Radar dynamic target flow velocity estimation method and system based on multi-frame joint optimization

Abstract

The invention discloses a radar dynamic target flow velocity estimation method and system based on multi-frame joint optimization. The method comprises the steps that radar signal data are collected and preprocessed; filtering the preprocessed radar signal data to obtain filtered signal data; performing target detection on the filtered signal data to obtain a detection result; and performing joint estimation on the target flow velocity based on a detection result. According to the method, robust detection of low-speed and small-signal targets can be realized, the separation capability in a multi-target environment is improved, and the influence of angle errors on speed estimation is remarkably reduced, so that high-precision and low-false-alarm flow velocity estimation is realized in a complex dynamic scene.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing, and in particular to a radar dynamic target flow velocity estimation method and system based on multi-frame joint optimization. Background Art

[0002] The continuous expansion of application scenarios such as smart cities, intelligent transportation, and water resource management has placed higher demands on real-time monitoring of the motion state of fluid targets (such as vehicles, pedestrians, and water flows). Traditional flow velocity measurement technologies rely on contact sensors (such as electromagnetic flowmeters and ultrasonic Doppler flowmeters). However, radar flow velocity detection technology, with its non-contact, all-weather, and high-resolution advantages, is gradually replacing traditional contact velocity measurement methods.

[0003] Currently, radar velocity detection technology has the following drawbacks: (1) Insufficient static clutter suppression: Fixed clutter results in a low signal-to-noise ratio for low-speed target detection (the clutter floor in measured data is 10-15dB higher than the target signal). (2) Poor multi-target separation capability: Conventional CFAR (constant false alarm rate) detection has an increased false alarm rate in dense target scenarios (OS-CFAR has a false alarm rate of >5% under the same conditions). (3) Large velocity estimation error: The single-frame estimation method is significantly affected by angle ambiguity (the measured error is as high as ±0.3m / s).

[0004] Therefore, in order to solve the above problems, a radar dynamic target flow velocity estimation method and system based on multi-frame joint optimization is needed, which can achieve robust detection of low-speed and small-signal targets, improve the separation capability in multi-target environments, and significantly reduce the impact of angle error on velocity estimation, thereby achieving high-precision and low false alarm flow velocity estimation in complex dynamic scenes. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to overcome the defects in the prior art and provide a radar dynamic target flow velocity estimation method and system based on multi-frame joint optimization, which can achieve robust detection of low-speed and small-signal targets, improve the separation capability in multi-target environments, and significantly reduce the impact of angle error on velocity estimation, thereby achieving high-precision and low false alarm flow velocity estimation in complex dynamic scenes.

[0006] The radar dynamic target flow velocity estimation method based on multi-frame joint optimization of the present invention includes:

[0007] Collect radar signal data and perform preprocessing;

[0008] Performing filtering on the preprocessed radar signal data to obtain filtered signal data;

[0009] Perform target detection on the filtered signal data to obtain detection results;

[0010] Based on the detection results, the target flow velocity is jointly estimated.

[0011] Furthermore, radar signal data is collected and preprocessed, specifically including:

[0012] Equipped with multiple receiving channels, the analog-to-digital converter of each channel samples at a set rate to obtain high-precision time series signals of target echoes;

[0013] The continuously collected echo data is stored in frames, each frame includes multiple chirp signals, and the buffer will retain the data of the most recent n frames;

[0014] Each Chirp signal is weighted by a window function and then subjected to a fast Fourier transform to analyze the distance information of the target. Multiple frames of Chirp signals are weighted by a window function and then subjected to a fast Fourier transform to analyze the speed information of the target and form a range-Doppler spectrum.

[0015] Accumulate the range-Doppler spectra of multiple channels.

[0016] Furthermore, the pre-processed radar signal data is filtered, specifically including:

[0017] Perform first-order difference on the fast Fourier transform results of each receiving channel;

[0018] Perform Doppler fast Fourier transform on the differential signal to generate the range-velocity spectrum of the moving target;

[0019] Accumulate the distance-velocity spectra of moving targets from multiple channels.

[0020] Furthermore, target detection is performed on the filtered signal data, specifically including:

[0021] An ordered statistical constant false alarm rate algorithm is used to set protection units and training units. The kth largest value in the training unit is selected as the noise baseline and multiplied by the threshold factor to generate a dynamic threshold. The filtered distance-velocity spectrum is compared with the threshold to generate a binary detection result.

[0022] Furthermore, based on the detection results, the target flow rate is jointly estimated, specifically including:

[0023] Perform multi-channel phase difference analysis on each detected target, calculate the target azimuth, and obtain the direction cosine;

[0024] A weighted least squares optimization model is constructed, and regularization terms and inter-frame consistency constraints are introduced. The direction cosine is used as the horizontal axis and the velocity is used as the vertical axis to estimate the flow velocity estimation line of the target, and the slope of the flow velocity estimation line is used as the final estimated velocity.

[0025] Furthermore, the objective function of the weighted least squares optimization model is determined according to the following formula:

[0026] minimize(sum(W1 k *|Y1 k -u1*B1 k |)+sum(W2 k *|Y2 k -u2*B2 k |)+γ*|u1-u_max1|);

[0027] Among them, W1 k Y1 is the confidence weight of the kth detection point in the previous frame; k is the radial velocity of the kth detection point in the previous frame; u1 is the global flow velocity estimated in the previous frame; B1 k W2 is the azimuth cosine value of the kth detection point in the previous frame; k Y2 is the confidence weight of the kth detection point in the next frame; k is the radial velocity of the kth detection point in the next frame; u2 is the global flow velocity estimated in the next frame; B2 k is the azimuth cosine value of the kth detection point in the next frame; u_max1 is the prior estimate of the maximum flow velocity in the previous frame; γ is the regularization factor.

[0028] Furthermore, the value of the regularization factor γ is determined according to the following method:

[0029] In γ∈[10 -3 ,10 1 ] Perform logarithmic uniform sampling within the interval and calculate the change in the estimated flow velocity;

[0030] The value of γ that makes the estimation result stable is selected through the variation curve.

[0031] A radar dynamic target flow velocity estimation system based on multi-frame joint optimization, including a data acquisition module, a dynamic filtering module, an intelligent detection module and a three-dimensional perception fusion module;

[0032] The data acquisition module is used to collect radar signal data and perform preprocessing;

[0033] The dynamic filtering module is used to filter the preprocessed radar signal data to obtain filtered signal data;

[0034] The intelligent detection module is used to perform target detection on the filtered signal data to obtain a detection result;

[0035] The three-dimensional perception fusion module is used to jointly estimate the target flow velocity based on the detection results.

[0036] The beneficial effects of the present invention are as follows: a radar dynamic target flow velocity estimation method and system based on multi-frame joint optimization disclosed by the present invention effectively suppresses static clutter and background noise through differential technology and improved constant false alarm rate detection, thereby improving the weak target detection rate; a multi-frame joint optimization method that introduces weighted least squares and regularization constraints significantly enhances the accuracy and stability of speed estimation; the present invention is adaptable to a variety of measurement scenarios, has the deployment advantages of low power consumption and high real-time performance, and provides technical support for applications such as smart transportation and environmental monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0038] Figure 1 Schematic diagram of the flow chart of the radar dynamic target flow velocity estimation method of the present invention;

[0039] Figure 2 This is the original data heat map of the present invention;

[0040] Figure 3 This is the MTI filter heat map of the present invention;

[0041] Figure 4 Schematic diagram of OSCA-CFAR detection results of the present invention;

[0042] Figure 5 It is a schematic diagram of flow velocity estimation of the present invention;

[0043] Figure 6 Schematic diagram of the influence of the regularization parameters of the present invention. DETAILED DESCRIPTION

[0044] The present invention is further described below with reference to the accompanying drawings, as shown in the drawings:

[0045] This embodiment discloses a radar dynamic target flow velocity estimation method based on multi-frame joint optimization, comprising the following steps:

[0046] Step S1. Collect radar signal data and preprocess it;

[0047] Step S2: filtering the preprocessed radar signal data to obtain filtered signal data;

[0048] Step S3. Perform target detection on the filtered signal data to obtain a detection result;

[0049] Step S4: Based on the detection results, jointly estimate the target flow rate.

[0050] The present invention introduces a multi-frame joint optimization strategy, combines filtering with target detection, effectively suppresses static clutter, reduces false alarm rate, and enhances the detection capability of low-speed and small-sized targets. It adopts weighted least squares and regularization constraint methods to improve the accuracy of flow velocity estimation and reduce the error caused by angular ambiguity. At the same time, it has the ability to process multi-frame data consistency, and enhances stability and adaptability in complex dynamic environments.

[0051] In this embodiment, in step S1, radar signal data is collected and preprocessed, specifically including:

[0052] Equipped with 4 receiving channels, supporting MIMO virtual array, the analog-to-digital converter of each channel samples at a rate of 6MHz to obtain high-precision time series signals of target echoes;

[0053] The continuously collected echo data is stored in frames, each frame includes 128 chirp signals, and the buffer will retain the data of the most recent 100 frames;

[0054] Each chirp signal is weighted by a window function and then subjected to a fast Fourier transform to analyze the target's distance information; the range resolution is c / 2B = 3.79 cm;

[0055] The multi-frame Chirp signal is weighted by a window function and then subjected to a fast Fourier transform to analyze the target's velocity information and form a range-Doppler spectrum. The window function weighting can use a Hanning window; the velocity resolution is λ / 2TcNchirp = 0.11 m / s, where Tc is the Chirp period.

[0056] The range-Doppler spectra of the four channels are accumulated to improve the signal-to-noise ratio of weak targets. The readDCA1000 function can be used to read the radar's raw ADC data and parse it into a complex signal matrix for multiple receiving antennas (4 channels).

[0057] Through large bandwidth (3.96GHz) and high sampling rate (6MHz), it achieves centimeter-level range resolution, significantly superior to traditional millimeter-wave radar. Through window function weighting and dual FFT analysis, it accurately extracts target range and velocity information, with centimeter-level range resolution and 0.11m / s velocity resolution. Continuous multi-frame buffering improves the system's temporal continuity and adapts to dynamic target detection requirements. By accumulating data from multiple receiving channels, the signal-to-noise ratio is increased by 10log10(4)≈6dB, enhancing the visibility of distant targets.

[0058] like Figure 2As shown, the horizontal axis is speed (m / s), ranging from approximately -5 to 5 m / s, covering typical low-speed targets (such as pedestrians and vehicles). The vertical axis is range (m), calculated from radar parameters (0 to approximately 19.4 m, based on maxRange = numADCSamples * rangeResolution). Color mapping: Color depth represents signal strength (dB), with the dynamic range set to a maximum of -60 dB to highlight valid targets.

[0059] Zero-speed highlight band: Static objects form a continuous highlight area at a speed of 0 m / s. Discrete peak points: Moving targets appear as isolated bright spots at non-zero speeds, such as water at 2 m / s and 5 m / s. Multipath effect: Repeated peaks may appear at a fixed distance (e.g., multiple speed points at 10 m), caused by differences in signal reflection paths.

[0060] In this embodiment, in step S2, filtering is performed on the pre-processed radar signal data, specifically including:

[0061] Perform first-order difference (diff(rangeFFT,1,1)) on the fast Fourier transform results of each receiving channel to eliminate static clutter (such as the ground and walls);

[0062] Perform Doppler fast Fourier transform on the differential signal to generate the range-velocity spectrum (rangeDopplerSumMTI) of the moving target;

[0063] The range-velocity spectra of the moving targets of the four channels or antennas are accumulated to further suppress the residual noise.

[0064] A stop band can be formed near zero speed by first-order differential filtering, with an attenuation of more than 20dB (such as Figure 3 The energy in the zero-speed region is significantly reduced). The differential operation retains the Doppler shift of the moving target, making the low-speed target (0.5–2 m / s) Figure 3 Through multi-antenna fusion, false alarm points caused by multipath effects (such as Figure 3 Reduce stray points).

[0065] Specifically, if Figure 3 Middle, horizontal and vertical axis: Figure 2 The color range is the same as the overall signal strength, but the filtering compresses the overall signal strength. In the area of ​​significant change, the zero-speed band disappears: static clutter is suppressed, and the energy near zero speed is significantly reduced (the color becomes darker); moving targets are enhanced: non-zero-speed targets (such as 1.5m / s and -3m / s) are relatively brighter, and the signal-to-noise ratio is improved.

[0066] In this embodiment, in step S3, target detection is performed on the filtered signal data, specifically including:

[0067] An ordered statistical constant false alarm rate algorithm is used. Protection units (8×8) are set to prevent target energy leakage, and background noise is counted together with training units (8×8). The 20th largest value in the training unit is selected as the noise baseline and multiplied by a threshold factor (1.5) to generate a dynamic threshold. The filtered range-velocity spectrum is compared with the threshold to generate a binary detection result.

[0068] The ordered statistics constant false alarm rate algorithm reduces the false alarm rate by approximately 30% compared to traditional CA-CFAR in clutter edge and interference scenarios. A dynamic threshold factor (1.5) and training unit optimization ensure that the detection probability of weak targets (such as drones) is greater than 90%.

[0069] In addition, the algorithm complexity is O(Ndoppler×Nrange×(Ttrain+Gguard)), and real-time processing can be achieved on embedded platforms (such as TI IWR1642).

[0070] Specifically, if Figure 4 As shown, only the points exceeding the threshold are retained, which appear as isolated bright spots; its coordinate axis range is: Figure 3 The color represents the binary result (1 for detection, 0 for failure). The detection points are concentrated at the true target location (e.g., at 1.5 m / s and 10 m), with a sparse distribution consistent with physical laws. There are no false alarms in background areas (e.g., long-range and high-speed areas), demonstrating noise immunity.

[0071] In this embodiment, in step S4, the target flow rate is jointly estimated based on the detection results, specifically including:

[0072] Perform multi-channel or antenna phase difference analysis (64-point FFT) on each detection target or detection point, calculate the target azimuth θ, and obtain the direction cosine B = cosθ;

[0073] A weighted least squares optimization model is constructed, and regularization terms and inter-frame consistency constraints are introduced. The direction cosine is used as the horizontal axis and the velocity is used as the vertical axis to estimate the flow velocity estimation line of the target, and the slope of the flow velocity estimation line is used as the final estimated velocity.

[0074] Among them, the objective function in the weighted least squares optimization model is:

[0075] Minimize the weighted velocity residual ∑Wi|Yi-u·Bi|, where the weight Wi is the normalized amplitude of the detection point.

[0076] Regularization constraint: γ·(|u1-umax,1|+|u2-umax,2|) is introduced to balance data fitting with the prior flow velocity estimation (u_max is taken as the 90th percentile value). Consistency constraint: The flow velocity of the two frames is forced to be consistent (u1=u2) to eliminate inter-frame jitter.

[0077] Specifically, the objective function of the weighted least squares optimization model is determined according to the following formula:

[0078] minimize(sum(W1 k *|Y1 k -u1*B1 k |)+sum(W2 k *|Y2 k -u2*B2 k |)+γ*|u1-u_max1|);

[0079] Among them, W1 k Y1 is the confidence weight of the kth detection point in the previous frame; k is the radial velocity of the kth detection point in the previous frame; u1 is the global flow velocity estimated in the previous frame; B1 k W2 is the azimuth cosine value of the kth detection point in the previous frame; k Y2 is the confidence weight of the kth detection point in the next frame; k is the radial velocity of the kth detection point in the next frame; u2 is the global flow velocity estimated in the next frame; B2 k is the azimuth cosine value of the kth detection point in the next frame; u_max1 is the prior estimate of the maximum flow velocity in the previous frame; γ is the regularization factor.

[0080] By calculating multi-antenna phase differences, the angular resolution reaches 180° / 64≈2.8°, avoiding velocity estimation bias caused by direction cosine errors. A weighted strategy (weight Wi) allows strong targets to dominate the optimization process, suppressing noise interference and achieving an estimation error of less than 5%. Multi-frame consistency: By constraining the velocity of two frames to be consistent, the problem of unstable single-frame estimation (such as temporary target occlusion) is resolved.

[0081] like Figure 5 As shown in the sub-graphs of frame 1 and frame 2, the horizontal axis is direction cosine (cosβ), the vertical axis is velocity (m / s), and the scatter color represents the normalized weight (W). The black fitting line is the velocity estimation line obtained by weighted least squares optimization, and the slope is the final estimated velocity (u_est).

[0082] In this embodiment, the value of the regularization factor γ is determined according to the following method:

[0083] In γ∈[10 -3 ,10 1 ]The logarithmic uniform sampling is performed within the interval, and the change of the estimated flow velocity is calculated as: |uγ-uest|;

[0084] The value of γ that stabilizes the estimation result is selected through the variation curve (the default value is γ = 0.1).

[0085] like Figure 6 As shown, horizontal axis: regularization parameter γ, logarithmic scale (10-3 to 10 1 ). Vertical axis: Change in velocity estimation (m / s), indicating the deviation of the estimation result from the baseline value (γ=0.1) under different γ values. Curve trend: drastic changes at low γ and asymptotically stable at high γ. The curve shows that when γ<0.1, the model overfits the noise; when γ>1, the estimation deviates from the true value. Selecting γ=0.1 minimizes the estimation error (<0.1m / s). Through parameter impact analysis, a reproducible parameter adjustment guide is provided to avoid relying on trial and error.

[0086] The present invention also relates to a radar dynamic target flow velocity estimation system based on multi-frame joint optimization. The estimation system corresponds to the estimation method of the above embodiment and can be understood as an estimation system that implements the above estimation method. The estimation system includes a data acquisition module, a dynamic filtering module, an intelligent detection module, and a three-dimensional perception fusion module.

[0087] The data acquisition module is used to collect radar signal data and perform preprocessing;

[0088] The dynamic filtering module is used to filter the preprocessed radar signal data to obtain filtered signal data;

[0089] The intelligent detection module is used to perform target detection on the filtered signal data to obtain a detection result;

[0090] The three-dimensional perception fusion module is used to jointly estimate the target flow velocity based on the detection results.

[0091] The data acquisition module includes a DCA1000EVM data acquisition board and a TI IWR1642 millimeter-wave radar.

[0092] The dynamic filtering module includes a configurable MTI filter bank and a dual-window joint processor; the configurable MTI filter bank adopts a first-order difference design with zero padding in the first row (such as mtiFiltered = [zeros(1,...)]); the dual-window joint processor adopts a dynamic switching circuit between a Hanning window (hann(numADCSamples)) and a Kaiser window.

[0093] The intelligent detection module includes an OSCA-CFAR processor; the OSCA-CFAR processor includes a two-dimensional programmable array dynamic sorting threshold generation circuit of a sliding protection unit (guard=[8,8]) and a training unit (train=[8,8]), which supports online configuration of the os_k parameter (range 1-32).

[0094] The three-dimensional perception fusion module includes a 64-point angle FFT accelerator and a regularized optimizer; the 64-point angle FFT accelerator corresponds to numAngles=64; the regularized optimizer integrates a CVX solver hardware core (supporting γ parameter log space optimization logspace(-3,1,20)).

[0095] The present invention provides an objective and scientific radar dynamic target flow velocity estimation method and system. Based on the millimeter wave Doppler effect, the method inverts the flow velocity through phase changes. It is aerially mounted (with adjustable height) with a maximum monitoring distance of 120m. It requires no physical contact and only requires annual lens cleaning. It supports operation at -40°C to +85°C (measured signal-to-noise ratio >12dB in rainstorm / nighttime scenes). It performs real-time processing (delay of 18ms / frame) and supports sudden flow velocity monitoring. It provides technical support for bridge collision warning and mountain torrent monitoring.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A radar dynamic target flow velocity estimation method based on multi-frame joint optimization, characterized by: include: Collect radar signal data and perform preprocessing; Performing filtering on the preprocessed radar signal data to obtain filtered signal data; Perform target detection on the filtered signal data to obtain detection results; Based on the detection results, the target flow velocity is jointly estimated.

2. The radar dynamic target flow velocity estimation method based on multi-frame joint optimization according to claim 1 is characterized in that: Collect radar signal data and perform preprocessing, including: Equipped with multiple receiving channels, the analog-to-digital converter of each channel samples at a set rate to obtain high-precision time series signals of target echoes; The continuously collected echo data is stored in frames, each frame includes multiple chirp signals, and the buffer will retain the data of the most recent n frames; Each Chirp signal is weighted by a window function and then subjected to a fast Fourier transform to analyze the distance information of the target. Multiple frames of Chirp signals are weighted by a window function and then subjected to a fast Fourier transform to analyze the speed information of the target and form a range-Doppler spectrum. Accumulate the range-Doppler spectra of multiple channels.

3. The radar dynamic target flow velocity estimation method based on multi-frame joint optimization according to claim 1 is characterized in that: The pre-processed radar signal data is filtered, specifically including: Perform first-order difference on the fast Fourier transform results of each receiving channel; Perform Doppler fast Fourier transform on the differential signal to generate the range-velocity spectrum of the moving target; Accumulate the distance-velocity spectra of moving targets from multiple channels.

4. The radar dynamic target flow velocity estimation method based on multi-frame joint optimization according to claim 1 is characterized in that: Perform target detection on the filtered signal data, specifically including: An ordered statistical constant false alarm rate algorithm is used to set protection units and training units. The kth largest value in the training unit is selected as the noise baseline and multiplied by the threshold factor to generate a dynamic threshold. The filtered distance-velocity spectrum is compared with the threshold to generate a binary detection result.

5. The radar dynamic target flow velocity estimation method based on multi-frame joint optimization according to claim 1 is characterized in that: Based on the detection results, the target flow rate is jointly estimated, including: Perform multi-channel phase difference analysis on each detected target, calculate the target azimuth, and obtain the direction cosine; A weighted least squares optimization model is constructed, and regularization terms and inter-frame consistency constraints are introduced. The direction cosine is used as the horizontal axis and the velocity is used as the vertical axis to estimate the flow velocity estimation line of the target, and the slope of the flow velocity estimation line is used as the final estimated velocity.

6. The radar dynamic target flow velocity estimation method based on multi-frame joint optimization according to claim 5 is characterized in that: The objective function of the weighted least squares optimization model is determined as follows: minimize(sum(W1 k *|Y1 k -u1*B1 k |)+sum(W2 k *|Y2 k -u2*B2 k |)+γ*|u1-u_max1|); Among them, W1 k Y1 is the confidence weight of the kth detection point in the previous frame; k is the radial velocity of the kth detection point in the previous frame; u1 is the global flow velocity estimated in the previous frame; B1 k W2 is the azimuth cosine value of the kth detection point in the previous frame; k Y2 is the confidence weight of the kth detection point in the next frame; k is the radial velocity of the kth detection point in the next frame; u2 is the global flow velocity estimated in the next frame; B2 k is the azimuth cosine value of the kth detection point in the next frame; u_max1 is the prior estimate of the maximum flow velocity in the previous frame; γ is the regularization factor.

7. The radar dynamic target flow velocity estimation method based on multi-frame joint optimization according to claim 6 is characterized in that: The value of the regularization factor γ is determined as follows: In γ∈[10 -3 ,10 1 ] Perform logarithmic uniform sampling within the interval and calculate the change in the estimated flow velocity; The value of γ that makes the estimation result stable is selected through the variation curve.

8. A radar dynamic target velocity estimation system based on multi-frame joint optimization, characterized by: Including data acquisition module, dynamic filtering module, intelligent detection module and three-dimensional perception fusion module; The data acquisition module is used to collect radar signal data and perform preprocessing; The dynamic filtering module is used to filter the preprocessed radar signal data to obtain filtered signal data; The intelligent detection module is used to perform target detection on the filtered signal data to obtain a detection result; The three-dimensional perception fusion module is used to jointly estimate the target flow velocity based on the detection results.

Images