizontal resolution of the image;

[0018] S3.2. After normalizing the pixel position coordinates (Ti_x, Ti_y) with a pixel grayscale of 1 in step S3.1, the target point coordinates are obtained as (Ti_x / W, Ti_y / H).

[0019] Further, step S4 is specifically as follows:

[0020] S4.1, use the target point coordinates (Ti_x / W, Ti_y / H) obtained in step S3.2 as the input of the clustering algorithm;

[0021] S4.2. Select a target point as the seed point X in the target point coordinates (Ti_x / W, Ti_y / H)i , which is the initial cluster center, and the remaining target points are recorded as X j , and take the seed point X i Determine a window at the center for density calculation;

[0022] The radius of the window is the same as the Gaussian bandwidth parameter h;

[0023] S4.3, within the determined window in step S4.2, calculate each target point X j Relative to the seed point X i Density estimation of ;

[0024] S4.4. Update the seed point X according to the mean shift vector i , get the new seed point X i1 ;

[0025] S4.5. Check new seed point X i1 With seed point X i Whether the distance between them is less than the set threshold;

[0026] If the new seed point X i1 With seed point X i If the distance between them is not less than the set threshold, the initial cluster center has not converged, and a new initial cluster center is selected, and the process returns to S4.2;

[0027] If the new seed point X i1 With seed point X i If the distance between them is less than the set threshold, the initial cluster center has converged, and the seed point X i As a final cluster center, and converge others to the seed point X i The target point X j Divide into the cluster of the final cluster center to form a group target;

[0028] S4.6. Determine whether all target points X j are all assigned to a cluster of a final cluster center. If not, the next target point that is not designated as the initial cluster center is selected as a new seed point, and the algorithm returns to S4.2. If yes, multiple group targets are obtained, and the algorithm ends.

[0029] S4.7, the output set includes the pixel coordinates of each group target (Ti_x j , Ti_y j ) and the final cluster center position (Ti_x0, Ti_y0) of the group target.

[0030] Furthermore, in step S4.3, each target point X j Relative to the seed point X i The density estimate satisfies the following formula:

[0031]

[0032] Where: f(X i ) is the density estimation function, K is the Gaussian kernel function, h is the Gaussian bandwidth parameter, and n is the size of the selected window data set;

[0033] Furthermore, in step S4.4, the mean shift vector M h (X) satisfies the following formula:

[0034]

[0035] Furthermore, in step S4.5, the threshold is set to 1e-3*h.

[0036] Furthermore, step S5 is specifically as follows:

[0037] S5.1. Set the sample matrix of a single group target as Y. For each group target, transform the sample points Y a Standardization is performed, that is

[0038] in: is the sample mean of the single group target;

[0039] S5.2. According to Calculate the covariance matrix C of the sample matrix Y;

[0040] According to Cb=λb (b≠0), the eigenvalue λ of the covariance matrix C and the eigenvector b corresponding to the eigenvalue λ are calculated, and the eigenvalues ​​λ are arranged in descending order, and the first eigenvalue λ1, the second eigenvalue λ2, and the first eigenvector b1 are selected;

[0041] S5.3. The elliptical shape equation of the group target is:

[0042] S5.4, according to the first eigenvector b1 and the cluster center position (Ti_x0, Ti_y0) of the group target, the elliptical shape of the group target is rotated and translated to obtain the fitting of the group target shape;

[0043] Obtain the final cluster center position (Ti_x0, Ti_y0), main axis direction b1, major and minor axis lengths λ1, λ2, and eccentricity of the group target Complete the initial grouping detection and group parameter estimation of drone swarm targets.

[0044] Beneficial effects of the present invention:

[0045] [1] The present invention's method for swarm detection and swarm parameter estimation based on photoelectric detection of drone swarm targets can effectively solve the problem of large blind spots in low-altitude flight of drone swarms and difficulty in detecting and tracking slow-speed flight. It can quickly and accurately classify and judge anti-drone swarm targets in complex environments, obtain multi-dimensional characteristic parameters of swarm targets, and achieve effective tracking of swarm targets, effectively improving the monitoring efficiency and accuracy of the system.

[0046] [2] The present invention uses a single-frame image to perform group detection on drone group targets at the initial stage of group target tracking, divides targets that meet the characteristic distance into the same group, and fits the spatial shape of each group target according to relevant characteristics to obtain information such as the center, major and minor axis lengths, eccentricity, and main axis angle of the group target shape, thereby providing spatial feature information for group target tracking and ensuring the effectiveness and reliability of monitoring parameters.

[0047] [3] The present invention converts a single-frame image of a drone swarm target into a surface target for segmentation, normalization, and principal component analysis. The position cluster center of the swarm target constructed by the present invention is more accurate and stable, which facilitates the servo system to stably track the drone swarm targets, thereby improving the detection capability of the system. By estimating the shape distribution parameters of the drone swarm targets, the accurate position information and real-time motion status of the swarm targets can be obtained, providing intuitive visual data results for monitoring personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of the method for group detection and group parameter estimation based on photoelectric detection of drone group targets of the present invention;

[0049] Figure 2 This is an original visible light simulation image of a drone swarm in an embodiment of the present invention;

[0050] Figure 3 It is a schematic diagram of the grouping detection and ellipse shape fitting estimation results in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] like Figure 1 As shown, the method for group detection and group parameter estimation based on photoelectric detection of drone group targets includes the following steps:

[0052] S1, such as Figure 2 As shown, a single-frame image of the drone group target is collected, and the pixel position coordinates and pixel grayscale value of each drone in the single-frame image are extracted;

[0053] S2, set the pixel grayscale of each drone in the single frame image to 1, set the grayscale of other background pixels to 0, and output the pixel position coordinates with pixel grayscale of 1;

[0054] S3, normalizing the pixel position coordinates with a pixel grayscale of 1 output in step S2 to obtain the target point coordinates;

[0055] S3.1. Set the normalized image size to W*H, and obtain the pixel position coordinates (Ti_x, Ti_y) with a pixel grayscale of 1 in a single-frame image, where i=1…m, m is the number of drones in a single-frame image, and m≥1; where W is the image vertical resolution, and H is the image horizontal resolution; S3.2. After normalizing the pixel position coordinates (Ti_x, Ti_y) with a pixel grayscale of 1 in step S3.1, the target point coordinates are obtained as (Ti_x / W, Ti_y / H);

[0056] S4, using the target point coordinates obtained in step S3 as the input of the clustering algorithm to perform clustering processing, obtaining the final clustering centers of multiple group targets, and dividing all target points into clusters corresponding to the final clustering centers to form multiple group targets;

[0057] S4.1, use the target point coordinates (Ti_x / W, Ti_y / H) obtained in step S3.2 as the input of the clustering algorithm;

[0058] S4.2. Select a target point as the seed point X in the target point coordinates (Ti_x / W, Ti_y / H) i , which is the initial cluster center, and the remaining target points are recorded as X j , and take the seed point X i Determine a window at the center for density calculation;

[0059] The radius of the window is the same as the Gaussian bandwidth parameter h;

[0060] S4.3, within the determined window in step S4.2, calculate each target point X j Relative to the seed point X i Density estimation of ;

[0061] Each target point X j Relative to the seed point X i The density estimate satisfies the following formula:

[0062]

[0063] Where: f(X i ) is the density estimation function, K is the Gaussian kernel function, h is the Gaussian bandwidth parameter, and n is the size of the selected window data set;

[0064] S4.4. According to the mean shift vector Update seed point X i , get the new seed point X i1 ;

[0065] S4.5. Check new seed point X i1 With seed point X i Whether the distance between them is less than the set threshold, the threshold is set to 1e-3*h;

[0066] If the new seed point X i1 With seed point X i If the distance between them is not less than the set threshold, the initial cluster center has not converged, and a new initial cluster center is selected, and the process returns to step S4.2;

[0067] If the new seed point X i1 With seed point X i If the distance between them is less than the set threshold, the initial cluster center has converged, and the seed point X i As a final cluster center, and converge others to the seed point X i The target point X j Divide into the cluster of the final cluster center to form a group target;

[0068] S4.6. Determine whether all target points X j are all assigned to a cluster of a final cluster center. If not, the next target point that is not designated as the initial cluster center is selected as a new seed point, and the algorithm returns to S4.2. If yes, multiple group targets are obtained, and the algorithm ends.

[0069] S4.7, the output set includes the pixel coordinates of each group target (Ti_x j , Ti_y j ) and the final cluster center position of the group target (Ti_x0, Ti_y0)

[0070] S5, such as Figure 3 As shown in the figure, the sample points of each group target are fitted into an elliptical shape through the principal component analysis method, and the center position, main axis direction, major and minor axis length, and eccentricity information of each group target are obtained, thereby completing the initial group detection and group parameter estimation of the UAV group target.

[0071] S5.1. Set the sample matrix of a single group target as Y. For each group target, transform the sample points Y a Standardization is performed, that is

[0072] in: is the sample mean of the single group target, and a is any group target among multiple group targets;

[0073] S5.2. According to Calculate the covariance matrix C of the sample matrix Y;

[0074] According to Cb=λb (b≠0), the eigenvalue λ of the covariance matrix C and the eigenvector b corresponding to the eigenvalue λ are calculated, and the eigenvalues ​​λ are arranged in descending order, and the first eigenvalue λ1, the second eigenvalue λ2, and the first eigenvector b1 are selected;

[0075] S5.3. The elliptical shape equation of the group target is:

[0076] S5.4, according to the first eigenvector b1 and the cluster center position (Ti_x0, Ti_y0) of the group target, the elliptical shape of the group target is rotated and translated to obtain the fitting of the group target shape;

[0077] Obtain the final cluster center position (Ti_x0, Ti_y0), main axis direction b1, major and minor axis lengths λ1, λ2, and eccentricity of the group target Complete the initial grouping detection and group parameter estimation of drone swarm targets.

[0078] This embodiment is applicable to targets of different scales on a two-dimensional image. It divides the group targets and establishes the elliptical shape parameters. The feature point accuracy is 97.42%. Compared with the cluster detection algorithm that directly uses the target center as input, the accuracy is improved by 15.74%, and spatial shape data is provided for subsequent target tracking. However, the method in this embodiment is not limited to cluster detection of targets of different scales on a two-dimensional image.

Claims

1. A method for group detection and group parameter estimation based on photoelectric detection of drone group targets, characterized in that: The following steps are involved: S1, collect a single frame image of the drone group target, and extract the pixel position coordinates and pixel grayscale value of each drone in the single frame image; S2, set the pixel grayscale of each drone in the single frame image to 1, set the grayscale of other background pixels to 0, and output the pixel position coordinates with pixel grayscale of 1; S3, normalizing the pixel position coordinates with a pixel grayscale of 1 output in step S2 to obtain the target point coordinates; S4, using the target point coordinates obtained in step S3 as the input of the clustering algorithm to perform clustering processing, obtaining the final clustering centers of multiple group targets, and dividing all target points into clusters corresponding to the final clustering centers to form multiple group targets; S5. The sample points of each group target are fitted into an elliptical shape through the principal component analysis method to obtain the center position, main axis direction, major and minor axis lengths, and eccentricity information of each group target, thereby completing the initial group detection and group parameter estimation of the drone group targets.

2. According to claim 1, the method for group detection and group parameter estimation based on photoelectric detection of drone group targets is characterized in that: Step S3 is specifically as follows: S3.

1. Set the normalized image size to W*H, and obtain the pixel position coordinates (Ti_x, Ti_y) with a pixel grayscale of 1 in a single-frame image, where i=1…m, m is the number of drones in a single-frame image, and m≥1; Where W is the vertical resolution of the image, and H is the horizontal resolution of the image; S3.

2. After normalizing the pixel position coordinates (Ti_x, Ti_y) with a pixel grayscale of 1 in step S3.1, the target point coordinates are obtained as (Ti_x / W, Ti_y / H).

3. The method for group detection and group parameter estimation based on photoelectric detection of drone group targets according to claim 2 is characterized in that: Step S4 is specifically as follows: S4.1, use the target point coordinates (Ti_x / W, Ti_y / H) obtained in step S3.2 as the input of the clustering algorithm; S4.

2. Select a target point as the seed point X in the target point coordinates (Ti_x / W, Ti_y / H) i , which is the initial cluster center, and the remaining target points are recorded as X j , and take the seed point X i Determine a window at the center for density calculation; The radius of the window is the same as the Gaussian bandwidth parameter h; S4.3, within the determined window in step S4.2, calculate each target point X j Relative to the seed point X i Density estimation of ; S4.

4. Update the seed point X according to the mean shift vector i , get the new seed point X i1 ; S4.

5. Check new seed point X i1 With seed point X i Whether the distance between them is less than the set threshold; If the new seed point X i1 With seed point X i If the distance between them is not less than the set threshold, the initial cluster center has not converged, and a new initial cluster center is selected, and the process returns to step S4.2; If the new seed point X i1 With seed point X i If the distance between them is less than the set threshold, the initial cluster center has converged, and the seed point X i As a final cluster center, and converge others to the seed point X i The target point X j Divide into the cluster of the final cluster center to form a group target; S4.

6. Determine whether all target points X j are all assigned to a cluster of the final cluster center. If not, the next target point that is not designated as the initial cluster center is selected as the new seed point and the algorithm returns to S4.

2. If yes, multiple group targets are obtained and the algorithm ends. S4.

7. The output set includes the pixel coordinates of each group target (Ti_x j , Ti_y j ) and the final cluster center position (Ti_x0, Ti_y0) of the group target.

4. The method for group detection and group parameter estimation based on photoelectric detection of drone group targets according to claim 3 is characterized by: In step S4.3, each target point X j Relative to the seed point X i The density estimate satisfies the following formula: Where: f(X i ) is the density estimation function, K is the Gaussian kernel function, h is the Gaussian bandwidth parameter, and n is the size of the selected window data set.

5. The method for group detection and group parameter estimation based on photoelectric detection of drone group targets according to claim 4 is characterized in that: In step S4.4, the mean shift vector M h (X) satisfies the following formula:

6. The method for group detection and group parameter estimation based on photoelectric detection of drone group targets according to claim 5 is characterized in that: In step S4.5, the threshold is set to 1e-3*h.

7. The method for group detection and group parameter estimation based on photoelectric detection of drone group targets according to claim 6 is characterized in that: Step S5 is specifically as follows: S5.

1. Set the sample matrix of a single group target as Y. For each group target, transform the sample points Y a Standardization is performed, that is in: is the sample mean of the single group target; S5.

2. According to Calculate the covariance matrix C of the sample matrix Y; According to Cb=λb (b≠0), the eigenvalue λ of the covariance matrix C and the eigenvector b corresponding to the eigenvalue λ are calculated, and the eigenvalues ​​λ are arranged in descending order, and the first eigenvalue λ1, the second eigenvalue λ2, and the first eigenvector b1 are selected; S5.

3. The elliptical shape equation of the group target is: S5.4, according to the first eigenvector b1 and the cluster center position (Ti_x0, Ti_y0) of the group target, the elliptical shape of the group target is rotated and translated to obtain the fitting of the group target shape; Obtain the final cluster center position (Ti_x0, Ti_y0), main axis direction b1, major and minor axis lengths λ1, λ2, and eccentricity of the group target Complete the initial grouping detection and group parameter estimation of drone swarm targets.

Images