Millimeter wave radar tunnel scene multipath ghost suppression method

Abstract

The application discloses a kind of multi-path ghost suppression method of millimeter wave radar tunnel scene, it is applied to millimeter wave radar target detection field, for the poor multi-path ghost suppression effect in tunnel scene of prior art;The application first constructs confidence criterion using the multidimensional information point cloud data generated by radar original echo, and carries out density-based clustering to point cloud data;Then linear fitting is carried out to zero doppler point cloud to obtain the slope information of tunnel wall, simultaneously, the median line slope of dynamic point cloud pair is matched with the slope set, and the target pair associated successfully is judged as multi-path ghost with lower confidence;Finally, considering the motion characteristics of target in tunnel, as well as multi-path reflection geometric model, the double rectangular region division of transverse and longitudinal is carried out to dynamic point cloud, and the point cloud data in region is subjected to Doppler information constraint, and the target satisfying the condition and with lower confidence is screened out and judged as multi-path ghost.The application can detect vehicle target in semi-closed environment, while retaining real target, can effectively identify and suppress the multi-path ghost generated by reflection of tunnel wall and railing etc..

Description

Technical Field

[0001] This invention belongs to the field of millimeter-wave radar target detection, and specifically relates to a multipath ghosting suppression technology. Background Technology

[0002] Millimeter-wave radar target detection technology detects, locates, and tracks targets by utilizing the echo signals generated from the reflected signals. In intelligent transportation systems, millimeter-wave radar is a crucial sensing unit, playing a vital role in target detection, autonomous driving environmental perception, and traffic information collection due to its unique advantages such as strong anti-interference capabilities in complex road environments and immunity to weather conditions. During radar detection, electromagnetic wave propagation involves reflection and diffraction, resulting in multipath signals in addition to the direct line-of-sight signal. The ghosting caused by multipath signals during environmental perception and target detection not only interferes with the detection and tracking of real targets but also increases computational load during data processing, leading to unnecessary waste of computing resources. This is particularly problematic in tunnel traffic radar scenarios, where the presence of multiple walls and railings in semi-enclosed spaces generates numerous multipath ghostings due to static reflections. Furthermore, the multiple reflections of electromagnetic waves between different vehicles in tunnel scenarios further exacerbate the interference with the detection and tracking of real targets. Therefore, suppressing multipath ghosting in tunnel scenarios is crucial.

[0003] Many research institutions both domestically and internationally have conducted research on multipath ghosting suppression methods in tunnel environments. A team from Xi'an University of Electronic Science and Technology proposed a method for removing multipath ghosting in tunnels based on multi-radar video in their patent, "A Method for Removing Multipath Ghosting in Tunnels Based on Multi-Radar Video." This method utilizes the fusion of millimeter-wave radar and visual information to obtain more comprehensive target information. It continues the system track ID and visual information by matching targets in overlapping areas of adjacent radars, and uses the presence or absence of visual information in the target to eliminate multipath ghosting. While this method achieves multipath interference suppression through multi-radar and visual fusion perception, the visual information in this method is significantly affected by lighting conditions, and the use of multiple radars for target matching requires high costs. A team from Tongji University proposed a method for removing mirror images from tunnel trajectory data based on millimeter-wave radar in their patent, "A Method for Removing Mirror Images from Tunnel Trajectory Data Based on Millimeter-Wave Radar." This method estimates the tunnel walls to eliminate mirrored vehicle trajectory data outside the tunnel wall area. It also establishes a trajectory credibility evaluation index, scoring the credibility of target trajectories and eliminating vehicle trajectory data with scores lower than the lane credibility score, thus achieving the identification and suppression of multipath ghosting. However, this method only identifies and suppresses mirror ghosting caused by tunnel walls and cannot suppress multipath interference caused by multiple reflections within the tunnel. Meanwhile, researchers from Xi'an University of Electronic Science and Technology proposed a method for removing multipath interference from tunnel tracks based on zero-Doppler clutter points in their patent, "A Method for Removing Multipath Interference from Tunnel Tracks Based on Zero-Doppler Clutter Points." This method obtains the distance and velocity of the target inside the tunnel, uses the maximum likelihood algorithm to obtain the angles of the left and right multipaths of the target, generates the zero-channel clutter point vector at each moment, and removes multipath interference in the track by using the target area and the multipath area. This method divides the multipath area by using the target information and uses certain target prior information. At the same time, it cannot interfere with or suppress ghosting in the tunnel. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a multipath ghosting suppression method for millimeter-wave radar tunnel scenarios.

[0005] The technical solution adopted in this invention is: a multipath ghosting suppression method for millimeter-wave radar tunnel scenes, comprising:

[0006] S1: The raw echo data received by the traffic millimeter-wave radar is sampled by an ADC (Analog to Digital Converter), and the data is subjected to a two-dimensional fast Fourier transform, constant false alarm rate target detection, and angle of arrival estimation to obtain radar point cloud data.

[0007] S2: Perform spatial projection transformation on radar point cloud data, taking into account the radar's installation height, to achieve coordinate and radial velocity projection transformation from three-dimensional space to two-dimensional space;

[0008] S3: The point cloud data unified into a two-dimensional space in step S2 is agglomerated using a density-based clustering algorithm.

[0009] S4: Construct a confidence evaluation criterion for radar point clouds based on multi-dimensional information, and calculate the confidence level of each point cloud data;

[0010] S5: For the point cloud data after point cloud aggregation in step S4, perform zero Doppler point cloud separation based on radial velocity information to obtain dynamic point cloud clusters and static point cloud clusters respectively; perform linear fitting on the static point cloud clusters to obtain the reflection boundary slope set.

[0011] S6: Based on the multipath propagation geometric model, according to the reflection boundary slope set obtained in step S5, perform dynamic target matching, evaluate the confidence of target point pairs that meet the conditions, identify multipath ghosts, and eliminate the identified multipath ghosts to obtain a dynamic point cloud cluster set.

[0012] S7: Divide the dynamic point cloud cluster set into candidate regions; use velocity feature information for constraints, complete the confidence parameter evaluation of the point cloud data, thereby re-identifying multipath ghosts in the dynamic point cloud cluster set, and eliminating the identified multipath ghosts to obtain point cloud data after suppressing multipath ghosts.

[0013] The beneficial effects of this invention are as follows: This invention proposes a multipath ghosting suppression method for millimeter-wave radar tunnel scenarios. This method can effectively identify second-order multipath ghosting caused by reflective surfaces such as walls and railings in semi-enclosed tunnel scenarios. Simultaneously, based on the motion characteristics of vehicle targets within the tunnel, it achieves the identification and suppression of first-order multipath ghosting within the tunnel. Based on the constructed confidence criterion, multipath ghosting is suppressed without causing the loss of real targets. Compared to the multipath suppression methods mentioned in the background, this invention, under the detection conditions of a single radar sensor, does not rely on auxiliary information from other sensors. It perceives the tunnel environment based on radar point cloud data and simultaneously identifies and suppresses multipath ghosting generated by reflections from both inside and outside the tunnel. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of a tunnel scene.

[0015] Figure 2 A top view of electromagnetic wave reflection in the tunnel environment;

[0016] Figure 3 Geometric model of the reflection from the left wall of the tunnel;

[0017] Figure 4 Geometric model of the tunnel right wall reflection;

[0018] Figure 5 A top view of electromagnetic wave reflection in the tunnel environment;

[0019] Figure 6 This is a schematic diagram of spatial projection transformation;

[0020] (a) is the top view; (b) is the side view.

[0021] Figure 7 A geometric model of reflection within the tunnel;

[0022] Figure 8 This is the result of processing measured data;

[0023] Among them, (a) is the experimental scene; (b) is the multi-frame accumulation result after clustering the original point cloud; (c) is the multi-frame multipath ghosting recognition result; and (d) is the multi-frame multipath ghosting suppression result. Detailed Implementation

[0024] To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with the accompanying drawings, further illustrates the invention.

[0025] This invention provides a method for suppressing multipath ghosting in millimeter-wave radar tunnel scenarios. First, point cloud data is generated based on the original radar echo, and a confidence criterion is constructed based on the multidimensional information of the point cloud data. Simultaneously, density-based clustering is performed on this data. For second-order multipath ghosting generated by tunnel walls and railings, a tunnel multipath ghosting suppression method based on slope set matching is proposed. Specifically, the slope of each cluster is obtained by linear fitting of the zero-Doppler point cloud. Simultaneously, the slope of the perpendicular bisector of any target point pair in the dynamic point cloud is obtained and matched with the slope set. Target point pairs with lower confidence among those successfully associated are identified as multipath ghosting. Second, for multipath ghosting formed by multiple reflections of electromagnetic waves between vehicles and walls within the tunnel, considering the motion characteristics of vehicles within the tunnel, the target is divided into horizontal and vertical double-rectangular regions. Doppler information constraints are applied to the point cloud data within these regions. Candidate targets that meet the conditions are evaluated for confidence, thereby achieving the identification and suppression of multipath ghosting within the tunnel.

[0026] like Figure 1 The image shown is a tunnel scene diagram, and the top view obtained by geometrically modeling this scene diagram is as follows. Figure 2As shown. Radar R is located at the tunnel entrance, and a pedestrian walkway is installed inside the semi-enclosed tunnel. The left and right walls of the tunnel, as well as the pedestrian walkway, are strong reflective surfaces. In addition to the direct signal to vehicles inside the tunnel, electromagnetic wave signals also exhibit multipath ghosting caused by multiple reflections from these strong reflective surfaces. An electromagnetic wave propagation path model is established based on ray tracing theory.

[0027] The left wall of the tunnel serves as a reflective surface, such as Figure 3 As shown, the specific propagation path of the electromagnetic wave signal is as follows:

[0028] Direct line of sight: R→T→R, the electromagnetic wave signal is transmitted from radar R to the real target T and then returns to radar R;

[0029] First-order multipath: R→T→Q1→R, after the electromagnetic wave signal is transmitted from radar R to the real target T, it is reflected back to radar R through the left wall of the tunnel Q1, generating a first-order multipath ghost image S1.

[0030] Second-order multipath: R→Q1→T→Q1→R, the electromagnetic wave signal is emitted from radar R, reflected to the left wall of the tunnel Q1, and after reaching the real target T, it passes through the left wall of the tunnel Q1 and returns to radar R, generating a second-order multipath ghost S.

[0031] The right wall of the tunnel serves as a reflective surface, such as Figure 4 As shown, the specific propagation path of the electromagnetic wave signal is as follows:

[0032] Direct line of sight: R→T→R, the electromagnetic wave signal is transmitted from radar R to the real target T and then returns to radar R;

[0033] First-order multipath: R→T→Q3→R, after the electromagnetic wave signal is transmitted from radar R to the real target T, it is reflected back to radar R through the right wall of the tunnel Q1, generating a first-order multipath ghost image L1.

[0034] Second-order multipath: R→Q3→T→Q3→R, the electromagnetic wave signal is emitted from radar R, reflected to the right wall of the tunnel Q3, and after reaching the real target T, it passes through the right wall of the tunnel Q3 and returns to radar R, generating a second-order multipath ghost image L.

[0035] The railings inside the tunnel act as reflective surfaces, and the specific propagation path of electromagnetic wave signals is as follows:

[0036] First-order multipath: R→T→Q2→R, after the electromagnetic wave signal is transmitted from radar R to the real target T, it is reflected back to radar R through the tunnel railing Q2, generating a first-order multipath ghost image G1.

[0037] Second-order multipath: R→Q2→T→Q2→R, the electromagnetic wave signal is emitted from radar R, reflected to the tunnel railing Q2, and after reaching the real target T, it passes through the right wall of the tunnel Q2 and returns to radar R, generating a second-order multipath ghost image G.

[0038] Considering the significant energy attenuation of millimeter-wave radar signals after multiple reflections, the weak energy of higher-order path signals from multiple reflections is negligible. Only the real targets formed by the first and second reflections, and the first and second-order ghost images, are considered. A top-down view of electromagnetic wave reflection in a tunnel environment is shown below. Figure 5 As shown.

[0039] Based on the electromagnetic wave propagation signal modeling in the above radar detection scenario, the specific implementation steps of the present invention are given below:

[0040] Step 1: Radar point cloud data generation

[0041] The received millimeter-wave radar echo signal is sampled by an ADC to obtain a data matrix. Represents a matrix, N R N represents the number of ADC sampling points. D N represents the number of chirp signals in millimeter-wave radar. A This represents the number of echo data antenna channels, and N is used to process the sampled digital signal. r A point-to-time fast Fourier transform yields a data matrix containing target distance information. Simultaneously perform N on this data matrix d A point-slow time-dimensional fast Fourier transform yields the distance Doppler data matrix. Considering that the energy of radar echoes is relatively weak during multiple reflections of electromagnetic waves, a two-dimensional data matrix is ​​obtained by incoherently accumulating the obtained range Doppler data matrix in the antenna channel dimension.

[0042]

[0043] Cell-averaged constant false alarm rate (CFAR) detection is performed on the matrix to obtain the target's range and Doppler frequency information. Finally, the target's angle of arrival is estimated to obtain the target's azimuth information, thereby generating a matrix containing the target's range R and radial velocity v. D , azimuth angle θ, radar cross section rcs, signal-to-noise ratio snr of point cloud data, and received power power of radar point cloud data.

[0044] Step 2: Spatial Projection Transformation

[0045] In road traffic scenarios, targets typically move in a two-dimensional plane, requiring relatively low resolution in the vertical direction. Therefore, most millimeter-wave radars used for detection in this scenario are uniform linear arrays, unable to obtain elevation angle information and requiring a certain installation height. The obtained point cloud data is based on the radar coordinate system, necessitating a projection transformation from three-dimensional space to two-dimensional space. Figure 6As shown. Assuming the radar's installation height is h, a coordinate transformation is performed on the radar point cloud from three-dimensional space to two-dimensional space. The radial distance R in three-dimensional space is projected onto two-dimensional space, yielding the radial distance r and elevation angle α as follows:

[0046]

[0047] Based on the azimuth angle θ of the target point cloud detected by radar, and the radial distance r projected onto the two-dimensional space, the target's position coordinates (x, y) are obtained as follows:

[0048]

[0049] The radial velocity v in three-dimensional space D' Projected into two-dimensional space as:

[0050] v d =v D' sinα

[0051] By performing a spatial projection transformation on the radar point cloud data, its spatial location information is unified onto a two-dimensional plane, resulting in a point cloud data set P:

[0052]

[0053] Where, x i y represents the lateral distance of the i-th point cloud. i θ represents the vertical distance of the i-th point cloud. i Represents azimuth, rcs i Indicates the target scattering area, power i Indicates signal power, snr i Indicates the signal-to-noise ratio. It represents the set of rational numbers.

[0054] Step 3: Density-based spatial clustering of point cloud data

[0055] Millimeter-wave radar features short wavelength, large bandwidth, and miniaturized antennas, resulting in lower range resolution and higher spatial angular resolution. This allows for more accurate resolution of target details and differentiation of different reflection points within a small spatial area. Since targets in a road environment, such as vehicles, railings, and walls, consist of multiple reflective surfaces, a high-resolution millimeter-wave radar can clearly distinguish multiple parts of a target, thus generating multiple reflection points for a single target, forming point cloud data. The point cloud data set P after spatial projection transformation in step 2 is clustered using a density-based spatial clustering method. Clustering parameters are determined: neighborhood radius eps and minimum neighborhood number Minpts. By setting reasonable clustering parameters, the point cloud set is clustered. In this embodiment, the neighborhood radius is set to 2m based on the width of vehicles in the tunnel. Considering the high resolution of millimeter-wave radar, which allows for some expansion of the measured target, the minimum neighborhood number is set to 5 in this embodiment, based on actual data. Assuming that density-based spatial clustering of the point cloud data yields l clusters, the cluster set generated from a single frame of radar data is:

[0056] L={cluster1,cluster2,cluster3,...cluster l}

[0057] Here, "cluster" represents a cluster, which contains point cloud data that meets the clustering criteria. Furthermore, for each clustered point cloud cluster... i Cluster center selection. Considering the uneven distribution of points within a cluster for each target in point cloud clustering, directly using the mean as the cluster center is easily affected by outliers within the cluster. Therefore, the center point of each cluster is selected as the cluster center. This center has the smallest sum of distances to all other points in the cluster and represents a real, existing data point. Assume a cluster... i The cluster contains N i Six-dimensional data points, Define any two points p in the cluster i and p j The Euclidean distance is:

[0058] d(p i ,p j )=||p i -p j ||

[0059] For each point p in the cluster i ∈cluster i Calculate the sum of distances D(p) from this point to all other points within the cluster. i )for:

[0060]

[0061] We can obtain the i-th cluster. i The vector D is the sum of the distances from each point to all other points. i for:

[0062] D i ={D(p1),D(p2),...,D(p Ni )}

[0063] The point with the smallest sum of distances is selected as the center point p of the cluster. medoid :

[0064]

[0065] Where, p medoid ={x i ,y i ,θ i ,rcs i power i ,snr i The cluster center of the cluster is represented by a multidimensional vector of the center point, and the number of point clouds in the cluster, num, is recorded. i The cluster center is a feature used to describe the density of the cluster. The updated cluster center points are:

[0066]

[0067] Step 4: Construct confidence evaluation criteria

[0068] Based on the multidimensional feature information of point cloud data, considering the significant differences between the real target and the multipath ghost in radar cross section (rcs), signal-to-noise ratio (SNR) and received power (power), and based on the radar equation, it is determined that there is a certain correlation between the multidimensional features. Therefore, a confidence evaluation criterion is constructed to further determine the candidate multipath ghost.

[0069] Define each point cloud cluster i M, the multidimensional feature vector of the midpoint cloud i =[rcs i ,snr i power i ] T Follows a joint multidimensional Gaussian distribution:

[0070] M i =[rcs i ,snr i power i ]~N(μ i ,Σ i )

[0071] Where, μ i and Σ i These are the feature mean vector and covariance matrix of the cluster, respectively. For each point cloud cluster, the corresponding feature mean vector can be calculated. for:

[0072]

[0073] Simultaneously targeting point cloud clusters i The correlation between different features of the data points is calculated to obtain the covariance matrix Σ of the Gaussian distribution. i for:

[0074]

[0075] The covariance is defined as follows:

[0076]

[0077] For each cluster i The joint probability density function for multidimensional information is defined as follows:

[0078]

[0079] Where d represents the dimension of the multidimensional vector, which is taken as 3 in this embodiment, Σ i and μ i Let f(i) represent the mean and covariance of the multidimensional features of the point cloud data for the cluster, respectively. The resulting probability density value f(i) is defined as the confidence level of the cluster. For each point cloud cluster... i (i = 1, 2, 3... N) i Calculate the confidence score to describe the degree to which the cluster represents the real target, and use it as the cluster center p. medoid The new one-dimensional feature, and the updated cluster center points are:

[0080]

[0081] Step 5: Tunnel Environment Perception

[0082] The semi-enclosed tunnel environment contains static architectural structures such as continuous walls and railings, but their surfaces are not completely smooth and have a certain degree of roughness. This leads to the scattering of electromagnetic wave reflection signals, forming multiple reflection points. Furthermore, considering factors such as radar beamwidth, the perception of the static environment often presents discrete point cloud data. The presence of numerous reflective surfaces in the static environment causes multipath ghosting during the detection of real targets. Therefore, zero-Doppler point cloud separation is performed on the set L obtained after clustering in step 3. Considering the velocity measurement error inherent in radar detection, a Doppler velocity error Δv is defined. D Perform point cloud separation to obtain dynamic point cloud clusters L. dy And static point cloud clusters L st Theoretically, millimeter-wave radar measures the Doppler velocity of a static object as 0 m / s. However, considering the inherent velocity measurement error in millimeter-wave radar, this invention selects a Doppler velocity error Δv. D It is 0.05 m / s.

[0083] There are k static point cloud clusters in a single frame of data. For each static point cloud cluster Linear fitting is performed using the least squares method. Assume the fitted line for this cluster is represented as:

[0084]

[0085] in, Let be the slope of the fitted line for this cluster. The intercept of the fitted line. Define the loss function for the slope and intercept of the clusters. for:

[0086]

[0087] The loss function with respect to variable a k b k Taking the partial derivative, the optimal solution is:

[0088]

[0089] Among them, (x i ,y i ) represents this cluster The static point coordinates are used to perform linear fitting on each cluster in the static point cloud set, forming a slope set.

[0090] Step 6: Matching and associating moving targets

[0091] The dynamic point cloud cluster L obtained in step 5 dyThe cluster centers are paired and associated. The dynamic target association matrix is ​​defined as follows:

[0092]

[0093] in, Let be any two cluster centers within the dynamic point cloud clusters. According to the typical multipath propagation geometry model, the real target and the second-order multipath ghost are mirror-symmetric about the reflecting surface. Calculate the dynamic target association pairs. The slope k of the perpendicular bisector ij for:

[0094]

[0095] Determine the slope k of the perpendicular bisector of the associated pair ij Does k satisfy? ij ∈A s If a target belongs to this set, it satisfies the "real target - multipath ghost" association pair. The association matrix formed by traversing all cluster centers of the dynamic point cloud clusters is used for judgment, and target association pairs that meet the conditions are selected. The confidence level decision yields the multipath ghosting identification result Re as follows:

[0096]

[0097] in, Let L' represent the confidence scores of the associated target pairs, and the dynamic point cloud cluster set obtained by eliminating the identified multipath ghosting is denoted as L'. dy .

[0098] Step 7: Candidate Region Division and Feature Constraints

[0099] In a semi-enclosed tunnel environment, there are multiple static reflective surfaces such as walls and railings. When detecting a real vehicle target, electromagnetic waves are easily reflected multiple times between the static building structure and the real target, forming multipath ghosting. However, due to the lateral distance R between the reflective surfaces and the vehicle target in the tunnel environment... th When the distance is small, according to the multipath propagation geometry model, when the distance is close, the real target and the first-order multipath ghost formed by the reflection are geometrically coincident in distance measurement. This is further supported by the characteristic that vehicles in tunnel scenes move in a straight line and the lane width R. road Information on the dynamic point cloud clusters obtained in step 6 Perform candidate region division as follows Figure 7 As shown.

[0100] Assuming the center of a dynamic point cloud cluster Define rectangular regions with width w1 and length h1 for the geometric center. for:

[0101]

[0102] Similarly, define a rectangular region with width w2 and length h2. for:

[0103]

[0104] In a real-world scenario, based on experimental data and considering the size and motion characteristics of vehicles within the tunnel, a rectangular area was defined. The width and length are 4m and 3m respectively. Considering the geometric characteristics of multipath ghosting caused by multiple reflections of electromagnetic waves within the tunnel, a rectangular area is set. Its width and length are 2m and 6m respectively.

[0105] The real target and the different velocities of the multipath ghost images formed by multiple reflections are correlated, and this can be verified through theoretical derivation and simulation:

[0106] |v real -v ghost |≤△v th

[0107] Among them, v real v ghost Let Δv represent the velocity of the real target and the multipath ghost, respectively. th As a threshold for velocity difference, this embodiment conducts a simulation experiment based on a typical multipath propagation scenario. The simulation conditions are: reflection boundary x = 2m, real target position x = 3m, y = 6m, real target velocity is 3m / s, and the simulation is performed for 30 frames. The maximum Doppler velocity difference between the real target and the first-order ghost is 0.3m / s, and the maximum Doppler velocity difference between the real target and the second-order ghost is 0.7m / s. In the tunnel scenario, considering the certain velocity measurement error of millimeter-wave radar and the characteristics of the actual scene, the velocity threshold is set to 1m / s in this embodiment.

[0108] For satisfying the dynamic point cloud cluster center candidate regions Point cloud data is used to constrain the velocity feature dimension, and the confidence level of points that meet the conditions is evaluated. By comparing the confidence levels, points with low confidence levels that also meet the conditions of being located within the candidate region and having velocity feature dimension constraints are considered multipath ghosts. Multipath ghosting is defined as a condition that satisfies the following criteria:

[0109]

[0110] The proposed algorithm was validated using real-world test data. The experimental scenario is as follows: Figure 8 As shown in (a). Figure 8(b) is the result of clustering the original radar point cloud data; Figure 8 (c) shows the recognition results of multipath ghosting in the tunnel scene; Figure 8 (d) is from Figure 8 Remove from (b) Figure 8 (c) The suppression result obtained after identifying multipath ghosting.

[0111] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A method for suppressing multipath ghosting in millimeter-wave radar tunnel scenes, characterized in that, include: S1: The raw echo data received by the traffic millimeter-wave radar is sampled by ADC, and the data is subjected to two-dimensional fast Fourier transform, constant false alarm target detection and angle of arrival estimation to obtain radar point cloud data. S2: Perform spatial projection transformation on radar point cloud data, taking into account the radar's installation height, to achieve coordinate and radial velocity projection transformation from three-dimensional space to two-dimensional space; S3: The point cloud data in the two-dimensional space obtained in step S2 is agglomerated using a density-based clustering algorithm. S4: Construct a confidence evaluation criterion for radar point clouds based on multidimensional information, and calculate the confidence score of each cluster. S5: For the point cloud data after point cloud aggregation in step S3, perform zero Doppler point cloud separation based on radial velocity information to obtain dynamic point cloud clusters and static point cloud clusters respectively; perform linear fitting on the static point cloud clusters to obtain the reflection boundary slope set; S6: Based on the multipath propagation geometric model, according to the reflection boundary slope set obtained in step S5, perform dynamic target matching, evaluate the confidence of target point pairs that meet the conditions, identify multipath ghosts, and eliminate the identified multipath ghosts to obtain a dynamic point cloud cluster set. S7: Divide the dynamic point cloud cluster set into candidate regions; use velocity feature information for constraints, complete the confidence parameter evaluation of the point cloud data, thereby re-identifying multipath ghosts in the dynamic point cloud cluster set, and eliminating the identified multipath ghosts to obtain point cloud data after suppressing multipath ghosts.

2. The multipath ghosting suppression method for millimeter-wave radar tunnel scenes according to claim 1, characterized in that, Step S1, the two-dimensional fast Fourier transform, includes: performing a fast time-dimensional fast Fourier transform on the sampled data matrix to obtain a data matrix N containing target distance information. r For N r Perform a slow-time dimension Fast Fourier Transform to obtain the distance Doppler data matrix N. d .

3. The multipath ghosting suppression method for millimeter-wave radar tunnel scenes according to claim 2, characterized in that, This also includes performing incoherent accumulation of the range Doppler data matrix in the antenna channel dimension to obtain a two-dimensional data matrix N. c : Where, N A Indicates the number of antenna channels.

4. The multipath ghosting suppression method for millimeter-wave radar tunnel scenes according to claim 3, characterized in that, Step S2 specifically includes: Assuming the radar's installation height is h, a coordinate transformation is performed on the radar point cloud from three-dimensional space to two-dimensional space. The radial distance R in three-dimensional space is projected onto two-dimensional space, yielding the radial distance r and elevation angle α as follows: Based on the azimuth angle θ of the target point cloud detected by radar, and the radial distance r projected onto the two-dimensional space, the target's position coordinates (x, y) are obtained as follows: The radial velocity v in three-dimensional space D' Projected into two-dimensional space: v d =v D' sinα By performing a spatial projection transformation on the radar point cloud data, its spatial location information is unified onto a two-dimensional plane, resulting in a point cloud data set P:

5. The multipath ghosting suppression method for millimeter-wave radar tunnel scenes according to claim 4, characterized in that, Step S3 selects a center point in the cluster as the cluster center. This center point has the smallest sum of distances to all other points in the cluster and is a real data point.

6. The multipath ghosting suppression method for millimeter-wave radar tunnel scenarios according to claim 5, characterized in that, The implementation process of step S4 is as follows: Define each point cloud cluster i M, the multidimensional feature vector of the midpoint cloud i =[rcs i ,snr i power i ] T Follows a joint multidimensional Gaussian distribution: M i =[rcs i ,snr i ,power i ]~N(μ i ,S i ) Where, μ i , Σ i These are the feature mean vector and covariance matrix of the cluster, respectively; For each point cloud cluster, the corresponding feature mean vector can be calculated. for: Simultaneously targeting point cloud clusters i The correlation between different features of the data points is calculated to obtain the covariance matrix Σ of the Gaussian distribution. i for: The covariance is defined as follows: For each cluster i The joint probability density function for multidimensional information is defined as follows: Where d represents the dimension of the multidimensional vector, Σ i and μ i These represent the mean and covariance of the multidimensional features of the point cloud data of this cluster, respectively. The obtained probability density value f(i) is defined as the confidence level of the cluster.

7. The multipath ghosting suppression method for millimeter-wave radar tunnel scenes according to claim 6, characterized in that, In step S5, linear fitting is performed on the static point cloud clusters to obtain the set of reflection boundary slopes. The specific process is as follows: The fitted straight line is represented as: in, The slope of the straight line fitted to the cluster of static point clouds. The intercept of the fitted line; Define the loss function for the slope and intercept of the cluster. for: The loss function with respect to variable a k b k Taking the partial derivative, the optimal solution is: Among them, (x i ,y i () represents a static point cloud cluster. The coordinates of a static point in the image; By performing linear fitting on each cluster in the static point cloud set, a slope set is constructed.

8. The multipath ghosting suppression method for millimeter-wave radar tunnel scenes according to claim 7, characterized in that, The implementation process of step S6 is as follows: Perform pairwise matching and association on the cluster centers of the dynamic point cloud clusters obtained in step S5; define the dynamic target association matrix as: in, Let i be any two cluster centers in the dynamic point cloud cluster, i ≠ j; Based on a typical multipath propagation geometry model, the real target and the second-order multipath ghost are mirror-symmetric about the reflecting surface; calculate the dynamic target association pair. The slope k of the perpendicular bisector ij for: The association matrix formed by the centroids of all clusters in the dynamic point cloud is traversed to make a judgment, and conditions k are satisfied. ij ∈A s Target association pairs After performing a confidence score determination, the multipath ghosting identification result Re is obtained as follows: in, These represent the confidence levels of the associated target pairs; The dynamic point cloud cluster set obtained by eliminating the identified multipath ghosting is L' dy .

9. The multipath ghosting suppression method for millimeter-wave radar tunnel scenes according to claim 8, characterized in that, The specific implementation process of step S7 is as follows: Considering the target's size and motion characteristics, the center of the dynamic point cloud cluster is used. Define the first rectangular region as the geometric center. Considering the geometric characteristics of the multipath ghost distribution generated by multiple reflections of electromagnetic waves within the tunnel, with the center of the dynamic point cloud cluster as an example. Define the second rectangular region as the geometric center. The velocities of the real target and the different multipath ghost images formed by multiple reflections are correlated, satisfying the following: |v real -v ghost |≤△v th Among them, v real v ghost Let Δv represent the velocity of the real target and the multipath ghost, respectively. th The threshold for the speed difference; For satisfying the dynamic point cloud cluster center candidate regions The point cloud data is constrained in terms of velocity feature dimension, and the confidence level of points that meet the conditions is evaluated. A multipath is defined as one that satisfies the following decision relationship:

Images