A target tracking method based on multi-channel detection

Abstract

The present invention discloses a target tracking method based on multi-channel detection, which belongs to the field of target tracking technology and solves the problem in the prior art that the tracking method easily loses the tracking target, resulting in unstable tracking. The target tracking method comprises: step 1, defining the radar coordinate system (X, Y, Z) in the dual-channel SAR system; step 2, sub-aperture carrier frequency removal and fast time matched filtering processing; step 3, compensating for the additional Doppler centroid deviation caused by the azimuth deviation of the dual-channel antenna; step 4, using azimuth reference function convolution to perform azimuth processing to obtain an imaging result; step 5, interpolating and aligning the two images to compensate for the phase error caused by the azimuth position deviation of the receiving aperture; step 6, obtaining micro-motion target energy, processing the micro-motion target energy to obtain a target image; step 7, tracking the detected target. The present invention can be used for target tracking and guidance on a rotating platform, achieving stable tracking of the target.

Description

Technical Field

[0001] The present invention relates to the technical field of target tracking, and in particular to a target tracking method based on multi-channel detection. Background Art

[0002] Traditional SAR is only suitable for detecting and imaging targets in static scenes. However, in actual reconnaissance areas, high-value targets often exhibit unusual motion, such as the rotation of the radar antenna and the pitch and roll of the ship. These micro-movements can also produce the Doppler effect, which modulates the phase of the radar echo, causing the Doppler modulation frequency to vary and the signal spectrum to broaden. This manifests itself in the SAR image as ghosting or defocusing at the target's location. This type of motion distorts the phase history of the target echo, appearing as an out-of-focus signal in the SAR image.

[0003] In addition, when tracking a target, the rotation of the target in the image often leads to target tracking loss. Existing target tracking algorithms cannot meet this application condition, which has a great impact on the performance of the guidance system. Existing technologies urgently need to be studied and improved. Summary of the Invention

[0004] In view of the above analysis, the present invention aims to provide a target tracking method based on multi-channel detection to solve the technical problem that the existing tracking method easily loses the tracking target and causes tracking instability.

[0005] The purpose of the present invention is mainly achieved through the following technical solutions:

[0006] The present invention provides a target tracking method based on multi-channel detection, comprising the following steps:

[0007] Step 1: Define the dual-channel SAR system where antenna 2 transmits the signal. The echo signal is received by antennas 1 and 2 simultaneously. The radar coordinate system (X, Y, Z) is established with antenna 2 as the origin. The origin is always fixed at the phase center of antenna 2.

[0008] Step 2: Remove the carrier frequency of the two sub-apertures and perform fast time matched filtering.

[0009] Step 3: Compensate for the additional Doppler centroid deviation caused by the azimuth deviation of the dual-channel antenna;

[0010] Step 4: Use the azimuth reference function to perform azimuth processing on the compensated echo signal convolution to obtain multiple imaging results;

[0011] Step 5: interpolate and register any two of the imaging results to obtain compensated imaging results to compensate for the phase error caused by the azimuth position deviation of the receiving aperture;

[0012] Step 6: Based on the compensated imaging result, obtain the micro-motion target energy, and determine the target position based on the micro-motion target energy;

[0013] Step 7: Track the detected target.

[0014] Furthermore, in step 7,

[0015] Step S71: Acquire an initial frame image for target tracking; there is a rotation angle between the initial frame image and subsequent frame images;

[0016] Step S72: Acquire a subsequent frame image, and rotate the image according to the rotation angle with respect to the initial frame image to obtain a rotated image with the rotation angle eliminated;

[0017] Step S73: Using the target point coordinates of the previous frame image as the center, a correlation filter model is used to perform target detection in the rotated image to obtain the target point coordinates in the rotated image;

[0018] Step S74: converting the target point coordinates in the rotated image to the image before rotation to obtain the target point coordinates in the image before rotation; and calculating the line of sight angular velocity of the rotating platform as a tracking output signal based on the target point coordinates;

[0019] Step S75: intercepting a search area with the target point coordinates of the rotated image as the center and updating the correlation filter model, and performing target tracking on the next frame of image according to the method of steps S72-S74;

[0020] Repeat steps S72-S75 until target tracking is performed on all subsequent frame images.

[0021] Furthermore, in step 72, the image is rotated according to the rotation angle with respect to the initial frame image, and the pixel coordinates (x′, y′) of the rotated image with the rotation angle eliminated are obtained as follows: Among them, (x, y) is the pixel coordinate of the image before rotation; (x r ,y r ) is the coordinate of any point on the image before rotation, and ξ is the image rotation angle.

[0022] Furthermore, in step 73, step S73 includes:

[0023] 1) Interpolating the rotated image to obtain a grayscale image including the grayscale value of each pixel in the image;

[0024] 2) Taking the target point coordinates of the previous frame image as the center, cut out an image of a set size from the grayscale image, normalize it, and extract the Hog features to obtain the feature matrix Z;

[0025] 3) Use the correlation filter model to perform target detection in the image of a set size to obtain the coordinates of the target point in the rotated image.

[0026] Further, in step 74, step S4 includes:

[0027] 1) Convert the coordinates of the target point in the rotated image to the image before rotation, and obtain the coordinates of the target point in the image before rotation (X T , Y T );

[0028] 2) According to the formula Calculate the line-of-sight angular velocity of the rotating platform as the tracking output signal;

[0029] Among them, V x and V y is the angular velocity of the line of sight in the X and Y directions, X c and Y c is the coordinate of the center point in the X and Y directions of the image before rotation, R x and R y is the equivalent value of the line of sight angular velocity in the X and Y directions.

[0030] Furthermore, in step 5, the compensation function C is used 12 (t a ) interpolates and registers the two complex images to compensate for the phase error caused by the position deviation of the receiving aperture. The compensation function C 12 (t a )for:

[0031]

[0032] Where j is the imaginary unit, k is the signal modulation frequency; R0 is the distance from the radar to the target; t a For slow time; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X-axis.

[0033] Furthermore, in step 2, the subaperture is de-carriered and processed with fast time matched filtering, and the obtained echo signal is:

[0034]

[0035]

[0036] in, They represent the Doppler center frequency and modulation rate of the uniform rotation passive micro-interference echo received by the middle aperture respectively; λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation rate; R0 is the distance from the radar to the target; r is the target rotation radius; r<<R o ;ω is the rotation speed; is the initial phase; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; Tp is the signal pulse width; c is the speed of light; t r Fast time; t a is the slow time, Jm((2kr) is the first kind of m-order Bessel function, m is the order; θ(t a ) is the transformation angle.

[0037] Furthermore, in step 3, the additional Doppler centroid deviation caused by the azimuth deviation of the dual-channel antenna is compensated, and the compensation function is:

[0038]

[0039] Where j is an imaginary number; k is the signal modulation frequency; R0 is the distance from the radar to the target; t a For slow time; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X-axis.

[0040] Furthermore, in step 3, the compensation function C1(t a ) is directly multiplied by J in step 2 p1 To compensate, J p2 The echo signal after compensation is unchanged:

[0041]

[0042]

[0043] in, They represent the Doppler center frequency and modulation rate of the uniform rotation passive micro-interference echo received by the middle aperture respectively; λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation rate; R0 is the distance from the radar to the target; r is the target rotation radius; r<<R o ;ω is the rotation speed; is the initial phase; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; Tp is the signal pulse width; c is the speed of light; tr Fast time; t a For slow time, J m ((2kr) is the first kind of m-order Bessel function, m is the order, θ(t a ) is the transformation angle.

[0044] Furthermore, in step 4, the azimuth reference function is used Respectively with J p1 and J p2 Perform convolution processing and obtain the imaging result after convolution processing:

[0045]

[0046]

[0047] Among them, A3(t r , t a ) is an additional modulation term; θ(t a ) is the transformation angle; λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation frequency; R0 is the distance from the radar to the target; r is the target rotation radius; f<<R o ;ω is the rotation speed; is the initial phase; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; Tp is the signal pulse width; c is the speed of light; tr is the fast time; ta is the slow time, and J m ((2kr) is the first kind of m-order Bessel function, m is the order; B d is the Doppler bandwidth of the echo signal.

[0048] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0049] (1) The target tracking method of the present invention has a wide range of applications and can be generally applied to photoelectric tracking and guidance in rotating platforms. The tracking method has a low computational complexity and good real-time performance, and can meet real-time computing requirements on commonly used embedded hardware platforms.

[0050] (2) The present invention can use the change in Doppler frequency of echoes from different antennas to offset stationary targets and extract moving targets. It can also use the relationship between the phase difference of two SAR images and the interferometric SAR system parameters and target motion parameters to obtain target motion parameter information. The method of the present invention is simple and highly operational, combining SAR's high-resolution imaging of ground targets and micro-motion target detection capabilities. In subsequent engineering transformation and practical applications, it can detect, locate, and track high-value targets, providing useful information for on-site situation assessment, command and control, and is of great significance for reconnaissance and on-site perception.

[0051] In the present invention, the above-mentioned technical solutions can be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages will become apparent from the description or be understood through practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the embodiments of the description and the contents particularly pointed out in the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like parts throughout the drawings.

[0053] Figure 1 Schematic diagram of the process of the inversion positioning method of the present invention;

[0054] Figure 2 is a schematic diagram of a dual-channel SAR system of the present invention;

[0055] Figure 3 This is a schematic diagram of the imaging results of a ship and a micro-moving target according to the present invention;

[0056] Figure 4 Schematic diagram of the micro-motion target interference detection result of the present invention;

[0057] Figure 5 This is a schematic diagram of the repositioning result of the micro-movement target of the present invention;

[0058] Figure 6 Schematic diagram of the target tracking process of the present invention. DETAILED DESCRIPTION

[0059] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein the accompanying drawings constitute a part of the present invention and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0060] The present invention provides a target tracking method based on multi-channel detection, comprising the following steps:

[0061] Step 1: Define the dual-channel SAR system where antenna 2 transmits the signal. The echo signal is received by antennas 1 and 2 simultaneously. The radar coordinate system (X, Y, Z) is established with antenna 2 as the origin. The origin is always fixed at the phase center of antenna 2.

[0062] Step 2: Remove the carrier frequency of the two sub-apertures and perform matched filtering in the fast time domain to obtain the echo signal;

[0063] Step 3: Compensate the echo signal for the additional Doppler centroid deviation caused by the azimuth deviation of the dual-channel antenna;

[0064] Step 4: Use the azimuth reference function to perform azimuth processing on the compensated echo signal convolution to obtain multiple imaging results;

[0065] Step 5: interpolate and register any two of the imaging results to obtain compensated imaging results to compensate for the phase error caused by the azimuth position deviation of the receiving aperture;

[0066] Step 6: acquiring the micro-motion target energy based on the compensated imaging result; and determining the target position based on the micro-motion target energy;

[0067] Step 7: Track the target whose target position is determined, that is, the detected target.

[0068] like Figure 1 As shown, the present invention first performs SAR imaging on the two echo signals respectively; then interpolates and aligns the two images; then conjugate multiplies the two image data to extract the phase; by setting a certain threshold phase, the interference phase is detected, and if the interference phase exceeds the threshold phase, it is determined that the target exists; finally, the target motion parameter information is obtained by using the relationship between the phase difference between the two SAR images and the interferometric SAR system parameters and the target motion parameters.

[0069] Compared to existing technologies, this invention uses the variation in Doppler frequency of echoes from different antennas to offset stationary targets and extract moving targets. It also uses the relationship between the phase difference between two SAR images, the interferometric SAR system parameters, and the target motion parameters to derive target motion parameter information. By combining SAR's high-resolution imaging of ground targets with micro-motion target detection, this invention addresses the effects of micro-motion effects on targets and maps them onto clear SAR images. This allows for the detection, location, and tracking of high-value targets, providing valuable information for on-site situation assessment, command, and control, and is of great significance for reconnaissance and on-site awareness.

[0070] In step 1 above, if Figure 2 As shown, R0 is the distance from the radar to the target, the target rotation radius is r, and f<<R o , the speed is ω, the initial phase is t a For slow time, D a is the phase center distance between the two antennas of the dual-channel SAR system. Antenna No. 2 transmits, and the echo signal is received by antenna No. 1 and antenna No. 2 at the same time. The radar coordinate system (X, Y, Z) is established with antenna No. 2 as the origin. The origin is always fixed on the phase center of antenna No. 2. The SAR carrier moves along the positive direction of the X axis at a speed of V aFlight; Assume that the position of the target P in the radar coordinate system is (x0, y0, -h), and the flight speed of the SAR platform is (V a , 0, 0), the phase center position of antenna No. 1 is (0, 0, D a ), Jm((2kr) is the first kind of m-order Bessel function, m is the order; Bd is the Doppler bandwidth of the echo signal, θ(t a ) is the transformation angle.

[0071] It should be noted that Figure 2 In the figure, the dotted line represents a plane parallel to the XOY plane.

[0072] In step 2 above, the two sub-apertures are orthogonally demodulated to remove the carrier frequency, and then matched filtering is performed in the fast time domain using a matched filter. The resulting echo signal is:

[0073]

[0074]

[0075] in, They represent the Doppler center frequency and modulation rate of the uniform rotation passive micro-interference echo received by the middle aperture respectively; λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation rate; R0 is the distance from the radar to the target; r is the target rotation radius; f<<R o ;ω is the rotation speed; is the initial phase; is the corresponding position history; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; T p is the signal pulse width; c is the speed of light; t r Fast time; t a For slow time, J m ((2kr) is the first kind of m-order Bessel function, m is the order, θ(t a ) is the transformation angle.

[0076] In step 3 above, due to the azimuth deviation between the dual-channel antennas, additional Doppler centroid deviation is caused, so compensation is required. The compensation function is as follows:

[0077]

[0078] Where j is an imaginary number; k is the signal modulation frequency; R0 is the distance from the radar to the target; t a For slow time; D a is the phase center distance between the two antennas of the dual-channel SAR system; Va is the flight speed of the SAR carrier along the positive direction of the X-axis.

[0079] Using the compensation function C1(t a ) Directly multiply J in the previous step p1 To compensate, J p2 The specific compensation process is as follows:

[0080]

[0081] That is, the echo signal of antenna 1 after compensation is:

[0082]

[0083]

[0084] in, They represent the Doppler center frequency and modulation rate of the uniform rotation passive micro-interference echo received by the middle aperture respectively; λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation rate; R0 is the distance from the radar to the target; r is the target rotation radius; r<<R o ;ω is the rotation speed; is the initial phase; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; T p is the signal pulse width; c is the speed of light; t r Fast time; t a is the slow time, θ(t a ) is the transformation angle, J m (2kr) is the first kind of m-order Bessel function, where m is the order; B d is the Doppler bandwidth of the echo signal.

[0085] In step 4 above, the azimuth reference function is:

[0086]

[0087] The above azimuth phase reference function is used to compare with J p1 and J p2 Perform convolution processing to obtain imaging results

[0088]

[0089]

[0090] Among them, A3(t r , t a) is an additional modulation term, which reflects the effect of the additional phase term on the expansion and blurring of the image, but does not affect the peak position of the interference; θ(t a ) is the transformation angle; λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation frequency; R0 is the distance from the radar to the target; r is the target rotation radius; r<<R o ;ω is the rotation speed; is the initial phase; t a For slow time; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; T p is the signal pulse width; c is the speed of light; tr is the fast time; ta is the slow time, J m ((2kr) is the first kind of m-order Bessel function, m is the order; B d is the Doppler bandwidth of the echo signal.

[0091] In step 5 above, the two images are interpolated and registered using the following compensation function to compensate for the phase error caused by the azimuth position deviation of the receiving aperture.

[0092] Specifically, after imaging processing, since the receiving aperture has an azimuth position deviation, it is necessary to compensate for the phase error caused by this position deviation before performing ground clutter cancellation.

[0093] The compensation function is shown below:

[0094]

[0095] Using the compensation function C 12 (t a ) multiplied by I P1 , the result is approximately equal to I P2 The specific calculation process is as follows:

[0096]

[0097]

[0098] I p1 (t r , t a )*C 12 (t a )≈I p2 (t r , t a )

[0099] It should be noted that in the above calculation process, x0=V a *t a .

[0100] Where, The term actually corresponds to the compensation function Compensation function C1(t a ) is because the time domain linear term corresponds to the frequency domain shift, so compensation is performed to align the center. However, the image obtained after imaging is a dual time domain image. Therefore, the phase before compensation corresponds to the phase error caused by the antenna aperture, which can be seen to be independent of the starting position x0 of the target point; where j is the imaginary unit, k is the signal modulation frequency; R0 is the distance from the radar to the target; t a For slow time; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X-axis.

[0101] In step 6 above, the phases of the two images are extracted, and the interference phase is detected by setting a certain threshold phase to cancel the stationary target. That is, the two images are subtracted and canceled, and then the cancellation result is modulo (see the following formula) to obtain the energy of the slightly moving target;

[0102]

[0103] Take the modulus value of the cancellation result to obtain the micro-motion target energy:

[0104]

[0105] Among them, A3(t r , t a ) is the additional modulation term; λ is the wavelength of the radar transmission signal; k is the signal modulation frequency; R0 is the distance from the radar to the target; r is the target rotation radius; r<<R o ;ω is the rotation speed; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; c is the speed of light; t r Fast time; t a For slow time, J m ((2kr) is the first kind of m-order Bessel function, m is the order; B d is the Doppler bandwidth of the echo signal.

[0106] After obtaining the energy of the micro-motion target, the maximum value of the micro-motion target energy is calculated, and the position of the maximum value is the position of the micro-motion target in the SAR image.

[0107] In one embodiment, through digital simulation, a ship is selected as an imaging target, wherein the ship warning radar has micro-motion target characteristics, and the simulation parameters used are shown in Table 1.

[0108] Table 1 Simulation parameters of micro-motion target

[0109]

[0110] According to the above simulation parameters, the ship target imaging results are as follows: Figure 3 As shown in Figure 2, we can see that the ship target can be well imaged and the outline of the ship is clear. At the same time, during the accumulation time, the radar rotation causes obvious Doppler stripes, such as Figure 3 As shown in the white circle.

[0111] The method provided by the present invention is used to perform dual-channel image interference cancellation, and to detect micro-movement targets based on the detection threshold, and to extract micro-movement target results such as Figure 4 According to the relationship between the dual-channel interferometric phase and the rotation parameters, the relocation of the rotating radar target is realized, and the results are shown in Figure 5 As shown in the figure, when the SNR is better than 15dB, the corresponding target positioning accuracy can reach 1 / 15 of the beam width, and is better than 0.5° when the beam width is less than 7°.

[0112] It should be emphasized that after steps 1 to 6 of the present invention detect and locate the target, a series of imaging results at consecutive moments are obtained; the previous image at adjacent moments is defined as the initial frame image, and the image at the next moment is defined as, based on the obtained series of imaging results at consecutive moments, the present invention can also track the target, that is, it also includes step 7, the process of tracking the target.

[0113] like Figure 6 As shown, tracking the detected target specifically includes:

[0114] Step S71: Select a certain moment and obtain an initial frame image corresponding to the moment for target tracking; the initial frame image and subsequent frame images have a rotation angle;

[0115] Step S72: Acquire a subsequent frame image, and rotate the image according to the rotation angle with respect to the initial frame image to obtain a rotated image with the rotation angle eliminated;

[0116] Step S73: Using the target point coordinates of the previous frame image as the center, a correlation filter model is used to perform target detection in the rotated image to obtain the target point coordinates in the rotated image; alternatively, the target point coordinates obtained in step 6 above are directly used;

[0117] Step S74: converting the target point coordinates in the rotated image to the image before rotation to obtain the target point coordinates in the image before rotation; and calculating the line of sight angular velocity of the rotating platform as a tracking output signal based on the target point coordinates;

[0118] Step S75: intercepting a search area with the target point coordinates of the rotated image as the center and updating the correlation filter model, and performing target tracking on the next frame of image according to the method of steps S72-S74;

[0119] Repeat steps S72-S75 until target tracking is performed on all subsequent frame images.

[0120] Specifically, since a rotating platform is used in this embodiment to track the target, there is a certain image rotation angle between each frame of image due to the selection of the platform.

[0121] Specifically, the calculation of the rotation angle in this embodiment includes:

[0122] 1) According to the posture data of the rotating platform, establish the point p in the world coordinate system w (x w ,y w , z w ) to point p in the imaging coordinate system m (x m ,y m , z m )’s transformation matrix T:

[0123]

[0124] Where φ, θ, and γ are the imaging heading angle, pitch angle, and roll angle, respectively. -1 is the transformation matrix from the imager coordinate system to the earth coordinate system:

[0125] 2) Find the inverse matrix T of the transformation matrix T -1 Obtain the transformation matrix from the imager coordinate system to the earth coordinate system;

[0126]

[0127] 3) Get the image rotation angle ξ:

[0128]

[0129] Specifically, in step S72, the image is rotated according to the rotation angle with respect to the initial frame image, and the pixel coordinates (x, y) of the rotated image after eliminating the rotation angle are obtained as follows: Among them, (x, y) is the pixel coordinate of the image before rotation; (x r ,y r ) is the coordinate of any point on the image before rotation as the rotation axis; ξ is the image rotation angle.

[0130] Specifically, step S73 includes:

[0131] Step S7301: interpolate the rotated image to obtain a grayscale image including the grayscale value of each pixel in the image;

[0132] The image coordinates obtained after rotation are generally floating-point numbers, while the image pixel coordinates are generally integers. Bilinear interpolation is used when rotating an image. The grayscale values ​​of the four neighboring points are linearly interpolated in two directions to obtain the grayscale value of the sampled point. That is, the grayscale value of the sampled point is calculated based on the corresponding weight determined by the distance between the sampled point and its neighboring points. Where (x, y) coordinates represent the position of the pixel, and f(x, y) represents the grayscale value of the pixel. Its mathematical expression is:

[0133] f(i+u,j+v)=(1-u)(1-v)f(i,j)+(1-u)v f (i,j+1)+u(1-v)f(i+1,j)+uvf(i+1,j+1);

[0134] In the above formula, u and v range from (0, 1), representing the difference between the x-coordinate and y-coordinate of the sampling point and the upper left corner of the four adjacent points; i and j represent the x-coordinate and y-coordinate of the sampling point and the upper left corner of the four adjacent points.

[0135] Step S7302: Taking the target point coordinates of the previous frame image as the center, intercept an image of a set size in the grayscale image, perform normalization, and extract Hog features to obtain a feature matrix Z;

[0136] Step S7303: Use a correlation filter model to perform target detection in the image of a set size to obtain the coordinates of the target point in the rotated image.

[0137] The target capture of the initial frame image can be performed by target edge extraction and target recognition, or by learning and training a classifier, or by using existing target recognition technologies such as training a neural network. Any existing target capture technology does not affect the creative solution of the target tracking system of the present invention.

[0138] Specifically, in step S7302, the rotated image is I, the initial target size is (w, h), and an image I′ of size (4w, 4h) is intercepted from the image I with the target point coordinates of the previous frame as the center. The image I′ is normalized to a 64*64 image I″, and the steps for extracting Hog features are as follows:

[0139] 1) Calculate the gradients in the X and Y directions, including the gradient amplitude and angle;

[0140] Gradient values ​​in x and y directions:

[0141]

[0142] Gradient magnitude and gradient angle:

[0143]

[0144] In the above formula, f(x+1, y), f(x-1, y), f(x, y+1), and f(x, y-1) represent the grayscale values ​​of the pixels at the corresponding positions, respectively. x (x, y) represents the gradient value in the x direction at the coordinate (x, y), Gy(x, y) represents the gradient value in the y direction at the coordinate (x, y), G(x, y) is the gradient amplitude, and θ(x, y) is the gradient angle.

[0145] 2) Divide the cells and construct the gradient direction histogram of each cell;

[0146] Specifically, a soft mapping strategy is adopted. When constructing the gradient direction histogram, for pixels at the intersection of multiple cells, the gradient of each pixel is distributed to the adjacent cells by linear interpolation.

[0147] More specifically, P1 to P4 represent four pixel positions, corresponding to four cases of linear interpolation:

[0148] P1 is shared by 4 cells;

[0149] P2 is shared by the upper and lower cells;

[0150] P3 is shared by the left and right cells;

[0151] P4 is mapped to only one cell.

[0152] The mapping relationship of point P1:

[0153] Assuming that the coordinates of P1 are (x, y), the mapping formula of the gradient histogram vector of point P1 to the adjacent 4 cells is:

[0154]

[0155] In the above formula, cellsize represents the number of pixels contained in each cell. and Represents the rounding down of a and b, h p1 Represents the gradient histogram vector of point P1, h A 、h B 、h C 、h D Respectively represent the gradient histogram vectors of the four cells adjacent to point P1; preferably, the value of cellsize is 4.

[0156] The mapping formula of points P2 and P3 is:

[0157]

[0158] In the above formula, the calculation method of a and b is the same as the previous formula, h p2 Represents the gradient histogram vector of point P2, h p3 Represents the gradient histogram vector of point P3, h E 、h F Represents the gradient histogram vectors of the upper and lower cells adjacent to point P2, h G 、h H Represent the gradient histogram vectors of the left and right cells adjacent to point P3 respectively.

[0159] The mapping formula for point P4 is:

[0160] h I + = h P4 ;

[0161] In the above formula, h p4 Represents the gradient histogram vector of point P4, h I Represents the gradient histogram vector of the cell where the P4 point is located.

[0162] 3) Normalized gradient histogram, connect the normalized gradient histogram vectors together to obtain the Hog feature matrix.

[0163] Specifically, when normalizing the gradient histogram, each cell is normalized, taking the cell as the basic unit, integrating 2*2 cells into 1 block, and using the L2 norm of the undirected gradient histogram vector to calculate the normalization factor of each block;

[0164] The cell unit E (coordinates (i, j)) belongs to the same four blocks B1, B2, B3, and B4 in space. The normalization factor of each block is:

[0165]

[0166] In the above formula, δ and γ are 1 or -1. Represents the gradient histogram vector of cell E, N δ，γ (i, j) is the block normalization factor of the cell unit E corresponding to the position (δ, γ).

[0167] Preferably, when δ and γ are both 1, N δ，γ (i, j) is the normalization factor of cell unit E in the lower right block, Represents the gradient histogram vector of the cell to the right of cell E, Represents the gradient histogram vector of the cell below cell E, Represents the gradient histogram vector of the cell to the lower right of cell E.

[0168] The normalization factor is used to normalize the elements of each gradient histogram vector, the set threshold is used to truncate the data, and the average of the four values ​​obtained after the normalization of each element is taken as the final result after the normalization of the current element.

[0169] Specifically, four normalization factors are used to normalize the elements of each gradient histogram vector, and a threshold of 0.2 is used for truncation. The four values ​​obtained after normalization of each element are averaged as the final result after normalization of the current element. The formula is as follows:

[0170]

[0171] In the above formula, is the gradient histogram vector of the cell unit E direction k in the kth quadrant with the coordinate (i, j) after taking the mean, Represents the gradient histogram vector of the cell unit E with coordinates (i, j) before taking the mean; N -1，1 (i, j), N 1，-1 、N -1，1 、N 1，1 Respectively represent the normalization factors of cell unit E in the upper left, upper right, lower left, and lower right blocks; T 0.2 (x) represents the threshold cutoff of the variable x in the brackets, and average() means taking the average of multiple values ​​in the brackets;

[0172] T 0.2 In (x), if x is less than or equal to 0.2, the value is x; otherwise, the value is 0.2.

[0173] In addition, the undirected gradients normalized by different normalization factors are accumulated to obtain the sum of four gradient energies, which are used as the texture distribution features of a cell. The formula is as follows:

[0174]

[0175] In the above formula, nbins represents the number of quadrants divided by all gradient directions. Represents the cumulative sum of all directional gradient values; preferably, nbins is 6.

[0176] The normalized histogram vectors described above are concatenated together to obtain the Hog feature matrix.

[0177] Specifically, in step S7303, the correlation filter model is used to detect the target in the image of the set size, and the filter response matrix R is calculated; the peak value is searched in the filter response matrix R to obtain the target point coordinate (X′ T , Y′ T );

[0178] in, F -1 is the inverse Fourier transform; is the Fourier transform of the correlation filter model A; is the Fourier transform of the Gaussian kernel function; λ is the regularization parameter; is the Fourier transform of the cross-correlation function; the regularization parameter λ is set to 0.1.

[0179] Related functions σ is the width parameter of the Gaussian kernel function; preferably, σ of the Gaussian kernel function is set to 0.125. X is the template data of the correlation filter model. is the conjugate of the Fourier transform of the template data; is the Fourier transform of the characteristic matrix Z.

[0180] Specifically, the step S74 includes:

[0181] 1) Convert the coordinates of the target point in the rotated image to the image before rotation, and obtain the coordinates of the target point in the image before rotation (X T , Y T );

[0182] The coordinates of the target point (X′ T , Y′ T ) is converted to the image before rotation, and the target point coordinates (X T , Y T ) is:

[0183] Where (x, y) is the coordinate of the point in the image before rotation; (x′, y′) is the pixel coordinate of the image after rotation; (x r ,y r ) is the coordinate of any point in the image before rotation; ξ is the image rotation angle.

[0184] 2) According to the formula Calculate the line-of-sight angular velocity of the rotating platform as the tracking output signal;

[0185] Among them, V x and V y is the angular velocity of the line of sight in the X and Y directions, X c and Yc is the coordinate of the center point in the X and Y directions of the image, R x and R y is the equivalent value of the line of sight angular velocity in the X and Y directions.

[0186] Specifically, the updating of the relevant filtering model in step S75 includes updating the template data and updating the relevant model.

[0187] Template data updates include:

[0188] Before the correlation filter model is updated, the initial value of the template data X of the correlation filter model is: with the target point coordinates of the initial frame image as the center, an image of a set size is intercepted from the grayscale image of the initial frame image for normalization, and the Hog feature is extracted to obtain a feature matrix.

[0189] The update formula of the relevant filtering model template data X is X new =(1-α)*X old +α*X update ; Where α is the update rate; X old is the template data of the last correlation filtering; X update The Hog feature matrix is ​​calculated by intercepting the search area with the target point coordinates of the rotated image of the latest tracked frame image as the center.

[0190] Among them, relevant model updates include:

[0191] Before the correlation filter model is updated, the correlation filter model is the Fourier transform of the Gaussian kernel function; λ is the regularization parameter; K XX is the autocorrelation function of the initial value of the template data X;

[0192] The update formula of the correlation filter model is A new =(1-α)*A old +α*A update ; Where α is the update rate; A old is the correlation filtering model of the last correlation filtering;

[0193] A update In order to obtain the relevant filtering model using the updated template data,

[0194]

[0195] α is the update rate, and its value is set to 0.012.

[0196] For example, when performing correlation filtering on an image after the initial frame image, the correlation filtering model is in the state before updating. At this time, the template data X of the correlation filtering model is the initial value; the correlation filtering model

[0197] When the initial frame image is tracked and the next frame image is tracked, the template data and the related model are updated. At this time, the template data of the related filtering model is updated so that the template data X new =(1-α)*X old +α*X update ; Among them, X old is the template data of the last correlation filtering, X update The Hog feature matrix is ​​calculated by intercepting the search area with the target point coordinates of the rotated image after the initial frame image as the center. new =(1-α)*A old +α*A update ; Where α is the update rate; A old is the correlation filtering model of the last correlation filtering; A update To use the updated template data X new Get the relevant filter model,

[0198] According to the above updating process, the updated correlation filter model is used to track the target for all subsequent frame images.

[0199] In summary, the present invention achieves stable tracking during image rotation.

[0200] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A target tracking method based on multi-channel detection, characterized in that: The steps include: Step 1: Define the dual-channel SAR system where antenna 2 transmits the signal. The echo signal is received by both antennas 1 and 2. Establish a radar coordinate system (X, Y, Z) with antenna 2 as the origin. The origin is always fixed at the phase center of antenna 2. Step 2: Remove the carrier frequency of the two sub-apertures and perform fast time matched filtering. Step 3: Compensate for the additional Doppler centroid deviation caused by the azimuth deviation of the dual-channel antenna; Step 4: Use the azimuth reference function to perform azimuth processing on the compensated echo signal convolution to obtain multiple imaging results; Step 5: interpolate and register any two of the imaging results to obtain compensated imaging results to compensate for the phase error caused by the azimuth position deviation of the receiving aperture; Step 6: acquiring the micro-motion target energy based on the compensated imaging result, and determining the target position based on the micro-motion target energy; Step 7: Track the detected target; The following sub-steps are included: Step S71: Acquire an initial frame image for target tracking; the initial frame image and subsequent frame images have a rotation angle; Step S72: Acquire a subsequent frame image, and rotate the image according to the rotation angle with respect to the initial frame image to obtain a rotated image with the rotation angle eliminated; Step S73: Using the target point coordinates of the previous frame image as the center, a correlation filter model is used to perform target detection in the rotated image to obtain the target point coordinates in the rotated image; Step S74: converting the target point coordinates in the rotated image to the image before rotation to obtain the target point coordinates in the image before rotation; and calculating the line of sight angular velocity of the rotating platform as a tracking output signal based on the target point coordinates; Step S75: intercepting a search area with the target point coordinates of the rotated image as the center and updating the correlation filter model, and performing target tracking on the next frame of image according to the method of steps S72-S74; Repeat steps S72-S75 until target tracking is performed on all subsequent frame images.

2. The target tracking method based on multi-channel detection according to claim 1, characterized in that: In step 72, the image is rotated according to the rotation angle with respect to the initial frame image, and the pixel coordinates (x′, y′) of the rotated image after eliminating the rotation angle are obtained as follows: Among them, (x, y) is the pixel coordinate of the image before rotation; (x r ,y r ) is the coordinate of any point on the image before rotation, and ξ is the image rotation angle.

3. The target tracking method based on multi-channel detection according to claim 2, characterized in that: In the step 73, the step S73 includes: 1) Interpolating the rotated image to obtain a grayscale image including the grayscale value of each pixel in the image; 2) Taking the target point coordinates of the previous frame image as the center, cut out an image of a set size from the grayscale image, normalize it, and extract the Hog features to obtain the feature matrix Z; 3) Use the correlation filter model to perform target detection in the image of a set size to obtain the coordinates of the target point in the rotated image.

4. The target tracking method based on multi-channel detection according to claim 1, characterized in that: In the step 74, the step S74 includes: 1) Convert the coordinates of the target point in the rotated image to the image before rotation, and obtain the coordinates of the target point in the image before rotation (X T , Y T ); 2) According to the formula Calculate the line-of-sight angular velocity of the rotating platform as the tracking output signal; Among them, V x and V y is the angular velocity of the line of sight in the X and Y directions, X c and Y c is the coordinate of the center point in the X and Y directions of the image before rotation, R x and R y is the equivalent value of the line of sight angular velocity in the X and Y directions.

5. The target tracking method based on multi-channel detection according to any one of claims 1 to 4, characterized in that: In step 5, the compensation function C is used 12 (t a ) interpolates and registers the two complex images to compensate for the phase error caused by the position deviation of the receiving aperture. The compensation function C 12 (t a )for: Where j is the imaginary unit, k is the signal modulation frequency; R0 is the distance from the radar to the target; t a For slow time; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X-axis.

6. The target tracking method based on multi-channel detection according to claim 1, characterized in that: In step 2, the subaperture is de-carriered and processed by fast time matched filtering, and the obtained echo signal is: in, represent the Doppler center frequency and modulation frequency of the uniform rotation passive micro-interference echo received by the middle aperture respectively; θ(t a ) is the transformation angle, λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation frequency; R0 is the distance from the radar to the target; r is the target rotation radius; r<<R o ;ω is the rotation speed; is the initial phase; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; T p is the signal pulse width; c is the speed of light; t r Fast time; t a is the slow time, m is a constant J m ((2kr) is the m-th order Bessel function of the first kind, where m is the order.

7. The target tracking method based on multi-channel detection according to claim 6, characterized in that: In step 3, the additional Doppler centroid deviation caused by the azimuth deviation of the dual-channel antenna is compensated, and the compensation function is: Where, j is an imaginary number; k is the signal modulation frequency; R0 is the distance from the radar to the target; t a For slow time; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X-axis.

8. The target tracking method based on multi-channel detection according to claim 7, characterized in that: In step 3, the compensation function C1(t a ) is directly multiplied by J in step 2 p1 To compensate, J p2 The echo signal after compensation is unchanged: in, They represent the Doppler center frequency and modulation rate of the uniform rotation passive micro-interference echo received by the middle aperture respectively; λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation rate; R0 is the distance from the radar to the target; r is the target rotation radius; r<<R o ;ω is the rotation speed; is the initial phase; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; T p is the signal pulse width; c is the speed of light; t r Fast time; t a For slow time, J m (2kr) is the first kind of m-order Bessel function, m is the order, θ(t a ) is the transformation angle.

9. The target tracking method based on multi-channel detection according to claim 8, characterized in that: In step 4, the azimuth reference function is used Respectively with J p1 and J p2 Perform convolution processing and obtain the imaging result after convolution processing: Among them, A3(t r ,t a ) is an additional modulation term; θ(t a ) is the transformation angle; λ is the wavelength of the radar transmission signal; j is an imaginary number; k is the signal modulation frequency; R0 is the distance from the radar to the target; r is the target rotation radius; r<<R o ;ω is the rotation speed; is the initial phase; D a is the phase center distance between the two antennas of the dual-channel SAR system; V a is the flight speed of the SAR carrier along the positive direction of the X axis; B is the signal bandwidth; T p is the signal pulse width; c is the speed of light; t r Fast time; t a For slow time, J m (2kr) is the first kind of m-order Bessel function, where m is the order; B d is the Doppler bandwidth of the echo signal.

Images