Carrier speed estimation method and device based on millimeter wave radar, and storage medium

Abstract

The invention discloses a carrier speed estimation method and device based on a millimeter wave radar, and a storage medium, and relates to the technical field of carrier speed estimation. The method comprises the following steps: firstly, carrying out interference suppression processing on a distance-velocity spectrum matrix to obtain an interference suppression spectrum matrix; the first path performs velocity dimension energy aggregation on the interference suppression spectrum matrix to generate a static background energy distribution spectrum and determine a first velocity estimation value according to the static background energy distribution spectrum; a second path screens and verifies an effective ground reflection target from the interference suppression spectrum matrix, and a second speed estimation value is obtained by combining the pitch angle of the effective ground reflection target; and finally, respectively carrying out time sequence filtering and weighted fusion on the outputs of the two paths, and outputting the final carrier speed. According to the method, the problem that a traditional point cloud method fails in an open scene is fundamentally solved, and the robustness, accuracy and stability of a spectrum analysis method in a complex interference environment are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of carrier velocity estimation technology, and particularly relates to a carrier velocity estimation method, device and storage medium based on millimeter-wave radar. Background Technology

[0002] Millimeter-wave radar, as an important environmental perception sensor, has been widely used in fields such as autonomous driving and intelligent transportation. Among these applications, achieving self-velocity measurement of a vehicle using radar (i.e., calculating the vehicle's own velocity solely through radar echoes without relying on an external reference frame) is a key technology. Currently, there are two main technical approaches to self-velocity measurement based on millimeter-wave radar, but both have significant technical bottlenecks, making it difficult to achieve robust and accurate velocity estimation in complex and ever-changing real-world scenarios.

[0003] (1) Indirect methods based on point cloud clustering and tracking and their inherent limitations

[0004] The first technical approach is an indirect method. This method first performs a series of processing steps on the raw radar echo signal (including two-dimensional fast Fourier transform, constant false alarm rate detection, clustering, etc.) to generate a discrete target point cloud; then, a tracking algorithm is used to separate the static target from the point cloud and estimate its motion state relative to the radar; finally, the absolute velocity of the carrier itself is deduced from the relative motion between the static target and the radar.

[0005] The core drawback of this method lies in its strong scene dependence. Its performance is highly dependent on the presence of a sufficient number of static point cloud targets with good reflectivity (such as guardrails, signs, and buildings) in the environment. In feature-rich scenes such as cities and urban blocks, the method works. However, in scenes with sparse static targets, such as rural roads and open spaces, the number of static point clouds that can be effectively detected and stably tracked decreases drastically, causing the method to fail or its velocity measurement accuracy to drop sharply. The fundamental reason is that point clouds are "high-level" abstract results obtained after multiple levels of processing and information filtering, and a large amount of weak scattering information from the original signal is lost during the generation process. Inferring the underlying carrier motion state from these incomplete high-level features carries an inherent risk of information loss.

[0006] (2) Direct methods based on range-velocity spectrum analysis and the shortcomings of their existing implementations

[0007] To overcome the scenario-dependent problem of the first technical approach, the industry has proposed a second approach—a direct method based on range-velocity (or Doppler) spectrum analysis. The physical basis of this method is that the velocity of a stationary background target (such as a road surface or a stationary guardrail) in the Doppler dimension of the radar echo is equal in magnitude and opposite in direction to the velocity of the radar carrier itself. Therefore, theoretically, the carrier velocity can be directly calculated simply by locating the velocity element containing the peak of static background energy accumulation in the range-velocity spectrum, an intermediate product of radar signal processing. This method avoids information loss in point cloud generation and theoretically possesses better environmental universality.

[0008] However, a typical existing implementation scheme under this approach has revealed insufficient robustness in practical applications, specifically manifested as follows:

[0009] Susceptible to interference from strong dynamic targets: When a large vehicle or other strong reflective dynamic target appears in the radar field of view, the energy generated by it in the range-velocity spectrum will seriously interfere with or even completely mask the energy peak of the static background.

[0010] Susceptible to noise and spectrum leakage contamination: System noise, DC components and spectrum leakage can contaminate critical velocity units at and near zero velocity, leading to incorrect static background peak localization.

[0011] Unstable estimation in sparse static scenes: When the overall reflection energy of the static background is low, its energy peak is weak and not prominent, and it is easily affected by random fluctuations, causing non-physical jumps in the velocity estimate between frames.

[0012] Analysis revealed that the root cause of the aforementioned shortcomings lies in the overly crude energy aggregation strategy employed in this typical approach. Specifically, for each velocity unit in the range-velocity spectrum, the energy values ​​of all its corresponding range units are summed indiscriminately to form an energy-velocity array, and the velocity unit with the highest summation energy is identified as the static background. This "global summation" operation essentially mixes useful static background signals, highly interfering dynamic target signals, and various types of noise indiscriminately. When the signal-to-interference ratio (SIR) is low, the peak of the summation result is easily dominated by the energy of strong interference, rather than reflecting the true static background velocity, thus causing the method to fail in complex scenarios.

[0013] In summary, existing point cloud indirect methods are limited by their inherent scenario-dependent nature, while more promising direct spectral analysis methods suffer from weak anti-interference capabilities and poor robustness due to deficiencies in energy utilization strategies in their current typical implementations. Therefore, there is an urgent need for a new method that can achieve a more intelligent and robust energy screening and focusing mechanism within the range-velocity spectral analysis framework, fundamentally improving the practicality and reliability of millimeter-wave radar self-velocity measurement technology in various real-world scenarios. Summary of the Invention

[0014] In view of the above-mentioned defects in the existing technology, the purpose of this invention is to provide a carrier velocity estimation method, device and storage medium based on millimeter-wave radar, so as to overcome the defects of existing direct velocity measurement methods, which are susceptible to interference and estimation failure in low signal-to-interference ratio scenarios due to the coarse energy aggregation of the range-velocity spectrum, and improve the robustness and universality of millimeter-wave radar self-velocity measurement technology in real complex environments.

[0015] This invention solves the above-mentioned technical problems through the following technical solution: a carrier velocity estimation method based on millimeter-wave radar, comprising:

[0016] Based on the range-velocity spectrum matrix of the current frame, interference spectrum energy is identified and suppressed to obtain an interference suppression spectrum matrix; wherein, the interference spectrum energy originates at least from a dynamic target whose potential motion velocity differs from that of the carrier by more than a preset tolerance;

[0017] Energy aggregation is performed along the velocity dimension on the interference suppression spectrum matrix to obtain the static background energy distribution spectrum; a first velocity estimate is determined based on the global peak value of the static background energy distribution spectrum.

[0018] Based on the installation geometry parameters of the millimeter-wave radar, candidate spectral elements that may correspond to stationary ground reflectors are identified and extracted from the interference suppression spectral matrix; the direction of arrival of the candidate spectral elements is estimated to obtain the elevation angle, and the elevation angle is used to verify whether it is an effective surface reflection target; based on the verified velocity information and elevation angle of the ground reflection target, a second velocity estimate is calculated.

[0019] The first velocity estimate and the second velocity estimate are subjected to time-series filtering to obtain the first tracking velocity and the second tracking velocity, respectively. The first tracking velocity and the second tracking velocity are then weighted and fused to output the final carrier velocity estimate.

[0020] This invention abandons the high-level feature of point clouds, which loses underlying information, and directly analyzes the raw energy in the range-velocity spectral matrix. By aggregating static background energy in the velocity dimension, even in open scenes with weak and scattered static scattering points, the accumulated energy can still form stable spectral peaks for detection. This completely eliminates the method's performance dependence on the presence of strong reflective or easily trackable static objects in the environment, achieving a fundamental shift from "relying on scene features" to "relying on physical principles." Therefore, it can work reliably in various road environments, completely overcoming the scene-dependent defects of point cloud methods.

[0021] This invention addresses interference through a progressive, three-tiered mechanism, ensuring the robustness of carrier velocity estimation: First, it actively identifies and suppresses near-field strong dynamic target energy that is mismatched with the carrier's motion state at the signal processing front end, eliminating the main source of interference. Second, it directly performs global peak detection on the static background energy distribution spectrum to determine the first velocity estimate, which is computationally simple, reduces reliance on post-processing algorithms, and improves estimation efficiency and real-time performance. Finally, it introduces a second estimation path based on geometric verification, utilizing ground reflection to determine the second velocity estimate, and eliminates interference from non-ground dynamic targets through a physical reflection mechanism. These three mechanisms work together to ensure accurate and stable extraction of static background velocity peaks in complex traffic scenarios, thereby improving the reliability of the final velocity estimation result.

[0022] This invention creatively constructs a dual-path parallel estimation and intelligent fusion architecture. The two paths (macroscopic analysis based on the overall energy spectrum and microscopic verification based on ground target geometry) are independent and mutually verifying, providing redundant information. Based on this, each path undergoes temporal filtering based on a motion model to smooth the output, and then dynamic weighted fusion is performed according to the real-time confidence levels of each path. This design not only improves the stability of individual paths but also achieves intelligent fault tolerance at the system level: when one path temporarily fails due to extreme interference, the system can automatically rely on the other path, thereby ensuring the continuity, reliability, and optimality of the final velocity estimate.

[0023] Furthermore, the identification and suppression of interfering spectral energy therein includes:

[0024] The range-velocity spectrum matrix is ​​subjected to constant false alarm rate detection to obtain potential target cells;

[0025] The potential target units are clustered to form at least one candidate target cluster;

[0026] Calculate the cluster features of each of the candidate target clusters;

[0027] Based on the cluster characteristics and interference determination conditions, interference target clusters are identified from the candidate target clusters;

[0028] Set the energy of the spectral units covered by the interference target cluster to zero.

[0029] This invention achieves a fundamental improvement in anti-interference capability through a progressive processing approach: constant false alarm rate detection and localization, clustering, multi-feature analysis and discrimination, and precise energy zeroing. First, it separates independent energy-gathering entities (candidate target clusters) from a complex spectrum. Then, it comprehensively utilizes multi-dimensional cluster features such as energy, distance, and velocity for highly reliable discrimination. Finally, it eliminates only specific spectral units identified as strong dynamic near-field interference. This process overcomes the shortcomings of traditional methods that indiscriminately mix signals, causing interference to overwhelm useful signals. While thoroughly suppressing the main interference sources, it maximizes the protection of the integrity of the global static background spectrum information, laying a clean data foundation for subsequent high-precision velocity estimation.

[0030] Furthermore, the cluster features include the total energy, centroid distance, and centroid velocity of the candidate target cluster;

[0031] The interference determination conditions include: the total energy exceeds the energy threshold, the centroid distance is less than the distance threshold, and the absolute value of the difference between the centroid velocity and the estimated carrier velocity of the previous frame is greater than the velocity threshold.

[0032] This invention provides a clear, quantifiable, and physically intuitive logic for identifying interference targets by setting a joint judgment condition of "high energy, close range, and unmatched speed": it uses total energy to filter out strong reflectors that are sufficient to cause interference, limits the distance to the key area of ​​the radar near field by the center of mass distance, and accurately distinguishes between dynamic interference and static background based on the significant difference between the center of mass speed and the historical speed of the vehicle. This achieves efficient and reliable identification and elimination of strong dynamic targets in the near field, and effectively avoids the error in vehicle speed estimation caused by misjudging interference such as vehicles in the adjacent lane as static reference objects in complex traffic scenarios.

[0033] Furthermore, the energy aggregation along the velocity dimension includes:

[0034] For each velocity index of the interference suppression spectrum matrix, the energy values ​​of all corresponding distance indices are summed to obtain the static background energy distribution spectrum.

[0035] This invention performs energy aggregation along the velocity dimension, summing the energy of the same velocity unit across all distance dimensions to form a static background energy distribution spectrum. This process effectively focuses and amplifies the static background energy dispersed at different distances throughout the detection scene, significantly enhancing the energy distinguishability of the static background in the velocity spectrum and overcoming the problem that the weak energy of static scattering points within a single distance unit makes reliable detection difficult.

[0036] Further, based on the global peak value of the static background energy distribution spectrum, a first velocity estimate is determined, including:

[0037] Calculate the sum of neighborhood energies corresponding to each velocity index in the static background energy distribution spectrum; wherein, the sum of neighborhood energies is the sum of the energy value of the velocity index itself and the energy value of at least one adjacent velocity index;

[0038] Determine the peak velocity index corresponding to the maximum value of the sum of the neighborhood energies;

[0039] The first speed estimate is calculated based on the peak speed index and the preset speed resolution.

[0040] This invention enhances the ability to identify real static background energy peaks by introducing "neighborhood energy sum" as a peak detection criterion. Even if the real peak is slightly wide and flat or has a small offset due to the sparseness of the target, its joint energy with the neighboring units can still form a significant extremum, thereby ensuring that the real peak representing the carrier velocity can still be reliably located in complex spectrum environments, and significantly reducing the inter-frame jitter and gross error of the velocity estimation.

[0041] Furthermore, based on the installation geometry parameters of the millimeter-wave radar, candidate spectral units that may correspond to stationary ground reflectors are determined and extracted from the interference suppression spectral matrix, including:

[0042] A radial distance range is determined based on the installation height of the millimeter-wave radar;

[0043] From the interference suppression spectrum matrix, extract all spectral units whose radial distance falls within the radial distance interval, and use them as candidate spectral units.

[0044] This invention achieves efficient initial screening of stationary ground reflectors by transforming the prior geometric parameter of radar installation height into an intelligent constraint on the radial distance range. By utilizing the geometric relationship between installation height and ground reflection, the radar's detection space is focused on a specific range band most likely to contain effective surface echoes. This eliminates a large number of irrelevant long-range aerial targets, high-altitude obstacles, and near-field clutter interference at the data processing front end, significantly improving the processing efficiency and signal quality of subsequent direction-of-arrival estimation and velocity calculation. This provides an accurate and clean data foundation for obtaining reliable second velocity estimates.

[0045] Further, verifying whether it is an effective surface reflection target based on the pitch angle includes:

[0046] Based on the radial distance and elevation angle of the candidate spectral units, and in conjunction with the radar installation height, the corresponding height estimation deviation is calculated.

[0047] If the absolute value of the height estimation deviation is less than or equal to the preset height tolerance threshold, it is determined to be an effective surface reflection target.

[0048] This invention achieves physical consistency verification of ground-reflecting targets by constructing a geometric verification model of "radial distance-pitch angle-installation height": using the radar-measured pitch angle and radial distance, the estimated height of the target is inferred and compared with the actual ground height (characterized by installation height) with tolerance. This allows for a strict distinction between truly stationary reflection points that are close to the ground and false targets with similar radial velocities but located at different heights. This fundamentally ensures the authenticity and reliability of the data source on which the second velocity estimation path depends, effectively avoiding systematic errors introduced by misusing non-ground target velocity information.

[0049] Further, time-series filtering is performed on the first velocity estimate and the second velocity estimate respectively to obtain the first tracking velocity and the second tracking velocity, including:

[0050] Based on the carrier motion model, a filter for velocity tracking is constructed;

[0051] The first velocity estimate of the current frame and multiple historical frames is input into the filter for filtering and tracking to obtain a smooth and continuous first tracking velocity.

[0052] The second velocity estimate of the current frame and multiple historical frames is input into the filter for filtering and tracking to obtain a smooth and continuous second tracking velocity.

[0053] This invention introduces a temporal filter based on a carrier motion model to improve isolated single-frame velocity estimation into smooth tracking velocity with temporal continuity. The filter utilizes the inertial constraints of carrier motion to optimally fuse the velocity values ​​of the current and historical multiple frames, effectively suppressing random fluctuations and jumps in single-frame estimation caused by sensor noise, target flicker, or instantaneous interference. This results in a more physically reasonable and stable velocity trajectory, providing high-quality and reliable input for subsequent fusion decisions and significantly improving the system's output stability and user experience in dynamically changing scenarios.

[0054] Based on the same concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program or instructions stored in the memory, wherein the processor executes the computer program or instructions to implement the carrier velocity estimation method based on millimeter-wave radar as described above.

[0055] Based on the same concept, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the carrier velocity estimation method based on millimeter-wave radar as described above.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] For scenarios with sparse static targets, this invention proposes a global aggregation of static background energy, independent of specific target identification and clustering. This overcomes the limitation of traditional point cloud clustering methods failing in open scenes due to target sparsity. By directly aggregating the energy of all distance cells along the velocity dimension, weak and dispersed static scattering energy can be effectively gathered. Even in environments with sparse static features, stable and detectable spectral peaks can still be formed, achieving a robust transition from "depending on specific environmental features" to "depending on universal physical energy distribution."

[0058] To address interference from highly dynamic targets and noise, this invention establishes a three-tiered progressive mechanism: front-end interference suppression, mid-stage energy aggregation and peak extraction, and back-end geometric verification. First, by identifying and suppressing the spectral energy of dynamic targets with speeds significantly different from the carrier's, the main sources of interference are weakened at the source. Second, on the static background energy distribution spectrum formed by energy aggregation, the stability of identifying the true velocity peak is improved through neighborhood energy and detection of global peaks. Finally, through pitch angle verification based on geometric relationships, information about stationary ground reflectors is independently extracted, physically eliminating non-ground dynamic interference. These three mechanisms complement each other, ensuring reliable extraction of static background velocity information in complex traffic scenarios.

[0059] Unlike traditional single-path estimation, this invention constructs a parallel estimation framework with two paths: "macroscopic energy spectrum analysis" and "microscopic ground target geometric analysis," integrating temporal filtering and dynamic weighting strategies. The two paths are physically independent and mutually verify each other, enhancing the system's fault tolerance; time-based filtering ensures the temporal continuity of velocity output; and the weighting fusion mechanism, dynamically adjusted based on path confidence, enables the system to adaptively select the more reliable estimation result. This system-level design significantly improves the overall stability, adaptability, and final accuracy of velocity estimation. Attached Figure Description

[0060] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a flowchart of the carrier velocity estimation method based on millimeter-wave radar in an embodiment of the present invention. Detailed Implementation

[0062] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0064] Example 1

[0065] This invention provides a method for estimating the speed of a vehicle (e.g., a vehicle) based on millimeter-wave radar. The method can be implemented by an electronic control unit integrated with or connected to the millimeter-wave radar, an onboard computing platform, or a dedicated processor. The millimeter-wave radar is preferably a frequency-modulated continuous-wave radar with an array antenna to support direction-of-arrival estimation. Figure 1 Here is an overall flowchart of a carrier velocity estimation method provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the method mainly includes the following steps:

[0066] Step S1: Perform two-dimensional fast Fourier transforms on the original echo signal received by the millimeter-wave radar in the range and velocity dimensions sequentially to generate the range-velocity spectrum matrix of the current frame.

[0067] The millimeter-wave radar receives the echo signal from the antenna, down-converts it using a mixer, and then performs analog-to-digital conversion to obtain a digital intermediate frequency (IF) signal. To generate the range-velocity spectrum matrix, the following steps are performed:

[0068] Fast time dimension processing (distance dimension): The sampling sequence (i.e., fast time dimension data) in each transmission cycle is padded with zeros and a Hanning window is added to suppress spectral leakage. A fast Fourier transform is then performed to transform the signal to the distance frequency domain to obtain the distance dimension information.

[0069] Slow-time dimension processing (velocity dimension): The sampled values ​​of multiple consecutive transmission cycles on the same distance cell are used to form a sequence (i.e., slow-time dimension data), and zero-padding, Hanning windowing, and fast Fourier transform are performed to analyze its Doppler frequency shift.

[0070] Two-dimensional spectrum generation: The results of the two-dimensional FFT above are combined and amplitude calculated to finally generate a two-dimensional complex matrix, namely the range-velocity spectrum matrix. In this matrix, the row index corresponds to the range element r (or range index), whose physical distance is r × ΔR, where ΔR is the range resolution; the column index v corresponds to the velocity element (or velocity index), whose physical velocity is v × ΔV, where ΔV is the velocity resolution. The squared amplitude (or magnitude) of the element RD(r,v) in the matrix represents the echo energy intensity at distance r × ΔR and velocity v × ΔV, where r = 0, 1, ..., N. r -1, v = 0, 1, ..., N v -1, N r N represents the number of range cells in the range-velocity spectrum matrix. v This represents the number of velocity elements in the range-velocity spectrum matrix.

[0071] Step S2: Based on the range-velocity spectrum matrix of the current frame, identify and suppress the interference spectrum energy to obtain the interference suppression spectrum matrix.

[0072] Step S2 aims to accurately identify and remove interference energy from the range-velocity spectrum matrix generated in step S1 that contaminates the static background velocity estimation, thereby obtaining a cleaner interference suppression spectrum matrix. Interference mainly originates from near-field strong dynamic targets (such as vehicles in adjacent lanes), the system's DC component, and near-field clutter. The specific implementation process is as follows:

[0073] Step S2.1: Pre-clipping of the distance dimension edge interference area.

[0074] Considering that interference such as DC offset and antenna coupling in the radar system is mainly concentrated in the very near field, while range gates beyond the effective range are mainly noise, this step first performs edge pruning on the range dimension to improve data quality. Two positive integers N1 and N2 are set (e.g., N1 = 3, N2 = 3), corresponding to the number of nearest (near field) and farthest (beyond range) range gates to be excluded, respectively. For the range-velocity spectrum matrix, only range gates from the N1th range cell to the Nth range cell are retained. r Data from -N²−1 range cells is used to obtain a cropped range-velocity spectrum matrix. This operation isolates the main noise sources before subsequent processing, improving the signal-to-noise ratio of subsequent energy and computation.

[0075] Step S2.2: Potential target cell detection based on constant false alarm rate (CFAR) detection.

[0076] Two-dimensional constant false alarm rate (CFAR) detection is applied to the cropped range-velocity spectrum matrix. Specifically, a cell-average CFAR detector is used. For each cell in the cropped range-velocity spectrum matrix, the local noise level is estimated using the amplitude (or energy) of its surrounding guard cells and background reference cells, and an adaptive detection threshold is set accordingly. Cells with amplitudes exceeding this threshold are marked as potential target cells; these cells represent scattering points in the current frame with energy significantly higher than the background noise.

[0077] Step S2.3: Clustering and feature extraction of potential target units.

[0078] To aggregate discrete potential target units into physically independent target entities, a density-based clustering algorithm is used to cluster the potential target units obtained in step S2.2. In a preferred embodiment, the DBSCAN algorithm is used, which aggregates spatially close and velocously continuous units into a candidate target cluster Ck based on the proximity of units in the distance-velocity two-dimensional space, where k is the cluster index.

[0079] For each candidate target cluster Ck, its multidimensional features are calculated as the basis for subsequent interference discrimination. Core features include:

[0080] Cluster total energy Ek: The sum of the energies (squared magnitudes) of all potential target cells within the cluster.

[0081] Centroid distance Rk: Based on the energy weighting of each potential target unit in the cluster, calculate its centroid position in the distance dimension, and multiply it by the distance resolution ΔR to obtain the centroid distance.

[0082] Centroid velocity Vk: Based on the energy weighting of each potential target unit in the cluster, calculate its centroid position in the velocity dimension, and multiply it by the velocity resolution ΔV to obtain the centroid velocity.

[0083] Step S2.4: Near-field strong dynamic interference target identification.

[0084] Based on the features extracted in step S2.3 and combined with historical information, a discrimination logic is established to identify the cluster of interfering targets that need to be suppressed (i.e., near-field strong dynamic targets). The judgment condition must simultaneously satisfy the following three points:

[0085] High-intensity energy condition: The total cluster energy Ek is greater than a preset energy threshold (e.g., corresponding to a squared amplitude of 20000). This condition filters out targets with signal strength sufficient to significantly mask the static background spectrum;

[0086] Close-range condition: The centroid distance Rk is less than a preset distance threshold (e.g., 10 meters). This condition limits the focus of interference suppression to the radar's near-field region.

[0087] High dynamic condition: The absolute value of the difference between the centroid velocity Vk and the estimated carrier velocity in the previous frame is greater than a preset velocity threshold (e.g., 2 m / s). This condition is used to distinguish between static / low-speed targets (such as road signs and guardrails) with similar motion states to the carrier and independently moving dynamic interference targets. If the current frame is the first frame and there is no historical velocity information, this condition is temporarily disabled.

[0088] Step S2.5: Set the interference energy to zero and generate the interference suppression spectrum matrix.

[0089] All range-velocity cells covered by the candidate target cluster Ck identified as near-field strong dynamic interference targets are set to zero (or their complex amplitudes are set to zero) in the cropped range-velocity spectral matrix. This operation directly eliminates the energy contribution of these strong interference sources to the spectral matrix. The matrix after this processing is denoted as the interference suppression spectral matrix, which retains the static background and the energy of other targets not identified as near-field strong dynamic interference, and is used for subsequent velocity estimation.

[0090] Through the series of processes in step S2, the present invention actively and accurately eliminates the main interference sources before energy aggregation, laying a key foundation for the subsequent robust extraction of static background velocity peaks.

[0091] Step S3: Perform energy aggregation along the velocity dimension on the interference suppression spectrum matrix to obtain the static background energy distribution spectrum.

[0092] Step S3 aims to process the interference suppression spectrum matrix obtained in step S2. Through aggregation, it highlights the energy distribution generated by the static background target (whose velocity is equal in magnitude and opposite in direction to the carrier's own velocity), forming a clear static background energy distribution spectrum, which lays the foundation for subsequent velocity peak detection.

[0093] Static background targets (such as road surfaces and stationary guardrails) may be dispersed across different distance cells in the distance dimension, but converge in the same velocity cell (corresponding to the negative of the vehicle's velocity) in the velocity dimension. To enhance this characteristic, the interference suppression spectrum matrix is ​​summed along the velocity dimension. Specifically, for each velocity index v, the sum of its energy across all effective distance indices is calculated, resulting in a one-dimensional vector, i.e., the static background energy distribution spectrum.

[0094] Step S4: Determine the first velocity estimate based on the global peak value of the static background energy distribution spectrum.

[0095] Step S4 aims to robustly locate the global energy peak representing the carrier velocity from the static background energy distribution spectrum generated in step S3, and calculate the first velocity estimate accordingly.

[0096] Directly searching for maximum values ​​in the static background energy distribution spectrum can be sensitive to residual local fluctuations. To enhance the robustness of the detection, the neighborhood energy sum for each velocity index is further calculated. This method essentially smooths the energy spectrum and ensures that peaks are located at the core of energy concentration regions.

[0097] For each velocity index v (v = 0, 1, ..., N) v -1), determine its neighborhood range.

[0098] In a preferred embodiment, the current velocity index v and its two adjacent indices (one to its left and one to its right), for a total of three indices, are considered. This is to handle the cyclic characteristics of the velocity spectrum (i.e., velocity indices 0 and N). v -1 is adjacent in the velocity domain), and the neighborhood index is determined using modular arithmetic:

[0099] Left Neighbor Index : ;

[0100] Right Neighbor Index : ;

[0101] The neighborhood energy of each velocity index v :

[0102] (1)

[0103] in, Indicates the static background energy distribution spectrum in the index The value at; Indicates the static background energy distribution spectrum in the index The value at; This represents the value of the static background energy distribution spectrum at index v.

[0104] This yields a new vector called the neighborhood energy and spectrum. This spectral line is smoother than the static background energy distribution spectrum, and its peak position tends to appear in the center of the energy concentration region.

[0105] In the neighborhood energy and spectrum, we search for the element with the maximum value. The velocity index corresponding to the element with the maximum value is the global peak velocity index. Converting the peak velocity index into physical velocity yields the first velocity estimate V1 for the current frame.

[0106] (2)

[0107] Global Peak Speed ​​Index It represents the location where the static background energy is most concentrated in the velocity dimension.

[0108] The conversion requires the velocity resolution ΔV of the millimeter-wave radar system, which is predetermined by the radar system configuration (such as carrier wavelength, number of frequency modulation cycles, frame time, etc.).

[0109] The physical meaning of the first velocity estimate V1 is the radial velocity of the radar carrier relative to the stationary background. Since the velocity of the static background is in the opposite direction to the velocity of the carrier, in subsequent processing, the sign can be adjusted according to the coordinate system definition, or its absolute value can be directly taken as the velocity magnitude.

[0110] Step S5: Based on the installation geometry parameters of the millimeter-wave radar, determine and extract candidate spectral units that may correspond to ground stationary reflectors from the interference suppression spectral matrix.

[0111] Step S5 aims to select spectral units from the interference suppression spectral matrix that are most likely to originate from near-field stationary ground reflectors, serving as a candidate set for subsequent fine-grained verification and velocity calculation. This method utilizes prior geometric information from radar installations to effectively narrow the processing scope and improve the efficiency and signal-to-noise ratio of subsequent steps.

[0112] Based on the actual installation geometry parameters of the millimeter-wave radar on the carrier, the radial distance range where the ground reflection point may appear in the near-field region is calculated. The core parameter is the vertical installation height H of the radar antenna phase center above the ground.

[0113] Considering the ground as a horizontal plane, for a reflecting point located on the ground at a horizontal distance D in front of radar 0, its radial (slant range) R and elevation angle to the radar are... Satisfies geometric relations:

[0114] (3)

[0115] (4)

[0116] This invention focuses on the near-field ground region, therefore defining a horizontal distance interval of interest [D]. min D max ]. Among them, D min This represents the minimum horizontal distance threshold used to exclude the area directly below the radar (which may be obstructed by mounting brackets or have complex reflections); D max This represents the maximum horizontal distance threshold, which is set based on the level of attention paid to the near-field ground area and the radar beam illumination range (e.g., 1 meter to 5 meters).

[0117] Based on the horizontal distance interval [D] min D max According to formula (3), the corresponding radial distance interval [R] can be calculated. min ,R max Radial distance interval [R] min ,Rmax This refers to the ground-based range of interest used for screening. It directly corresponds to the distance dimension of the interference suppression spectral matrix.

[0118] For each element in the interference suppression spectrum matrix, if its distance index r corresponds to a physical radial distance Rr within the radial distance interval [R... min ,R max If a cell is found to be within the specified range, it is added to the ground stationary target candidate spectral cell set. The interference suppression spectral matrix is ​​then traversed to generate the ground stationary target candidate spectral cell set. Each cell in the ground stationary target candidate spectral cell set is a candidate spectral cell that may correspond to a ground stationary reflector.

[0119] This step narrows the system's focus from the entire range-velocity spectrum to a specific radial range band that is more likely to contain effective surface stationary reflection signals, providing a high-quality data candidate set for subsequent high-precision verification based on direction of arrival.

[0120] Step S6: Estimate the direction of arrival for the candidate spectral cell to obtain the pitch angle, and verify whether it is an effective surface reflection target based on the pitch angle.

[0121] Step S6 aims to perform fine verification on the candidate spectral unit set of ground stationary targets extracted in step S5. Through high-precision direction-of-arrival estimation and geometric consistency verification, it rigorously screens out stationary reflection points that actually originate from the near-field horizontal ground in order to obtain a highly reliable second velocity estimation source.

[0122] For each candidate spectral element in the set of candidate spectral elements for stationary ground targets, two-dimensional direction-of-arrival (AOA) estimation is performed using the multi-channel antenna array signal from the millimeter-wave radar. Specifically, for the multi-channel complex signal corresponding to each candidate spectral element, a super-resolution algorithm or digital beamforming method is used to estimate the horizontal azimuth angle θ and elevation angle of the candidate spectral element relative to the radar. Among them, pitch angle Defined as the angle between the radar line-of-sight direction and the horizontal plane, it is a core parameter for subsequent geometric verification. This step outputs a set of angle estimates (θ, ...) for each candidate spectral unit. ).

[0123] To distinguish between real ground-reflecting targets and false targets at other heights (such as bridges, the top of signs, or aerial debris), this invention constructs a physical consistency verification model based on "radial distance-pitch angle-installation height".

[0124] For a stationary point on the ground, its geometric relationship satisfies the formula H = R × sin Where H is the installation height of the radar phase center above the ground, and R is the radial distance of the target measured by the radar (calculated from the cell distance index). The measured pitch angle.

[0125] Based on the measured radial distance R and pitch angle of the target Calculate the height estimation bias of the candidate spectral unit. That is, its deviation from the theoretical ground height:

[0126] (5)

[0127] Considering system measurement errors (distance, angular resolution) and minor ground undulations, a reasonable ground height tolerance threshold Δh is set (e.g., Δh = 0.3m). The decision logic is as follows:

[0128] If | If |≤Δh, then the candidate spectral unit is determined to originate from a valid ground stationary reflective target.

[0129] If | If | > Δh, then the candidate spectral unit is determined not to meet the ground geometric constraints and is excluded.

[0130] All candidate spectral units verified in step S6 and their associated information (including distance index r, velocity index v, radial distance R, and pitch angle) will be included. These targets are collected to form an effective set of surface-reflecting targets. The targets in this set have high physical reliability, confirmed as stationary points close to the horizontal ground, and their velocity information will be used for subsequent accurate calculation of the carrier's velocity. This step fundamentally ensures the purity and reliability of the second estimated path data source, avoiding systematic errors introduced by using spurious velocity values ​​from aerial or high-altitude targets.

[0131] Step S7: Based on the verified velocity information and pitch angle of the ground-reflecting target, calculate the second velocity estimate.

[0132] Step S7 aims to accurately calculate one or more independent second velocity estimates using the velocity and geometric information of each target in the valid ground reflection target set verified in step S6, providing crucial input for subsequent multi-source fusion. The core of this calculation process lies in the geometric projection correction of the radial velocity measured by radar.

[0133] For each target in the effective surface-reflecting target set, its velocity index v in the interference suppression spectrum matrix is ​​known. Calculate the radial apparent velocity v of the target relative to the radar based on the predetermined velocity resolution ΔV of the millimeter-wave radar system. radial = v × ΔV. This radial apparent velocity v radial It is the projected component of the target velocity along the radar line of sight (radial). For a stationary ground target, its actual velocity is zero, therefore the observed radial apparent velocity vradial It was entirely caused by the movement of the radar carrier itself.

[0134] Because the radar's line of sight (radial) forms an angle with the horizontal ground (i.e., elevation angle) The projection of the carrier's actual horizontal velocity onto the radar radial direction is the apparent radial velocity v. radial The two satisfy a geometric relationship:

[0135] (6)

[0136] Therefore, given the known radial apparent velocity v of the target relative to the radar... radial and pitch angle In this case, the true horizontal velocity of the carrier can be calculated according to formula (6), which is the second velocity estimate based on a single effective surface reflecting the target. Formula (6) corrects for the attenuation of the radial velocity component caused by the pitch angle, thus obtaining the velocity estimate of the carrier in the horizontal direction.

[0137] When the effective surface reflection target set contains M targets, M independent second velocity estimates will be obtained. To improve the robustness and accuracy of the estimation, the M independent second velocity estimates can be fused.

[0138] In a preferred embodiment, a robust statistical method is employed for fusion. For example, the median of M independent second velocity estimates is calculated as the final second velocity estimate. The median is insensitive to potential outliers (e.g., abnormal solutions due to angle estimation errors) and provides a more reliable estimate.

[0139] In another embodiment, the average or weighted average of M independent second velocity estimates can also be calculated (the weights can be set according to the signal-to-noise ratio or distance of each target).

[0140] Through step S7, the system calculates one or more second velocity estimates that are independent of the first estimated path (based on the global energy spectrum) and have clear physical meaning from geometrically verified ground stationary reflection points, laying a solid foundation for subsequent dual-source information fusion.

[0141] Step S8: Perform time-series filtering on the first velocity estimate and the second velocity estimate respectively to obtain the first tracking velocity and the second tracking velocity, and perform weighted fusion on the first tracking velocity and the second tracking velocity to output the final carrier velocity estimate.

[0142] Step S8 aims to perform time-domain smoothing and information fusion on the two independent velocity sources obtained in the previous steps: the first velocity estimate and the second velocity estimate, so as to finally output a stable, continuous and reliable carrier velocity estimation curve.

[0143] To suppress single-frame estimation noise and random jumps, a temporal filter based on a linear carrier motion model (such as a uniform velocity model or a low-order acceleration model) is employed. In a preferred embodiment, a Kalman filter is used. The state variables of the Kalman filter typically include velocity and its rate of change (acceleration), and its state transition equation is based on the assumption of uniform velocity or uniform acceleration. The observations of the Kalman filter are the first or second velocity estimates input. The Kalman filter is updated recursively based on model predictions and observations, outputting the optimal state estimate (i.e., the smoothed velocity).

[0144] The sequence of first velocity estimates for the current frame and multiple historical frames is used as observations and input into a constructed Kalman filter. The Kalman filter performs prediction and update iterations, outputting a smooth and continuous velocity value corresponding to the current frame, which is denoted as the first tracking velocity. This process effectively filters out high-frequency jitter between frames caused by spectral leakage, noise, or static background energy fluctuations in the first velocity estimate.

[0145] Similarly, the sequence of second velocity estimates from the current frame and multiple historical frames is used as observations and input into another identical Kalman filter. The Kalman filter outputs a smooth, continuous velocity value corresponding to the current frame, denoted as the second tracking velocity. This process filters out random variations in the second velocity estimate caused by ground target detection fluctuations, pitch angle estimation errors, or solution anomalies.

[0146] After obtaining two smoothed tracking velocities, the first and second tracking velocities are weighted and fused to generate the final carrier velocity estimate. The weighting coefficients can be dynamically allocated based on the real-time confidence level of each path.

[0147] The confidence level of the first path can be based on the peak values ​​in the static background energy distribution spectrum. The significance of the second path is determined by factors such as the ratio of the peak value to the second-highest value. The confidence of the second path can be determined based on the number of targets in the effective surface reflection target set or the consistency of the solution speed of each target (such as the inverse of variance).

[0148] When the confidence of a certain path is extremely low (such as in an open scene where the confidence of the second path may be 0), the weighted fusion automatically degenerates into a path that completely relies on the high-confidence path, thus achieving the system's adaptive fault tolerance.

[0149] Through step S8, this invention not only provides a smooth estimate in the time dimension, but also combines the advantages of the two paths of "macro background energy analysis" and "micro ground geometry verification" through an intelligent fusion strategy, and finally outputs a carrier velocity estimate that is significantly better than a single technical route in terms of accuracy, continuity and robustness.

[0150] The technical solution provided by this invention achieves significant improvements in universality, anti-interference, reliability, and stability through a systematic design, specifically manifested in the following ways:

[0151] The invention significantly enhances environmental versatility by abandoning traditional indirect methods that rely on discrete point cloud features and directly aggregates and analyzes the raw energy in the range-velocity spectrum. Through global summation and neighborhood smoothing of the velocity dimension energy, even in open scenes with weak and scattered static scattering points, the accumulated energy can form stable and significant spectral peaks for detection. This method fundamentally overcomes the dependence of point cloud methods on specific strongly reflective targets in the environment, ensuring that vehicle speed estimation remains stable and reliable even in scenarios with sparse static features such as overpasses and rural roads, thus achieving broad adaptability of the technical solution to various driving environments.

[0152] Systematic Improvement in Anti-interference Capability: Addressing the vulnerability of existing direct spectral analysis methods to strong dynamic targets and noise interference, this invention constructs a multi-layered, collaborative anti-interference mechanism. At the signal processing front end, a feature-based clustering-based discrimination method actively identifies and suppresses near-field strong dynamic targets, eliminating major interference sources at their source; simultaneously, distance-dimensional clipping isolates DC offset and near-field clutter. In the signal processing middle stage, global energy aggregation along the velocity dimension generates a static background energy distribution spectrum, and peak location is performed using neighborhood energy sums, enhancing the significance and stability of the static background energy peaks. At the signal processing back end, a second independent estimation path based on geometric verification is introduced, utilizing ground reflection to determine the second velocity estimate and eliminating non-ground target interference through physical reflection mechanisms. These measures collectively ensure that in real-world traffic scenarios with dense traffic and complex road conditions, the system can perform peak detection based on a "clean" static background energy distribution, reliably avoiding interference from dynamic targets such as vehicles in adjacent lanes, significantly improving the accuracy of velocity estimation.

[0153] System reliability is ensured through a multi-source fusion architecture: This invention innovatively constructs a dual-path parallel estimation and fusion architecture for "macroscopic static background energy spectrum analysis" and "microscopic ground reflection target geometric analysis." The two paths operate independently based on different physical principles (energy statistics and geometric projection), and their outputs can be mutually verified and used as backups. When one independent path temporarily fails due to extreme interference (such as global strong dynamic target masking or severe near-field ground occlusion), the system can automatically rely on the other path to provide an effective estimate. Subsequent dynamic weighted fusion strategy based on confidence levels adaptively integrates the dual-path information. This design endows the system with inherent fault tolerance, greatly enhancing the robustness and reliability of the overall solution.

[0154] Significantly optimized output smoothness and stability: In the information fusion stage, this invention applies temporal filtering (such as Kalman filtering) based on the carrier motion model to the velocity estimates of the two paths, effectively filtering out random noise and jumps in single-frame estimation, resulting in a smooth and continuous tracking velocity. Finally, by intelligently fusing the two smooth paths, a physically reasonable and temporally continuous final velocity curve is output. This makes the system not only superior to existing solutions in accuracy, but also achieves a qualitative improvement in output smoothness, stability, and user experience, especially demonstrating robust performance in scenarios with sparse static targets or continuous interference.

[0155] Example 2

[0156] This invention also provides an electronic device, which includes: a memory, a processor, and a computer program or instructions stored in the memory. The processor executes the computer program or instructions to implement the carrier velocity estimation method based on millimeter-wave radar in this invention.

[0157] Although not shown, the electronic device includes a processor that can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) or loaded from a storage portion into random access memory (RAM). The processor can be a multi-core processor or may contain multiple processors. In some embodiments, the processor may include a general-purpose main processor and one or more specialized coprocessors, such as a central processing unit, graphics processing unit (GPU), neural network processor (NPU), digital signal processor (DSP), etc. Various programs and data required for device operation are also stored in RAM. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0158] The processor and memory described above are used together to execute programs / instructions stored in the memory. When the program / instructions are executed by the computer, they can implement the methods, steps, or functions described in the above embodiments.

[0159] Although not shown, embodiments of the present invention also provide a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implements the carrier velocity estimation method based on millimeter-wave radar in embodiments of the present invention.

[0160] Readable storage media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0161] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A carrier velocity estimation method based on millimeter-wave radar, characterized in that, The estimation method includes: Based on the range-velocity spectrum matrix of the current frame, interference spectrum energy is identified and suppressed to obtain an interference suppression spectrum matrix; wherein, the interference spectrum energy originates at least from a dynamic target whose potential motion velocity differs from that of the carrier by more than a preset tolerance; Energy aggregation is performed along the velocity dimension on the interference suppression spectrum matrix to obtain the static background energy distribution spectrum; a first velocity estimate is determined based on the global peak value of the static background energy distribution spectrum. Based on the installation geometry parameters of the millimeter-wave radar, candidate spectral elements that may correspond to stationary ground reflectors are identified and extracted from the interference suppression spectral matrix; the direction of arrival of the candidate spectral elements is estimated to obtain the elevation angle, and the elevation angle is used to verify whether it is an effective surface reflection target; based on the verified velocity information and elevation angle of the ground reflection target, a second velocity estimate is calculated. The first velocity estimate and the second velocity estimate are subjected to time-series filtering to obtain the first tracking velocity and the second tracking velocity, respectively. The first tracking velocity and the second tracking velocity are then weighted and fused to output the final carrier velocity estimate.

2. The carrier velocity estimation method based on millimeter-wave radar according to claim 1, characterized in that, The identification and suppression of interfering spectral energy includes: The range-velocity spectrum matrix is ​​subjected to constant false alarm rate detection to obtain potential target cells; The potential target units are clustered to form at least one candidate target cluster; Calculate the cluster features of each of the candidate target clusters; Based on the cluster characteristics and interference determination conditions, interference target clusters are identified from the candidate target clusters; Set the energy of the spectral units covered by the interference target cluster to zero.

3. The carrier velocity estimation method based on millimeter-wave radar according to claim 2, characterized in that, The cluster features include the total energy, centroid distance, and centroid velocity of the candidate target cluster; The interference determination conditions include: the total energy exceeds the energy threshold, the centroid distance is less than the distance threshold, and the absolute value of the difference between the centroid velocity and the estimated carrier velocity of the previous frame is greater than the velocity threshold.

4. The carrier velocity estimation method based on millimeter-wave radar according to claim 1, characterized in that, The energy aggregation along the velocity dimension includes: For each velocity index of the interference suppression spectrum matrix, the energy values ​​of all corresponding distance indices are summed to obtain the static background energy distribution spectrum.

5. The carrier velocity estimation method based on millimeter-wave radar according to claim 1, characterized in that, Based on the global peak value of the static background energy distribution spectrum, a first velocity estimate is determined, including: Calculate the sum of neighborhood energies corresponding to each velocity index in the static background energy distribution spectrum; wherein, the sum of neighborhood energies is the sum of the energy value of the velocity index itself and the energy value of at least one adjacent velocity index; Determine the peak velocity index corresponding to the maximum value of the sum of the neighborhood energies; The first speed estimate is calculated based on the peak speed index and the preset speed resolution.

6. The carrier velocity estimation method based on millimeter-wave radar according to claim 1, characterized in that, Based on the installation geometry parameters of the millimeter-wave radar, candidate spectral units that may correspond to stationary ground reflectors are identified and extracted from the interference suppression spectral matrix, including: A distance range is determined based on the installation height of the millimeter-wave radar; From the interference suppression spectrum matrix, extract all spectral units whose radial distance falls within the distance interval, and use them as candidate spectral units.

7. The carrier velocity estimation method based on millimeter-wave radar according to claim 1, characterized in that, Verifying whether it is an effective surface reflection target based on the stated pitch angle includes: Based on the radial distance and elevation angle of the candidate spectral units, and in conjunction with the radar installation height, the corresponding height estimation deviation is calculated. If the absolute value of the height estimation deviation is less than or equal to the preset height tolerance threshold, it is determined to be an effective surface reflection target.

8. The carrier velocity estimation method based on millimeter-wave radar according to any one of claims 1 to 7, characterized in that, The first velocity estimate and the second velocity estimate are subjected to time-series filtering to obtain the first tracking velocity and the second tracking velocity, respectively, including: Based on the carrier motion model, a filter for velocity tracking is constructed; The first velocity estimate of the current frame and multiple historical frames is input into the filter for filtering and tracking to obtain a smooth and continuous first tracking velocity. The second velocity estimate of the current frame and multiple historical frames is input into the filter for filtering and tracking to obtain a smooth and continuous second tracking velocity.

9. An electronic device comprising a memory, a processor, and a computer program or instructions stored in the memory, characterized in that, The processor executes the computer program or instructions to implement the carrier velocity estimation method based on millimeter-wave radar as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by the processor, they implement the carrier velocity estimation method based on millimeter-wave radar as described in any one of claims 1 to 8.

Images