A method for phased array radar dimension reduction four-channel and difference beam angle measurement

Abstract

The application discloses a method for phased array radar dimension reduction four-channel and difference beam angle measurement, relates to the technical field of angle measurement, and comprises the following steps: collecting a depth map and an image sequence of a target scene, pre-processing the depth map, generating a target depth map, establishing a phased array radar angle domain mapping relationship through the target depth map, generating a heat map of the target in the angle domain through the phased array radar angle domain mapping relationship, dividing phased array radar subarrays by using the heat map, optimizing the boundaries of the phased array radar subarrays through clustering, and outputting phased array radar subarray configurations; collecting phased array radar four-channel signals based on the phased array radar subarray configurations, synthesizing subarray levels and beams, azimuth difference beams and elevation difference beams, calculating target motion optical flow fields in combination with the image sequence and mapping the target motion optical flow fields to the phased array radar angle domain, and correcting the azimuth difference beams and the elevation difference beams; and the method corrects signal deviation caused by high-speed targets through motion compensation, and improves angle measurement accuracy in a dynamic scene.

Description

Technical Field

[0001] This invention relates to the field of angle measurement technology, and in particular to a method for angle measurement using a phased array radar with reduced-dimensional four-channel sum-difference beam. Background Technology

[0002] Phased array radar technology, due to its high-precision angle measurement capabilities and flexible beam control characteristics, has been widely used in fields such as air traffic management, unmanned aerial vehicles, military reconnaissance, and multi-target tracking. In recent years, with advancements in radar signal processing technology and multi-modal sensor fusion, the four-channel sum-difference beam angle measurement method for phased array radar has seen significant development. Traditional methods achieve high angular resolution by synthesizing beams from the four quadrants of the antenna array, including azimuth and elevation difference beams, and calculating target angles using beam ratios. For example, the phase difference of the difference beam signals can be used to estimate the target's azimuth and elevation angles. Furthermore, the integration of computer vision technology with radar has also improved angle measurement performance, enhancing target resolution in complex scenarios by using target detection through image sequences to assist radar angle measurement.

[0003] Despite significant progress in phased array radar angle measurement, there are still areas for improvement. For example, while existing vision-radar fusion technologies provide auxiliary information through target detection, they lack precise mapping between visual data and the radar angle domain, limiting the accuracy of cross-modal fusion, especially in low signal-to-noise ratio or high-speed target scenarios, where the fusion effect is limited. Furthermore, existing angle measurement methods do not adequately handle target motion compensation, and signal deviations caused by high-speed targets are not effectively corrected, leading to increased angle measurement errors. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar to solve the problems of lack of accurate mapping and insufficient processing of target motion compensation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar, comprising,

[0008] The depth map and image sequence of the target scene are acquired, the depth map is preprocessed to generate the target depth map, and the phased array radar angle domain mapping relationship is established through the target depth map;

[0009] A heatmap of the target in the angle domain is generated by mapping the phased array radar angle domain relationship. The heatmap is used to divide the phased array radar subarrays. The phased array radar subarray boundary is optimized by clustering, and the phased array radar subarray configuration is output.

[0010] Based on the phased array radar subarray configuration, the four-channel signals of the phased array radar are collected, and the subarray level and beam, azimuth difference beam and elevation difference beam are synthesized. The target motion optical flow field is calculated by combining the image sequence and mapped to the phased array radar angle domain to correct the azimuth difference beam and elevation difference beam.

[0011] Based on the corrected azimuth difference beam and elevation difference beam, a feature matrix is ​​constructed and its dimension is reduced. After obtaining the dimension-reduced feature matrix, it is weighted and fused to output the target width estimate.

[0012] Based on the target width estimate and dimensionality reduction features, the target angles at the subarray level are calculated through subarray level and beam, and the target angle set is output to the corresponding target width.

[0013] As a preferred embodiment of the method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar described in this invention, the step of establishing the phased array radar angle domain mapping relationship through the target depth map specifically includes the following steps.

[0014] The depth image and RGB image sequence of the target scene are captured. After denoising the depth image, the outline of the target object is extracted. The target region is segmented by combining the depth information to generate the target depth map.

[0015] The pixel coordinates of the target depth map are converted into phased array radar angle domain coordinates, forming a phased array radar angle domain mapping relationship.

[0016] As a preferred embodiment of the method for dimensionality reduction four-channel sum-difference beam angle measurement for phased array radar described in this invention, the method includes: generating a heatmap of the target in the angle domain through the phased array radar angle domain mapping relationship; using the heatmap to divide the phased array radar into subarrays; and optimizing the subarray boundaries through clustering to output the phased array radar subarray configuration. Specifically, this includes the following steps:

[0017] Based on the mapping relationship between the target depth map and the phased array radar angle domain, the distribution density of target points in the angle domain coordinates is statistically analyzed to generate a heat map;

[0018] Based on the density values ​​of the heatmap, the phased array radar antenna array is divided into phased array radar subarrays. The subarray region boundaries are optimized by clustering algorithm. The four-channel weights are assigned according to the subarray region size and the heatmap density values ​​to generate the phased array radar subarray configuration.

[0019] As a preferred embodiment of the method for angle measurement of four channels and difference beams for phased array radar described in this invention, the synthesized subarray level and beam, azimuth difference beam and elevation difference beam refer to the acquisition of four channel signals through the phased array radar subarray configuration, which correspond to the four quadrants of the antenna array, and the synthesis of the subarray level and beam, azimuth difference beam and elevation difference beam.

[0020] As a preferred embodiment of the method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar described in this invention, the step of calculating the target motion optical flow field by combining image sequences and mapping it to the angular domain of the phased array radar, and correcting the azimuth difference beam and elevation difference beam, specifically includes the following steps.

[0021] The target motion trajectory is calculated based on RGB image sequences, the motion velocity distribution is generated, and the phased array radar angle domain mapping relationship is used to convert it into an angle domain motion velocity distribution.

[0022] The Doppler frequency shift is calculated using four channel signals and converted into radial velocity. The visual optical flow radial velocity is calculated by combining the angular domain motion velocity distribution. The visual optical flow radial velocity and the radar radial velocity are compared to generate correction coefficients. The correction coefficients are then used to correct the azimuth difference beam and the elevation difference beam.

[0023] As a preferred embodiment of the method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar described in this invention, the step of constructing a feature matrix based on the corrected azimuth difference beam and elevation difference beam and performing dimension reduction to obtain the dimension-reduced feature matrix specifically includes the following steps.

[0024] Extract the sum beam, corrected azimuth difference beam and elevation difference beam of each phased array radar subarray to form a feature vector, combine the feature vectors of all phased array radar subarrays to construct a high-dimensional feature matrix.

[0025] Principal component analysis is applied to the high-dimensional feature matrix to generate a mean-centered feature matrix and an eigenvector matrix. The mean-centered feature matrix is ​​then projected onto the transpose of the eigenvector matrix to obtain the dimensionality-reduced feature matrix.

[0026] As a preferred embodiment of the method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar described in this invention, the step of performing weighted fusion to output a target width estimate specifically includes the following steps.

[0027] The target pixel width is extracted from the RGB image sequence, and the angular domain width is obtained by combining it with the target depth map.

[0028] Extract subarray feature vectors from the dimensionality-reduced feature matrix, construct fused feature vectors by combining the angular domain width, calculate the signal-to-noise ratio through subarray level and beam, and construct fused weight matrix;

[0029] Based on the fused feature vector and the fused weight matrix, matrix multiplication is used to weight the fused feature vector to generate a weighted fused feature vector;

[0030] The inter-class scatter matrix and intra-class scatter matrix are calculated using pre-labeled calibration data. The projection vector is obtained based on the inter-class scatter matrix and intra-class scatter matrix. The weighted fusion feature vector is mapped to the projection vector to generate a one-dimensional fusion feature value.

[0031] A linear regression model is trained based on pre-labeled calibration data. The trained linear regression model is obtained. One-dimensional fused feature values ​​are input into the linear regression model to obtain an initial width estimate. Post-processing is applied to the initial width estimate to limit the width range, and the target width estimate is output.

[0032] As a preferred embodiment of the method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar described in this invention, the method includes the following steps: based on the target width estimate and dimension reduction characteristics, the target angle at the subarray level is calculated through subarray level and beam, and the target angle set and corresponding target width are output.

[0033] Based on the dimensionality-reduced feature matrix, the feature vector of each phased array radar subarray is extracted, and the subarray level and beam are combined to calculate the subarray level target angle.

[0034] The density values ​​of the heatmap are used to weight the subarray-level target angles to generate weighted subarray-level target angles. The weighted subarray-level target angles are then clustered to classify multiple targets and generate a set of target angles.

[0035] By comparing the target angle with the position of the phased array radar subarray with the fused feature vector, the target width estimate is correlated with the corresponding target angle, and the target angle set and the corresponding target width estimate are output.

[0036] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the method for dimensionality reduction four-channel sum difference beam angle measurement for phased array radar as described in the first aspect of the present invention.

[0037] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for dimension reduction four-channel sum-difference beam angle measurement for phased array radar as described in the first aspect of the present invention.

[0038] The beneficial effects of this invention are as follows: By generating a target depth map, pixel coordinates are converted into phased array radar angle domain coordinates, forming an angle domain mapping relationship, achieving precise alignment between visual data and radar signals, and providing a high-quality foundation for cross-modal fusion. Subsequently, by extracting pixel width and converting it into angle domain width, and dynamically adjusting the visual width weight in combination with the signal-to-noise ratio, a weighted fusion feature vector is generated, realizing dynamic fusion of radar and visual data, and enhancing the fusion effect in low signal-to-noise ratio scenarios. In addition, by calculating Doppler frequency shift using fast Fourier transform to generate correction coefficients, the azimuth difference beam and elevation difference beam are corrected, realizing multimodal motion compensation of visual and radar data, directly solving the problem of high-speed target signal deviation. Through motion compensation, the signal deviation caused by high-speed targets is corrected, improving the angle measurement accuracy in dynamic scenarios. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a method for dimension reduction four-channel sum-difference beam angle measurement for phased array radar.

[0041] Figure 2 This is a schematic diagram of heatmap generation and subarray partitioning.

[0042] Figure 3 This is a flowchart of signal acquisition and motion compensation.

[0043] Figure 4 A flowchart for target estimation and angle output. Detailed Implementation

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0047] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar, comprising the following steps:

[0048] S1: Acquire depth maps and image sequences of the target scene, preprocess the depth maps to generate target depth maps, and establish phased array radar angle domain mapping relationships through target depth maps;

[0049] Specifically, the steps include the following:

[0050] A high-resolution depth camera and an RGB camera are coaxially mounted with the phased array radar to ensure that the fields of view of the depth camera and RGB camera are consistent with the field of view of the phased array radar, thus achieving synchronous data acquisition. Specifically, the depth camera captures the distance information of target objects in the target scene and generates a depth image containing RGB color information and depth information. The RGB camera acquires a continuous sequence of RGB images of the target scene.

[0051] A 3×3 median filter is applied to the depth image to remove abnormal points caused by ambient light or sensor noise. After denoising, the contour of the target object in the depth image is extracted using the Canny edge detection algorithm to generate an edge image. Combined with the depth information in the depth image, the edge image is segmented to separate the target object from the background and generate a target depth map, which represents the spatial distribution of the target object in the scene.

[0052] To further explain, through depth image preprocessing and target region segmentation, the generated target depth map accurately reflects the spatial distribution of the target, providing high-quality input for subsequent heatmaps and dynamic subarray division, and significantly improving the resolution in multi-target scenes.

[0053] Based on the target depth map, a calibration board is used to calibrate the relative position of the depth camera and the phased array radar, and obtain the intrinsic and extrinsic parameters of the depth camera. The intrinsic parameters include focal length and pixel size, while the extrinsic parameters are calculated by the calibration board. This describes the spatial transformation relationship between the depth camera and the phased array radar. The calibration process ensures that the pixel coordinates of the target depth map are aligned with the viewpoint of the phased array radar, forming a unified coordinate reference system.

[0054] Based on the intrinsic and extrinsic parameters of the depth camera, the pixel coordinates in the target depth map are converted into the angular domain coordinates of the phased array radar. For each pixel in the target depth map, the corresponding azimuth and elevation angles are calculated based on the intrinsic parameters of the depth camera, forming the angular domain mapping relationship of the phased array radar.

[0055] To further explain, through calibration board calibration and pixel coordinate to angle domain conversion, the visual data and phased array radar signal were precisely aligned, laying the foundation for subsequent visual-assisted dynamic subarray reconstruction and multimodal motion compensation, and reducing the angle measurement error caused by sensor deviation.

[0056] Extract consecutive frame data from the RGB image sequence, retain the complete RGB image sequence, containing image data at 60 frames per second and a resolution of 1920×1080, and ensure that its timestamp is synchronized with the target depth map.

[0057] S2: Generate a heat map of the target in the angle domain through the phased array radar angle domain mapping relationship, use the heat map to divide the phased array radar subarray, optimize the edges of the phased array radar subarray through clustering, and output the phased array radar subarray configuration.

[0058] Specifically, the steps include the following:

[0059] S2.1: Based on the mapping relationship between the target depth map and the phased array radar angle domain, analyze the density distribution of targets in the scene. Specifically, count the number of target points corresponding to each pixel in the target depth map, and calculate the distribution density of targets in the phased array radar angle domain coordinates. Using the Gaussian kernel density estimation method, generate a heatmap reflecting the spatial distribution of targets. The heatmap uses azimuth and elevation angles as coordinates. The expression for generating the target heatmap in the phased array radar angle domain is:

[0060]

[0061] Where H(θ,φ) represents the density value of the heatmap at the angular coordinates (θ,φ), θ represents the azimuth angle, φ represents the elevation angle, N represents the total number of target points in the target depth map, i represents the index of the target point, K represents the Gaussian kernel function, and θ i φ represents the azimuth angle of the i-th target point. i Let represent the pitch angle of the i-th target point, and σ represent the bandwidth of the Gaussian kernel function.

[0062] Heatmaps identify high values ​​for densely packed target areas and low values ​​for sparsely packed areas. Dense areas indicate that target objects are concentrated in one area, while sparse areas indicate that there are few or no target objects. Heatmaps provide spatial information on target density for subarray division.

[0063] To further explain, by using heatmaps to identify high-value areas as dense target areas and low-value areas as sparse areas, it is possible to effectively distinguish the distribution of targets in complex scenes. This capability is particularly suitable for multi-target tracking scenarios, improving the accuracy of target identification.

[0064] S2.2: Based on the heat map, the subarray division of the phased array radar is dynamically adjusted. Specifically, the phased array radar antenna array consists of 16×16 elements. The density threshold is set based on half of the maximum density value of the heat map. Areas with density higher than the density threshold in the heat map are divided into smaller 2×2 or 4×4 subarrays to improve target resolution. Areas with density lower than or equal to the density threshold are merged into larger 8×8 subarrays to reduce the amount of computation. Each subarray corresponds to a set of antenna elements, forming a preliminary subarray division configuration.

[0065] Based on the heat map and preliminary subarray division configuration, the initial number of subarrays in the phased array radar angular domain coordinates is determined as follows: (1) The initial number of subarrays is set to 4. The K-means clustering algorithm is applied to randomly select 4 points from the density distribution data in the heat map as the initial cluster center points. Each center point represents the initial center of a subarray region, and the coordinates are the azimuth and elevation angles in the corresponding phased array radar angular domain. The selection of the initial cluster center points is based on the density value of the heat map, and the coordinates of high-density regions are selected first to ensure that the initial center is close to the dense target region.

[0066] (2) Calculate the Euclidean distance between each angular coordinate (θ, φ) in the heatmap and all initial cluster centers. Based on the minimum distance principle, assign each angular coordinate to its corresponding sub-region, forming preliminary sub-regions. Using the preliminary sub-regions and the heatmap, recalculate the cluster centers for each sub-region. For each sub-region, take the average of all angular coordinates assigned to that sub-region as the new cluster center. The expression for updating the cluster centers is:

[0067]

[0068] Where, μ k C represents the cluster center of the k-th subarray region. k This represents the k-th subarray region.

[0069] (3) Repeatedly assign the angle domain coordinates of the heat map to the subarray region and perform iterative optimization until the cluster center point changes less than the preset clustering threshold or reaches the maximum number of iterations. The preset clustering threshold is set according to the requirements, such as 0.01°. The optimized subarray region boundary is more in line with the distribution characteristics of dense and sparse areas of the target in the heat map.

[0070] (4) Based on the optimized subarray regions and heatmaps, assign corresponding four-channel weights to each subarray region. The specific process is as follows: Each subarray region corresponds to a set of antenna elements in the phased array radar antenna array. According to the size of the subarray region (2×2, 4×4 or 8×8) and the density value of the heatmap, assign weights to adjust the sensitivity of subsequent signal processing. High-density regions are assigned higher weights, and low-density regions are assigned lower weights. For example, if the heatmap density value is 0.8, assign a weight of 1.0 to enhance the signal processing sensitivity, so that the 2×2 subarray can prioritize processing dense targets and improve resolution; if the heatmap density value is 0.2, assign a weight of 0.5 to reduce the signal processing intensity, so that the 8×8 subarray can cover sparse targets and reduce the amount of computation. Through this weight allocation, high-density regions obtain more refined angle measurement capabilities, while low-density regions optimize computational efficiency and generate an optimized phased array radar subarray configuration, including the boundary of each subarray region and the four-channel weights.

[0071] To further explain, by assigning four-channel weights to the optimized subarray region and adjusting the signal processing sensitivity based on the subarray region size and heatmap density, the signal processing sensitivity can be dynamically adjusted, and resources can be allocated in a targeted manner, thereby improving overall performance.

[0072] S3: Based on the phased array radar subarray configuration, collect the four-channel signals of the phased array radar, synthesize the subarray level and beam, azimuth difference beam and elevation difference beam, combine the image sequence to calculate the target motion optical flow field and map it to the phased array radar angle domain, and correct the azimuth difference beam and elevation difference beam.

[0073] Specifically, the steps include the following:

[0074] By configuring the phased array radar subarrays, signals from four channels are acquired from the phased array radar antenna array, corresponding to the four quadrants of the antenna array (upper left, upper right, lower left, and lower right), generating four channel signals, denoted as S1, S2, S3, and S4, corresponding to the four quadrants of each subarray. Each subarray and beam is obtained by adding the four channel signals. The azimuth difference beam is generated by subtracting the signals from the left and right quadrants, and the elevation difference beam is generated by subtracting the signals from the upper and lower quadrants, thus obtaining the subarray-level beam, azimuth difference beam, and elevation difference beam. The subarray-level beam, azimuth difference beam, and elevation difference beam provide support for subsequent motion compensation.

[0075] Based on RGB image sequences, the Lucas-Kanade optical flow algorithm is applied to analyze pixel displacement between adjacent frames and calculate the target's motion trajectory. The process is as follows: Select two adjacent frames from the RGB image sequence. Using the Lucas-Kanade optical flow algorithm, it is assumed that pixels within the target region have similar motion speeds and constant brightness. By calculating the grayscale gradient of pixels in the target region in each frame and combining it with pixel intensity changes over time, an optical flow constraint equation is constructed. Within the target region, with a small window (e.g., 5×5 pixels) as the center, the optical flow vector (horizontal and vertical velocity components) is solved. The optical flow constraint of pixels within the window is optimized using the least squares method to obtain the motion speed estimate of each pixel. The optical flow vectors of all pixels within the target region are summarized to form the target's motion trajectory, generating a preliminary motion speed distribution. Gaussian filtering is applied to the preliminary motion speed distribution for smoothing, reducing errors caused by image noise or illumination changes. The resulting motion speed distribution of the target on the image plane is represented as a vector field containing horizontal and vertical velocity components, with units of pixels per second.

[0076] By analyzing the velocity distribution, and utilizing the phased array radar's angular domain mapping relationship, the velocity distribution on the image plane is transformed into a velocity distribution in the phased array radar's angular domain. This transformation process uses the intrinsic parameters of the depth camera to convert horizontal and vertical velocity components into angular domain velocity components, completing the mapping and obtaining an angular domain velocity distribution representing the target's motion direction information in the phased array radar's angular domain. Based on the four channel signals S1, S2, S3, and S4, a Fast Fourier Transform (FFT) is applied to the signals of each subarray. The Doppler frequency shift is calculated using the FFT, expressed as:

[0077] f d =argmax f |FFT(S i (t))|;

[0078] Among them, f d Let f represent the frequency component in the spectrum of the Fast Fourier Transform output, and FFT represent the Fast Fourier Transform. i Let t represent the time-domain signal of the i-th channel of the phased array radar, where t represents time.

[0079] The calculated Doppler frequency shift is converted into radial velocity. Then, based on the angular domain motion velocity distribution, the radial velocity component of the visual optical flow is calculated, and the expression is as follows:

[0080] v ' r =V ' ·cos(α);

[0081] Among them, v ' rV represents the radial velocity component of the visual optical flow. ' Let α represent the velocity distribution in the angular domain, cos represent the cosine, and α represent the angle between the motion vector and the radar line of sight.

[0082] By comparing the radial velocity of the visual optical flow and the radial velocity of the radar, correction coefficients are generated to compensate for signal deviations caused by high-speed targets. Based on the azimuth difference beam, the elevation difference beam, and the correction coefficients, the difference beam of each subarray is corrected to generate the corrected azimuth difference beam and the elevation difference beam, thereby obtaining the corrected signal set of each subarray, which includes the sum beam, the corrected azimuth difference beam, and the elevation difference beam.

[0083] To further explain, by acquiring four-channel signals from a phased array radar, synthesizing subarray-level sum and difference beams, calculating the target motion optical flow field, mapping it to the phased array radar angular domain, extracting Doppler frequency shift and generating correction coefficients, and correcting the difference beam, multi-modal fusion and motion compensation of visual and radar data are achieved, providing a high-precision signal processing foundation for the phased array radar dimension-reduced four-channel sum and difference beam angle measurement method.

[0084] S4: Construct a feature matrix based on the corrected azimuth difference beam and elevation difference beam and perform dimensionality reduction. After obtaining the dimensionality-reduced feature matrix, perform weighted fusion to output the target width estimate.

[0085] Specifically, the steps include the following:

[0086] Based on the corrected signal set of each subarray, the sum beam, corrected azimuth difference beam and elevation difference beam of each subarray are extracted. Auxiliary channels are ignored to form feature vectors. The feature vectors of all subarrays are combined to construct a high-dimensional feature matrix, where each row corresponds to the feature vector of a subarray. The high-dimensional feature matrix contains the angle measurement information of each subarray.

[0087] Principal component analysis (PCA) is applied to reduce the dimensionality of the high-dimensional feature matrix. Specifically, based on the high-dimensional feature matrix, the mean of each column of eigenvalues ​​is subtracted to generate a mean-centered feature matrix. The covariance matrix of the mean-centered feature matrix is ​​calculated, which captures the variability and correlation structure of the features. The covariance matrix is ​​sorted by eigenvalues ​​from largest to smallest, and the eigenvectors corresponding to the first two eigenvalues ​​are selected. The two selected eigenvectors represent the directions of the largest variance in the feature space and are combined to form an eigenvector matrix. The mean-centered feature matrix is ​​projected onto the transpose of the eigenvector matrix to generate the dimensionality-reduced feature matrix. The projection process transforms the eigenvectors of each subarray from 3D space to 2D space. The dimensionality-reduced feature matrix retains the angular measurement information of the beam, the corrected azimuth difference beam, and the corrected elevation difference beam, while significantly reducing computational complexity by eliminating redundant feature dimensions.

[0088] S4.1: By applying the YOLOv8 target detection algorithm to the image sequence, the bounding box of the target is identified, the position and range of the target on the image plane are determined, and the pixel width of the target is extracted from the bounding box to represent the horizontal dimension of the target on the image plane. Based on the pixel width and the target depth map, the distance information of the target is extracted. Using the intrinsic parameters of the depth camera, the pixel width is converted into the angular domain width to represent the width information of the target in the angular domain of the phased array radar.

[0089] The feature vectors of the corresponding rows are extracted from the dimensionality-reduced feature matrix. Each vector contains two principal components after dimensionality reduction, representing the angle measurement information of the radar signal. The fused feature vectors are constructed by directly splicing the angular domain width. This process integrates radar signal features (dimensionality-reduced sum and difference beam information) and visual width information (angular domain size of the target). Before splicing, the feature vectors of the corresponding rows and the angular domain width are normalized to ensure that the dimensions of each component are consistent.

[0090] Based on the subarray level and beam, for each subarray, the current signal-to-noise ratio (SNR) is estimated by calculating the power of the subarray level and beam. The power calculation is based on the sum of squares of the time-domain signal amplitudes of the subarray level and beam, and the noise power is estimated from the signal in the background region. The SNR is defined as the ratio of signal power to noise power, and the calculation formula is as follows:

[0091]

[0092] Among them, SNR k P represents the signal-to-noise ratio of the k-th subarray. signal,k P represents the signal power of the k-th subarray. noise,k This represents the noise power of the k-th subarray.

[0093] A sliding window averaging method is used to smooth the obtained signal-to-noise ratio in order to reduce the estimation error caused by instantaneous noise fluctuations.

[0094] Based on the current signal-to-noise ratio (SNR), a fusion weight matrix is ​​constructed. This is a diagonal matrix containing three weight values, corresponding to two radar signal feature components and one visual width prior component in the fused feature vector. Weight 3 is dynamically adjusted according to the SNR. The adjustment process is as follows: the maximum SNR is determined through calibration experiments. When the SNR is lower than the maximum SNR, the weight of the visual width prior is increased to enhance the contribution of visual information in low SNR scenarios. When the SNR is close to or higher than the maximum SNR, weight 3 is decreased, prioritizing the reliance on radar signal features. Weights 1 and 2 remain equal, with a value of (1 - weight 3) / 2, ensuring that the total weight sum is 1.

[0095] Based on the fused feature vector and fused weight matrix, the fused feature vector is weighted through matrix multiplication to generate a weighted fused feature vector. The weighting process is as follows: two radar signal feature components and a visual width prior component are multiplied by their corresponding weights 1, 2, and 3, respectively, so that the contributions of the radar signal features and the visual width prior are dynamically adjusted according to the signal-to-noise ratio (SNR). In low SNR scenarios (e.g., SNR below 50% of the preset maximum SNR), weight 3 is increased, for example, set to 0.6, to enhance the contribution of the visual width prior and compensate for noise interference in the radar signal. In high SNR scenarios (e.g., SNR close to or higher than the preset maximum SNR), weights 1 and 2 are increased (e.g., each set to 0.45), so that the radar signal features dominate, to fully utilize the high-precision angle measurement information of the radar signal. Before weighting, each component of the fused feature vector is standardized using z-score standardization to eliminate the influence of dimensional differences on the weighting result. After weighting, the numerical stability of the weighted fused feature vector is checked to ensure that all components are within a reasonable range, such as [-10, 10], to avoid outliers affecting subsequent processing. The weighted fused feature vector integrates the dynamic contributions of radar signal features and visual width prior. The L2 norm normalization method is applied to the weighted fused feature vector to scale its magnitude to 1, eliminating potential scale differences after weighting and improving the stability of subsequent dimensionality reduction.

[0096] S4.2: First, acquire pre-labeled calibration data. This pre-labeled calibration data includes training datasets for various target width categories. The data is collected by phased array radar and depth cameras in various scenarios, including RGB image sequences of different target types, target depth maps, and corresponding four-channel phased array radar signals. Each sample is labeled with the target's true width, expressed in angular domain width in degrees. It is divided into multiple width categories and follows the same format as the fused feature vector. The calibration data covers both high and low signal-to-noise ratio (SNR) scenarios to ensure adaptability to different environmental conditions, such as sunny days, foggy days, and nighttime.

[0097] Using pre-labeled calibration data and normalized weighted fused feature vectors, the inter-class scatter matrix and intra-class scatter matrix are calculated. The inter-class scatter matrix reflects the degree of separation between different target width classes by calculating the weighted covariance between the mean vector of each width class and the overall mean vector. The intra-class scatter matrix reflects the feature dispersion within the same class by calculating the sum of the covariances of samples within each width class relative to the class mean. The intra-class scatter matrix is ​​regularized by adding a small perturbation to ensure matrix invertibility and prevent singular matrix problems.

[0098] Based on the inter-class scatter matrix and the intra-class scatter matrix, the optimization objective of linear discriminant analysis is defined as maximizing the ratio of inter-class scatter to intra-class scatter, formalized as an objective function on the projection vector. The optimization objective is rewritten as a generalized eigenvalue problem. By utilizing the relationship between the inter-class scatter matrix and the intra-class scatter matrix, the eigenvalues ​​and eigenvectors are solved. The generalized eigenvalue problem is solved by calculating the product of the inverse of the intra-class scatter matrix and the inter-class scatter matrix, and selecting the eigenvector corresponding to the largest eigenvalue as the projection vector. The normalized weighted fusion eigenvector is projected onto the projection vector to generate one-dimensional fusion eigenvalues. The projection process is completed through matrix multiplication, reducing the 3-dimensional weighted fusion eigenvector to a one-dimensional scalar, retaining the key information of the cross-modal fusion features, and using pre-labeled calibration data for validity verification, resulting in verified one-dimensional fusion eigenvalues.

[0099] S4.3. Train a linear regression model using pre-labeled calibration data. The input to the linear regression model is a one-dimensional fusion feature value, and the target width is the output. The optimization objective is to minimize the squared error between the predicted width and the true width. The training process uses samples from the pre-labeled calibration data and optimizes the slope and intercept of the fitted linear regression model using the least squares method to ensure that the linear regression model can accurately map the one-dimensional fusion feature value to the target width, thus obtaining the trained linear regression model.

[0100] The latest one-dimensional fusion feature value is input into the trained linear regression model to generate an initial width estimate. The initial width estimate is then post-processed to ensure that the result is within a reasonable range. Specifically, calibration data analysis is performed based on the angular domain resolution capability of the phased array radar to determine the width range limit. If the initial width estimate exceeds the width range limit, it is truncated to the boundary value to avoid the impact of outliers on subsequent analysis and improve the reliability of the estimation result.

[0101] To further explain, by constructing a high-dimensional feature matrix, performing dimensionality reduction processing, extracting visual width priors, fusing dynamic weights, performing dimensionality reduction through linear discriminant analysis, and estimating through linear regression, cross-modal fusion of radar and visual data was achieved, providing a high-precision and high-accuracy target width estimation for the dimensionality reduction four-channel sum-difference beam angle measurement method of phased array radar.

[0102] S5: Based on the target width estimate and dimensionality reduction features, calculate the target angle at the subarray level and the beam through subarray level and output the target angle set with the corresponding target width.

[0103] Specifically, the steps include the following:

[0104] Based on the dimensionality-reduced feature matrix and heatmap, the target angles at the subarray level are calculated and clustered. Combined with the target width estimate, a set of target angles and corresponding target width estimates are generated. The specific process is as follows: Based on the dimensionality-reduced feature matrix, the feature vector of each subarray is extracted from the dimensionality-reduced feature matrix. It contains two principal components after dimensionality reduction, representing the angle measurement information of the corrected azimuth difference beam and elevation difference beam. Combining the subarray level and beam, the target azimuth and elevation angles of each subarray are calculated. The azimuth angle is obtained by dividing the first component of the feature vector of each subarray by the subarray level and beam and multiplying it by a calibration constant. The calibration constant is measured through the calibration scenario of the phased array radar. The elevation angle is obtained by dividing the second component of the feature vector of each subarray by the subarray level and beam and multiplying it by another calibration constant. The target azimuth and elevation angles of the subarray are also the target angles at the subarray level.

[0105] Based on the subarray-level target angles and heatmaps, the subarray-level target angles are weighted according to the density values ​​of the heatmaps. Subarray angles in high-density areas are given higher weights, while those in low-density areas are given lower weights. The weighted subarray-level target angles are generated by multiplying the subarray-level target angles by the heatmap density values.

[0106] By weighting the target angles at the subarray level, the DBSCAN algorithm is applied to cluster the azimuth and elevation angles of all subarrays, setting the cluster radius and minimum number of points to separate multiple targets and generate a preliminary target angle set. Based on the preliminary target angle set, the weighted average of the subarray angles within each cluster group is calculated, with the weights determined by the heatmap density values, generating a refined target angle set containing the azimuth and elevation angles of each target. According to the refined target angle set and the target width estimate, the target width estimate is associated with the corresponding target angle by comparing the target angle with the subarray position of the fused feature vector. The association process is as follows: the fused feature vector comes from the dimensionality reduction features and visual width prior of each subarray, and the subarray position is associated with the angular domain coordinates through the spatial index of the phased array radar, mapping the target angle to the nearest subarray position, thereby assigning the target width estimate to the corresponding target angle, forming a target angle set and a corresponding target width estimate.

[0107] This embodiment also provides a computer device applicable to the method of reducing the dimension of four channels and differential beam angle measurement for phased array radar, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method of reducing the dimension of four channels and differential beam angle measurement for phased array radar as proposed in the above embodiment.

[0108] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0109] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for dimension reduction four-channel sum-difference beam angle measurement for phased array radar as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0110] In summary, this invention achieves precise alignment between visual data and radar signals by generating a target depth map, converting pixel coordinates into phased array radar angle domain coordinates, and establishing an angle domain mapping relationship. This provides a high-quality foundation for cross-modal fusion. Subsequently, by extracting pixel widths and converting them into angle domain widths, and dynamically adjusting the visual width weights based on the signal-to-noise ratio, a weighted fusion feature vector is generated. This enables dynamic fusion of radar and visual data, enhancing the fusion effect in low signal-to-noise ratio scenarios. Furthermore, by calculating Doppler frequency shifts using fast Fourier transform to generate correction coefficients, the azimuth and elevation difference beams are corrected, achieving multimodal motion compensation between visual and radar data. This directly solves the problem of high-speed target signal deviation. Through motion compensation, the signal deviation caused by high-speed targets is corrected, improving the angle measurement accuracy in dynamic scenarios.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar, characterized in that: include, The depth map and image sequence of the target scene are acquired, the depth map is preprocessed to generate the target depth map, and the phased array radar angle domain mapping relationship is established through the target depth map; A heatmap of the target in the angle domain is generated by mapping the phased array radar angle domain relationship. The heatmap is used to divide the phased array radar subarrays. The phased array radar subarray boundary is optimized by clustering, and the phased array radar subarray configuration is output. Based on the phased array radar subarray configuration, the four-channel signals of the phased array radar are collected, and the subarray level and beam, azimuth difference beam and elevation difference beam are synthesized. The target motion optical flow field is calculated by combining the image sequence and mapped to the phased array radar angle domain to correct the azimuth difference beam and elevation difference beam. The process of calculating the target's motion optical flow field by combining image sequences and mapping it to the phased array radar's angular domain, and correcting the azimuth and elevation difference beams, specifically includes the following steps. The target's motion trajectory is calculated based on RGB image sequences, a motion velocity distribution is generated, and then converted into an angle domain motion velocity distribution using the phased array radar angle domain mapping relationship; The Doppler frequency shift is calculated using four channel signals and converted into radial velocity. The visual optical flow radial velocity is calculated by combining the angular domain motion velocity distribution. The visual optical flow radial velocity and radar radial velocity are compared to generate correction coefficients. The correction coefficients are then used to correct the azimuth difference beam and elevation difference beam. Based on the corrected azimuth difference beam and elevation difference beam, a feature matrix is ​​constructed and its dimension is reduced. After obtaining the dimension-reduced feature matrix, it is weighted and fused to output the target width estimate. Based on the target width estimate and dimensionality reduction features, the target angles at the subarray level are calculated through subarray level and beam, and the target angle set is output to the corresponding target width.

2. The method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar as described in claim 1, characterized in that: The establishment of the phased array radar angle domain mapping relationship through the target depth map specifically includes the following steps. The depth image and RGB image sequence of the target scene are captured. After denoising the depth image, the outline of the target object is extracted. The target region is segmented by combining the depth information to generate the target depth map. The pixel coordinates of the target depth map are converted into phased array radar angle domain coordinates, forming a phased array radar angle domain mapping relationship.

3. The method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar as described in claim 2, characterized in that: A heatmap of the target in the angle domain is generated by mapping the phased array radar angle domain relationship. The heatmap is then used to divide the phased array radar into subarrays. The subarray boundaries are optimized by clustering, and the phased array radar subarray configuration is output. The specific steps include the following: Based on the mapping relationship between the target depth map and the phased array radar angle domain, the distribution density of target points in the angle domain coordinates is statistically analyzed to generate a heat map; Based on the density values ​​of the heatmap, the phased array radar antenna array is divided into phased array radar subarrays. The subarray region boundaries are optimized by clustering algorithm. The four-channel weights are assigned according to the subarray region size and the heatmap density values ​​to generate the phased array radar subarray configuration.

4. The method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar as described in claim 3, characterized in that: The synthesized subarray level and beam, azimuth difference beam and elevation difference beam refer to the acquisition of four channel signals through the phased array radar subarray configuration, which correspond to the four quadrants of the antenna array, and the synthesized subarray level and beam, azimuth difference beam and elevation difference beam.

5. The method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar as described in claim 4, characterized in that: The step of constructing a feature matrix based on the corrected azimuth and elevation difference beams and then performing dimensionality reduction to obtain a dimensionality-reduced feature matrix includes the following steps. Extract the sum beam, corrected azimuth difference beam and elevation difference beam of each phased array radar subarray to form a feature vector, combine the feature vectors of all phased array radar subarrays to construct a high-dimensional feature matrix. Principal component analysis is applied to the high-dimensional feature matrix to generate a mean-centered feature matrix and an eigenvector matrix. The mean-centered feature matrix is ​​then projected onto the transpose of the eigenvector matrix to obtain the dimensionality-reduced feature matrix.

6. The method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar as described in claim 5, characterized in that: The weighted fusion process to output the target width estimate includes the following steps: The target pixel width is extracted from the RGB image sequence, and the angular domain width is obtained by combining it with the target depth map. Extract subarray feature vectors from the dimensionality-reduced feature matrix, construct fused feature vectors by combining the angular domain width, calculate the signal-to-noise ratio through subarray level and beam, and construct fused weight matrix; Based on the fused feature vector and the fused weight matrix, matrix multiplication is used to weight the fused feature vector to generate a weighted fused feature vector; The inter-class scatter matrix and intra-class scatter matrix are calculated using pre-labeled calibration data. The projection vector is obtained based on the inter-class scatter matrix and intra-class scatter matrix. The weighted fusion feature vector is mapped to the projection vector to generate a one-dimensional fusion feature value. A linear regression model is trained based on pre-labeled calibration data. The trained linear regression model is obtained. One-dimensional fused feature values ​​are input into the linear regression model to obtain an initial width estimate. Post-processing is applied to the initial width estimate to limit the width range, and the target width estimate is output.

7. The method for dimension-reduced four-channel sum-difference beam angle measurement for phased array radar as described in claim 6, characterized in that: Based on the target width estimate and dimensionality reduction features, the target angles at the subarray level are calculated through subarray level and beamforming, outputting a set of target angles and corresponding target widths. Specifically, the process includes the following steps. Based on the dimensionality-reduced feature matrix, the feature vector of each phased array radar subarray is extracted, and the subarray level and beam are combined to calculate the subarray level target angle. The density values ​​of the heatmap are used to weight the subarray-level target angles to generate weighted subarray-level target angles. The weighted subarray-level target angles are then clustered to classify multiple targets and generate a set of target angles. By comparing the target angle with the position of the phased array radar subarray with the fused feature vector, the target width estimate is correlated with the corresponding target angle, and the target angle set and the corresponding target width estimate are output.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for dimension reduction four-channel sum difference beam angle measurement for phased array radar as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for dimension reduction four-channel sum difference beam angle measurement for phased array radar as described in any one of claims 1 to 7.

Images