Millimeter wave radar data processing method, electronic device, and program product

Abstract

The application relates to the technical field of radar data processing, in particular to a millimeter wave radar data processing method, an electronic device and a program product. The method comprises the following steps: acquiring point cloud data detected by a millimeter wave radar; clustering the point cloud data to obtain at least one scatterer point cluster; determining scatterer parameters of each scatterer point cluster, wherein the scatterer parameters comprise a backscattering coefficient. The method solves the defects of traditional millimeter wave point intensity, i.e., no unified physical meaning and incapability of quantitatively inverting target scattering characteristics, outputs a scatterer parameter set carrying position, speed, backscattering coefficient and reliability, and can be used for intelligent driving road surface state identification, vehicle and pedestrian material differentiation and target RCS estimation, thereby effectively improving the automatic driving environment perception quantization capability. The application synchronously provides a vehicle-mounted electronic device and a computer program product for executing the method.

Description

Technical Field

[0001] This application relates to the field of radar data processing technology, specifically to a millimeter-wave radar data processing method, electronic equipment, and program product. Background Technology

[0002] Millimeter-wave radar, with its advantages of all-weather operation, resistance to rain, snow, fog, and dust, penetration of non-metallic obstructions, ability to detect subtle movements / vital signs, small size, and high accuracy in speed and distance measurement, is widely used in automotive, intelligent transportation, industrial, security, home, medical and health care, robotics, aerospace, and defense fields. Existing millimeter-wave radar point clouds are mainly used for target detection, display, clustering, and tracking. The points in their output typically include position, velocity, angle, and point intensity. However, these point intensities mostly originate from range-Doppler heatmap amplitude, beamforming spectrum peak values, constant false alarm rate (CFAR) detection unit outputs, or angular spectrum intensity, and generally lack a unified, stable, and comparable physical meaning. Therefore, the point intensity in millimeter-wave radar point clouds is mainly used for display and filtering, and it is difficult to use it directly for quantitative analysis. For example, based on the point intensity, the radar cross section (RCS) and backscattering coefficient, which characterize the reflectivity of the millimeter-wave radar detected object (hereinafter referred to as the scatterer), cannot accurately characterize the physical scattering ability of the real scatterer, and cannot be used for quantitative inversion. For example, it is impossible to accurately distinguish the material of the scatterer, and it is impossible to quantitatively analyze the scattering characteristics of environmental changes such as the slippery state of the road surface.

[0003] There are several reasons for the above problems: For example, current millimeter-wave radar point cloud data uses single points as the calculation unit, which cannot correspond to continuous planar targets such as roads, vegetation, and walls, and lacks a calculation carrier suitable for planar targets. Another example is that, due to radar beam main lobe expansion, side lobe leakage, and resolution limitations, the same planar scattering body is broken down into a large number of discrete points. Directly accumulating the energy will result in artificially high scattering intensity, violating the physical definition of the backscattering coefficient as "scattering power per unit area," meaning that a physically meaningful backscattering coefficient cannot be obtained. Summary of the Invention

[0004] This application provides a millimeter-wave radar data processing method, electronic device, and program product.

[0005] In a first aspect, this application provides a millimeter-wave radar data processing method, which includes: acquiring point cloud data detected by millimeter-wave radar; clustering the point cloud data to obtain at least one scatterer point cluster; and determining the scatterer parameters of each scatterer point cluster, wherein the scatterer parameters include the backscattering coefficient of the scatterer point cluster.

[0006] Therefore, this application clusters discrete radar point clouds into physically meaningful scattering point clusters, and combines source information for sidelobe separation and system compensation to accurately retrieve the target's true backscattering coefficient and reliability. This solves the problems of traditional point clouds lacking physical dimensions and being susceptible to sidelobe interference, significantly improving the radar's ability and accuracy in perceiving target material and environment, and enhancing the user experience.

[0007] In one possible implementation of the first aspect above, the above-mentioned clustering of point cloud data to obtain at least one scatterer point cluster includes: determining source information associated with the point cloud data, wherein the source information includes the original radar data and intermediate data that generated the point cloud data; and obtaining at least one scatterer point cluster based on the source information associated with the point cloud data.

[0008] In one possible implementation of the first aspect described above, the intermediate data includes distance cells, Doppler cells, angle information, data channel identifiers, heatmap cells, and local complex observation indexes.

[0009] In one possible implementation of the first aspect above, obtaining at least one scatterer point cluster based on the source information associated with the point cloud data includes: determining the aggregation similarity between any two points in the point cloud data based on the source information; and obtaining at least one scatterer point cluster through a clustering algorithm based on the aggregation similarity between any two points.

[0010] In one possible implementation of the first aspect above, determining the aggregation similarity between any two points in the point cloud data based on the source information includes: calculating the aggregation similarity between any two points based on the spatial Euclidean distance, radial velocity difference, and observation domain distance between any two points in the point cloud data.

[0011] In one possible implementation of the first aspect above, determining the scatterer parameters of each scatterer cluster includes: separating the sidelobe repetition points of the scatterer cluster to obtain the first scattering intensity of the scatterer cluster; performing propagation and system response compensation on the first scattering intensity to obtain the second scattering intensity; and mapping the second scattering intensity to a backscattering coefficient.

[0012] In one possible implementation of the first aspect above, the above-mentioned separation of sidelobe repetition points from the scatterer point cluster to obtain the first scattering intensity of the scatterer point cluster includes: identifying the sidelobe repetition points in each scatterer point cluster, dividing each scatterer point cluster into a set of repetition points and a set of independent candidate points; setting different weighting values ​​for the main response point of the scatterer point cluster, the repetition points in the set of repetition points, and the independent candidate points in the set of independent candidate points, respectively, to obtain the first scattering intensity.

[0013] In one possible implementation of the first aspect described above, different weighting values ​​are set for the main response point of the scatterer point cluster, the repeating points in the repeating point set, and the independent candidate points in the independent candidate point set to obtain the first scattering intensity, including: calculating the first scattering intensity based on the point cloud intensity of each point in the main response point of the scatterer point cluster, the repeating point set, and the independent candidate point set; or calculating the first scattering intensity based on the original multi-channel complex observation data in the source information of each point in the main response point of the scatterer point cluster, the repeating point set, and the independent candidate point set.

[0014] In one possible implementation of the first aspect above, the above-mentioned propagation and system response compensation of the first scattering intensity to obtain the second scattering intensity includes: calculating the second scattering intensity based on the following formula. :

[0015]

[0016] in, The straight-line distance from millimeter-wave radar to the scatterer cluster, For the first scattering intensity, For the total system gain, For the transmission pattern gain, For receiving pattern gain, To handle chain gain and loss terms, Array channel compensation.

[0017] In one possible implementation of the first aspect above, determining the scatterer parameters of each scatterer cluster further includes mapping the second scattering intensity to a radar cross section.

[0018] In one possible implementation of the first aspect above, the scatterer parameters further include the inversion confidence of the backscattering coefficients, and the scatterer parameters of each scatterer cluster are determined. The inversion confidence is calculated based on at least one of the following: the signal-to-noise ratio confidence of the echo signal of the scatterer cluster, the unique expression confidence, the multi-frame stability confidence, the calibration confidence, and the sidelobe residual contamination confidence.

[0019] In one possible implementation of the first aspect described above, the scatterer parameters also include the position parameters and velocity parameters of the scatterer point cluster.

[0020] Secondly, this application provides an electronic device including one or more processors and one or more memories; the memories store computer programs / instructions, which, when executed by the processor, cause the electronic device to perform the millimeter-wave radar data processing method described in the first aspect and any possible implementation thereof. For example, the electronic device may be an in-vehicle device such as an in-vehicle domain controller (vehicle radar controller or in-vehicle chip), or it may be an embedded processor chip (such as a digital signal processor or microcontroller unit) integrated inside a millimeter-wave radar sensor.

[0021] Thirdly, this application provides a computer-readable storage medium including a computer program / instructions that, when executed by a processor, implement the millimeter-wave radar data processing method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the millimeter-wave radar data processing method described in the first aspect and any possible implementation thereof. Attached Figure Description

[0023] Figure 1 The diagram shown is a schematic representation of an application scenario for a millimeter-wave radar provided in an embodiment of this application.

[0024] Figure 2 The diagram shown is a flowchart illustrating a millimeter-wave radar data processing method provided in an embodiment of this application.

[0025] Figure 3 The diagram shown is a schematic diagram illustrating the process for determining the scatterer parameters of a scatterer cluster according to an embodiment of this application.

[0026] Figure 4 The diagram shown is a structural composition schematic of an electronic device provided in an embodiment of this application.

[0027] Figure 5 The diagram shown is a schematic representation of the structural composition of a vehicle according to an embodiment of this application. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0029] To facilitate understanding of the solutions in the embodiments of this application by those skilled in the art, some concepts and terms involved in the embodiments of this application will be explained below.

[0030] (1) Radar Cross Section (RCS)

[0031] Represents the ability of a target to scatter electromagnetic waves toward the radar under far-field conditions. It is equivalent to the virtual cross-section of an ideal isotropic scatterer and is usually used to characterize the overall echo strength of a single discrete target (such as a vehicle, pedestrian, aircraft, corner reflector, or sign). In other words, it describes the total scattering capability (overall characteristics) of an entire target.

[0032] (2) Backscattering coefficient

[0033] Representing the average RCS per unit area of ​​a surface / area medium, it is specifically used to describe the scattering ability of continuously distributed area targets (ground, vegetation, walls, sea surface, road surface, etc., extended clutter regions), and can distinguish the material of the scatterer (e.g., slippery road surface, scatterer being metallic or non-metallic, etc.). In other words, it describes the average scattering ability (local and distributed characteristics) of a medium per unit area.

[0034] As mentioned earlier, millimeter-wave radar has many application scenarios. For ease of understanding, the application scenario section of this application uses the field of intelligent driving as an example to describe the technical solution. However, it is understood that the technical solution of this application is also applicable to other application fields of millimeter-wave radar. For example, Figure 1 This illustrates one application scenario for millimeter-wave radar. Specifically, such as... Figure 1 As shown, vehicle 10 is equipped with a millimeter-wave radar 20 (illustrated only). This millimeter-wave radar 20 emits electromagnetic waves into the vehicle's driving environment and receives echo signals reflected by targets (such as the vehicle 30 ahead, the roadside guardrail 40, the sidewalk 50, the lane 60, etc.). The processor on vehicle 20 can process the received echo signals to generate millimeter-wave radar point cloud data containing target position, speed, and intensity information, and perform tasks such as target detection, environmental perception, and driving decision-making based on this point cloud data.

[0035] However, the point intensities in current millimeter-wave radar point cloud data lack a unified, stable, and comparable physical meaning. It is impossible to calculate the absolute values ​​of the RCS and backscattering coefficient that can accurately characterize the physical scattering ability of a real scatterer based on these point intensities. Consequently, quantitative inversion is impossible. For example, it is impossible to accurately distinguish the material of the scatterer or to quantitatively analyze the scattering characteristics of environmental changes such as wet road conditions.

[0036] To address the aforementioned issues, this application discloses a method for processing millimeter-wave radar data. Specifically, point cloud data detected by millimeter-wave radar is clustered to obtain at least one scatterer cluster, and the scatterer parameters of each scatterer cluster are determined. These scatterer parameters include the backscattering coefficient of the scatterer cluster. For example, points in the point cloud data that are highly similar in spatial location, velocity, and signal characteristics are grouped together to form scatterer clusters. Each scatterer cluster essentially corresponds to a real physical scatterer in the real world, such as a car, a road sign, or a local area of ​​a road surface. The final determined backscattering coefficient of the scatterer cluster is a parameter determined based on the scatterer cluster as a whole; it is a physical quantity that measures the inherent scattering capability per unit area of ​​an actual physical scatterer or local area, and can be used for quantitative inversion.

[0037] This method clusters discrete millimeter-wave radar point cloud data into scatterer point clusters, realizing the transformation from single-point representation to scatterer unit representation. It can address continuous planar targets such as roads, walls, and vegetation, providing a suitable computational platform for calculating backscattering coefficients. Furthermore, it outputs scatterer parameters including backscattering coefficients, overcoming the limitations of traditional millimeter-wave radar point clouds that only provide position, velocity, and relative intensity. It can directly obtain the inherent scattering capability per unit area of ​​the target, achieving a quantitative characterization of the target's scattering characteristics.

[0038] It is understood that in some embodiments of this application, point cloud data is associated with source information containing the original radar data and intermediate data that generated the point cloud data. Based on the point cloud data and source information, a clustering algorithm is used to divide the points in the point cloud data into at least one cluster of scattering points. Then, based on the point cloud data and source information, sidelobe repetition point separation and propagation and system response compensation are performed on the scattering point clusters, finally obtaining the backscattering coefficient and RCS. This solves the problem that, due to radar beam main lobe expansion, sidelobe leakage, and resolution limitations, the same planar scattering object is decomposed into a large number of discrete points, and directly accumulating energy will cause the scattering intensity to be artificially high, violating the physical definition of the backscattering coefficient as "scattering capability per unit area". At the same time, it solves the problem that the point cloud intensity is superimposed with non-target factors such as propagation distance loss, antenna pattern, system gain, and array channel error, and the intensity at different distances and angles cannot be unified, failing to reflect the inherent scattering properties of the target.

[0039] The technical solution of this application will be further described in detail below with reference to specific embodiments.

[0040] Figure 2A flowchart illustrating a method for processing millimeter-wave radar data is shown. This method can be executed by a processor or controller of a millimeter-wave radar system mounted on a vehicle, a processor or controller of a vehicle-mounted infotainment system, or a server associated with the millimeter-wave radar system or the vehicle-mounted infotainment system. Specifically, as... Figure 2 As shown, the millimeter-wave radar data processing method includes:

[0041] S201: Acquire point cloud data detected by millimeter-wave radar.

[0042] During vehicle operation, the millimeter-wave radar continuously transmits frequency-modulated continuous-wave (FMCW) signals and receives reflected echoes from all targets ahead, including vehicle 30, guardrail 40, and lane 50. The onboard processor performs a two-dimensional fast fourier transform (FFT) operation on the intermediate-frequency signal to generate a range-Doppler heatmap. The heatmap is a two-dimensional complex matrix, with the horizontal axis corresponding to range cells and the vertical axis corresponding to Doppler cells. Each cell in the matrix contains in-phase and quadrature (I / Q) complex amplitude and phase components. Constant false alarm rate (CFAR) detection is used to screen candidate cells whose energy exceeds a threshold. Then, multiple-input multiple-output (MIMO) array digital beamforming, multiple signal classification (MUSIC) algorithm, and signal parameter estimation algorithm based on rotational invariance subspace technology are used to estimate signal parameters. The ESPRIT (Technology) super-resolution angle measurement algorithm calculates the azimuth and elevation angles of each detection unit, and outputs a point cloud dataset (i.e., a collection of point cloud data) after coordinate transformation.

[0043] For example, point cloud datasets can be uniformly expressed as the following formula (1).

[0044] (1).

[0045] Among them, in formula (1) For point In the three-dimensional coordinates of the radar coordinate system, For point The target's radial velocity (positive for distance from radar). For point The original echo intensity, where N is the number of points in the point cloud dataset.

[0046] S202: Cluster the point cloud data to obtain at least one scatterer point cluster.

[0047] In some embodiments of this application, the above-described S202 can determine the source information associated with the point cloud data; based on the source information associated with the point cloud data, at least one scatterer point cluster is obtained.

[0048] It is understandable that this application can synchronously and completely record the source information corresponding to each point when collecting point cloud data, providing underlying support for subsequent clustering and intensity correction.

[0049] The source information is divided into two categories: the raw radar data used to generate point clouds and the intermediate data from the entire radar signal processing chain. The intermediate data includes range cell indexes, Doppler cell indexes, angle information, data channel identifiers, heat map cell or local complex observation indexes, etc.

[0050] For example, point The complete source information is defined as shown in Formula (2) below. Formulas (1) and (2) are used to characterize the correlation model between millimeter-wave point clouds and original observations.

[0051] (2).

[0052] In formula (2) Indicates the distance cell index, marking the horizontal position of the point in the distance-Doppler heatmap; Doppler cell index, marking the vertical position of a point in the distance-Doppler heatmap; It can represent angle information, such as the target azimuth and elevation angle; For associated indexes, data channel identifiers, heatmap cells, or local complex observation indexes from which points originate can be associated.

[0053] In addition, a single point Complete source information can also include five types of raw data: range-Doppler heatmap peaks, range-azimuth heatmap local peaks, beamforming complex amplitudes, multi-channel complex vector observations, and raw intermediate frequency local data. For example, the correlation index in formula (2) It also allows for tracing back to these five types of raw data at the underlying level. Through this correlation, this application enables the tracing back from point-level results to signal-level observation results, providing a basis for subsequent scatterer-level reconstruction and physical intensity correction.

[0054] It is understandable that by tracing the source information, it is possible to trace the entire signal chain from the point cloud output end to ADC sampling, FFT, and angle estimation, solving the problem of cross-target mismerging caused by existing clustering methods that rely solely on spatial coordinates and velocity. Thus, it enables backtracking from a single point to the underlying signal layer, providing underlying support for subsequent accurate clustering and complex coherent energy calculation.

[0055] In some embodiments of this application, the aggregation similarity between any two points in point cloud data is determined based on source information; and at least one scatterer point cluster is obtained through a clustering algorithm based on the aggregation similarity between any two points.

[0056] As an example, consider any two points in a point cloud dataset. , It can simultaneously integrate three dimensions: three-dimensional Euclidean distance, radial velocity difference, and source observation domain distance, and calculate aggregate similarity through Gaussian kernel function.

[0057] For example, two points , The aggregation similarity can be calculated using the scatterer-level aggregation model characterized by the following formula (3).

[0058] (3).

[0059] In formula (3), the formula is as follows: The three-dimensional coordinates of the point; The similarity is the Euclidean distance between two points in three-dimensional space. For the same entity, the spatial distance is close, and the greater the distance, the lower the similarity. The radial velocity difference between two points indicates that the velocities of the same rigid body target are basically the same, while large velocity differences indicate that they are different scattering bodies. To trace the distance of the information observation domain, it is necessary to distinguish independent targets that are spatially and velocally close but belong to different detection peaks. , , The scale tolerance parameters are specifically the matching scales for space, velocity, and observation domain, and their values ​​are proportional to the radar range, velocity, and angular resolution. The cluster similarity range is (0,1], with higher values ​​indicating a higher probability that two points originate from the same real scatterer; that is, the closer the value is to 1, the more likely the two points originate from the same real scatterer. This design can avoid the erroneous merging of heterogeneous points with similar spatial velocities but originating from different signal detection peaks.

[0060] In some embodiments of this application, a point association graph is constructed based on the aggregation similarity of points in the point cloud dataset, and a segmentation algorithm is used to divide the candidate scatterer point cluster set; furthermore, basic features are calculated for each candidate scatterer point cluster in the candidate scatterer point cluster set.

[0061] As an example, an association graph is constructed using all points in the point cloud dataset as graph nodes and the aggregation similarity between two points as edge weights. Any segmentation algorithm, such as graph segmentation, connected component analysis, spectral clustering, or density clustering, can be used to complete the graph partitioning and output a set of candidate scatterer point clusters. This set of candidate scatterer point clusters contains at least one of the above-mentioned scatterer point clusters.

[0062] For example, the set of candidate scatterer point clusters It can be expressed using the following formula (4).

[0063] (4).

[0064] In formula (4), each cluster This refers to a cluster of scattering points, where m takes values ​​of 1, 2, ..., M. Specifically, each cluster... This can correspond to a single physical scattering object in the real world (a whole vehicle, a section of road, a single guardrail), such as each cluster. This represents a local region of a candidate scatterer.

[0065] As an example, this application considers the candidate scatterer point cluster set. Each scatterer point cluster Unified computing fundamental characteristics.

[0066] Among them, scatterer point clusters The fundamental features include: the principal response point corresponding to the maximum intensity within the cluster. (Scatterer energy anchor point, subsequent sidelobe discrimination criterion), cluster center three-dimensional coordinates (Mean coordinates within the cluster, smoothed noise), cluster center velocity (Cluster velocity median, excluding outlier sidelobe velocities), cluster average intensity, cluster peak intensity, and three-dimensional covariance and spatial morphology characteristics within the cluster. Covariance characteristics are used to distinguish between a single compact scatterer (vehicle, wall) and a multi-target, cohesive, elongated cluster, providing a basis for subsequent sidelobe stripping and energy weighting.

[0067] Thus, this application relies on full-link traceability information to complete the division of scatterer point clusters, realizing the tracing of point cloud from the final output to the original radar observation signal. The boundary of the scatterer point cluster closely matches the distribution of real scattering entities such as road surface, vehicles, and guardrails, greatly reducing the probability of cross-target erroneous clustering.

[0068] S203: Determine the scatterer parameters of each scatterer point cluster.

[0069] As an example, the scatterer parameters (also known as the scatterer parameter set) of a scatterer cluster may include backscattering coefficients. Alternatively, the scatterer parameters of a scatterer cluster may include not only backscattering coefficients, but also at least one of the following: radar cross section (RCS), cluster position parameters, velocity parameters, and backscattering coefficient inversion confidence level.

[0070] In some embodiments, this application can determine the scatterer parameters of all or part of the scatterer point clusters in at least one clustered scatterer point cluster. For example, this application can determine a set of scatterer point clusters. The scattering parameters of each scatterer point cluster in the set of scatterer point clusters. It can be the above candidate scatterer point cluster set It itself or a subset thereof.

[0071] For example, the set of scatterer point clusters Set of scatterer parameters for each scatterer point cluster This is expressed by the following formula (5).

[0072] (5).

[0073] Among them, in formula (5) express The position parameters, i.e., the three-dimensional center coordinates; Represents a cluster of scattering points The velocity parameter, i.e., the average radial velocity of the scatterer point cluster; Represents a cluster of scattering points The backscattering coefficient (linear / dB format), such as the radar cross section (RCS); Represents a cluster of scattering points The inversion confidence level. The inversion confidence level is the scattering point cluster. The reliability score of the backscattering coefficient (typically a value between 0 and 1) is a comprehensive evaluation index of the measurement quality of the target's physical properties in the quantitative inversion process. This elevates radar perception from simple geometric location detection to a level of reliable physical property perception, significantly enhancing the system's robustness and security in complex scenarios.

[0074] In addition, in formula (5) The values ​​are 1, 2, ... ,and The value is less than or equal to This application uses a standardized set of scatterer parameters for the output scatterer point clusters of point cloud data, replacing the traditional discrete point cloud data input to autonomous driving perception algorithms, which helps improve the accuracy of algorithm execution.

[0075] Thus, the millimeter-wave radar data processing method provided in this application clusters discrete millimeter-wave radar point cloud data into scatterer point clusters, realizing the transformation from single-point representation to scatterer unit representation. This method can correspond to continuous planar targets such as roads, walls, and vegetation, providing a suitable computational platform for calculating backscattering coefficients. Furthermore, it outputs scatterer parameters including backscattering coefficients, overcoming the limitations of traditional millimeter-wave radar point cloud data that only provides position, velocity, and relative intensity. It can directly obtain the inherent scattering capability per unit area of ​​the target, achieving a quantitative characterization of the target's scattering characteristics.

[0076] Reference Figure 3 ,right Figure 2 The process for determining the scatterer parameters of each scatterer point cluster defined in the document is explained. For example... Figure 3 As shown, S203 above includes the following steps:

[0077] S301: Separate the sidelobe repeating points of the scatterer point cluster to obtain the first scattering intensity.

[0078] The first scattering intensity is the original energy at the scatterer level.

[0079] In some embodiments of this application, S301 can identify the sidelobe repeating points in each scatterer point cluster, divide each scatterer point cluster into a repeating point set and an independent candidate point set; and set different weighting values ​​for the main response point of each scatterer point cluster, the repeating points in the repeating point set, and the independent candidate points in the independent candidate point set, to obtain the corresponding first scattering intensity.

[0080] As an example, for scatterer point clusters The process for identifying and classifying sidelobe repeat points includes: selecting scatterer point clusters. intra-cluster master response point (Strongest point), for other points in the cluster Determine whether it is a sidelobe repetition point of the main response. Specifically, consider the scatterer point cluster. Using the main response point within the cluster as the judgment criterion, and substituting it into the radar system's point spread function and array beam response, other points are calculated. Theoretical sidelobe attenuation Simultaneously calculate the sidelobe matching error at the corresponding point. To determine the point Is it a sidelobe repetition point of the main response?

[0081] For example, using scatterer point clusters intra-cluster master response point As a benchmark, points within the cluster Theoretical sidelobe attenuation It can be determined based on the following formula (6).

[0082] (6).

[0083] Among them, in formula (6) The relative distance difference represents the point. With the main response point within the cluster Deviation in the distance dimension; The difference in relative velocity represents the point. With the main response point within the cluster Deviation in the Doppler velocity dimension; The relative azimuth difference represents the point. With the main response point within the cluster Deviation in the horizontal azimuth dimension; The relative pitch angle difference represents the point. With the main response point within the cluster Deviation in the vertical pitch angle dimension.

[0084] These four parameters Together, they form a four-dimensional relative deviation vector. (Function) Essentially, it's a multidimensional spatial sidelobe attenuation model. In an ideal radar system response, the farther a point deviates from the main response point (i.e., the larger these four deviation values ​​are), the greater the theoretical sidelobe attenuation it should experience. The larger the value, the better. In the cluster purification stage of quantitative inversion, the algorithm uses formula (6) to calculate the theoretical attenuation of each point and compares it with the actual measured intensity of that point; if the actual intensity is highly consistent with the theoretical sidelobe attenuation, it is determined to be a false sidelobe and is removed, thereby ensuring the uniqueness and high reliability of the inversion data.

[0085] For example, scatterer point clusters interior point Sidelobe matching error The following formula (7) can be used for calculation.

[0086] (7).

[0087] Among them, in formula (7) For point The actual strength, that is, the signal strength (or power, signal-to-noise ratio) in actual measurement. For scattering point clusters Internal main response point The actual measured signal strength of (i.e., the main peak of the cluster); For the points calculated based on formula (6) Theoretical sidelobe attenuation.

[0088] It can be understood that formula (7) is essentially a mathematical expression of pattern matching and anomaly detection. It calculates points... The actual relative intensity ( / ) and theoretical sidelobe attenuation ( The absolute difference between the points is used to determine whether the point represents a natural leakage (sidelobe) of the main peak energy in multidimensional space, or a real physical target independent of the main peak. This is a crucial step in achieving cluster purification and ensuring the reliability of backscattering coefficient inversion.

[0089] As an example, this application calculates the point. Theoretical sidelobe attenuation ( ) and sidelobe matching error Simultaneously, spatial distance thresholds and velocity difference thresholds are verified; points that satisfy the triple constraints of error, distance, and velocity are divided into a set of sidelobe repeating points. The remaining points are divided into a set of independent candidate points. .

[0090] For example, if point If all three threshold conditions expressed by the following formulas (8) to (10) are met, it is determined to be a sidelobe repetition point, that is, the point is more likely to be a repetition point formed by the extension of the main response, rather than an independent scatterer.

[0091] (8).

[0092] in, The preset error threshold is used to measure the strictness of energy matching, and its value can be 0.1~0.3. Formula (8) is used to determine candidate points. Does the energy decay conform to the sidelobe law? If This indicates that the actual performance at this point highly matches the theoretical sidelobe model, and the algorithm will identify it as spurious sidelobe interference and prepare to remove it; conversely, if it is greater than If the value does not conform to the sidelobe rule, it is likely a real independent scatterer and should be retained.

[0093] (9).

[0094] in, Represents a cluster of scattering points interior point Coordinates in three-dimensional space (usually including spatial information such as distance, azimuth, and pitch); Cluster Internal main response point The three-dimensional spatial coordinates of the (main peak); It represents the Euclidean distance or linear distance between two spatial coordinates. The preset spatial distance threshold is usually determined by the radar's spatial resolution and antenna beamwidth. Formula (9) is used to delineate the physical spatial range where sidelobes may exist. Real independent scatterers are usually physically separated from the main peak in space, while sidelobes are always attached to the main peak and generated in a local space. When candidate points... With the main response point Spatial distance less than At that time, this point possesses the spatial conditions to become a side lobe of the main peak.

[0095] (10).

[0096] in, For scattering point clusters interior point Relative to radar speed, such as radial Doppler velocity; For clusters Internal main response point The speed, such as radial Doppler velocity; The preset velocity difference threshold is usually determined by the radar's velocity resolution. Formula (10) is used to ensure the point... With the main response point Consistency in motion state. Since side lobes are the leakage of energy from the main peak in multidimensional space, they must possess motion characteristics that are the same as or very similar to those of the main peak. If point With the main response point The speed difference is less than If the velocity difference is too large, it indicates that the point may be background clutter or another independently moving object.

[0097] As an example, this application addresses scatterer point clusters. The set of points can be divided into two categories: the set of sidelobe repeating points. and the set of independent candidate points after removing duplicates This process completes the purification of scatterer point clusters, distinguishing between false sidelobes and real sub-scatterers. Thus, through this step, this application achieves a preliminary transformation from point-level output to scatterer-level purification at the point cloud level.

[0098] In some embodiments of this application, a weighted fusion of two modes is used to calculate the scatterer point cluster. First scattering intensity This refers to the scatterer-level primordial energy (primordial fusion energy). The device can select either Mode 1 or Mode 2 based on its data storage capacity.

[0099] As an example, in Mode 1, the first scattering intensity is calculated based on the point cloud intensity of each point in the main response point, the set of repeating points, and the set of independent candidate points of the scatterer point cluster. Mode 1 is suitable for low-cost mass-produced vehicle-mounted radar.

[0100] For example, in Mode 1, the scatterer point cluster First scattering intensity The energy merging model of the scatterer level, characterized by formula (11), is used for calculation.

[0101] (11).

[0102] In formula (11) Represents a cluster of scattering points Central main response point The signal strength; Indicates the main response point The primary response weight, This represents the set of candidate points within the cluster that are determined to be representative peaks, i.e., the set of sidelobe repetition points; Represents the set of sidelobe repetition points One sidelobe repetition point That is, a representative sub-peak; Indicates side lobe repetition point Regression weights for repetition points; Represents the set of independent candidate points An independent candidate point in the peak, i.e., an independent sub-peak; For the set of independent candidate points Points in The weights of independent candidate points.

[0103] For example, the main response weights in formula (11) =0.8 1. High signal-to-noise ratio, fixed at 1. Sidelobe repetition point weight. In adaptive schemes The value range is 0~0.2; in the simplified scheme, it is uniformly set to 0.1; if the signal strength of the sidelobe repetition point exceeds 50% of the signal strength of the main response point, it is forcibly set to 0. Independent candidate point weights. The value is set to 1 in spherical single clusters, 0.5 to 0.8 in long strip multi-adhesive clusters, and 0 if it is below the noise threshold.

[0104] It can be understood that formula (11) is a multi-source energy weighted summation model. It abandons the crude approach of traditional radar that relies solely on the strongest reflection point to assess the target size, and instead treats the target as a composite body composed of a main peak, representative secondary reflective surfaces, and independent reflective components. By weighted fusion of these real and effective scattered energies, the algorithm can more completely and stably restore the true physical size and reflection characteristics of the target, thereby providing high-fidelity data support for subsequent quantitative inversion.

[0105] In some embodiments of this application, in Mode 2, the first scattering intensity is calculated based on the original multi-channel complex observation data from the source information of the main response point, the set of repeating points, and the set of independent candidate points in the scatterer point cluster. This Mode 2 is applicable to high-precision imaging radar.

[0106] As an example, this application uses the associated index in the traceability information. Extract the complex amplitude of the raw multi-channel complex I / Q data at each point. Phase alignment compensation is performed based on the phase of the principal response point to construct a complex composite quantity. As shown in the following formula (12).

[0107] (12).

[0108] in, For scattering point clusters midpoint The corresponding local complex amplitude; For phase alignment and direction compensation weights.

[0109] For example, in Mode 2, the scatterer point cluster First scattering intensity That is, the original intensity of the scatterer level can be calculated using formula (13).

[0110] (13).

[0111] Among them, scatterer point clusters First scattering intensity Complex composite quantity It is obtained by taking the square of the modulus.

[0112] Therefore, the method for determining the first scattering intensity in Mode 2 is superior to simply summing the point intensities, because it can avoid distortion caused by phase inconsistency or repeated counting of information to a certain extent.

[0113] Thus, this application classifies scatterer point clusters into three categories of data: main response points, sidelobe repetition points, and independent candidate points, and fuses their energy differentially. It abandons the traditional method of direct summation and simply taking the peak value, avoiding the energy duplication counting problem caused by antenna sidelobe leakage and multiple points derived from a single scatterer. Corresponding fusion rules are set for points with different attributes, weakening the energy proportion of spurious sidelobe points while retaining effective scattering information. The final output of the scatterer-level original intensity more closely resembles the echo energy level of a real scatterer, providing accurate basic data for subsequent intensity correction and scattering coefficient inversion.

[0114] S302: Propagate the first scattering intensity and compensate for the system response to obtain the second scattering intensity.

[0115] Due to multiple interferences from the original first scattering intensity superposition distance, antenna, hardware, and algorithm, this application can also perform propagation and system response compensation on the first scattering intensity to calculate the corrected second scattering intensity. .

[0116] As an example, this application calculates the second scattering intensity based on the propagation and system response compensation model characterized by the following formula (14). .

[0117] (14).

[0118] in, The straight-line distance from millimeter-wave radar to the scatterer cluster, For the first scattering intensity, For the total system gain, For the transmission pattern gain, For receiving pattern gain, To handle chain gain and loss terms, Array channel compensation or residual system correction factor.

[0119] For monostatic radars, the propagation loss generally approximately satisfies Therefore, the intensity of the scatterer after propagation compensation can be obtained. Thus, we can obtain the above formula (14).

[0120] Therefore, after correction by this formula (14), the intensity deviation of the scatterer under different detection distances and different observation angles is completely eliminated. After correction, the second scattering intensity is no longer affected by radar hardware and propagation environment, but is determined only by the scattering characteristics of the target itself. Targets at different distances and at different angles can be directly compared.

[0121] Thus, this application constructs an integrated compensation model, sequentially correcting propagation loss, antenna pattern response, overall system gain, array channel error, and signal processing link gain loss, thereby eliminating the superimposed influence of non-target factors on echo intensity. After comprehensive compensation, the scatterer intensity deviation at different distances and observation angles is effectively eliminated. The corrected scatterer intensity is no longer constrained by radar hardware, signal processing, and spatial propagation conditions, and can objectively characterize the target's own scattering capability, laying a reliable foundation for the subsequent accurate mapping and quantitative inversion of backscattering coefficients.

[0122] S303: Maps the second scattering intensity to the backscattering coefficient.

[0123] As an example, this application maps the second scattering intensity to the radar cross section.

[0124] For example, the corrected second scattering intensity of this application The intensity to scattering coefficient is linearly related to the target scattering capability and can be represented by the basic mapping formula (15).

[0125] (15).

[0126] in, These are global scattering calibration coefficients. Calibration coefficients The radar algorithm can be calibrated and solidified offline using four methods: standard corner reflector, standard asphalt pavement, multi-target least squares fitting, and electromagnetic simulation combined with field measurement. It also supports two lightweight output modes: relative scattering coefficient and normalized scattering intensity, making it suitable for vehicle-mounted equipment with incomplete calibration.

[0127] For example, for area targets (road surfaces, walls, vegetation). Represents the backscattering coefficient, dimensionless, and its engineering equivalent is a logarithmic expression in the dB domain. .

[0128] For example, for discrete point targets (vehicles, pedestrians, guardrails and signs). Represents the radar cross section (RCS), in units of The project is converted into a logarithmic expression in the dBsm domain. .

[0129] S304: Calculate the inversion confidence of the scatterer point cluster.

[0130] Not all scatterer reconstruction results are suitable for direct use in quantitative analysis; that is, not all scatterer parameters can be used for quantitative perception in intelligent driving. Specifically, S304 constructs a quantitative inversion confidence model, outputting the inversion confidence level in the 0-1 interval. .

[0131] As an example, this application can calculate the inversion confidence based on at least one of the following: signal-to-noise ratio confidence, unique representation confidence, multi-frame stability confidence, calibration confidence, and sidelobe residual contamination confidence of the echo signal of the scatterer point cluster.

[0132] For example, to differentiate the quality of inversion data, the inversion confidence can be calculated based on the quantitative inversion confidence model characterized by the following formula (16). .

[0133] (16).

[0134] Among them, the weighting coefficient The sum is 1, and it can be adjusted offline according to the scenario; all five indicators are normalized to 0~1 based on the intermediate data mentioned above.

[0135] To assess the reliability of the signal-to-noise ratio (SNR), the SNR is calculated based on the distance-Doppler heatmap signal and noise power, and then linearly mapped piecewise to 0~1. To uniquely express credibility, i.e. whether the stripping of duplicate points is sufficient, the completeness of the stripping is evaluated based on the proportion of sidelobe duplicate points and the matching error. The reliability of multi-frame stability is characterized by statistically analyzing the scattering intensity and coordinate fluctuation amplitude of consecutive multi-frame frames; the smaller the fluctuation, the higher the score. To determine the reliability of the calibration, targets within the radar's standard calibration range receive higher scores. The lower the percentage of residual sidelobe energy, the higher the score for the confidence level of residual sidelobe contamination.

[0136] As an example, The rules for classifying engineering projects include: ≥0.7 High reliability, suitable for quantitative analysis of materials and road surfaces; 0.3 < <0.7 Low confidence level, only suitable for target visualization and tracking; Clusters with a density of ≤0.3 are directly removed from the scattering calculation.

[0137] In other embodiments, the above Figure 3 S304 above can be omitted, meaning that the inversion confidence level is not included in the scatterer parameter set.

[0138] Traditional inversion methods cannot evaluate the quality of output results and are prone to using distorted and invalid data as valid physical parameters. This application combines multiple evaluation indicators to calculate the inversion reliability, objectively characterizing the true reliability level of each scatterer inversion result. By classifying data based on inversion reliability, valid data can be screened out and contaminated data eliminated, ensuring that only reliable results are used in subsequent analysis and application stages, significantly reducing various problems caused by data distortion.

[0139] S305: Outputs complete scatterer parameters.

[0140] As an example, the final output of the complete scatterer parameter set for each scatterer point cluster can be referenced from the example in formula (5) above. At this time, the scatterer parameter set includes the scatterer's three-dimensional position, motion velocity, backscattering coefficient (equivalent RCS), and inversion confidence.

[0141] Then, the autonomous driving domain controller reads these scatterer parameters to realize dry and wet road surface recognition, vehicle or pedestrian material differentiation, obstacle scattering feature modeling, and optimize advanced driving functions such as adaptive cruise control (ACC), automatic emergency braking (AEB), and automatic parking.

[0142] In summary, this application constructs a complete processing chain of "original point cloud - scatterer reconstruction - quantitative scattering parameter output," forming a three-layer millimeter-wave sensing system of "display - reconstruction - quantification." At the data processing level, discrete single points are clustered into scatterer clusters of real objects, solving the problem that traditional point clouds cannot calculate the backscattering coefficient of surface targets. Simultaneously, duplicate points derived from the antenna's main lobe and side lobes are explicitly removed, and the energy repetition counting and intensity inflation issues are avoided through differentiated weighted fusion of the three types of points. In terms of physical correction, an integrated compensation model is constructed to remove system interference, ensuring that scattering intensity under different conditions can be compared laterally. Furthermore, offline calibration outputs RCS with real physical dimensions, enabling quantitative identification of target materials and road surface conditions. In addition, the system adds a five-dimensional weighted inversion credibility evaluation mechanism to improve robustness, is compatible with amplitude radar and 4D complex imaging radar, and supports various optimization schemes such as multi-frame joint extension, fully adapting to application scenarios such as vehicles, security, and robotics, providing new physical dimension data support for high-precision road environment modeling.

[0143] In some embodiments, in practical applications of radar detection systems, due to environmental complexity, equipment aging, or limitations of the calibration site, it is often difficult to obtain accurate absolute system calibration parameters (such as absolute system gain, atmospheric attenuation coefficient, etc.). To address this issue, this embodiment proposes a normalized comparison strategy based on relative scattering intensity. This method does not require the inversion of the radar cross section (RCS) with absolute physical dimensions (such as standard square meters), but rather focuses on extracting the relative scattering characteristics of the target or region. Specifically, the system uses the second scattering intensity, after propagation loss correction and system response compensation, as a benchmark, and maps it to a relative scattering coefficient or normalized scattering intensity through normalization processing. This processing method effectively decouples the dependence on absolute calibration parameters, enabling the output results to objectively reflect the comparative relationship of scattering capabilities between different targets (such as the material differences between pedestrians and vehicles) or between different areas of the same scene (such as the reflectivity differences between road surfaces and obstacles) under a unified relative scale. By outputting relative scattering intensity, the system can still provide high-value feature information for upper-layer applications, supporting tasks such as target classification, anomaly detection, and environmental semantic segmentation, significantly improving the robustness and practical value of the algorithm under insufficient calibration conditions.

[0144] In some embodiments, to address the problem that single-frame radar data is susceptible to noise interference, multipath effects, and instantaneous fading, leading to large fluctuations in scatterer intensity estimation and limited inversion accuracy, this embodiment introduces a multi-frame joint scatterer reconstruction mechanism. This method overcomes the limitations of single-frame processing, fully utilizing the trajectory continuity characteristics of the target in the time domain to spatiotemporally correlate and align observation data of the same physical scatterer across multiple consecutive frames. Based on this, the system employs multi-frame coherent or incoherent accumulation techniques to enhance the scatterer energy: for phase-stable signals, coherent accumulation significantly improves the signal-to-noise ratio; for scenarios with random or rapidly fluctuating phases, incoherent accumulation smooths instantaneous fluctuations. This multi-frame joint reconstruction strategy effectively suppresses random noise and clutter interference, and significantly smooths intensity flicker caused by dynamic environmental changes, thereby greatly improving the stability and consistency of scatterer-level intensity. Ultimately, this mechanism achieves deep information fusion in the time dimension, effectively improving the inversion accuracy of the backscattering coefficient and radar cross section, providing more reliable data support for subsequent target identification and behavior analysis.

[0145] In some embodiments, in a multi-channel radar detection system, amplitude and phase inconsistencies between channels of the antenna array can severely distort the received beam pattern, leading to differences in signal gain at different incident angles and thus affecting the accuracy of scattering intensity inversion. To address this, this embodiment proposes a joint processing mechanism combining subarray consistency and online channel correction. This method uses the online channel correction results obtained in the aforementioned steps (i.e., the real-time amplitude and phase error estimates for each channel) as prior information input to perform refined preprocessing of the received data. Specifically, the system uses correction parameters to perform reverse compensation on the received signals of each subarray and channel, eliminating amplitude imbalance and phase deviation caused by hardware link differences, thereby restoring the ideal beam pattern characteristics of the antenna array. After this correction process, the system can accurately determine the true antenna gain value at the target's angle, and then perform high-precision pattern compensation and system gain compensation. This mechanism effectively solves the pattern distortion problem caused by channel inconsistencies, ensures a unified benchmark for scattering intensity measurements at different field-of-view angles, significantly improves the accuracy of system gain compensation, and lays a solid foundation for obtaining high-fidelity scattering characteristic data.

[0146] In some embodiments, in complex radar detection environments, not all detected point cloud data is suitable for high-precision scattering inversion. Low signal-to-noise ratio, severe clutter interference, or sparse point cloud distribution often lead to significant deviations or even complete errors in the inversion results. To address this issue, this embodiment introduces an adaptive inversion strategy deeply integrated with the point cloud quality evaluation module. Before performing specific scattering coefficient calculations, this method first conducts a multi-dimensional quality assessment of the point cloud data within the current detection area, considering indicators such as point cloud density, spatial distribution uniformity, signal-to-noise ratio level, and trajectory continuity. Based on the evaluation results, the system divides the detection area into "high reversibility regions" and "low reversibility regions." Complex scattering inversion algorithms are only executed on high reversibility regions whose quality indicators meet preset thresholds, while regions with poor quality are filtered out or marked as unreliable. This cascaded processing mechanism of "evaluation first, inversion later" eliminates the interference of low-quality data on the inversion results from the source, greatly reduces the probability of erroneous inversion, and ensures that the final output scattering intensity data has extremely high reliability and physical consistency, thereby improving the output quality of the entire radar perception system.

[0147] The hardware architecture and system composition involved in this application will be described next.

[0148] This application provides an electronic device 4000 for executing the above-described millimeter-wave radar data processing method.

[0149] Reference Figure 4 The hardware components of the electronic device 4000 include: one or more processors 401, one or more memories 4002, etc. Figure 4 (Only one is shown), as well as the optional radar interface module 4003 and vehicle communication module 4004. The processor 4001 and memory 4002 exchange data via the device's internal bus.

[0150] The processor 4001 can be an onboard central processing unit (CPU), digital signal processor (DSP), field-programmable gate array (FPGA), neural processing unit (NPU), or application-specific integrated circuit (ASIC). It is used to load programs and sequentially execute processes such as point cloud source tracing, similarity clustering, sidelobe stripping, energy merging, system compensation, scattering coefficient mapping, and confidence calculation.

[0151] The memory 4002 may include flash memory (non-volatile flash memory) for storing calibration coefficients and the complete algorithm program, and random access memory (RAM) for caching point clouds, source information, and intermediate scattering parameters. Alternatively, the memory may employ read-only memory (ROM), RAM, or automotive solid-state storage. Its communication interface supports Controller Area Network Flexible Data Rate (CAN FD), automotive Ethernet, or 4G / 5G wireless communication. When the program code in the memory is loaded and executed by the processor, the electronic device can completely perform all the above processing steps and output scatterer parameters with true physical dimensions.

[0152] The radar interface module 4003 is used to connect to the vehicle-mounted millimeter-wave radar 20 via a hardware bus to receive the radar's raw intermediate frequency signal or pre-processed point cloud.

[0153] The vehicle communication module 4004 supports controller area network (CAN) or Ethernet communication and is used to transmit the final set of scatterer parameters to the autonomous driving decision unit or vehicle network cloud server.

[0154] In another implementation of this embodiment, the electronic device can also be an on-board domain controller (vehicle radar controller or on-board chip), which can independently carry out the entire scattering inversion calculation process, effectively sharing the computing power pressure of the main autonomous driving chip.

[0155] In another implementation of this embodiment, the electronic device can also be an embedded processor chip (such as a digital signal processor or microcontroller unit) integrated within the millimeter-wave radar sensor. In this scheme, the radar antenna, radio frequency front-end, signal processing circuit, and the processor of this embodiment are integrated into the same module to form an integrated radar sensor. This sensor directly outputs a set of physically meaningful scatterer parameters (including the position, velocity, backscattering coefficient, and optional confidence level of each scatterer), rather than point cloud data in the traditional sense, thereby simplifying the processing burden of the vehicle's upper-level control system and enabling it to make decisions directly using data with quantitative physical meaning.

[0156] The memory can be volatile memory (such as dynamic random access memory) or non-volatile memory (such as flash memory, read-only memory, or programmable read-only memory), or any combination thereof. The processor can be one or more of a general-purpose central processing unit, a digital signal processor, a graphics processor, a field-programmable gate array, or an application-specific integrated circuit. The program can be stored in the memory in the form of software, firmware, or a hardware description language.

[0157] This embodiment provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, they implement the millimeter-wave radar data processing method described above.

[0158] In one implementation of this embodiment, the computer program product is a non-volatile computer-readable storage medium, such as a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk. The storage medium stores computer program code, which, when loaded and executed by the processor of the millimeter-wave radar signal processing device or the vehicle controller, enables [the following functionality / function]: Figure 2 Steps S201 to S203 described herein (including) Figure 3 The millimeter-wave radar data processing method is shown in the optional sub-steps S301, S302 to S305.

[0159] In another implementation of this embodiment, the computer program product can also be in the form of a carrier of program instructions. For example, the computer program instructions can be transmitted to the vehicle's receiving module in the form of signals through wired or wireless communication networks, and received and executed by the on-board processor, thus achieving the above method. For example, after leaving the factory, the vehicle can download the program product of this embodiment from the server via over-the-air (OTA) technology to update the signal processing software of the on-board millimeter-wave radar system, enabling the system to obtain the ability to reconstruct scatterers and perform intensity correction, thereby upgrading the traditional point cloud output to a scatterer parameter output with quantitative physical meaning.

[0160] The computer program product can exist independently of the hardware, or it can be pre-installed in the memory of the aforementioned electronic device as part of firmware or software. When the processor executes the program, the program product drives the electronic device to perform each step of the millimeter-wave radar data processing method provided in this application, realizing a complete processing flow from point cloud acquisition to scatterer parameter output.

[0161] In some embodiments, the above-described electronic equipment may be deployed in vehicle 10. The hardware structure of the vehicle used in the embodiments of this application will be described below.

[0162] As an example, Figure 5This is a schematic diagram of a possible functional framework for a vehicle 10 provided in an embodiment of this application. For example... Figure 5 As shown, the functional framework of vehicle 10 may include various subsystems, such as the sensor system 100, control system 200, one or more peripheral devices 300 (one is shown as an example), power supply 400, and computer system 500. In some embodiments, vehicle 10 may also include other functional systems, such as an engine system that provides power to vehicle 10, etc., which are not limited herein.

[0163] The sensor system 100 may include several detection devices that can sense the measured information and convert the sensed information into electrical signals or other required forms of information output according to a certain rule. These detection devices may include a global positioning system 11 (GPS), a vehicle speed sensor 12, an inertial measurement unit 13 (IMU), a radar unit 14, a laser rangefinder 15, a camera unit 16, a wheel speed sensor 17, a steering sensor 18, a gear position sensor 19, or other components for automatic detection, etc., and this application is not limited thereto.

[0164] Radar unit 14, also known as a radar system, may include vehicle-mounted millimeter-wave radar 20. The radar unit uses wireless signals to sense objects in the current environment where the vehicle 10 is traveling. Camera unit 16 is used to capture images, such as pictures and videos.

[0165] The autonomous driving system 24 may include hardware and software systems for processing and analyzing data input to the autonomous driving system 24 to obtain actual control parameters of each component in the control system 200, such as the desired braking pressure of the brake controller in the braking unit and the desired torque of the engine.

[0166] Several functions of vehicle 10 are controlled and implemented by computer system 500. Computer system 500 may include one or more processors 51 (the figure shows one processor as an example), memory 52 (also called storage device), and router 53. In practical applications, the memory 52 may be located inside computer system 500 or outside computer system 500, such as as a cache in vehicle 10, etc., and this application does not limit this. For example, computer system 500 may include the processor or chip where the electronic device 400 mentioned above is located.

[0167] The processor 51 may include one or more general-purpose processors, such as a graphics processing unit (GPU). The processor 51 can be used to execute relevant programs or instructions corresponding to programs stored in the memory 52 to implement the corresponding functions of the vehicle 10. As an example, the processor 51 may include an on-board chip 511, for example, the on-board chip 511 can be... Figure 4 The vehicle-mounted chip where the electronic device 4000 is located.

[0168] The memory 52 can be used to store a set of program code or instructions corresponding to the program code, so that the processor 51 can call the program code or instructions stored in the memory 52 to implement the corresponding functions of the vehicle 10. This function includes, but is not limited to, […]. Figure 5 The functional framework diagram of vehicle 10 shown includes some or all of the functions.

[0169] Among them, this application Figure 5 The sensor system 100, control system 200, and computer system 500 shown are merely examples and do not constitute a limitation. In practical applications, the vehicle 10 can combine several components according to different functions to obtain subsystems with corresponding functions.

[0170] In this specification, the reference to "an embodiment" or "an embodiment" means that a specific feature, structure, or characteristic described in connection with the embodiment is included in at least one exemplary implementation or technology disclosed according to an embodiment of this application. The appearance of the phrase "in an embodiment" in various places in the specification does not necessarily refer to the same embodiment.

[0171] Furthermore, the language used in this specification has been primarily chosen for readability and instructional purposes and may not have been chosen to depict or limit the disclosed subject matter. Therefore, the embodiments disclosed herein are intended to illustrate, and not limit, the scope of the concepts discussed herein.

Claims

1. A millimeter-wave radar data processing method, characterized in that, include: Acquire point cloud data detected by millimeter-wave radar; The source information associated with the point cloud data is determined, wherein the source information includes the original radar data and intermediate data that generated the point cloud data; Based on the source information, determine the aggregation similarity between any two points in the point cloud data; Based on the aggregation similarity between any two points, at least one cluster of scatterer points is obtained through a clustering algorithm; Identify the sidelobe repetition points within each scatterer point cluster, and divide each scatterer point cluster into a set of repetition points and a set of independent candidate points; Different weighting values ​​are assigned to the main response point of the scatterer point cluster, the repeating points in the repeating point set, and the independent candidate points in the independent candidate point set, respectively, to obtain the first scattering intensity; The first scattering intensity is propagated and compensated for by system response to obtain the second scattering intensity; The second scattering intensity is mapped to the backscattering coefficient; The scatterer parameters of each scatterer cluster include the backscattering coefficient of the scatterer cluster.

2. The method according to claim 1, characterized in that, The intermediate data includes distance cells, Doppler cells, angle information, data channel identifiers, heatmap cells, and local complex observation indexes.

3. The method according to claim 1, characterized in that, The determination of the aggregate similarity between any two points in the point cloud data based on the source information includes: Based on the spatial Euclidean distance, radial velocity difference, and observation domain distance between any two points in the point cloud data, the aggregation similarity between the two points is calculated.

4. The method according to claim 1, characterized in that, The step of setting different weighting values ​​for the main response point of the scatterer point cluster, the repeating points in the repeating point set, and the independent candidate points in the independent candidate point set to obtain the first scattering intensity includes: The first scattering intensity is calculated based on the point cloud intensity of each point in the main response point, the set of repeating points, and the set of independent candidate points of the scatterer point cluster; or The first scattering intensity is calculated based on the original multi-channel complex observation data from the source information of the main response point, the set of repeating points, and the set of independent candidate points in the scatterer point cluster.

5. An electronic device, characterized in that, include: One or more processors; One or more memories; the one or more memories store one or more programs that, when executed by the one or more processors, cause the electronic device to perform the millimeter-wave radar data processing method according to any one of claims 1 to 4.

6. A computer program product, characterized in that, It includes a computer program / instruction that, when executed by a processor, implements the millimeter-wave radar data processing method according to any one of claims 1 to 4.

Images