A Wall Behind Concealed Target Tracking Method Applicable to Low Signal-to-Noise Ratio Environments

Abstract

The present invention discloses a method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments, which is applied to the field of radar tracking technology. Aiming at the problem in the prior art that in a complex electromagnetic environment, there is a lot of clutter and noise in the echo, resulting in poor imaging quality and thus difficult target tracking. The present invention first reconstructs the scene by using the back-projection method. Then, it extracts the target area features and establishes a target template based on the ridge regression idea, establishes a scale pool with small scale change in the range direction and large scale change in the azimuth direction based on the through-wall radar target imaging mechanism and the characteristics of the target after imaging. Then, it tracks the target in the next frame based on the correlation filtering idea, updates the target template for each frame to adapt to the changes in the through-wall radar target features, and finally dynamically estimates the target area and obtains the tracking trajectory. The present invention can achieve precise tracking of hidden targets behind walls in low signal-to-noise ratio scenarios.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar tracking, and particularly relates to a hidden target tracking technology. Background Art

[0002] Ultra-wideband radar tracking technology has received extensive research interest due to its ability to track hidden targets behind multiple walls, and it has a wide range of applications in fields such as disaster rescue, monitoring, and urban warfare. For ultra-wideband radar tracking technology, stably and accurately tracking targets can not only obtain the movement position of the targets at any time, but also predict the next actions of the targets, providing support for our behavior strategies. Therefore, it has also attracted great attention in recent years.

[0003] Many research institutions at home and abroad have carried out research on target tracking based on ultra-wideband radar and achieved rich research results. In ultra-wideband radar tracking technology, post-imaging tracking is used as a basic strategy because it is not easily affected by channel quality and has a low computational load in the case of multiple targets compared with pre-imaging tracking. The literature "Multiple moving targets tracking for through-wall imaging radar, Proc. IEEE Radar Conf., pp. 501-505, 2015." obtains the target position through an image erosion detection method and updates the target motion state using the IMM method to achieve target tracking. The literature "Scale-Adaptive Human Target Tracking for Through-Wall Imaging Radar, IEEE Geosci. Remote Sensing Lett., pp. 1348-1352, 2020." uses the mean shift method to achieve target tracking and combines the method of image moments to dynamically estimate the scale and direction changes of the target image. However, in practical applications, the shapes of target images after ultra-wideband radar imaging are variable, and fixed detection methods are difficult to detect targets. Using the mean shift method based on image moments is an effective solution, but the electromagnetic wave propagation path is complex in a complex environment, and the echo signal contains a large amount of clutter and noise, resulting in a low signal-to-noise ratio of the final imaging image. The accuracy of the mean shift method and the image moment method is greatly reduced due to the influence of the signal-to-noise ratio. Therefore, it is of great practical significance to study a method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments, which can accurately track targets.

[0005] The technical solution adopted by the present invention is as follows: A method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments, including:

[0006] S1. Data acquisition: Use a two-transmitter and four-receiver ultra-wideband radar to perform single-view scanning on hidden targets behind walls, and the receiving antenna receives the reflected signals.

[0007] S2. Image the signals received in step S1: Use the back-projection algorithm to focus and image the received signals.

[0008] S3. Preprocess the imaging results obtained in step S2. Use the energy coherence factor algorithm to suppress grating lobes, and then use the Gaussian filtering algorithm to smooth the imaging results.

[0009] S4. According to the urban and rural results processed in step S3, when the target first appears, use the visual annotation method to frame the target area. In view of the characteristic that the target response of the through-wall radar is a point spread function, use the amplitude distribution and edge features of the target area to jointly represent the target features, and establish a target template based on the ridge regression algorithm.

[0010] In view of the characteristic that the scale of the through-wall radar target changes significantly with distance, establish a scale pool S with large azimuthal changes but small range changes.

[0011] S5. Use the tracking algorithm based on correlation filtering to establish the relationship between the candidate area template and the target template, perform target tracking and estimate the scale of the target area.

[0012] The suppression of the grating lobes in the imaging results reconstructed in step S2 described in step S3 is specifically to suppress the grating lobes by using the energy difference caused by the phase difference between channels at the grating lobes according to the characteristic that the echoes of different channels are in phase only at the target.

[0013] The establishment of the scale pool is specifically to establish a scale pool S with different azimuthal and range change rates. For example, the azimuthal scale pool is:

[0014] [1, 1.05, 1.1, 1.15]

[0015] The range scale pool is:

[0016] [1, 1.02, 1.04, 1.06]

[0017] Step S5 is specifically as follows:

[0018] When a new frame of radar imaging result is obtained, extract the target candidate area centered on the target position of the previous frame according to the size of the pre-set scale pool, establish a candidate area template, and calculate the responses of candidate area templates with different scales and the target template.

[0019]

[0020] The scale corresponding to the maximum response is the current target image scale, and the position where the maximum response value is located is the target position.

[0021] Advantages of the present invention: A method for tracking hidden targets behind an ultra-wideband radar wall applicable to a low signal-to-noise ratio environment in the present invention can achieve precise target tracking in a low signal-to-noise ratio environment, can effectively match the targets in the front and rear frames in a strong noise background, and solve the problem that it is difficult to precisely track targets in a low signal-to-noise ratio environment. Therefore, the present invention has the advantage of adapting to a complex electromagnetic environment and can be directly applied to ultra-wideband imaging radar equipment. Description of the Drawings

[0022] Figure 1 It is a schematic diagram of the observation scene;

[0023] Figure 2 It is the electromagnetic simulation imaging result in a noise environment;

[0024] Figure 3 It is the tracking result using the improved mean shift method;

[0025] Among them, (a) is the tracking result of the 51st frame, (b) is the tracking result of the 70th frame, and (c) is the tracking trajectory result;

[0026] Figure 4 It is the tracking result using the proposed method;

[0027] Among them, (a) is the tracking result of the 51st frame, (b) is the tracking result of the 70th frame, and (c) is the tracking trajectory result. Detailed Embodiment

[0028] To facilitate those skilled in the art to understand the technical content of the present invention, the content of the present invention will be further explained below with reference to the drawings.

[0029] The working schematic diagram of the radar node of the present invention is as Figure 1 shown. The radar node includes two transmitting antennas and four receiving antennas. The radar is closely attached to the wall, and the receiving antennas receive the reflected signals after interacting with the scene.

[0030] Taking the example of using a through-wall radar to monitor the target behind a single-layer wall in the present invention, as Figure 1 shown, the radar is set in front of the wall to monitor the target behind the wall. The method of the present invention specifically includes the following steps:

[0031] Step 1: Back-projection imaging of the echo signal

[0032] The signal emitted by the mth transmitting antenna and received by the nth antenna at the kth time after interacting with the scene can be expressed as:

[0033]

[0034] Among them, \(t\) is the time, \(s(t)\) is the transmitted signal, and \(\tau\) mnk is the time delay, and \(\sigma\) mnk is the scattering coefficient, and \(\omega\) mnk is the direct coupling signal, and \(n\) is the clutter and noise.

[0035] The method of moving target indication is adopted to eliminate the direct coupling signal, fixed clutter and noise, and the expression is:

[0036]

[0037] Among them, \(r\) mnk (t) represents the signal received by the \(n\)-th antenna at the \(k\)-th moment from the \(m\)-th transmitting antenna, and \(r\) mn(k-1) (t) represents the signal received by the \(n\)-th antenna at the \(k - 1\) moment from the \(m\)-th transmitting antenna.

[0038] Using the multi-channel echo signal for back-projection focusing imaging, the imaging scene is gridded, and the energy of each grid is calculated:

[0039]

[0040] Among them is the calculation time delay of the coordinates \((x, y)\), \(N\) represents the total number of receiving antennas, and \(M\) represents the total number of transmitting antennas.

[0041] Step 2: Preprocessing of the imaging result

[0042] Since the array arrangement does not satisfy the spatial Nyquist sampling law, grating lobes will appear in the imaging result and contaminate the imaging result. It is necessary to suppress the grating lobes. The method based on the energy coherence factor is adopted, and the energy coherence factor of the pixel point \(x\) q is:

[0043]

[0044] Among them, \(I\) i (x q ) represents the result of back-projection imaging of the \(i\)-th channel.

[0045] Multiply the imaging result by the corresponding energy coherence factor. After weighting by the energy coherence factor, there are many holes and burrs in the target area. A two-dimensional Gaussian kernel is used to smooth the imaging result. After back-projection imaging and image preprocessing, the real-valued radar image of the detection area is finally obtained.

[0046] Step 3: Establishment of the target template

[0047] Step 3-1: Target feature extraction

[0048] After obtaining the real-value image, the imaging result of the monitoring area at each moment can be obtained. When there is no target in the monitoring area, only randomly distributed environmental noise is shown in the real-value image; when a target appears, obvious high-energy aggregation points will appear in the real-value image, such as Figure 2 shown. When the target first appears in the real-value image, the target area is framed by the method of visual annotation. Let the scale size of the target area be s T =(s x , s y ), s x , s y are the length and width pixel values of the target area, and a set of scale pools S = {t1, t2,..., t k} is defined, where t1, t2,..., t k are multiples of s T . The energy distribution feature and the edge feature of the target area are concatenated as the feature of the target. Specifically:

[0049] The pixel values of the target area are unified between 0 and 1 as the energy feature of the target area:

[0050]

[0051] where pix(i, j) is the pixel value of the coordinate (i, j) in the target area, and max(pix) is the maximum pixel value in the target area.

[0052] To extract the edge feature, first calculate the gradient of each pixel point, including the magnitude and direction. The gradient of the pixel point (x, y) is:

[0053] G x = H(x + 1, y) - H(x - 1, y)

[0054] G y = H(x, y + 1) - H(x, y - 1)

[0055] G x , G y , and H are the horizontal gradient, vertical gradient, and pixel value of the pixel point (x, y) respectively. The gradient magnitude and gradient direction of the pixel point are:

[0056]

[0057]

[0058] Then the area is divided into multiple small cells, and the gradient histogram of each cell is statistically calculated. The 0-180 degrees is divided into 9 intervals, and the gradient direction histogram is calculated as the feature vector.

[0059] Normalize and truncate the eigenvectors in nine adjacent cells. Let C(i, j) denote the nine-dimensional eigenvector of cell (i, j).

[0060]

[0061] Define

[0062] N δ,γ (i, j) = (||C(i, j)|| 2 + ||C(i + δ, j)|| 2

[0063] + ||C(i, j + γ)|| 2 + ||C(i + δ, j + γ)|| 2 )

[0064] δ, γ ∈ {-1, 1}

[0065] T α T(x) = min(x, α)

[0066] Thus, a 36-dimensional eigenvector corresponding to the cell is obtained.

[0067] When the gradient is in the range of 0 - 180, the gradient is considered unsigned. According to the above process, a 36-dimensional eigenvector can be extracted, regarded as a 4×9 matrix, and the 13-dimensional eigenvector is obtained by adding the elements row by row and column by column. When the gradient is in the range of 0 - 360, the gradient is considered signed. According to the above process, it can be inferred that a 4×18-dimensional eigenvector is extracted, and the 18-dimensional eigenvector is obtained by summing the elements row by row. The 18-dimensional eigenvector and the 13-dimensional eigenvector obtained when unsigned together form a 31-dimensional eigenvector, which is the HOG feature and serves as the edge feature of the target region.

[0068] Concatenate the energy distribution feature of the target region and the edge feature of the target region as the feature c of the target.

[0069] Step 3 - 2: Establishment of the target template

[0070] To distinguish the target from the background, project the target region in the real-valued image into the kernel space and establish a classification function where c i is a row of c, w is a parameter, and the classification function is trained using the target template of the starting frame. The objective function is to minimize the squared error between the target template of the starting frame and the regression target u, and the expression is:

[0071]

[0072] where \(u\) is the regression target, which is generally set in the form of a Gaussian vector, \(u\) i is the regression target corresponding to \(c\) i ; \(\lambda\) is the regularization parameter for controlling overfitting. Define \(a\) i as the weight parameter. Similar to the method in linear classification, here we use a linear combination to represent the parameter \(w\), and the inner product of the kernel function is written as: Here, the Gaussian kernel is adopted as the inner product of the kernel function, which is expressed as:

[0073]

[0074] where \(\sigma\) 2 is the parameter for controlling the range of action of the Gaussian kernel function. The larger its value, the larger the local influence range of the Gaussian kernel function, and \(c\) j is the \(j\)-th row of \(c\).

[0075] Then the objective function can be written as:

[0076]

[0077] The target template is expressed as \(m\) represents the number of cyclic shifts, where \(a\) i is the \(i\)-th row of \(a=(K + \lambda I)\) -1 \(u\), and \(I\) is the identity matrix.

[0078] From the imaging result Figure 2 it can be seen that the target image has an extension in both the range and azimuth directions. Among them, the extension in the range direction is mainly determined by the range resolution, and the range resolution is inversely proportional to the signal bandwidth. The calculation is as follows:

[0079]

[0080] where \(c\) light is the speed of light, \(3\times10\) 8 m / s, and \(B\) is the stepped-frequency signal bandwidth. The extension of the target image in the azimuth direction is mainly determined by the azimuth resolution, and the azimuth resolution is defined as follows:

[0081]

[0082] where \(\lambda\) c is the signal wavelength, \(D\) is the equivalent virtual array aperture, \(\psi\) is the angle between the target and the array normal, and \(r\) is the distance between the target and the array center. From the definitions of the range resolution and azimuth resolution of the target image, it can be seen that the range resolution does not change with the position of the target, while the azimuth resolution is related to the target position.

[0083] In view of the fact that the scale of the through-wall radar target changes significantly with the target position, mainly in the azimuth direction, a scale pool S with large azimuth change but small range change is established. The scale pool parameters should not be set too large. For example, the azimuth scale pool is:

[0084] [1, 1.05, 1.1, 1.15]

[0085] The range scale pool is:

[0086] [1, 1.02, 1.04, 1.06]

[0087] Among them, the range scale pool values are obtained according to empirical values, and the azimuth scale pool values are obtained by debugging according to the azimuth resolution definition formula, the imaging area grid size and empirical values.

[0088] Step 4: Target tracking and scale estimation

[0089] After obtaining the imaging result of the k-th frame, assume that the target position of the (k - 1)-th frame is (x k-1 , y k-1 ). Select a rectangular candidate area with scale sizes {t k-1 , y k-1} centered on (x i s T |t i ∈ S}. Extract the target features for each scale candidate area to form a candidate area template z. The steps are as in 3-1. The target template and the candidate area template form the following response:

[0090]

[0091] To simplify the calculation process, using the circularity of the sample x, the response equation can be written as where ⊙ is the dot product, is the two-dimensional Fourier transform of a, is the two-dimensional discrete Fourier transform of the matrix obtained by taking the inner product of the target template x and the candidate area z to be measured. Perform the inverse Fourier transform of the response to the spatial domain, and the position of the maximum response can be regarded as the movement position of the target. The templates formed by candidate areas of different scale sizes are respectively used to make responses with the target template, and the scale of the candidate area corresponding to the maximum response is the estimated scale of the target area in this frame.

[0092]

[0093] where is the candidate area template to be measured with size t i s t , and F -1 is the inverse Fourier transform.

[0094] To enable the template Model to adapt to the changes in the target with scale and continuously learn the features of the new frame of the target, that is, the template Model obtained from the new frame (the k-th frame) now is linearly combined with the template Model obtained from the previous k - 1 frames k-1 to obtain the updated template Model k :

[0095] Model k = θModel now +(1 - θ)Model k-1

[0096] Model = [a T , c T T

[0097] where, a T represents the transpose of a, θ represents the learning rate, and its value is 0.1.

[0098] Finally, the updated template is used to locate the position of the target.

[0099] The effects of the present invention are further illustrated by the following simulation verification:

[0100] Simulation results:

[0101] In the simulation, the radar is located at the center of the front wall and close to the wall. The wall thickness is 0.2 m, the relative permittivity and conductivity of the wall are 4 and 0.1 respectively, the transmitted signal is a Ricker wavelet signal, and the center frequency is 2 GHz. To create a low signal-to-noise ratio environment, Gaussian white noise of -1 dB is added to the echo signal.

[0102] The results after imaging and preprocessing are as Figure 2 shown. It can be found that there are a large number of high-energy points in the imaging results, and the target is mixed with the background and difficult to distinguish. The proposed correlation filtering tracking method of the present invention can track the target more accurately compared with the improved mean shift algorithm.

[0103] The imaging results are as Figure 4 shown. After adding noise, problems such as cluttered background and low signal-to-noise ratio appear in the imaging results. Figure 4 ​The result of the algorithm proposed in the literature "Scale-Adaptive Human Target Tracking for Through-Wall Imaging Radar, IEEE Geosci. Remote Sensing Lett., pp. 1348-1352, 2020." The target tracking method based on correlation filtering proposed by the present invention effectively improves the tracking accuracy in a low signal-to-noise ratio environment and avoids track breaks caused by failure to track.

[0104] The simulation results show that the present invention can achieve precise tracking of hidden targets behind walls and can adapt to a low signal-to-noise ratio environment.

[0105] Those of ordinary skill in the art will realize that the embodiments described herein are to assist the reader in understanding the principles of the present invention and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. For those skilled in the art, the present invention can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of the claims of the present invention.

Claims

1. A method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments, characterized in that, Including: S1. Data acquisition: A two-transmit and four-receive ultra-wideband radar is used to perform a single-view scan on the concealed target behind the wall, and the receiving antenna receives the reflected signal. S2. Image the signal received in step S1: The received signal is focused and imaged using the back-projection algorithm. S3. Preprocess the imaging result obtained in step S2. The energy coherence factor algorithm is used to suppress grating lobes, and then the Gaussian filtering algorithm is used to smooth the imaging result. S4. According to the imaging result processed in step S3, when the target first appears, the target area is framed by the method of visual annotation. Considering the characteristic that the target response of the through-wall radar is a point spread function, the energy distribution feature and edge feature of the target area are jointly used to represent the target feature, and a target template is established based on the ridge regression algorithm. In view of the characteristic that the scale of the through-wall radar target changes significantly with distance, a scale pool S with large azimuthal variation but small range variation is established. S5. Select candidate areas according to the scale pool, and extract target features for each scale candidate area to form a candidate area template. S6. Use the tracking algorithm based on correlation filtering to establish the relationship between the candidate area template and the target template, and perform target tracking and estimate the scale of the target area.

2. The method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments according to claim 1, characterized in that, The energy distribution feature of the target area is specifically: The pixel values of the target area are unified between 0 and 1 as the energy distribution feature of the target area: where pix(i,j) is the pixel value at the coordinate (i,j) in the target area, and max(pix) is the maximum pixel value in the target area.

3. The method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments according to claim 2, characterized in that, The edge feature of the target area is specifically calculated using the histogram of oriented gradients.

4. The method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments according to claim 3, characterized in that, The process of establishing the target template is: Project the target region onto the kernel space and establish a classification function where the superscript T represents transpose, c i is a row of c, c is the target feature described in step S4, and w is a parameter; Adopt The parameter w is represented by a linear combination of, and the target template is represented as m represents the cyclic shift number, a i is a = (K + λI) -1 The i-th row of u, where u is the regression target, I is the identity matrix, and K represents The inner product of, and λ is the regularization parameter for controlling overfitting.

5. The method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments according to claim 4, characterized in that, The target template and the candidate area template form the following response: z is the candidate region template, and κ(c i , z) represents c i taking the inner product with z in the kernel.

6. The method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments according to claim 5, characterized in that, The candidate area templates composed of candidate areas of different scale sizes are respectively made to respond with the target template, and the response is inverse Fourier transformed to the spatial domain, and the position of the maximum response is regarded as the moving position of the target.

7. The method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments according to claim 6, characterized in that, After tracking the target in each frame, it also includes updating the target template; the update expression is: Model k = θModel now +(1 - θ)Model k-1 Model=[a T ,c T T ​ Among them, Model k-1 is the target template updated at the (k - 1)-th frame, and Model k is the target template updated at the k-th frame. Model now is the target template created from the target region features extracted with the target position as the center and the optimal scale as the region size after the target at the k-th frame has been tracked. The superscript T represents the transpose, and θ represents the learning rate.

8. The method for tracking hidden targets behind walls applicable to low signal-to-noise ratio environments according to claim 7, characterized in that, The scale pool S in step S4 specifically includes: The azimuth scale pool is: [1,1.05,1.1,1.15] The range scale pool is: [1,1.02,1.04,1.06]。

Images