A portable fatigue driving monitoring and early warning system

Abstract

The application discloses a portable fatigue driving monitoring and early warning system and relates to the technical field of fatigue early warning. The system comprises a collecting module, a control module, a reminding module and a power module. The input end of the control module is connected with the output end of the collecting module, and the output end of the control module is connected with the reminding module. The application determines and intervenes the fatigue state by collecting and analyzing the frontal lobe single-channel electroencephalogram signals of the user based on an integrated independent device, and the process does not need to depend on intelligent terminal devices such as intelligent vehicles, smart phones or cloud servers, thereby reducing power consumption and system complexity. The application proposes a non-electroencephalogram signal component identification and elimination method based on independent component energy spectrum, spectrum entropy and correlation analysis, effectively suppresses the interference of artifacts such as eye movement and electromyography, and improves the accuracy and reliability of fatigue monitoring of the portable device.

Description

Technical Field

[0001] This invention relates to the field of fatigue warning technology, and in particular to a portable fatigue driving monitoring and warning system. Background Technology

[0002] Fatigue can cause drivers to have difficulty concentrating, longer reaction time, weakened judgment, and operational errors, thereby inducing traffic accidents and posing a great threat to the personal safety of drivers, passengers, and other road users.

[0003] Current methods for monitoring driver fatigue include monitoring the driver's behavior, facial and eye features, various physiological signals, and vehicle operation. These methods can be used independently or in combination. Monitoring various physiological signals, such as electroencephalogram (EEG), electrooculogram (EOG), heart rate, and skin conductance, is achieved through sensors installed in the steering wheel, driver's seat, or on the driver's body. Signal processing techniques are used to extract the characteristics of these physiological signals, reflecting the driver's fatigue state. When driver fatigue is detected, warnings and interventions are initiated through vehicle control devices such as speakers. Alternatively, data analysis and warnings can be performed using smartphone processors and cloud servers.

[0004] While existing technologies can monitor fatigue, their algorithms lack robustness. Specifically, existing technologies are still insufficient in dealing with non-EEG signal interference (such as artifacts from eye movements, blinking, and muscle activity) commonly found in real driving environments. In particular, when only a small number of prefrontal channels are used, these interferences can seriously affect the accuracy of fatigue determination. Summary of the Invention

[0005] Therefore, it is necessary to provide a portable fatigue driving monitoring and early warning system to address the aforementioned technical problems.

[0006] This invention provides a portable fatigue driving monitoring and early warning system, comprising:

[0007] The acquisition module is used to acquire electroencephalogram (EEG) signals from the user's prefrontal cortex.

[0008] The control module, configured in a separate portable device, performs the following steps: performing two-step preprocessing on the EEG signal to obtain a clean EEG signal; extracting EEG feature values ​​from the clean EEG signal to characterize the fatigue state; and comparing the EEG feature values ​​with a set fatigue threshold to determine whether the user is in a fatigue state.

[0009] The alert module is used to trigger an early warning when the control module determines that the user is in a state of fatigue.

[0010] The power module is used for system power supply, charging, power display, and device switching;

[0011] The step of performing two-step preprocessing on the EEG signal to obtain a pure EEG signal specifically includes: the two-step preprocessing includes filtering and independent component reconstruction; the filtered EEG signal is decomposed into multiple independent components, and non-EEG signal components in the multiple independent components are identified and excluded according to the preset rules of independent component reconstruction, and the remaining components are reconstructed into a pure EEG signal.

[0012] Optionally, the preset rule for independent component reconstruction is: based on the frequency points, spectral entropy, and linear correlation coefficients with the filtered EEG signal corresponding to the top N energies in the energy spectrum of each independent component.

[0013] Optionally, non-EEG signal components among multiple independent components are identified and excluded according to preset rules for independent component reconstruction, specifically including:

[0014] Non-EEG components are eliminated by analyzing the frequency points corresponding to the top three energy levels in the energy spectrum. Specifically, this includes:

[0015] If the frequencies corresponding to the top three energy levels are all lower than the low-frequency threshold, then the non-EEG component is excluded.

[0016] If the frequencies corresponding to the top three energy levels are all greater than the high-frequency threshold, then the non-EEG component is excluded.

[0017] If the number of remaining components is greater than 2, they are excluded based on the magnitude of the spectral entropy, specifically including:

[0018] Determine the mean spectral entropy EM of the remaining components;

[0019] Non-EEG components with spectral entropy less than the mean spectral entropy EM are excluded.

[0020] If the number of remaining components is greater than 1, they are excluded based on correlation, specifically including:

[0021] The EEG signal after wavelet decomposition and filtering is appropriately smoothed to obtain the smoothed EEG signal.

[0022] Determine the linear correlation coefficient between each remaining component and the smoothed EEG signal. If the absolute value of the linear correlation coefficient is greater than a set threshold, then exclude the non-EEG component.

[0023] Optionally, the EEG characteristic value θ The relative energy values ​​of the frequency band specifically include:

[0024] The sum of the energy corresponding to all frequency points within the set frequency band is taken as the current pure EEG signal level. θ The energy value of the frequency band;

[0025] The relative energy value of a pure EEG signal is calculated using the following formula:

[0026] Prel ( i,θ ) = P ( i,θ ) / P ( i,all );

[0027] in, P ( i,all () represents the total energy at all frequencies. Prel ( i,θ () represents the relative energy value of a pure EEG signal. P ( i,θ The current pure EEG signal is in θ The energy value of the frequency band.

[0028] Optionally, the relative energy values ​​of the pure EEG signal are subjected to temporal recursive smoothing based on the following formula:

[0029] Prel_sm ( i,θ ) = Prel_sm ( i -1 ,θ )*α+ Prel ( i,θ )*(1-α;

[0030] in, Prel_sm ( i -1 ,θ ) represents the smoothed relative energy value of the previous moment, and α is the smoothing factor; the current pure EEG signal is then set to... θ relative energy value of frequency band smoothing Prel_sm ( i,,θ ) as an eigenvalue.

[0031] Optionally, a fatigue threshold can be set, specifically including:

[0032] Obtain the fatigue status of multiple users θ The mean and standard deviation of the relative energy values ​​of the frequency band;

[0033] The upper and lower thresholds are determined based on the mean and standard deviation using the following formula:

[0034] THlo = MN -2* STD ;

[0035] THhi = MN +2* STD ;

[0036] in,MN The mean, STD Standard deviation, THhi The upper limit threshold, THlo The lower limit threshold;

[0037] Values ​​greater than the upper threshold and less than the lower threshold are identified as outliers and removed. The average of the remaining values ​​is then calculated to obtain the set fatigue threshold. THR .

[0038] Optionally, the filtered EEG signal is decomposed into multiple independent components, specifically including:

[0039] The filtered EEG signal was decomposed into N-level signals using wavelet transform.

[0040] The N-layer signal is decomposed into M IMF functions step by step using EMD;

[0041] PCA dimensionality reduction is performed on N*M IMF functions, retaining the top X important components;

[0042] The first X important components are decomposed using fastICA to obtain X independent components and the corresponding mixture matrix A.

[0043] Optionally, the remaining components are reconstructed into pure EEG signals, specifically including:

[0044] The amplitudes of the non-EEG independent components are set to zero to obtain the processed independent component mixing matrix A.

[0045] The processed independent component mixing matrix A is subjected to inverse ICA transform to obtain the reconstructed wavelet coefficients IMF.

[0046] The reconstructed wavelet coefficients IMF are integrated to obtain a pure EEG signal.

[0047] Optionally, the alert module includes one or a combination of alert sounds, voice alerts, and vibration alerts, and supports manual control of volume and vibration intensity.

[0048] The portable driver fatigue monitoring and early warning system provided in this invention has the following advantages compared with the prior art:

[0049] This invention proposes an artifact removal method, improving system robustness. Addressing the issue of interference with prefrontal EEG signals, a method for identifying and removing non-EEG signal components based on energy spectrum, spectral entropy, and correlation analysis is proposed. This method effectively suppresses interference from artifacts such as eye movements and electromyography, significantly improving the accuracy and reliability of fatigue monitoring. This represents a key technological breakthrough for achieving portable, high-precision monitoring. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the processing flow of a portable fatigue driving monitoring and early warning system provided in one embodiment;

[0051] Figure 2 This is a preprocessing flowchart of a portable fatigue driving monitoring and early warning system provided in one embodiment;

[0052] Figure 3 This is an electroencephalogram (EEG) of a portable fatigue driving monitoring and early warning system provided in one embodiment. Figure 3 In the figure, 'a' represents the original acquired EEG signal. Figure 3 In the diagram, 'b' represents the EEG signal after the first preprocessing step. Figure 3 In this context, 'c' represents the purified EEG signal after the second preprocessing step.

[0053] Figure 4 This is a flowchart of a simulated driving experiment for a portable fatigue driving monitoring and early warning system provided in one embodiment. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] Existing driver fatigue monitoring technologies suffer from high power consumption, making them unsuitable for prolonged wear. The complex algorithms and reliance on external devices for wireless communication result in high overall power consumption, hindering portable applications requiring continuous monitoring over extended periods. This invention addresses this issue by employing a lightweight algorithm for low-power operation. A lightweight fatigue assessment algorithm based on specific frequency band relative energy and recursive smoothing is proposed, with computationally significantly lower than complex models such as SVM and AHP. This allows the algorithm to run directly on a low-power microcontroller (MCU) in real-time, substantially reducing device power consumption and meeting the requirements for extended wear.

[0056] Existing driver fatigue monitoring technologies are highly dependent on systems and cannot operate independently: Current solutions typically employ a distributed system architecture, requiring EEG acquisition devices to work in conjunction with smartphones, in-vehicle control systems, or cloud servers. Signal processing and decision-making rely on high-performance processors, making the system unusable as a standalone portable device. This invention, however, achieves high integration and device independence. Signal acquisition, processing, analysis, and early warning functions are all integrated into a single portable device, eliminating reliance on smartphones, in-vehicle systems, or cloud networks. This reduces costs and improves ease of use and versatility.

[0057] In one embodiment, a portable fatigue driving monitoring and early warning system is provided, such as...Figure 1 As shown, the system includes:

[0058] The acquisition module is used to acquire electroencephalogram (EEG) signals from the user's prefrontal cortex.

[0059] The control module, configured within a separate portable device, performs the following steps: It performs a two-step preprocessing of the EEG signal to obtain a clean EEG signal. It then extracts EEG feature values ​​from the clean EEG signal to characterize fatigue. Finally, it compares these EEG feature values ​​with a set fatigue threshold to determine if the user is fatigued.

[0060] The alert module is used to trigger an early warning when the control module determines that the user is in a state of fatigue.

[0061] The power module is used for system power supply, charging, power display, and device switching.

[0062] The step of performing two-step preprocessing on the EEG signal to obtain a pure EEG signal specifically includes: the two-step preprocessing includes filtering and independent component reconstruction; the filtered EEG signal is decomposed into multiple independent components, and non-EEG signal components in the multiple independent components are identified and excluded according to the preset rules of independent component reconstruction, and the remaining components are reconstructed into a pure EEG signal.

[0063] Specifically, it acquires the user's brainwave signals and converts them into digital brainwave signals.

[0064] The first step of preprocessing the digital EEG signal yields EEG signals within a defined frequency band. The second step of preprocessing then yields a clean EEG signal. A Fast Fourier Transform (FFT) is used to transform the clean EEG signal from the time domain to the frequency domain. Feature extraction is then performed on the clean EEG signal in the frequency domain to obtain eigenvalues. These eigenvalues ​​characterize the clean EEG signal's frequency range. θ The relative energy value of the frequency band. Compare the characteristic value with a set threshold. THR If the feature value is greater than the set threshold, a comparison is made. THR If so, the user is determined to be in a state of fatigue.

[0065] The second preprocessing step for the EEG signal within a defined frequency band includes: decomposing the EEG signal within the defined frequency band into multiple independent components; using Fourier transform to obtain the energy spectrum of each independent component, and arranging the corresponding frequency points of each energy spectrum in descending order of energy to obtain the spectral sequence of each energy spectrum; determining the frequency points corresponding to the top three energies in the energy spectrum based on the spectral sequence of each energy spectrum and calculating the spectral entropy; excluding non-EEG components from the multiple independent components using the frequency points corresponding to the top three energies in the energy spectrum, the spectral entropy, and the linear correlation coefficient with the EEG signal; and reconstructing the remaining components to obtain a clean EEG signal.

[0066] The user's fatigue level is inferred from electroencephalogram (EEG) signals. If the user is determined to be fatigued, they are alerted through sound and mechanical vibration to intervene in their mental state and achieve the goal of waking them up. The specific data transmission and control process is as follows: Figure 1 As shown.

[0067] 1. Data Acquisition Module

[0068] EEG signals are extremely weak, on the order of microvolts, and are highly susceptible to interference from other physiological signals and external noise. Therefore, the performance requirements for acquisition equipment are high, demanding both high acquisition accuracy and strong anti-interference capabilities. Traditional EEG acquisition systems employ complex circuit designs to amplify and filter analog signals to meet EEG acquisition needs, but this results in bulky devices with high power consumption, hindering portability and inconvenience. This invention preferably utilizes the ADS1299, a low-noise analog-to-digital converter chip specifically designed for EEG signal acquisition, coupled with a simplified front-end preprocessing circuit, achieving portable, high-precision, and interference-resistant acquisition. For ease of wear, this invention preferably acquires EEG signals from the bald prefrontal cortex, such as the left frontal electrode Fp1 and right frontal electrode Fp2. The reference electrode REF is placed at the left and right mastoid processes or earlobes as a reference electrode for the left and right frontal electrodes Fp1 and Fp2.

[0069] 2. Control Module

[0070] The algorithms for EEG signal processing and fatigue state detection in this invention are both based on a microcontroller (MCU). Preferably, a 32-bit embedded microcontroller from STMicroelectronics' STM32F4 series can be used, which has powerful computing capabilities (reaching 210 DMIPS at 168MHz high speed) and low power consumption (230μA / MHz, with a power consumption of 38.6mA at 168MHz high speed).

[0071] 2.1 Preprocessing of EEG Signals

[0072] Given the weakness of EEG signals, they are highly susceptible to interference from various physiological and non-physiological signals, especially physiological artifacts such as electrocardiogram (ECG), electrooculogram (EOG), and electromyogram (EMG) spectral components, which overlap with EEG signals, extracting pure EEG data is extremely difficult. Although circuit design has preprocessed analog signals to suppress some noise or artifacts, further preprocessing and noise reduction of the acquired data at the digital signal processing level are still necessary to obtain pure EEG signals. This directly affects the reliability of subsequent feature extraction and fatigue state detection, and is therefore crucial.

[0073] After acquiring the user's EEG signal, preprocessing is performed including detrending, mean removal, filtering, time-frequency analysis, Independent Principal Component Analysis (ICA), and Empirical Mode Decomposition (EMD).

[0074] First, the EEG signal is low-pass filtered to remove high-frequency signals above the high-frequency threshold Fhi, resulting in an EEG signal free of high-frequency artifacts. Then, the EEG signal free of high-frequency artifacts is high-pass filtered to remove low-frequency signals below the low-frequency threshold Flo, resulting in an EEG signal free of low-frequency artifacts. The EEG signal free of low-frequency artifacts is defined as an EEG signal within a specified frequency band, with the minimum value being the low-frequency threshold Flo and the maximum value being the high-frequency threshold Fhi. Next, the EEG signal within the specified frequency band is decomposed into N layers using wavelet transform, resulting in N layers of signals. These N layers of signals are then progressively decomposed into M IMF functions using EMD. Principal Component Analysis (PCA) is performed on the N*M IMF functions to reduce their dimensionality, retaining the top X important components. Fast ICA decomposition is then performed on the top X important components to obtain X independent components and their corresponding mixing matrix A. Finally, Fourier transforms are performed on each of the X independent components to obtain X energy spectra. For each frequency point in the X energy spectra, arrange them in descending order of energy and calculate the spectral entropy. Non-EEG independent components are excluded by considering the frequency points corresponding to the top three energy levels, the magnitude of the spectral entropy, and the linear correlation with the filtered EEG signal. The amplitudes of the non-EEG independent components are set to zero, resulting in the processed independent component matrix. The processed independent component matrix is ​​then combined with the mixing matrix A and subjected to an inverse ICA transform to obtain the reconstructed wavelet coefficients IMF. Finally, the reconstructed wavelet coefficients IMF are integrated to obtain the clean EEG signal.

[0075] Specifically, the ICA method can separate mixed signals into independent components with relatively little prior knowledge and is commonly used in artifact removal of multi-channel EEG signals. However, this invention is based on signals acquired from two prefrontal lobe leads (Fp1 and Fp2), which has a limited number of channels. Directly using the ICA method will result in overcomplete ICA (the number of observed signals is less than the number of source signals), making it difficult to achieve ideal noise reduction results. EMD can efficiently decompose any one-dimensional complex signal into a finite number of Intrinsic Mode Functions (IMFs), freeing it from the constraints of Fourier transform and wavelet transform, which require setting basis functions. The decomposition process can be performed according to the time scale of the data itself, thus it can adapt to the signal and is suitable for processing nonlinear and non-stationary signals such as EEG and ECG. This is useful for removing artifacts such as DC trends, sweating baseline drift artifacts, and power frequency interference. However, some physiological artifacts, such as eye movement (EMG) signals and EEG signals, have significant spectral overlap. Several IMFs after EMD decomposition may simultaneously contain both EMG artifacts and EEG signals. Directly resetting the weights of these components to zero during signal synthesis can damage even normal EEG signals. Wavelet transform, compared to traditional Fourier transform, can analyze signals simultaneously in both time and frequency dimensions, thus offering a clear advantage in processing non-stationary signals. Given the limited number of EEG acquisition channels in the portable device of this invention, and the fact that prefrontal EEG signals are particularly susceptible to EMG artifacts, this invention employs a combination of wavelet transform, EMD decomposition, and ICA decomposition to separate and reconstruct the acquired signals, obtaining a relatively pure EEG signal.

[0076] EEG data preprocessing workflow as follows Figure 2 As shown. First, the raw EEG signals acquired over a certain period of time undergo a preprocessing step, including channel selection, mean removal, and filtering. Given the potential interference from factors such as electrode detachment, the first step in preprocessing is the selection of the EEG channel. Preferably, the electrode lead disconnection detection function built into the acquisition chip is used to exclude data from detached electrode channels. If no detachment is detected in either channel, any channel can be selected for subsequent processing. The EEG frequency band analysis range for fatigue detection is generally below 30Hz. A low-pass filter is preferred to filter out high-frequency signals >40Hz, thus reducing high-frequency artifacts such as 50Hz power line interference and electromyography artifacts. A high-pass filter is then used to filter out lower-frequency signals <1Hz, reducing low-frequency artifacts caused by scalp sweating.

[0077] Next, the filtered signal is decomposed into N layers using wavelet transform to obtain the decomposed N-layer signals. EEG signals and artifact signals such as electrooculogram (EOG) will appear simultaneously on several detail layers, and each layer can be regarded as a subset of the original EEG signal and artifact signals.

[0078] Subsequently, EMD or other EMD methods such as EEMD, VMD, CEEMD, etc., are used to further decompose the wavelet transform detail layer into M IMF functions (the number of IMFs varies in the decomposition results). Each IMF function satisfies the following conditions: 1) The number of extreme points and zero points within the data range is equal or differs by at most 1; 2) At any given time, the mean of the envelopes of the local maximum and the local minimum must be zero.

[0079] Next, the FastICA method is used to separate the N*M IMF functions into independent components, thus avoiding the overcomplete ICA problem.

[0080] Before decomposition, the matrix data composed of N*M IMFs is first subjected to mean removal, whitening, and PCA dimensionality reduction. Then, the selected top X important components are decomposed using FastICA. FastICA uses a fixed-point iterative method to find the maximum non-Gaussianity, with the goal of maximizing negative entropy. Each component of the ICA result can be regarded as a signal source. By setting certain judgment criteria, non-EEG signal source components are automatically identified and their amplitudes are set to zero. Finally, through inverse ICA transformation and wavelet coefficient reconstruction, a pure EEG signal can be obtained.

[0081] In this invention, the preferred data analysis window duration is 5 seconds. The preliminary preprocessing uses a 3rd-order Butterworth filter, the wavelet decomposition has 5 layers, the wavelet basis is the bidirectional orthogonal sym7, and the EMD-like decomposition uses the original EMD method. Before ICA decomposition, the dimensions are reduced to 7 dimensions using PCA.

[0082] Non-EEG components among multiple independent components are excluded by using the frequency points, spectral entropy, and linear correlation coefficients of the top three energies in the energy spectrum. Specifically, this includes:

[0083] (1) Non-EEG components are eliminated by using the frequency points corresponding to the top three energies in the energy spectrum, specifically including:

[0084] If the frequencies corresponding to the top three energy levels are all lower than the low-frequency threshold (such as 5Hz), then the non-EEG component is excluded.

[0085] If the frequencies corresponding to the top three energy levels are all greater than the high-frequency threshold (such as 20Hz), then the non-EEG component is excluded.

[0086] (2) If the number of remaining components is greater than 2, further exclusions are made based on the magnitude of the spectral entropy, specifically including:

[0087] Determine the mean spectral entropy EM of the remaining components;

[0088] Non-EEG components with spectral entropy less than the mean spectral entropy EM were excluded.

[0089] (3) If the number of remaining components is greater than 1, further elimination is performed from a correlation perspective, specifically including:

[0090] The EEG signal after filtering before wavelet decomposition is appropriately smoothed to obtain the smoothed EEG signal.

[0091] Determine the linear correlation coefficient between each remaining component and the smoothed EEG signal. If the absolute value of the linear correlation coefficient is greater than a set threshold (e.g., 0.1), then exclude the non-EEG component.

[0092] EEG signal image as follows Figure 3 As shown. Figure 3 In the figure, 'a' represents the original acquired EEG signal. Figure 3 In the diagram, 'b' represents the EEG signal after the first preprocessing step. Figure 3 In the diagram, 'c' represents the purified EEG signal after the second preprocessing step. It is evident that eye movement artifacts and high-frequency interference in the original data are significantly suppressed, while the EEG components are preserved.

[0093] 2.2 EEG Feature Extraction

[0094] Numerous studies have reported on brain electrical activity during fatigue; however, the conclusions are not entirely consistent. θ The energy increase was significant and the conclusions were consistent, therefore it can be considered a powerful biological signal indicating mental fatigue.

[0095] The digital EEG signal is converted from the time domain to the frequency domain using a Fast Fourier Transform (FFT) to obtain the current state of the digital EEG signal. θ energy value of frequency band P ( i,θ Energy value P ( i,θ ) is the sum of the energy corresponding to all frequency points in the 4~8Hz frequency band.

[0096] Then, the relative energy value of the digital EEG signal was obtained. Prel ( i,θ The calculation formula is as follows:

[0097] Prel ( i,θ ) = P ( i,θ ) / P ( i,all );

[0098] in, P ( i,all ) represents the total energy at all frequencies.

[0099] The relative energy value at the current moment is smoothed using a time-domain recursive smoothing method. Prel ( i,θ The smoothed relative energy value at the current moment is obtained by performing a smoothing process. Prel_sm ( i,,θ The formula is:

[0100] Prel_sm ( i,θ ) = Prel_sm ( i -1 ,θ )*α+ Prel ( i,θ )*(1-α;

[0101] in, Prel_sm ( i -1 ,θ ) represents the smoothed relative energy value at the previous time step, and α is the smoothing factor. In this embodiment, the smoothing factor α is set to 0.8.

[0102] The current digital EEG signal is in θ relative energy value of frequency band smoothing Prel_sm ( i,,θ ) as an eigenvalue.

[0103] 2.3 Fatigue State Assessment

[0104] The feature value is compared with the system's preset threshold THR. If the feature value is greater than the threshold, it is determined to be a fatigue state, and subsequent warnings and interventions are initiated. Relative values ​​are used here instead of absolute energy values ​​to minimize the impact of individual differences on the judgment results. Using absolute energy values ​​is more susceptible to individual differences, thus reducing the accuracy of fatigue state identification.

[0105] 2.4 Fatigue State Judgment Threshold Setting Method

[0106] Threshold setting is crucial for fatigue detection. To obtain a suitable threshold, a large number of subjects (more than 100) were recruited, and a fatigue state was induced in the subjects through simulated driving (e.g., Figure 4 As shown in the image, during a long and monotonous simulated driving session, the subjects' electroencephalogram (EEG) signals were simultaneously collected. Additionally, a camera captured the subjects' facial expressions to help identify their fatigue levels. A 10-minute test drive was conducted before the formal experiment to help subjects familiarize themselves with the system operation.

[0107] During the experiment, a short fatigue status questionnaire, such as the widely accepted and used Karolinska Sleepiness Scale or Samn-Perelli seven-point fatigue scale, was used every 5 to 10 minutes.

[0108] The Karolinska Sleepiness Scale (KSS) ranges from 1 to 9, with a score of 6 or higher indicating a state of fatigue. See Table 1 for details.

[0109] The Samn-Perelli seven-point fatigue scale (SPS) is scored from 1 to 7, with a score of 4 or higher indicating a state of fatigue. See Table 2 for details.

[0110] To obtain the subject's subjective fatigue status at different stages.

[0111] Table 1. Contents of the KSS Scoring Questionnaire

[0112]

[0113] Table 2 SPS Scoring Questionnaire Content

[0114]

[0115] The entire driving process lasted 20-30 minutes, based on questionnaire results indicating that the subjects had entered a state of subjective mental fatigue (the experiment could be terminated early if the subjects indicated excessive fatigue and could not continue). Post-processing of each subject's EEG data, combined with facial video and subjective questionnaire results, determined the subject's fatigue level and calculated their fatigue status. θ The mean of the relative energy values ​​of the frequency band. Then, outliers were excluded. The method was as follows: First, the fatigue state of all subjects was statistically analyzed. θ The mean of the relative energy values ​​of the frequency band MN and standard deviation STD Then, set upper and lower thresholds for outlier detection. THlo and THhi ,in,

[0116] THlo = MN -2* STD ;

[0117] THhi = MN +2* STD ;

[0118] greater than THhi and less than THloThe values ​​are identified as outliers and removed. Then, the remaining values ​​are averaged to obtain the threshold value used by the system to determine fatigue status. THR .

[0119] 3. Reminder Module

[0120] The reminder module is used to provide reminders based on the assessment of fatigue levels. It can include a combination of one or more reminder methods, such as using a voice chip and a miniature vibration motor for voice prompts and vibration intervention, respectively. Optionally, the volume and vibration intensity can be manually adjusted via buttons. In situations where it is inconvenient to use public address systems, such as classrooms or conference rooms, the volume can be pre-adjusted to a very low level while retaining the vibration function for reminders.

[0121] The application scenarios of this invention are not limited to driving, but can also be extended to other life scenarios that require keeping the brain alert, such as meetings, high-altitude operations, classrooms, etc., to monitor the user's fatigue and drowsiness, provide timely warnings and interventions, improve work and study efficiency, and prevent safety accidents.

[0122] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A portable fatigue driving monitoring and early warning system, characterized in that, include: The acquisition module is used to acquire electroencephalogram (EEG) signals from the user's prefrontal cortex. The control module, configured in a separate portable device, performs the following steps: performing two-step preprocessing on the EEG signal to obtain a clean EEG signal; Extract EEG feature values ​​to characterize fatigue state from the pure EEG signal; The EEG characteristic values ​​are compared with a set fatigue threshold to determine whether the user is in a state of fatigue. The reminder module is used to trigger an early warning when the control module determines that the user is in a state of fatigue. The power module is used for system power supply, charging, power display, and device switching; The step of performing a two-step preprocessing on the EEG signal to obtain a pure EEG signal specifically includes: the two-step preprocessing includes filtering and independent component reconstruction; the filtered EEG signal is decomposed into multiple independent components, and non-EEG signal components in the multiple independent components are identified and excluded according to the preset rules of independent component reconstruction, and the remaining components are reconstructed into a pure EEG signal. The preset rule for the reconstruction of independent components is: based on the frequency points, spectral entropy, and linear correlation coefficients with the filtered EEG signal of the top N energies in the energy spectrum of each independent component. The step of identifying and excluding non-EEG signal components among the multiple independent components according to preset rules for independent component reconstruction specifically includes: Non-EEG components are excluded by using the frequency points corresponding to the top three energies in the energy spectrum. Specifically, if the frequency points corresponding to the top three energies are all lower than the low-frequency threshold, the non-EEG component is excluded; if the frequency points corresponding to the top three energies are all higher than the high-frequency threshold, the non-EEG component is excluded. If the number of remaining components is greater than 2, they are excluded based on the magnitude of the spectral entropy. Specifically, this includes: determining the mean spectral entropy EM of the remaining components; and excluding non-EEG components whose spectral entropy is less than the mean EM. If the number of remaining components is greater than 1, they are excluded from the correlation perspective. Specifically, this includes: appropriately smoothing the EEG signal after filtering before wavelet decomposition to obtain a smoothed EEG signal; determining the linear correlation coefficient between each remaining component and the smoothed EEG signal; and excluding non-EEG components if the absolute value of the linear correlation coefficient is greater than a set threshold. The EEG characteristic value θ The relative energy values ​​of the frequency band specifically include: The sum of the energy corresponding to all frequency points within the set frequency band is taken as the current pure EEG signal level. θ The energy value of the frequency band; The relative energy value of a pure EEG signal is calculated using the following formula: Prel ( i,θ ) = P ( i,θ ) / P ( i,all )； in, P ( i,all () represents the total energy at all frequencies. Prel ( i,θ () represents the relative energy value of a pure EEG signal. P ( i,θ The current pure EEG signal is in θ The energy value of the frequency band; This also includes time-domain recursive smoothing of the relative energy values ​​of pure EEG signals based on the following formula: Prel_sm ( i,θ ) = Prel_sm ( i -1 ,θ )*α+ Prel ( i,θ )*（1-α）； in, Prel_sm ( i -1 ,θ ) represents the smoothed relative energy value of the previous moment, and α is the smoothing factor; the current pure EEG signal is then set to... θ relative energy value of frequency band smoothing Prel_sm ( i,,θ ) as an eigenvalue.

2. The portable fatigue driving monitoring and early warning system as described in claim 1, characterized in that, The set fatigue threshold specifically includes: Obtain the fatigue status of multiple users θ The mean and standard deviation of the relative energy values ​​of the frequency band; The upper and lower thresholds are determined based on the mean and standard deviation using the following formula: THlo = MN -2* STD ； THhi = MN +2* STD ； in, MN The mean, STD Standard deviation, THhi The upper limit threshold, THlo The lower limit threshold; Values ​​greater than the upper threshold and less than the lower threshold are identified as outliers and removed. The average of the remaining values ​​is then calculated to obtain the set fatigue threshold. THR .

3. The portable fatigue driving monitoring and early warning system as described in claim 1, characterized in that, The process of decomposing the filtered EEG signal into multiple independent components specifically includes: The filtered EEG signal was decomposed into N-level signals using wavelet transform. The N-layer signal is decomposed into M IMF functions step by step using EMD; PCA dimensionality reduction is performed on N*M IMF functions, retaining the top X important components; The first X important components are decomposed using fastICA to obtain X independent components and the corresponding mixture matrix A.

4. The portable fatigue driving monitoring and early warning system as described in claim 1, characterized in that, The process of reconstructing the remaining components into a pure EEG signal specifically includes: The amplitudes of the non-EEG independent components are set to zero to obtain the processed independent component mixing matrix A. The processed independent component mixing matrix A is subjected to inverse ICA transform to obtain the reconstructed wavelet coefficients IMF. The reconstructed wavelet coefficients IMF are integrated to obtain a pure EEG signal.

5. The portable fatigue driving monitoring and early warning system as described in claim 1, characterized in that, The reminder module includes one or a combination of warning sounds, voice prompts, and vibration alerts, and supports manual control of volume and vibration intensity.

Images