Vehicle-mounted human body heart rate monitoring method, system and equipment based on UWB radar and medium
By combining UWB radar with mean filtering, frequency band filtering, and VMD algorithm to separate heartbeat signals, the problem of vehicle bump noise interference is solved, and high-precision heart rate monitoring is achieved in complex vehicle environments.
Patent Information
- Application Number
- CN202511929068.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-03-03
AI Technical Summary
Traditional UWB physiological monitoring technology has difficulty effectively suppressing noise interference in bumpy and noisy vehicle environments, and long-term signal integration leads to real-time defects, making it impossible to achieve accurate heart rate monitoring in highly dynamic driving behaviors.
Signal preprocessing is performed using mean filtering and frequency band filtering. The intrinsic mode function (IMF) components related to heart rate are separated by variational mode decomposition (VMD) algorithm. Heart rate estimation is then performed by feature splicing and frequency domain adaptive filtering with a multilayer perceptron.
It significantly improves the accuracy and robustness of heart rate estimation, enabling reliable heart rate monitoring in complex in-vehicle scenarios and reducing errors.
Smart Images
Figure CN121587700A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of UWB physiological monitoring in vehicle environments, specifically to a method, system, device, and medium for vehicle-mounted human heart rate monitoring based on UWB radar. Background Technology
[0002] In recent years, breakthroughs in UWB radar technology have opened up entirely new possibilities for the field of physiological monitoring. Its high temporal resolution and non-invasive nature make it an ideal tool for analyzing vital signs in dynamic scenarios. In the field of intelligent vehicles, UWB radar, by emitting broadband pulses and capturing Doppler shift signals caused by subtle human movements, can detect physiological activities such as breathing and heartbeat in real time, making it particularly suitable for applications in complex in-vehicle environments. For example, by analyzing millimeter-level changes in chest displacement, UWB can extract heartbeat waveform features; by inverting the amplitude of chest cavity fluctuations using signal phase shift, respiratory rate can be calculated. This contactless monitoring method not only avoids the discomfort of wearable devices but also overcomes the performance limitations of cameras in low-light or obstructed environments.
[0003] However, existing UWB physiological monitoring technologies still face significant challenges in real-world driving scenarios. First, the vibration noise generated by vehicle movement is highly coupled with the frequency bands of physiological signals. Body swaying caused by engine vibration, road bumps, etc., produces interference with amplitudes far exceeding those of biological micro-movements (typically less than 1 mm). Traditional fixed-threshold filtering algorithms struggle to distinguish these interferences from the characteristics of real physiological signals, leading to an increased false detection rate. For example, low-frequency mechanical vibrations (0.5-5 Hz) caused by changes in vehicle speed may completely cover the dominant frequency range of respiratory signals (0.1-0.5 Hz). Second, existing systems rely on long-time signal integration (typically ≥10 seconds) to achieve high-precision measurements. However, in high-dynamic driving behaviors such as emergency braking and lane changing, sudden and dramatic changes in physiological state (such as a sudden increase in heart rate) may be averaged out due to the excessively long time window, severely reducing real-time warning capabilities.
[0004] While some improvements have been attempted to address the aforementioned issues, their effectiveness remains limited. For instance, noise reduction methods based on Empirical Mode Decomposition (EMD) can suppress specific broadband noise, but they are overly sensitive to higher-frequency noise and amplitude abrupt changes caused by bumps, failing to effectively separate noise in driving environments. Since the signal frequency domain resolution varies with the sampling window length, traditional frequency domain analysis methods based on Fourier transform and wavelet decomposition cannot generate accurate physiological monitoring data when the time window is small. Furthermore, existing research largely focuses on static or low-speed simulation tests under laboratory conditions, lacking sufficient robustness verification for real-world high-speed bumpy scenarios, leading to algorithm failure in real-world driving environments.
[0005] Therefore, there is an urgent need to propose a method for monitoring human heart rate to solve the above problems. Summary of the Invention
[0006] The technical problem to be solved by this invention is how to solve the real-time defects caused by insufficient suppression of bump noise and long-term signal integration dependence in traditional solutions.
[0007] This invention solves the above-mentioned technical problems through the following technical means: a vehicle-mounted human heart rate monitoring method based on UWB radar, comprising: S1. Acquire signals through UWB radar and form an original signal matrix. Then, perform noise reduction processing through mean filtering and first bandpass filtering. Extract human body reflection information vectors based on the noise-reduced signal matrix. S2. Perform second-band bandpass filtering, detrending processing and moving average filtering on the human body reflection information vector in sequence, and use the variational mode decomposition (VMD) algorithm to decompose the phase of the human body reflection signal vector to obtain IMF components. Calculate the center frequency of each IMF component and select the IMF components within the preset frequency range and preset conditions. S3. The amplitude and phase of the signal vector after the moving average filtering in step S2 are concatenated in the channel dimension to form feature matrix F1, and the filtered IMF components are concatenated in the channel dimension to form feature matrix F2. F1 and F2 are input into the block embedding module to obtain the embedded representation of the features, which are then processed by the channel interactive shuffling module and the interactive fusion module in sequence. After the fusion result is input into the frequency domain adaptive filtering module to complete the frequency domain processing, it is input into the multilayer perceptron to output the estimated heart rate values of the driver and passengers.
[0008] Furthermore, the calculation formula for the mean filter is as follows:
[0009] Where n is the number of frames contained in the signal matrix. , These represent the nth elements in the original signal matrix. OK Column, No. OK Column elements, This represents the element in the p-th row and q-th column of the signal matrix obtained after mean filtering.
[0010] Furthermore, the calculation formula for extracting the human body reflection information vector based on the denoised signal matrix is as follows:
[0011] Where r is the human body reflection information vector, and n is the number of frames contained in the signal matrix. Let t be the signal vector in the 0th column of the matrix after bandpass filtering in the first frequency band, and t be the row index of the element in the signal vector, with a value from 0 to n.
[0012] Furthermore, the calculation formulas for the detrending processing and moving average filtering are as follows:
[0013]
[0014] in, This is the signal vector after detrending r. To The signal vector obtained after performing a moving average filter. These are the low-frequency approximation coefficients obtained from wavelet decomposition, M is the window size selected in the moving average filtering, and t is the column index of the element in the signal vector. This is the signal vector after bandpass filtering in the second frequency band. This is the index of the time step within the sliding window.
[0015] Furthermore, when using the Variational Mode Decomposition (VMD) algorithm, the preset number of modes K=8, and the penalty factor... . Furthermore, the formula for calculating the center frequency of each IMF component is as follows:
[0016]
[0017]
[0018] Where imf represents the modal component, t is the vector index of the modal component IMF, X is the frequency domain signal, and k is the index of the calculated frequency domain signal. This represents the actual frequency of the k-th frequency domain signal. Indicates the sampling rate of the signal. This represents the final center frequency, where j is the imaginary unit.
[0019] Furthermore, the filtering of IMF components within a preset frequency range and under preset conditions specifically includes: Calculate the center frequency of each modal component and remove modal components whose center frequency is not in the range of 0.8Hz to 2.0Hz; If the number of remaining modal components is greater than 3, the kurtosis of each modal component is calculated, modal components with kurtosis greater than 6 are removed, and the three modal components with the smallest kurtosis are selected as target modal components.
[0020] This invention also provides a vehicle-mounted human heart rate monitoring system based on UWB radar, comprising: Signal acquisition and preprocessing module: used to acquire signals through UWB radar and form an original signal matrix, and to perform noise reduction processing through mean filtering and first bandpass filtering, and to extract human body reflection information vector based on the noise-reduced signal matrix; The reflection information depth processing module is used to sequentially perform second-band bandpass filtering, detrending processing, and moving average filtering on the human body reflection information vector, and use the variational mode decomposition (VMD) algorithm to decompose the phase of the human body reflection signal vector to obtain IMF components, calculate the center frequency of each IMF component, and filter out the IMF components within the preset frequency range and preset conditions. Feature fusion and heart rate estimation module: This module concatenates the amplitude and phase of the signal vector after moving average filtering in the multimodal feature extraction module along the channel dimension to form feature matrix F1, and concatenates the filtered IMF components along the channel dimension to form feature matrix F2. F1 and F2 are then input into the block embedding module to obtain the embedded representation of the features, which is then processed sequentially by the channel interactive shuffling module and the interactive fusion module. The fusion result is then input into the frequency domain adaptive filtering module to complete frequency domain processing, and finally input into the multilayer perceptron to output the heart rate estimates for the driver and passengers.
[0021] The present invention also provides a processing device, including at least one processor and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor can execute the above-described method steps by calling the program instructions.
[0022] The present invention also provides a computer-readable storage medium storing computer instructions that cause the computer to perform the above-described method steps.
[0023] The advantages of this invention are: This invention employs mean filtering and bandpass filtering for signal preprocessing, effectively suppressing environmental noise and multipath reflection interference, providing a high-quality signal foundation for subsequent analysis. Secondly, by using variational mode decomposition (VMD) with preset mode numbers and penalty factors, it adaptively separates the intrinsic mode function (IMF) components related to heartbeat, effectively solving the problem of weak heartbeat signals easily masked by respiratory harmonics and subtle body movements. Finally, the signal amplitude and phase features are concatenated into an F1 matrix, and the filtered IMF components are concatenated into an F2 matrix. Feature enhancement processing is then performed through a block embedding module, a channel interactive shuffling module, and an interactive fusion module. Combined with frequency domain adaptive filtering and multilayer perceptron classification, feature discriminative power is dynamically optimized. This significantly improves the accuracy and robustness of heart rate estimation under complex conditions such as vehicle vibration. The entire methodology works synergistically to ultimately achieve reliable heart rate monitoring in in-vehicle scenarios. Attached Figure Description
[0024] Figure 1 This is a flowchart of the vehicle-mounted human heart rate monitoring method based on UWB radar according to Embodiment 1 of the present invention; Figure 2 This is a diagram of the reasoning model for Embodiment 1 of the present invention; Figure 3 This is a graph showing the algorithm performance evaluation of Embodiment 1 of the present invention. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, 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.
[0026] Before describing the technical solutions provided in this application, we will first explain some of the technical terms involved in this application.
[0027] UWB radar is a wireless sensing technology based on extremely short pulses (nanosecond-level narrow pulses). Its signal spectrum ranges from hundreds of megahertz to several gigahertz, exhibiting high time resolution and strong resistance to multipath interference. UWB signals, by transmitting wideband short pulses, can penetrate non-metallic obstacles (such as fabrics and plastics) and simultaneously capture high-frequency phase changes caused by subtle target movements. This invention utilizes the sensitivity of UWB radar to sub-millimeter-level chest wall undulations, analyzing the time delay and phase differences of reflected pulses to achieve the interpretation and monitoring of physiological signals.
[0028] The IQ plane is a two-dimensional coordinate system describing the complex values of radar baseband signals. The horizontal axis represents the in-phase component (I, real part), and the vertical axis represents the quadrature components. UWB radar downconverts the received radio frequency signal to baseband through quadrature demodulation, decomposing it into two quadrature components, I and Q, while retaining the amplitude, phase, and frequency information of the signal. The periodic chest wall displacement caused by the human heartbeat modulates the IQ phase difference of the reflected signal, manifesting as a spiral trajectory change on the IQ plane (such as a periodic shift in phase angle).
[0029] The Fourier transform is a mathematical tool that decomposes a time-domain signal into integrals of its frequency-domain sine / cosine components. Its discrete form (DFT) is used in digital signal processing, while the fast algorithm (FFT) significantly improves computational efficiency. In traditional methods, the Fourier transform converts the time-domain micro-motion signal captured by UWB radar into a frequency-domain energy distribution, and heart rate estimation is achieved by detecting spectral peaks (such as heart rate of 1-3 Hz).
[0030] Wavelet decomposition is a time-frequency analysis tool that uses the scaling and translation of the mother wavelet to perform multi-resolution decomposition of signals, taking into account the ability to extract local features in both the time and frequency domains. Wavelet decomposition can overcome the limitation of the Fourier transform on the assumption of global stationarity, making it suitable for signal analysis in dynamic environments such as vehicle bumps and sudden noise.
[0031] Variational mode decomposition is an adaptive signal decomposition method that decomposes complex signals into multiple essential mode functions (IMFs) through constrained optimization. Each IMF has a specific center frequency and bandwidth, avoiding the mode aliasing defects of traditional empirical mode decomposition (EMD).
[0032] Principal component analysis is a linear dimensionality reduction technique that uses methods such as calculating the data covariance matrix and eigenvalue decomposition to achieve orthogonal transformation of the original data, projecting the original high-dimensional data into a low-dimensional space, retaining the principal components with the largest variance, and eliminating redundant information.
[0033] Example 1 like Figure 1 The diagram shows a flowchart of an in-vehicle human heart rate monitoring method based on UWB radar. A commercial UWB radar (XETHRU X4M05) is fixed to the center of the inside of the vehicle's windshield, enabling it to capture weak signals from multiple body parts. The radar's transmit pulse width is 0.4 ns, center frequency is 7.3 GHz, bandwidth is 1.4 GHz, and sampling frequency is 23.328 GHz. The program sets its frame rate to 20 frames per second. The UWB radar is connected to a PC via the XEP MATLAB Connector. The PC starts the UWB radar and receives signal frames in real time via I / O commands.
[0034] UWB radar-based methods for monitoring human heart rate in vehicles include: S1. Signals are acquired by UWB radar and formed into an original signal matrix. Noise reduction is performed by mean filtering and first-band bandpass filtering. Human body reflection information vectors are extracted based on the noise-reduced signal matrix.
[0035] Specifically, the UWB radar collects all signal frames acquired within the past 5 seconds and assembles them into an original signal matrix in chronological order. The mean filter for the original signal matrix is calculated using the following formula:
[0036] Where n is the number of frames contained in the signal matrix. , These represent the nth elements in the original signal matrix. OK Column, No. OK Column elements, This represents the element in the p-th row and q-th column of the signal matrix obtained after mean filtering.
[0037] right After performing bandpass filtering noise reduction from 0.1Hz to 0.8Hz, the matrix is obtained. All filters can be implemented using built-in Matlab library functions.
[0038] The formula for extracting the human body reflection information vector based on the denoised signal matrix is as follows:
[0039] Where r is the human body reflection information vector, and n is the number of frames contained in the signal matrix. Let t be the signal vector in the 0th column of the matrix after bandpass filtering in the first frequency band, and t be the row index of the element in the signal vector, with a value from 0 to n.
[0040] S2. Perform second-band bandpass filtering, detrending processing, and moving average filtering on the human body reflection information vector in sequence, and use the variational mode decomposition (VMD) algorithm to decompose the phase of the human body reflection signal vector to obtain IMF components. Calculate the center frequency of each IMF component and select the IMF components within the preset frequency range and preset conditions.
[0041] Specifically, the human body reflex information vector r is processed using a 0.8Hz-2.0Hz bandpass filter to remove respiration-related waveforms and high-frequency noise. The processed signal vector is denoted as... .
[0042] The signal vector after second bandpass filtering Detrending and moving average filtering are performed, and the signal vector after moving average filtering is denoted as... The calculation formulas for detrending processing and moving average filtering are as follows:
[0043]
[0044] in, This is the signal vector after detrending r. To The signal vector obtained after performing a moving average filter. These are the low-frequency approximation coefficients obtained from wavelet decomposition, M is the window size selected in the moving average filtering, and t is the column index of the element in the signal vector. This is the signal vector after bandpass filtering in the second frequency band. This is the index of the time step within the sliding window.
[0045] The phase component of the human body reflectance information vector r is decomposed using the variational mode decomposition (VMD) algorithm. The parameters are set as follows: number of modes. Punishment factor The center frequency of each IMF is calculated using FFT, and IMFs with center frequencies outside the range of 0.8Hz-2Hz are removed. The formula for calculating the center frequency is as follows:
[0046]
[0047]
[0048] Where imf represents the modal component, t is the vector index of the modal component IMF, X is the frequency domain signal, and k is the index of the calculated frequency domain signal. This represents the actual frequency of the k-th frequency domain signal. Indicates the sampling rate of the signal. This represents the final center frequency, where j is the imaginary unit.
[0049] If there are more than three remaining IMFs, calculate the kurtosis of all IMFs and retain the three IMFs with the lowest kurtosis. The formula for calculating kurtosis is as follows:
[0050] in, For mathematical expectation, For modal components, The mean of the modal components. denoted as the standard deviation of the modal components.
[0051] S3. The amplitude and phase of the signal vector after the moving average filtering in step S2 are concatenated in the channel dimension to form feature matrix F1, and the filtered IMF components are concatenated in the channel dimension to form feature matrix F2. F1 and F2 are input into the block embedding module to obtain the embedded representation of the features, which are then processed by the channel interactive shuffling module and the interactive fusion module in sequence. After the fusion result is input into the frequency domain adaptive filtering module to complete the frequency domain processing, it is input into the multilayer perceptron to output the estimated heart rate values of the driver and passengers.
[0052] Specifically, such as Figure 2 As shown, the signal vector after moving average filtering is... The phase and amplitude components are spliced together in the channel dimension to form a single... Feature matrix Simultaneously, the three selected IMF components are concatenated along the channel dimension to obtain a shape of... Feature matrix .
[0053] Will and The data are input into the block embedding module. The main idea of the block embedding module is to divide the long time series into several shorter time blocks, extract the temporal features of each block, and then summarize them into an overall embedding representation. In this module, each channel is divided into 5 parts in time, and each part is input into a Transformer layer (head number = 4, hidden layer dimension = 128) to obtain the embedding representation of the corresponding features.
[0054] The embedded representations of the corresponding features are input into the channel interaction shuffling module. This module integrates the ability to enhance local temporal features and interact across modalities, enabling simultaneous modeling in both the temporal and channel dimensions. First, the input embedded representations are subjected to depthwise separable convolutions in the temporal dimension to capture local temporal dependencies and expand the receptive field; this is the DWConv part, which includes a GELU activation function, batch normalization layers, and depthwise separable convolutional layers with kernel size k=15 and stride=1, used to extract local temporal patterns. To enhance intermodal and interchannel interactions, the channel interaction shuffling module introduces two layers of convolutional feedforward networks (ConvFFN). ConvFFN1 uses grouped convolutions to model intramodal features, with each modality corresponding to a convolutional group; ConvFFN2 uses pointwise convolutions to fuse features from different modalities, achieving cross-modal interaction. Finally, the outputs of ConvFFN2 and DWConv are concatenated in the channel dimension to obtain the output of the channel interaction shuffling module.
[0055] The median and variance of the kurtosis of all effective modal components are calculated and input into the learnable gating unit in the interactive fusion module. The gating unit consists of two linear layers and an activation function, and outputs a weight scalar, which is used to perform weighted addition fusion of the outputs of the previously obtained two-channel interactive shuffling module to obtain the output of the module.
[0056] The output of the interactive fusion module is input into the frequency domain adaptive filtering module. This module first performs a Fourier transform on the input to obtain the frequency domain signal, which is then fed into a linear layer. This linear layer outputs a threshold, which is used to implement a low-pass filter that only allows frequencies below the threshold to pass through, filtering the previously obtained frequency domain signal. Finally, the frequency domain signal undergoes an inverse Fourier transform to obtain its time domain representation, and the result is input into the multilayer perceptron to obtain the model's inference result, namely the driver's and passenger's heart rate information.
[0057] like Figure 3As shown in the figure, this demonstrates the superior performance of the implemented UWB radar-based vehicle-mounted human heart rate monitoring method TCN compared to traditional methods. The final implemented method achieves an error within 5 bpm for 90% of the samples, while other methods such as FFT, VMD, and EMD perform poorly, with errors within 20 bpm for 90% of the samples.
[0058] Example 2 Based on Embodiment 1, Embodiment 2 of the present invention also provides an in-vehicle human heart rate monitoring system based on UWB radar, including: Signal acquisition and preprocessing module: used to acquire signals through UWB radar and form an original signal matrix, and to perform noise reduction processing through mean filtering and first bandpass filtering, and to extract human body reflection information vector based on the noise-reduced signal matrix.
[0059] Specifically, the signal acquisition unit is used to fix a commercial UWB radar (XETHRU X4M05) in the center of the inside of the vehicle's windshield, enabling it to capture weak signals from multiple body parts.
[0060] The radar's transmit pulse width is 0.4 ns, center frequency is 7.3 GHz, bandwidth is 1.4 GHz, and sampling frequency is 23.328 GHz. The program sets its frame rate to 20 frames per second. The UWB radar is connected to a PC via a module, and the PC starts the UWB radar and receives signal frames in real time via I / O commands. Because the received signal frames are affected by multipath effects and reflections from static objects in the background, mean filtering is first used to remove background noise from the signal matrix.
[0061] Signal preprocessing unit: used to perform noise reduction processing on the original signal through mean filtering and first bandpass filtering.
[0062] The formula for calculating mean filtering is:
[0063] Where n is the number of frames contained in the signal matrix. , These represent the nth elements in the original signal matrix. OK Column, No. OK Column elements, This represents the element in the p-th row and q-th column of the signal matrix obtained after mean filtering. After noise removal, a signal matrix is obtained that stores dynamic information of the scene. In the current driving scenario, the centimeter-level movement of the chest cavity during breathing is reflected in the signal matrix, manifested as the periodic movement of the signal sampling points in the column containing the human body on the IQ plane. Due to the previous mean filtering, the changes in other columns of the matrix are not significant. The energy of each column in the signal matrix is calculated, and the column with the highest energy corresponds to the location of the human body. The system saves the column number and the corresponding signal vector for further processing.
[0064] Human body positioning unit: used to extract human body reflection information vectors based on the noise-reduced signal matrix.
[0065] The formula for extracting the human body reflection information vector based on the noise-reduced signal matrix is as follows:
[0066] Where r is the human body reflection information vector, and n is the number of frames contained in the signal matrix. Let t be the signal vector in the 0th column of the matrix after bandpass filtering in the first frequency band, and t be the row index of the element in the signal vector, with a value from 0 to n.
[0067] The deep processing module for reflection information is used to sequentially perform second-band bandpass filtering, detrending processing, and moving average filtering on the human body reflection information vector. It also uses the variational mode decomposition (VMD) algorithm to decompose the phase of the human body reflection signal vector to obtain IMF components, calculates the center frequency of each IMF component, and selects IMF components within the preset frequency range and preset conditions.
[0068] Specifically, the signal smoothing unit is used to sequentially perform second-band bandpass filtering, detrending processing, and moving average filtering on the human body reflection information vector.
[0069] The calculation formulas for detrending processing and moving average filtering are as follows:
[0070]
[0071] in, This is the signal vector after detrending r. To The signal vector obtained after performing a moving average filter. These are the low-frequency approximation coefficients obtained from wavelet decomposition, M is the window size selected in the moving average filtering, and t is the column index of the element in the signal vector. This is the signal vector after bandpass filtering in the second frequency band. This is the index of the time step within the sliding window.
[0072] VMD Mode Decomposition and Filtering Unit: Used to decompose the phase of the human body reflection signal vector using the Variational Mode Decomposition (VMD) algorithm to obtain IMF components, calculate the center frequency of each IMF component, and filter out IMF components within a preset frequency range and under preset conditions. When using the Variational Mode Decomposition (VMD) algorithm, the preset number of modes K=8, and the penalty factor... .
[0073] The formula for calculating the center frequency of each IMF component is as follows:
[0074]
[0075]
[0076] Where imf represents the modal component, t is the vector index of the modal component IMF, X is the frequency domain signal, and k is the index of the calculated frequency domain signal. This represents the actual frequency of the k-th frequency domain signal. Indicates the sampling rate of the signal. This represents the final center frequency, where j is the imaginary unit.
[0077] The IMF components within a preset frequency range and under preset conditions are selected as follows: Calculate the center frequency of each modal component and remove modal components whose center frequency is not in the range of 0.8Hz to 2.0Hz; If the number of remaining modal components is greater than 3, the kurtosis of each modal component is calculated, modal components with kurtosis greater than 6 are removed, and the three modal components with the smallest kurtosis are selected as target modal components.
[0078] Feature fusion and heart rate estimation module: This module concatenates the amplitude and phase of the signal vector after moving average filtering in the multimodal feature extraction module along the channel dimension to form feature matrix F1, and concatenates the filtered IMF components along the channel dimension to form feature matrix F2. F1 and F2 are then input into the block embedding module to obtain the embedded representation of the features, which is then processed sequentially by the channel interactive shuffling module and the interactive fusion module. The fusion result is then input into the frequency domain adaptive filtering module to complete frequency domain processing, and finally input into the multilayer perceptron to output the heart rate estimates for the driver and passengers.
[0079] Example 3 Based on Embodiment 1, Embodiment 3 of the present invention also provides a processing device, including at least one processor and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor can execute the method steps of Embodiment 1 by calling the program instructions.
[0080] Example 4 Based on Embodiment 1, Embodiment 4 of the present invention also provides a computer-readable storage medium storing computer instructions that cause the computer to perform the steps of the method described in Embodiment 1.
[0081] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for monitoring human heart rate in a vehicle based on UWB radar, characterized in that, include: S1. Acquire signals through UWB radar and form an original signal matrix. Then, perform noise reduction processing through mean filtering and first bandpass filtering. Extract human body reflection information vectors based on the noise-reduced signal matrix. S2. Perform second-band bandpass filtering, detrending processing and moving average filtering on the human body reflection information vector in sequence, and use the variational mode decomposition (VMD) algorithm to decompose the phase of the human body reflection signal vector to obtain IMF components. Calculate the center frequency of each IMF component and select the IMF components within the preset frequency range and preset conditions. S3. The amplitude and phase of the signal vector after the moving average filtering in step S2 are concatenated in the channel dimension to form feature matrix F1, and the filtered IMF components are concatenated in the channel dimension to form feature matrix F2. F1 and F2 are input into the block embedding module to obtain the embedded representation of the features, which are then processed by the channel interactive shuffling module and the interactive fusion module in sequence. After the fusion result is input into the frequency domain adaptive filtering module to complete the frequency domain processing, it is input into the multilayer perceptron to output the estimated heart rate values of the driver and passengers.
2. The vehicle-mounted human heart rate monitoring method based on UWB radar according to claim 1, characterized in that, The formula for calculating the mean filter is: Where n is the number of frames contained in the signal matrix. , These represent the nth elements in the original signal matrix. OK Column, No. OK Column elements, This represents the element in the p-th row and q-th column of the signal matrix obtained after mean filtering.
3. The vehicle-mounted human heart rate monitoring method based on UWB radar according to claim 1, characterized in that, The calculation formula for extracting the human body reflection information vector based on the denoised signal matrix is as follows: Where r is the human body reflection information vector, and n is the number of frames contained in the signal matrix. Let t be the signal vector in the 0th column of the matrix after bandpass filtering in the first frequency band, and t be the row index of the element in the signal vector, with a value from 0 to n.
4. The vehicle-mounted human heart rate monitoring method based on UWB radar according to claim 1, characterized in that, The calculation formulas for the detrending process and the moving average filter are as follows: in, This is the signal vector after detrending r. To The signal vector obtained after performing a moving average filter. These are the low-frequency approximation coefficients obtained from wavelet decomposition, M is the window size selected in the moving average filtering, and t is the column index of the element in the signal vector. This is the signal vector after being processed by the second frequency band bandpass filter. This is the index of the time step within the sliding window.
5. The vehicle-mounted human heart rate monitoring method based on UWB radar according to claim 1, characterized in that, When using the Variational Mode Decomposition (VMD) algorithm, the preset mode number K=8, and the penalty factor... .
6. The vehicle-mounted human heart rate monitoring method based on UWB radar according to claim 1, characterized in that, The formula for calculating the center frequency of each IMF component is as follows: Where imf represents the modal component, t is the vector index of the modal component IMF, X is the frequency domain signal, and k is the index of the calculated frequency domain signal. This represents the actual frequency of the k-th frequency domain signal. Indicates the sampling rate of the signal. This represents the final center frequency, where j is the imaginary unit.
7. The vehicle-mounted human heart rate monitoring method based on UWB radar according to claim 1, characterized in that, The filtering of IMF components within a preset frequency range and under preset conditions specifically involves: Calculate the center frequency of each modal component and remove modal components whose center frequency is not in the range of 0.8Hz to 2.0Hz; If the number of remaining modal components is greater than 3, the kurtosis of each modal component is calculated, modal components with kurtosis greater than 6 are removed, and the three modal components with the smallest kurtosis are selected as target modal components.
8. A vehicle-mounted human heart rate monitoring system based on UWB radar, characterized in that, include: Signal acquisition and preprocessing module: used to acquire signals through UWB radar and form an original signal matrix, and to perform noise reduction processing through mean filtering and first bandpass filtering, and to extract human body reflection information vector based on the noise-reduced signal matrix; The reflection information depth processing module is used to sequentially perform second-band bandpass filtering, detrending processing, and moving average filtering on the human body reflection information vector, and use the variational mode decomposition (VMD) algorithm to decompose the phase of the human body reflection signal vector to obtain IMF components, calculate the center frequency of each IMF component, and filter out the IMF components within the preset frequency range and preset conditions. Feature fusion and heart rate estimation module: This module concatenates the amplitude and phase of the signal vector after moving average filtering in the multimodal feature extraction module along the channel dimension to form feature matrix F1, and concatenates the filtered IMF components along the channel dimension to form feature matrix F2. F1 and F2 are then input into the block embedding module to obtain the embedded representation of the features, which is then processed sequentially by the channel interactive shuffling module and the interactive fusion module. The fusion result is then input into the frequency domain adaptive filtering module to complete frequency domain processing, and finally input into the multilayer perceptron to output the heart rate estimates for the driver and passengers.
9. A processing device, characterized in that, The method includes at least one processor and at least one memory communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor can execute the method as described in any one of claims 1 to 7 by invoking the program instructions.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause the computer to perform the method as described in any one of claims 1 to 7.