Driver psychological monitoring method and system based on electroencephalogram signals

Abstract

According to the driver psychological monitoring method and system based on the electroencephalogram signals, the electroencephalogram signals of the head of a driver are collected in real time through a multi-channel electroencephalogram device, multi-dimensional features are extracted after preprocessing, and emotion states such as tension, fatigue and anger are accurately recognized through a support vector machine model. The system adopts a dynamic threshold judgment mechanism, combines sliding time window statistical analysis and composite rule verification, adaptively adjusts an emotional anomaly standard, and significantly reduces the environmental interference influence. And based on a real-time monitoring result, a multi-stage safety intervention strategy is triggered, step-by-step response from voice warning, vehicle speed limitation to vehicle forced take-over is achieved, and driving autonomy and safety are balanced. The method has the advantages of high precision, strong robustness and scene adaptability, can effectively reduce the traffic accident risk caused by the psychological abnormality of the driver, and provides an innovative solution for intelligent traffic management and active safety technology development.

Description

Technical Field

[0001] The present invention relates to electroencephalogram (EEG) data processing, especially a method and system for driver psychological monitoring based on EEG signals. Background Art

[0002] In today's rapidly developing society, road traffic safety issues are like a sword hanging over the process of social development, constantly threatening the safety of human life and property. With the booming development of the automotive industry and the sharp increase in the number of private cars, the road traffic network has become increasingly complex, traffic flow has surged, and the incidence of traffic accidents has also risen accordingly. Statistical data reveals that the number of deaths caused by road traffic accidents every year is astonishing and this figure remains persistently high. In-depth analysis shows that although road facilities are constantly improving and vehicle safety technologies are increasingly advanced, as a core component of the traffic system, the fluctuations and changes in the driver's mental state are still one of the important reasons for the frequent occurrence of traffic accidents.

[0003] EEG signals can capture the neural activity patterns in the brain during the process of processing external information, generating emotional responses, and performing cognitive processing by recording the weak potential changes on the scalp surface. These patterns not only reveal the real-time state of brain function but also reflect the driver's psychological responses and behavioral tendencies in specific situations. Therefore, the driver psychological monitoring technology based on EEG signals is expected to provide objective and quantitative indicators for evaluating the driver's mental state and provide a scientific basis for realizing personalized and precise driving safety intervention. Summary of the Invention

[0004] To solve the above technical problems, the technical solution adopted by the driver psychological monitoring method based on EEG signals of the present invention includes the following steps: S1: EEG signal acquisition and preprocessing, real-time acquisition of raw EEG signals through a multi-channel EEG acquisition device deployed on the driver's head. The device includes at least 8 electrodes, covering the prefrontal lobe, parietal lobe, and temporal lobe regions, and signal acquisition is performed at a sampling frequency of not less than 256 Hz, and preprocessing operations are performed on the raw signals.

[0005] S2: EEG signal feature extraction, multi-dimensional feature extraction is performed on the preprocessed EEG signals, specifically including: (a) Calculate the power spectral density of each channel based on the fast Fourier transform, respectively extract the energy values of the δ wave, θ wave, α wave, β wave, and γ wave frequency bands, and calculate the percentage of the energy of each frequency band in the total frequency band energy; (b) Perform wavelet decomposition on the signal, extract the time-frequency energy distribution of the third-level detail coefficients of the θ wave and the fourth-level detail coefficients of the β wave, and calculate their variances as time-frequency domain features; (c) According to the channels in the prefrontal lobe region and Channel Wave energy difference, parietal region Channel sum Of the channel Wave energy difference, calculating the electroencephalogram signal asymmetry index: , where And Are for the prefrontal region Channel sum Of the channel Wave energy difference And Are for the parietal region Channel sum Of the channel Wave energy difference

[0006] S3: Mental state classification. The proportion of δ-wave energy, the proportion of θ-wave energy, the proportion of β-wave energy, the proportion of γ-wave energy, the time-frequency energy variance of θ-wave, the time-frequency energy variance of β-wave and the FAA index are used to form a feature vector, which is input into a pre-trained mental state classification model to output the current mental state category of the driver, where: (1) The classification model adopts the support vector machine (SVM) architecture, and the training data set contains electroencephalogram samples labeled with four emotions: tense, fatigued, angry, and calm; (2) The output of the classification model is the probability distribution of the four emotions. When the probability value of a certain emotion category ≥ 0.75 and the classification results are consistent for 6 seconds in a cumulative 3 time windows, it is determined that the emotion state takes effect; (3) When the angry or fatigued state takes effect, trigger the dynamic emotion threshold determination process.

[0007] S4: Dynamic emotion threshold determination. Dynamically adjust the emotion abnormality threshold according to the real-time classification results and historical data, specifically including: S41: Establish a sliding time window with a length of 60 seconds, store the intensity index of the current effective emotion (such as the proportion of β-wave energy corresponding to anger), and calculate the mean μ and standard deviation σ of the data in the window; S42: Define the emotion abnormality condition as the real-time emotion intensity index exceeding μ + 2σ, and set a composite determination rule according to different emotion types: Tense state: The proportion of θ-wave energy ≥ 35% and the FAA index ≤ -0.2 for 10 seconds; Angry state: The proportion of β-wave energy ≥ 40% and the growth rate of γ-wave energy in the temporal lobe region ≥ 20% within 1 second; Fatigued state: The proportion of α-wave energy ≥ 50% and the eyelid closing frequency detected by the on-vehicle camera ≥ 0.3 Hz; S43: When the composite determination rule is satisfied, preset multi-level intervention strategies are matched according to the emotion category.

[0008] S5: The multi-level intervention strategy is executed, and hierarchical control instructions are triggered according to the emotion category and the abnormal level, including: Level 1 intervention: When the α-wave energy ratio is 50%-60%, it is detected as being tense or slightly fatigued, and mild perception intervention is performed, and a voice prompt is sent to the driver. Level 2 intervention: When the α-wave energy ratio is 60%-70%, it is detected as being angry or moderately fatigued. On the basis of the level 1 intervention, environmental constraints are added, and at the same time, an instruction is sent to the engine control unit ECU through the vehicle CAN bus to limit the vehicle speed to 80% of the current road speed limit value. Level 3 intervention: When the β-wave energy ratio ≥ 40% exceeds 30 seconds, it is detected as being continuously angry or severely fatigued and when the α-wave energy ratio ≥ 70%, the control authority of the driver over the accelerator pedal is locked, the braking function is retained, the automatic driving system is started to forcibly take over the vehicle control, and an alarm signal including the driver ID, geographical location and emotion data is sent to the cloud monitoring platform through the vehicle-mounted communication module.

[0009] As a further solution of the present invention, the preprocessing operation in step S1 includes: (a) A band-pass filter with a frequency range of 0.5 Hz to 60 Hz is used to eliminate low-frequency motion artifacts and high-frequency noise, and a 50 Hz or 60 Hz notch filter is superimposed to suppress power frequency interference. (b) The electrooculogram artifact and electromyogram artifact parts in the signal are separated, and the electroencephalogram signal is reconstructed; the reconstructed signal is divided into data segments with a length of 2 seconds according to a fixed time window, and the mean value of each data segment is subtracted from the baseline correction for each data segment, and the normalization process is scaled to the range of [-1, 1].

[0010] As a further solution of the present invention, in step S2, the power spectral bands of each channel are: δ wave: 0.5 Hz to 4 Hz, θ wave: 4 Hz to 8 Hz, α wave: 8 Hz to 13 Hz, β wave: 13 Hz to 30 Hz, and γ wave: 30 Hz to 60 Hz. As a further solution of the present invention, for the driver psychological monitoring system based on electroencephalogram signals, the system module includes: Electroencephalogram signal acquisition module: responsible for collecting the original electroencephalogram signals through a multi-channel electroencephalogram acquisition device deployed on the driver's head, and performing preprocessing of baseline correction and normalization on the data. Feature extraction module: performing multi-dimensional feature extraction on the preprocessed electroencephalogram signals, including calculating the power spectral density of each channel, extracting the energy values and ratios of different frequency bands, performing wavelet decomposition to extract the time-frequency energy distribution and calculating the variance, and calculating the electroencephalogram asymmetry index. Mental state classification module: Input the extracted feature vectors into a pre-trained mental state classification model using the support vector machine (SVM) architecture, output the probability distribution of the current mental state category of the driver, and determine whether the emotional state takes effect according to the set rules; Dynamic emotional threshold determination module: According to the real-time classification results and historical data, by establishing a sliding time window, calculate the mean and standard deviation of the data within the window, define the emotional abnormality conditions and composite determination rules, and determine whether the emotional abnormality conditions are met; Multi-level intervention strategy execution module: Trigger hierarchical control instructions according to the emotional category and abnormality level, and execute different levels of intervention operations.

[0011] Beneficial effects: By collecting and analyzing the driver's EEG signals in real time, combining dynamic threshold determination and multi-level intervention strategies, the present invention effectively improves the driving safety and the accuracy of mental state monitoring. The system can capture the subtle changes in brain nerve activities, accurately identify emotional states such as tension, fatigue, and anger using multi-dimensional feature fusion technology, and reduce environmental noise interference through an adaptive dynamic threshold mechanism, significantly reducing the misjudgment rate. Based on the real-time monitoring results, the system intelligently triggers hierarchical intervention measures, from voice reminders, speed limits to vehicle forced takeover, gradually strengthening safety guarantees, while retaining the driver's autonomy and actively avoiding risks in extreme emotions. At the same time, by integrating in-vehicle cameras, vehicle control buses, and cloud communication modules, the collaborative analysis and remote monitoring of multi-modal data are realized, providing comprehensive protection for driving behaviors. The system has good environmental adaptability and expansion capabilities, can be flexibly applied to different driving scenarios, provides important support for the development of intelligent traffic management and active safety technologies, and has significant social value in reducing the incidence of traffic accidents and improving road safety levels. Description of the Drawings

[0012] Figure 1 It is the step flow chart of the driver mental monitoring method based on EEG signals of the present invention; Figure 2 It is the module schematic diagram of the driver mental monitoring system based on EEG signals of the present invention. Detailed Embodiments

[0013] The present invention will be further described in detail below in conjunction with embodiments.

[0014] Please refer to Figure 1 as shown in Figure 1 It is the step flow chart of the driver mental monitoring method based on EEG signals of the present invention, and the specific steps include: S1: Electroencephalogram (EEG) signal acquisition and preprocessing. The raw EEG signals are obtained in real time through a multi-channel EEG acquisition device deployed on the driver's head. The device includes at least 8 electrodes, covering the prefrontal, parietal, and temporal lobe regions, and the signals are acquired at a sampling frequency of not less than 256 Hz. Preprocessing operations are performed on the raw signals.

[0015] S2: EEG signal feature extraction. Multidimensional feature extraction is performed on the preprocessed EEG signals, specifically including: (a) Calculate the power spectral density of each channel based on the fast Fourier transform, extract the energy values of the δ-wave, θ-wave, α-wave, β-wave, and γ-wave frequency bands respectively, and calculate the percentage of the energy of each frequency band in the total frequency band energy; (b) Perform wavelet decomposition on the signal, extract the time-frequency energy distribution of the third-level detail coefficients of the θ-wave and the fourth-level detail coefficients of the β-wave, and calculate their variances as time-frequency domain features; (c) According to the channel and channel wave energy difference in the prefrontal region, and the channel and channel wave energy difference in the parietal region, calculate the EEG signal asymmetry index: , where and are the channel and channel wave energy difference in the prefrontal region, and are the channel and channel wave energy difference in the parietal region.

[0016] S3: Psychological state classification. The proportion of δ-wave energy, the proportion of θ-wave energy, the proportion of β-wave energy, the proportion of γ-wave energy, the variance of θ-wave time-frequency energy, the variance of β-wave time-frequency energy, and the FAA index are used to form a feature vector, which is input into a pre-trained psychological state classification model to output the current psychological state category of the driver. Among them: (1) The classification model adopts the support vector machine (SVM) architecture, and the training data set includes EEG samples labeled with four emotions: tense, fatigued, angry, and calm; (2) The output of the classification model is the probability distribution of the four emotions. When the probability value of a certain emotion category ≥ 0.75 and the classification results are consistent for 6 seconds in a cumulative of 3 time windows, it is determined that the emotion state takes effect; (3) When the angry or fatigued state takes effect, a dynamic emotion threshold determination process is triggered.

[0017] S4: Dynamic emotion threshold determination, dynamically adjusting the emotion abnormality threshold according to the real-time classification results and historical data, specifically including: S41: Establish a sliding time window with a length of 60 seconds, store the intensity index of the currently effective emotion (such as the proportion of β-wave energy corresponding to anger), and calculate the mean μ and standard deviation σ of the data within the window; S42: Define the emotion abnormality condition as the real-time emotion intensity index exceeding μ + 2σ, and set compound determination rules according to different emotion types: Tense state: The proportion of θ-wave energy ≥ 35% and the FAA index ≤ -0.2 for 10 seconds; Angry state: The proportion of β-wave energy ≥ 40% and the growth rate of γ-wave energy in the temporal lobe region ≥ 20% within 1 second; Fatigue state: The proportion of α-wave energy ≥ 50% and the eyelid closing frequency detected by the in-vehicle camera ≥ 0.3 Hz; S43: When the compound determination rule is satisfied, match the preset multi-level intervention strategy according to the emotion category.

[0018] S5: Execution of multi-level intervention strategy, triggering hierarchical control instructions according to the emotion category and abnormality level, including: First-level intervention: When the proportion of α-wave energy is 50% - 60%, detected as tense or mildly fatigued, perform mild perception intervention and send a voice prompt to the driver; Second-level intervention: When the proportion of α-wave energy is 60% - 70%, detected as angry or moderately fatigued, add environmental constraints on the basis of the first-level intervention, and at the same time send an instruction to the engine control unit ECU through the vehicle CAN bus to limit the vehicle speed to 80% of the current road speed limit value; Third-level intervention: When the proportion of β-wave energy ≥ 40% for more than 30 seconds, detected as continuously angry or severely fatigued and when the proportion of α-wave energy ≥ 70%, lock the driver's control authority over the accelerator pedal, retain the braking function, start the automatic driving system to forcibly take over vehicle control, and send an alarm signal containing the driver ID, geographical location, and emotion data to the cloud monitoring platform through the in-vehicle communication module.

[0019] Furthermore, the preprocessing operation in step S1 includes: (a) Use a band-pass filter from 0.5 Hz to 60 Hz to eliminate low-frequency motion artifacts and high-frequency noise, and superimpose a 50 Hz or 60 Hz notch filter to suppress power frequency interference; (b) Separate the electrooculogram artifact and electromyogram artifact parts in the signal and reconstruct the electroencephalogram signal; the reconstructed signal is divided into data segments with a length of 2 seconds according to a fixed time window, and the mean of each data segment is subtracted from each data segment for baseline correction and normalized to the range [-1, 1].

[0020] Further, in step S2, the power spectral bands of each channel are as follows: delta wave: 0.5 Hz to 4 Hz, theta wave: 4 Hz to 8 Hz, alpha wave: 8 Hz to 13 Hz, beta wave: 13 Hz to 30 Hz, and gamma wave: 30 Hz to 60 Hz; Further, please refer to Figure 2 as shown in Figure 2 the schematic diagram of the driver psychological monitoring system module based on electroencephalogram signals of the present invention. The system module includes: Electroencephalogram signal acquisition module: responsible for collecting raw electroencephalogram signals through a multi-channel electroencephalogram acquisition device deployed on the driver's head, and performing preprocessing of baseline correction and normalization on the data; Feature extraction module: performing multi-dimensional feature extraction on the preprocessed electroencephalogram signals, including calculating the power spectral density of each channel, extracting the energy values and ratios of different frequency bands, performing wavelet decomposition to extract the time-frequency energy distribution and calculating the variance, and calculating the electroencephalogram asymmetry index; Psychological state classification module: inputting the extracted feature vectors into a pre-trained psychological state classification model using the support vector machine (SVM) architecture, outputting the probability distribution of the current psychological state category of the driver, and determining whether the emotional state is valid according to the set rules; Dynamic emotional threshold determination module: according to the real-time classification results and historical data, by establishing a sliding time window, calculating the mean and standard deviation of the data within the window, defining the emotional abnormality conditions and composite determination rules, and determining whether the emotional abnormality conditions are met; Multi-level intervention strategy execution module: triggering hierarchical control instructions according to the emotional category and abnormality level, and executing different levels of intervention operations.

[0021] The above are only the embodiments of this specification and are not used to limit this specification. For those skilled in the art, various changes and modifications can be made to this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the scope of the claims of this application.

Claims

1. A driver psychological monitoring method based on electroencephalogram signals, characterized in that, It includes the following steps: S1: Electroencephalogram (EEG) signal acquisition and preprocessing. The original EEG signals are obtained in real time through a multi-channel EEG acquisition device deployed on the driver's head. The device includes at least 8 electrodes, covering the prefrontal lobe, parietal lobe, and temporal lobe regions, and the signals are acquired at a sampling frequency of not less than 256 Hz. Preprocessing operations are performed on the original signals; S2: EEG signal feature extraction. Multidimensional feature extraction is performed on the preprocessed EEG signals, specifically including: (a) Based on the fast Fourier transform, calculate the power spectral density of each channel, extract the energy values of the δ-wave, θ-wave, α-wave, β-wave, and γ-wave frequency bands respectively, and calculate the percentage of the energy of each frequency band in the total frequency band energy; (b) Perform wavelet decomposition on the signals, extract the time-frequency energy distribution of the third-level detail coefficients of the θ-wave and the fourth-level detail coefficients of the β-wave, and calculate their variances as time-frequency domain features; (c) According to the channel and channel wave energy difference, the parietal region channel and channel's wave energy difference, calculate the electroencephalogram signal asymmetry index: , where and are the prefrontal regions channel and the wave energy difference, and are the parietal regions channel and the wave energy difference; S3: Mental state classification. The proportion of δ-wave energy, the proportion of θ-wave energy, the proportion of β-wave energy, the proportion of γ-wave energy, the time-frequency energy variance of the θ-wave, the time-frequency energy variance of the β-wave, and the FAA index are used to form a feature vector, which is input into a pre-trained mental state classification model to output the current mental state category of the driver. Among them: (1) The classification model adopts the support vector machine (SVM) architecture, and the training data set includes EEG samples labeled with four types of emotions: tense, fatigued, angry, and calm; (2) The output of the classification model is the probability distribution of the four types of emotions. When the probability value of a certain emotion category ≥ 0.75 and the classification results are consistent for 6 seconds in a cumulative 3 time windows, it is determined that this emotion state takes effect; (3) When the angry or fatigued state takes effect, trigger the dynamic emotion threshold determination process; S4: Dynamic emotion threshold determination. Dynamically adjust the emotion abnormality threshold according to the real-time classification results and historical data, specifically including: S41: Establish a sliding time window with a length of 60 seconds, store the intensity index of the current effective emotion (such as the proportion of β-wave energy corresponding to anger), and calculate the mean μ and standard deviation σ of the data in the window; S42: Define the emotion abnormality condition as the real-time emotion intensity index exceeding μ + 2σ, and set a composite determination rule according to different emotion types: Tense state: The proportion of θ-wave energy ≥ 35% and the FAA index ≤ -0.2 for 10 seconds; Angry state: The proportion of β-wave energy ≥ 40% and the growth rate of γ-wave energy in the temporal lobe region ≥ 20% within 1 second; Fatigued state: The proportion of α-wave energy ≥ 50% and the eyelid closing frequency detected by the in-vehicle camera ≥ 0.3 Hz; S43: When the composite determination rule is satisfied, match the preset multi-level intervention strategy according to the emotion category; S5: Execution of multi-level intervention strategy. Trigger hierarchical control instructions according to the emotion category and abnormality level, including: First-level intervention: When the proportion of α-wave energy is 50% - 60%, it is detected as tense or mildly fatigued, perform mild perception intervention, and send a voice prompt to the driver; Second-level intervention: When the proportion of α-wave energy is 60% - 70%, it is detected as angry or moderately fatigued. On the basis of the first-level intervention, add environmental constraints, and at the same time send an instruction to the engine control unit (ECU) through the vehicle CAN bus to limit the vehicle speed to 80% of the current road speed limit value; Tertiary intervention: When the proportion of β-wave energy ≥ 40% for more than 30 seconds, it is detected as continuous anger or severe fatigue. And when the proportion of α-wave energy ≥ 70%, the control authority of the driver over the accelerator pedal is locked, the braking function is retained, the automatic driving system is activated to forcibly take over the vehicle control, and an alarm signal containing the driver ID, geographical location, and emotional data is sent to the cloud monitoring platform through the vehicle-mounted communication module.

2. The driver psychological monitoring method based on electroencephalogram signals according to claim 1, wherein The preprocessing operation in the step S1 includes: (a) A band-pass filter with a frequency range of 0.5 Hz to 60 Hz is used to eliminate low-frequency motion artifacts and high-frequency noise, and a 50 Hz or 60 Hz notch filter is superimposed to suppress power frequency interference; (b) The electrooculogram artifact and electromyogram artifact parts in the signal are separated, and the electroencephalogram signal is reconstructed; the reconstructed signal is divided into data segments with a length of 2 seconds according to a fixed time window, and the mean value of each data segment is subtracted for baseline correction and normalized to the range of [-1, 1].

3. The driver psychological monitoring method based on electroencephalogram signals according to claim 1, characterized in that In the step S2, the power spectral frequency bands of each channel are as follows: δ-wave: 0.5 Hz to 4 Hz, θ-wave: 4 Hz to 8 Hz, α-wave: 8 Hz to 13 Hz, β-wave: 13 Hz to 30 Hz, and γ-wave: 30 Hz to 60 Hz.

4. A driver psychological monitoring system based on electroencephalogram signals, characterized in that, The system module includes: Electroencephalogram signal acquisition module: responsible for collecting the original electroencephalogram signal through a multi-channel electroencephalogram acquisition device deployed on the driver's head, and performing preprocessing of baseline correction and normalization on the data; Feature extraction module: perform multi-dimensional feature extraction on the preprocessed electroencephalogram signal, including calculating the power spectral density of each channel, extracting the energy values and proportions of different frequency bands, performing wavelet decomposition to extract the time-frequency energy distribution and calculating the variance, and calculating the electroencephalogram asymmetry index; Mental state classification module: input the extracted feature vectors into a pre-trained mental state classification model using the support vector machine (SVM) architecture, output the probability distribution of the driver's current mental state category, and determine whether the emotional state is valid according to the set rules; Dynamic emotion threshold determination module: according to the real-time classification results and historical data, by establishing a sliding time window, calculate the mean value and standard deviation of the data within the window, define the emotion abnormality conditions and composite determination rules, and judge whether the emotion abnormality conditions are met; Multi-level intervention strategy execution module: trigger hierarchical control instructions according to the emotion category and abnormality level, and execute different levels of intervention operations.

Images