Intelligent In-vehicle Driving Assistance Method, Device and Equipment Based on Electroencephalogram Signals

Abstract

The present application relates to the technical field of brain-computer interfaces, and discloses an intelligent vehicle driving assistance method, device and equipment based on electroencephalogram signals. The method includes obtaining a target driving operation and corresponding target electroencephalogram data as a training set, and performing a benchmark comparison analysis on the target driving operation and the target electroencephalogram data to determine the electroencephalogram interaction range of each target driving operation, and generating corresponding heterogeneous features and inputting them into a preset interaction mapping table to determine the electroencephalogram interaction parameters under the preset driving operation index and the driving operation interaction parameters under the preset electroencephalogram index; generating a feature code of the heterogeneous features according to the electroencephalogram interaction parameters and the driving operation interaction parameters; obtaining the interaction mapping table filled with data according to the predefined driving operation description and electroencephalogram data description in the interaction mapping table; extracting the electroencephalogram interaction parameters and the driving operation interaction parameters in the interaction mapping table to obtain the current condition of the driver, and generating a vehicle auxiliary control instruction.

Description

Technical Field

[0001] This application relates to the technical field of brain-computer interfaces, and particularly to an intelligent in-vehicle driving assistance method, device, and equipment based on electroencephalogram (EEG) signals. Background Art

[0002] With the continuous development of automotive technology, driving assistance systems have become an important means to improve road safety. Traditional driving assistance systems mainly rely on external vehicle sensors, such as cameras, radars, and lidar, to perceive the surrounding environment and make corresponding decisions. In recent years, researchers have begun to focus on driver state monitoring technology based on electroencephalogram (EEG), attempting to incorporate the physiological and psychological states of drivers into the consideration of driving assistance systems. EEG signals can directly reflect the brain activities of drivers, including key factors such as attention, fatigue, and emotional states, all of which are closely related to driving safety.

[0003] However, existing driver state monitoring systems based on EEG signals have problems such as low standardization of data collection, single-dimensional feature extraction, and insufficient accuracy of state judgment. These systems lack reliable benchmark control analysis methods, making it difficult to establish a two-way mapping relationship between EEG activities and driving operations, and they also fail to fully consider the complex non-linear relationships between EEG features. At the same time, existing intelligent driving assistance systems mainly rely on preset rules to make decisions, lacking an in-depth understanding of the real-time state of drivers, making it difficult to adjust control strategies in a timely manner according to the dynamic changes of driver states, and unable to provide truly personalized and intelligent assistance.

[0004] Therefore, there is an urgent need for a method to solve at least one of the above problems. Summary of the Invention

[0005] Embodiments of this application provide an intelligent in-vehicle driving assistance method, device, and equipment based on EEG signals. The method aims to solve the problems of low standardization of data collection, single-dimensional feature extraction, and insufficient accuracy of state judgment in existing driver state monitoring systems based on EEG signals.

[0006] In a first aspect, embodiments of this application provide an intelligent in-vehicle driving assistance method based on EEG signals, including:

[0007] Obtain a target driving operation and corresponding target EEG data within a preset time period as a training set, and perform a benchmark control analysis on the target driving operation and the target EEG data;

[0008] According to the analysis result corresponding to the benchmark control analysis, determine the EEG interaction range of each target driving operation, and generate corresponding heterogeneous features;

[0009] Input the feature code corresponding to the inhomogeneous feature into a preset interactive mapping table to determine the EEG interaction parameters under preset driving operation indicators and the driving operation interaction parameters under preset EEG indicators;

[0010] According to the EEG interaction parameters and driving operation interaction parameters, generate the feature code of the inhomogeneous feature based on the feature coding generation model, and the feature codes of all inhomogeneous features are combined into a feature code set;

[0011] According to the predefined driving operation description and EEG data description in the interactive mapping table, split the feature code and insert it into the data unit of the interactive mapping table in sequence to obtain the interactive mapping table filled with data;

[0012] Extract the EEG interaction parameters and driving operation interaction parameters from the interactive mapping table according to the predefined driving operation category and EEG wave type;

[0013] Obtain the current condition of the driver according to the EEG interaction parameters and driving operation interaction parameters, and generate a vehicle auxiliary control instruction according to the current condition of the driver.

[0014] In a second aspect, the present application also provides an intelligent vehicle driving assistance device based on EEG signals, including:

[0015] An operation analysis module, configured to obtain a target driving operation and corresponding target EEG data within a preset time period as a training set, and perform a benchmark comparison analysis on the target driving operation and the target EEG data;

[0016] A result acquisition module, configured to determine the EEG interaction range of each target driving operation according to the analysis result corresponding to the benchmark comparison analysis, and generate a corresponding inhomogeneous feature;

[0017] A feature input module, configured to input the feature code corresponding to the inhomogeneous feature into a preset interactive mapping table to determine the EEG interaction parameters under preset driving operation indicators and the driving operation interaction parameters under preset EEG indicators;

[0018] A parameter acquisition module, configured to generate the feature code of the inhomogeneous feature based on the feature coding generation model according to the EEG interaction parameters and driving operation interaction parameters, and the feature codes of all inhomogeneous features are combined into a feature code set;

[0019] A feature coding module, configured to split the feature code according to the predefined driving operation description and EEG data description in the interactive mapping table, and insert it into the data unit of the interactive mapping table in sequence to obtain the interactive mapping table filled with data;

[0020] An operation locking module, configured to extract electroencephalogram (EEG) interaction parameters and driving operation interaction parameters from the interaction mapping table according to predefined driving operation categories and EEG types;

[0021] An instruction generation module, configured to obtain the current condition of the driver according to the EEG interaction parameters and driving operation interaction parameters, and generate a vehicle auxiliary control instruction according to the current condition of the driver.

[0022] In a third aspect, the present application further provides a computer device, including a processor and a memory, where the memory is used to store a computer program, and when the computer program is executed by the processor, the intelligent vehicle driving assistance method based on EEG signals as described in the first aspect is implemented.

[0023] This method is an intelligent driving assistance system that combines EEG signal analysis and vehicle control. Its core lies in establishing a dynamic mapping relationship between EEG signals and driving operations. By collecting driving operations (such as braking, steering) within a preset time period and synchronized EEG signals (such as waveforms related to attention and fatigue), a training set is formed and a benchmark comparison is performed to determine the EEG characteristics in normal and abnormal driving states. By analyzing the correlation between EEG signals and driving operations, features with spatio-temporal heterogeneity (such as changes in EEG energy in specific frequency bands) are extracted, and codes are generated to distinguish different driving scenarios. Using the feature codes, EEG parameters (such as the intensity of alpha waves and beta waves) are associated with driving operation parameters (such as the steering wheel angle and braking pressure) to form a two-way mapping relationship table, which supports real-time two-way parameter retrieval. Decoding the driver's state (such as distraction, fatigue) according to the real-time EEG signal, and matching the preset vehicle assistance strategies (such as automatic deceleration, lane keeping) through the mapping table to generate control instructions.

[0024] Triggering active intervention by predicting the driver's state (such as fatigue or distraction) to reduce the accident risk. Optimizing the driving assistance parameters according to individual EEG characteristics to improve the system adaptability. The non-homogeneous feature coding and mapping table mechanism shortens the data processing delay and meets the real-time requirements of driving scenarios. Combining physiological signals and mechanical control to promote the technological development of human-vehicle collaborative driving.

[0025] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic flowchart of the intelligent vehicle driving assistance method based on EEG signals shown in the embodiments of the present application;

[0027] Figure 2 It is a schematic structural diagram of the intelligent vehicle driving assistance device shown in the embodiments of the present application;

[0028] Figure 3 The structural schematic diagram of the computer device shown in the embodiments of the present application. Detailed implementation manners

[0029] In the following description, specific details such as specific system architectures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present application. However, those skilled in the art should understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present application.

[0030] It should be understood that when used in the specification of the present application and the appended claims, the term "comprising" indicates the presence of the described features, wholes, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0031] It should also be understood that the term "and / or" used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0032] As used in the specification of the present application and the appended claims, the term "if" can be interpreted as "when", "once", "in response to determining", or "in response to detecting" according to the context. Similarly, the phrase "if determined" or "if [the described condition or event] is detected" can be interpreted as meaning "once determined", "in response to determining", "once [the described condition or event] is detected", or "in response to detecting [the described condition or event]" according to the context.

[0033] In addition, in the description of the specification of the present application and the appended claims, the terms "first", "second", "third", etc. are only used for differentiating descriptions and cannot be understood as indicating or implying relative importance.

[0034] The reference to "one embodiment" or "some embodiments" etc. described in the specification of the present application means that specific features, structures, or characteristics described in connection with the embodiment are included in one or more embodiments of the present application. Thus, statements such as "in one embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments", etc. that appear in different places in this specification do not necessarily all refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "comprising", "including", "having", and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0035] The technical solutions of the embodiments of the present application will be introduced below.

[0036] With the continuous development of automotive technology, driver assistance systems have become an important means to improve road safety. Traditional driver assistance systems mainly rely on external vehicle sensors, such as cameras, radars, and lidars, to perceive the surrounding environment and make corresponding decisions. In recent years, researchers have begun to focus on electroencephalogram (EEG)-based driver state monitoring technology, attempting to incorporate the physiological and psychological states of drivers into the consideration of driver assistance systems. EEG signals can directly reflect the brain activities of drivers, including key factors such as attention, fatigue level, and emotional state, all of which are closely related to driving safety.

[0037] However, existing EEG-based driver state monitoring systems have problems such as low standardization of data collection, single-dimensional feature extraction, and insufficient accuracy of state judgment. These systems lack reliable benchmark comparison analysis methods, making it difficult to establish a two-way mapping relationship between brain electrical activities and driving operations, and also failing to fully consider the complex non-linear relationships between EEG features. At the same time, existing intelligent driver assistance systems mainly rely on preset rules to make decisions, lacking a deep understanding of the real-time state of drivers, making it difficult to adjust control strategies in a timely manner according to the dynamic changes of driver states, and unable to provide truly personalized and intelligent assistance.

[0038] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of an intelligent in-vehicle driving assistance method based on EEG signals provided by the embodiments of the present application. The intelligent in-vehicle driving assistance method based on EEG signals of the embodiments of the present application can be applied to computer devices, which include but are not limited to devices such as smartphones, laptop computers, tablet computers, desktop computers, physical servers, and cloud servers. As Figure 1 shown, the intelligent in-vehicle driving assistance method based on EEG signals of this embodiment includes steps S101 to S107, which are described in detail as follows:

[0039] Step S101, obtain the target driving operations and corresponding target EEG data within a preset time period as a training set, and perform benchmark comparison analysis on the target driving operations and target EEG data.

[0040] Specifically, first, based on the requirements of scientificity and representativeness, no less than 30 drivers aged between 25 and 50 years old, with a driving license for more than 3 years and corrected vision not lower than 1.0, are selected. During a preset time period (usually 30 consecutive days), their driving operations and electroencephalogram (EEG) data are synchronously collected. To ensure the diversity and representativeness of the data, the collection process needs to cover different driving environments (such as urban roads, highways, rural roads, with each type of road accumulating no less than 2 hours), weather conditions (sunny days, rainy days, foggy days, etc., with each condition accumulating no less than 1 hour), driving periods (6:00 - 10:00 in the morning, 10:00 - 14:00 at noon, 14:00 - 18:00 in the evening, 18:00 - 22:00 at night, with each period accumulating no less than 1 hour), and driver states (fully rested, mildly fatigued, severely fatigued, etc.).

[0041] In some embodiments, before performing the benchmark comparison analysis on the target driving operation and the target EEG data, it further includes: for the driving operation data, identifying and removing outliers based on the 3σ criterion, performing smoothing processing using a moving average filter, and performing Z - score normalization processing; for the target EEG data, using notch filtering to remove power frequency interference, using independent component analysis to remove artifacts such as electrooculogram (EOG) and electromyogram (EMG), and performing band - pass filtering to retain the main EEG activity frequency bands, ensuring that the signal - to - noise ratio of the target EEG data is greater than a preset signal - to - noise ratio; using a high - precision timestamp to accurately align the driving operation data and the target EEG data in time.

[0042] For the target driving operation data, a high - precision sensor (the steering wheel angle sensor has an accuracy better than 0.1°, and the pedal pressure sensor has an accuracy better than 0.1 N) is used to record various operations of the driver, including the steering angle and speed of the steering wheel, the position and pressure of the accelerator pedal, the position and pressure of the brake pedal, gear - shifting operations, the use of turn signals, and vehicle speed changes. The sampling frequency is set to 100 times per second to ensure capturing subtle operation changes. At the same time, a 64 - lead electroencephalogram (EEG) device is used to collect the EEG signals of the driver, with a sampling frequency of 1000 Hz, a signal - to - noise ratio not lower than 60 dB, and a frequency band range of 0.1 - 100 Hz, recording EEG wave types such as delta waves (0.5 - 4 Hz), theta waves (4 - 8 Hz), alpha waves (8 - 13 Hz), beta waves (13 - 30 Hz), and gamma waves (30 - 100 Hz). All devices are time - synchronized using GPS time - keeping or the Network Time Protocol (NTP) to ensure that the synchronization error is controlled within 1 ms.

[0043] After data acquisition, preprocessing is performed to improve data quality. For driving operation data, the 3σ criterion is used to identify and remove outliers, and a moving average filter with a window size of 5 sampling points is used for smoothing, followed by Z-score normalization. For EEG data, a 50Hz (or 60Hz) notch filter is used to remove power frequency interference, and the independent component analysis (ICA) method is used to remove artifacts such as electrooculogram and electromyogram. A band-pass filter of 0.5 - 45Hz is used to retain the main EEG activity frequency band, ensuring that the signal-to-noise ratio is greater than 20dB. To ensure data synchronization, high-precision timestamps are used to accurately align the driving operation data and EEG data in time, with the error controlled within 1ms. At the same time, an integrity check is performed to ensure no data loss.

[0044] In some embodiments, the benchmark comparison analysis of the target driving operation and target EEG data includes: obtaining the driving events corresponding to the driving operation data; the driving events at least include straight driving, acceleration, deceleration, and emergency braking; according to the EEG data segments corresponding to the driving events in the target EEG data, time domain features, frequency domain features, and time-frequency features are extracted; and benchmark comparison analysis is performed on the driving events and the corresponding time domain features, frequency domain features, and time-frequency features.

[0045] Next, the continuous driving operation data is divided into discrete driving events. Straight driving is defined as the steering wheel angle change being less than 5° and lasting for more than 3 seconds, turning is defined as the steering wheel angle change being greater than 30° and lasting for more than 2 seconds, acceleration is defined as the throttle pedal pressure increasing by more than 20% and lasting for more than 1 second, deceleration is defined as the brake pedal pressure exceeding 10% and lasting for more than 0.5 seconds, and emergency braking is defined as the brake pedal pressure exceeding 80% and the deceleration being greater than 5m / s².

[0046] Exemplarily, the benchmark comparison analysis of the driving events and the corresponding time domain features, frequency domain features, and time-frequency features includes: constructing EEG features based on the time domain features, frequency domain features, and time-frequency features; calculating the average EEG feature vector and covariance matrix of the EEG features corresponding to the driving events; performing variance analysis based on the average EEG feature vector and covariance matrix to determine abnormal EEG features; the abnormal EEG features are those with significant differences among multiple driving events; and using principal component analysis or linear discriminant analysis to reduce the feature dimension of the abnormal EEG features to generate the analysis results corresponding to the benchmark comparison analysis.

[0047] For each EEG data segment corresponding to a driving event, extract time-domain features (such as mean amplitude, standard deviation, skewness, kurtosis, zero-crossing rate), frequency-domain features (such as energy of each frequency band, frequency band energy ratio, dominant frequency, power spectral density), and time-frequency features (such as wavelet transform coefficients, instantaneous frequency obtained by Hilbert-Huang transform). Then, conduct statistical analysis on each type of driving event and the corresponding EEG features, calculate the average EEG feature vector and covariance matrix, and perform analysis of variance (ANOVA) to determine which EEG features have significant differences among different driving events. Use principal component analysis (PCA) or linear discriminant analysis (LDA) to reduce the feature dimension and extract the most discriminative feature combination.

[0048] Based on the analysis results, establish a preliminary association model between driving operations and EEG features. For continuous driving operation parameters, use a multiple linear regression model: where Y is the driving operation parameter (such as steering wheel angle), are the selected EEG features. For discrete driving events, use logistic regression for classification: where Y is the driving event category. Calculate the goodness of fit of the model, such as the R² value, AIC (Akaike information criterion), etc., to evaluate the model performance.

[0049] Finally, conduct a comprehensive quality assessment on the collected dataset, checking data integrity (ensuring that the missing rate is controlled within 5%), consistency (ensuring that data from different drivers and different time periods are comparable), representativeness (evaluating whether the dataset covers a sufficient variety of driving scenarios and driver states), and signal-to-noise ratio (ensuring that the signal-to-noise ratio of the EEG data is greater than 20 dB). After the assessment, divide the processed dataset into a training set (70%), a validation set (15%), and a test set (15%), ensuring that each subset contains an equal number of samples of each type of driving event. Through benchmark comparison analysis, the characteristic statistics, significance test results, and preliminary mathematical association models of each type of driving event are obtained, and these analysis results clearly show the basic correspondence between different driving operations and EEG features.

[0050] Step S102, according to the analysis results corresponding to the benchmark comparison analysis, determine the EEG interaction range of each target driving operation and generate corresponding heterogeneous features.

[0051] Specifically, based on the comparative analysis results obtained in the above steps, the EEG characteristics of each driving operation are analyzed in depth. For each driving operation (such as straight driving, turning, accelerating, decelerating, etc.), statistical measures such as the mean, standard deviation, and quartiles of its corresponding EEG characteristics under this operation are calculated and compared with the baseline state (such as the resting state). Taking the "turning" operation as an example, the analysis finds that it mainly causes the following characteristic changes: 200 ms before the start of turning, the energy of the α wave (8 - 13 Hz) in the occipital lobe region begins to increase, and the increase amplitude reaches 30% ± 5% when it reaches the peak; within 100 ms after the start of turning, the energy of the β wave (13 - 30 Hz) in the frontal and parietal lobe regions rises rapidly and maintains at a level of increasing by 50% ± 8%; at the same time, the energy ratio of the θ wave (4 - 8 Hz) to the β wave within the whole brain range continues to decrease, and finally decreases by 20% ± 3%; and during the whole turning process, the α wave coherence between the frontal lobe and the parietal lobe increases by 40% ± 6%, and this enhanced coherence lasts until 300 ms after the completion of turning.

[0052] To standardize these changes and facilitate cross - feature and inter - individual comparisons, the Z - score is used for feature standardization. For feature i, its Z - score calculation formula is: Z_i = (X_i - μ_i) / σ_i, where X_i is the observed value, μ_i is the mean of this feature, and σ_i is the standard deviation. The EEG interaction range is defined as the 95% confidence interval of the Z - score, that is, [-1.96, 1.96]. Any EEG activity falling within this range is considered to have a significant correlation with this driving operation. For specific scenarios that require higher - sensitivity detection (such as emergency braking), the confidence interval can be adjusted to 90%, and the corresponding Z - score range is [-1.645, 1.645].

[0053] After determining the EEG interaction range, heterogeneous features are generated through the following five analysis methods:

[0054] The first category is time - frequency joint analysis. The continuous wavelet transform is used to perform time - frequency decomposition on the EEG signal. The Morlet wavelet is selected as the mother wavelet, the scale range is set to 1 - 128 to cover the frequency range of 0.5 - 100 Hz, and the time resolution is 2 ms. Taking emergency braking as an example, the time - frequency diagram of the prefrontal θ - wave energy is extracted, and the energy change curve within 0 - 500 milliseconds after the start of braking is mainly analyzed, establishing a θ - wave energy change map from the generation of braking intention to the actual execution of the braking operation, and capturing the feature of the increase in θ - wave energy starting 200 ms before braking.

[0055] The second category is spatial distribution analysis, which calculates the spatial distribution pattern of EEG activities in specific frequency bands across the whole brain. Using the spherical spline interpolation method, the spatial resolution is set to 2 cm, and a topographic map of 64 channels of the whole brain is constructed. For example, a whole-brain topographic map analysis of the β-wave energy during the braking process is carried out, and a sequence of continuous topographic maps is calculated at intervals of 20 ms. The activation pattern of the motor cortex area is extracted with emphasis. It is found that during the period from 150 ms before the braking operation to 300 ms after the operation, the β-wave energy in the motor cortex area shows a process of enhancement from local to diffusion.

[0056] The third category is functional connectivity analysis, which calculates the interaction strength between different brain regions. Specifically, a brain network based on θ-wave coherence is constructed, the wavelet coherence algorithm is used, and the coherence threshold is set to 0.6. The clustering coefficient and path length of the network are calculated. Through the analysis of the continuous network topological changes within a 500-ms time window, the dynamic changes of information flow between different brain regions during the driving operation are reflected.

[0057] The fourth category is complexity index analysis, which quantifies the complexity of EEG signals by calculating indexes such as sample entropy and Lempel-Ziv complexity. For sample entropy calculation, the embedding dimension m = 2, the tolerance r = 0.2 × standard deviation, the time window length is 200 ms, and the sliding step is 50 ms. By comparing the complexity indexes of different time windows, the changing law of the cognitive complexity of the driver at different operation stages is reflected.

[0058] The fifth category is cross-frequency coupling analysis, which studies the interaction between different frequency bands, such as phase-amplitude coupling (PAC) or phase-phase coupling (PPC). The Hilbert transform is used to extract the phase information. The time window for phase-amplitude coupling analysis is set to 400 ms, and the overlap rate is 50%. For example, the modulation index between the θ-wave phase in the prefrontal lobe and the γ-wave amplitude is calculated, and a time-varying curve of the coupling strength between different frequency bands is established, reflecting the changing characteristics of the cognitive state of the driver during complex operations.

[0059] Taking the "emergency braking" operation as an example, its EEG interaction range is determined as follows: the Z-score range of prefrontal theta wave energy is [1.2, 2.5], the Z-score range of motor cortex beta wave energy is [1.8, 3.0], the Z-score range of prefrontal-parietal theta wave coherence is [1.5, 2.8], and the Z-score range of whole-brain sample entropy is [-0.5, -2.0]. Based on these ranges, the time evolution curve of theta wave energy, the spatial distribution map of beta wave energy, the theta wave coherence network characteristics, the change rate curve of sample entropy, and the theta-gamma phase-amplitude coupling index are generated. These heterogeneous features form a multi-dimensional feature vector F = [f_theta_evolution, f_beta_distribution, f_theta_coherence, f_sample_entropy, f_theta_gamma_coupling], where each component represents the feature generated by the corresponding method.

[0060] In some embodiments, before inputting the feature encoding corresponding to the heterogeneous features into a preset interaction mapping table, it further includes: constructing a deviation distribution map of the target driving operation according to the EEG interaction range, including distribution columns corresponding to the target EEG data and driving deviation types; comparing the deviation distribution map with the standard driving operation distribution map, determining the amplitude change amount of each distribution column, and screening the minimum amplitude change amount from the amplitude change amounts to correct the remaining each amplitude change amount; re-determining the EEG interaction range based on the corrected deviation distribution map; obtaining the feature expression of the corrected EEG interaction range and inputting the feature expression into a preset heterogeneous analysis model to obtain optimized heterogeneous features.

[0061] Based on the EEG interaction range determined in the above steps, construct a deviation distribution map corresponding to the target driving operation, where the deviation distribution map includes distribution columns in one-to-one correspondence with the EEG data and driving deviation types; compare the deviation distribution map with the standard driving operation distribution map, determine the amplitude change amount of each distribution column, and screen the minimum amplitude change amount from all the amplitude change amounts to correct the remaining each amplitude change amount in turn; re-determine the EEG interaction range corresponding to the target driving operation based on the corrected distribution map; obtain the feature expression of the corrected EEG interaction range and input it into the heterogeneous analysis model to obtain optimized heterogeneous features;

[0062] Based on the multi-dimensional feature vector F = [f_θ_evolution, f_β_distribution, f_θ_coherence, f_sample_entropy, f_θγ_coupling] obtained from the above steps, a deviation distribution map is constructed for each driving operation. In a complex road condition that requires continuous steering, such as a mountain road driving scenario, this deviation distribution map, as a visualization tool, shows the dynamic changes of various EEG features during the steering process. When the driver performs a steering operation in a continuous curve, each distribution column represents a specific EEG feature, such as the θ-wave energy of a specific brain region, the spatial distribution of β-waves, etc. The height of the distribution column represents the degree of deviation of this feature at different steering stages. Each distribution column is also associated with a specific type of driving deviation, such as "steering timing delay", "steering angle deviation", etc. Each component of the feature vector is converted into a corresponding distribution column: when the driver observes the upcoming curve and prepares to steer, the f_θ_evolution component is converted into a prefrontal θ-wave energy deviation of +2.3 (corresponding to the "spatial attention allocation" state); when the arm starts to turn the steering wheel, the f_β_distribution component is converted into a motor cortex β-wave energy deviation of +1.9 (corresponding to the "fine motor control" state); during the cornering process, the f_θ_coherence component is converted into a prefrontal-parietal θ-wave coherence deviation of +2.1 (corresponding to the "spatial navigation load" state); when the steering is completed, the f_sample_entropy component is converted into a whole-brain sample entropy deviation of -1.8 (corresponding to the "action coordination simplification" state); and when the driver evaluates the steering effect, the f_θγ_coupling component is converted into a θ-γ coupling strength deviation of +1.7 (corresponding to the "operation result evaluation" state).

[0063] Compare the constructed deviation distribution map with the standard driving operation distribution map. During standard driving operations such as making minor path corrections on a daily straight road, each EEG index shows a relatively stable baseline state. Through comparison, calculate the amplitude change amount of each distribution column. The amplitude change amount (ΔA) is calculated using the formula: ΔA = (A_observed - A_standard) / A_standard, where A_observed is the observed amplitude value and A_standard is the standard amplitude value. When the driver steers during a continuous curve, the EEG activities related to attention and motion control show significant changes compared to the standard state. For example, for the f_θ_evolution component, when the Z-score of the prefrontal θ-wave energy is usually around 1.0 during standard driving operations, and the observed value during continuous steering operations is 2.3, the calculated result of the amplitude change amount is: (2.3 - 1.0) / 1.0 = 1.3, indicating that in the case of continuous steering, this EEG feature is 130% higher than during normal driving. Similarly, for the other components of the eigenvector F, namely f_β_distribution, f_θ_coherence, f_sample_entropy, and f_θγ_coupling, the calculated amplitude change amounts are 0.9, 1.1, -0.8, and 1.0 respectively.

[0064] To eliminate the influence of natural fluctuations that may occur during steering and highlight key changes, correct the calculated amplitude change amounts. First, screen out the minimum value (ΔA_min) from the amplitude change amounts corresponding to all components of the eigenvector F. During continuous steering operations, the minimum amplitude change amount is -0.8 (-80%) corresponding to the f_sample_entropy component. Then use the formula ΔA_corrected = ΔA - ΔA_min to correct all other change amounts. This correction method ensures that natural fluctuations during driving are regarded as baseline noise, and significant changes caused by steering operations are evaluated relative to this baseline. After correcting each component, the corrected amplitude change amounts are: for the f_θ_evolution component, 1.3 - (-0.8) = 2.1; for the f_β_distribution component, 0.9 - (-0.8) = 1.7; for the f_θ_coherence component, 1.1 - (-0.8) = 1.9; for the f_sample_entropy component, -0.8 - (-0.8) = 0; for the f_θγ_coupling component, 1.0 - (-0.8) = 1.8.

[0065] Based on these corrected amplitude change amounts, reconstruct the deviation distribution map, and update the value ranges of the components of the feature vector F accordingly. For continuous steering operations, the new value ranges are updated as follows: the range of the f_θ_evolution component is [1.8, 3.1], which reflects the continuous attention allocation requirements during steering; the range of the f_β_distribution component is [1.5, 2.7], which reflects the fine steering wheel control requirements; the range of the f_θ_coherence component is [1.7, 2.9], which indicates the cognitive load of spatial navigation; the range of the f_sample_entropy component is [-0.2, -1.7], which shows the coordination of the action pattern; the range of the f_θγ_coupling component is [1.6, 2.8], which reflects the continuous operation evaluation requirements. This corrected range more accurately reflects the actual change characteristics of each feature component relative to normal driving during continuous steering operations.

[0066] Use these corrected value ranges as feature representations and re-enter them into the heterogeneous analysis model. This model adopts a deep neural network structure, including multiple convolutional layers and fully connected layers, to capture the non-linear relationships between feature components. By analyzing the dynamic change laws of each feature during continuous steering, the optimized heterogeneous feature F' = [f'_θ_evolution, f'_β_distribution, f'_θ_coherence, f'_sample_entropy, f'_θγ_coupling] is finally obtained. This optimized feature vector better describes the cognitive load changes, action control characteristics, and attention allocation patterns of the driver during continuous steering, providing a reliable feature index for real-time evaluation of the driver's operation status.

[0067] Step S103: Input the feature encoding corresponding to the heterogeneous feature into a preset interactive mapping table to determine the electroencephalogram interaction parameters under the preset driving operation indicators and the driving operation interaction parameters under the preset electroencephalogram indicators.

[0068] Specifically, feature encoding is generated for the optimized heterogeneous features F' = [f'_θ_evolution, f'_β_distribution, f'_θ_coherence, f'_sample_entropy, f'_θγ_coupling] obtained in the above steps. Based on the high-dimensional vector encoding method, each feature is encoded as a 100-dimensional vector, which can preserve the complexity and nuances of the features. In a continuous turning scenario, taking the cornering operation as an example, for the feature f'_θ_evolution that reflects attention allocation, the generated encoding vector form is: E_θβ = [0.75, 1.2, 0.9, ..., 1.1, 0.8]. Each dimension of this encoding vector corresponds to different aspects of the attention allocation state, such as features like the intensity, duration, and transfer speed of spatial attention.

[0069] To process these high-dimensional feature encodings and generate interaction parameters, a multi-input multi-output deep neural network structure is constructed. The input layer of the network contains 100 nodes, corresponding to an optimized heterogeneous feature encoding; the hidden layer contains 4 fully connected layers, with 512 nodes in each layer, using the ReLU activation function; the output layer is divided into two branches, respectively outputting driving operation interaction parameters and EEG interaction parameters. For example, in a continuous turning scenario, when the driver enters a continuous curve, the input layer receives the encoded attention allocation feature, and after being processed by multiple layers of the network, the output layer can simultaneously predict the driver's steering operation trend and the corresponding change in brain cognitive load. The mathematical expression of the neural network is: H = ReLU(W_4 * ReLU(W_3 * ReLU(W_2 * ReLU(W_1 * X + b_1) + b_2) + b_3) + b_4), Y_drive = W_drive * H + b_drive, Y_eeg = W_eeg * H + b_eeg, where X is the input feature encoding, W_i and b_i are the weight matrices and bias vectors of each layer, H is the output of the last hidden layer, Y_drive is the driving operation interaction parameter, and Y_eeg is the EEG interaction parameter.

[0070] In actual driving scenarios, the system focuses on five key driving operation indicators and five key electroencephalogram (EEG) indicators. The driving operation indicators include: Steering Angle (SA), which reflects the amplitude and accuracy of steering; Accelerator Pedal Position (AP), which embodies speed control during cornering; Brake Force (BF), which indicates speed adjustment during corner entry and exit; Speed Control Precision (SCP), which shows the smoothness of vehicle speed; and Lane Keeping Ability (LKA), which reflects the accuracy of the steering trajectory. The corresponding EEG indicators include: Frontal Theta Wave Power (FTP), which reflects the level of spatial attention; Motor Cortex Beta Wave Power (MCBP), which indicates the intensity of action control; Parietal Alpha Wave Power (PAP), which embodies visual-spatial processing; Frontal-Parietal Theta Wave Coherence (FPTC), which shows the spatial cognitive load; and Frontal Theta-Gamma Phase-Amplitude Coupling (FTGPAC), which reflects high-level cognitive processing.

[0071] The neural network parameters are optimized through a large amount of training data from steering scenarios, enabling it to accurately map the relationship between heterogeneous features and interaction parameters. For example, during a continuous S-shaped curve driving, the system simultaneously records the driver's EEG activity data and vehicle operation data. The outputs Y_drive and Y_eeg of the neural network each form a 5-dimensional vector, which represents the influence degree of input features on these indicators in real time. To comprehensively consider the influence of multiple heterogeneous features, the interaction parameters generated by all features are weighted and averaged, and the weights are determined according to the feature importance.

[0072] Finally, two interaction parameter matrices are obtained: the driving operation interaction parameter matrix M_drive (5x5) and the EEG interaction parameter matrix M_eeg (5x5). The value range of these interaction parameters is [-1, 1]. A positive value indicates a positive correlation, a negative value indicates a negative correlation, and the larger the absolute value, the stronger the correlation. In the continuous steering scenario, the specific manifestations are as follows: when the driver is about to enter a sharp turn, there is a significant positive correlation between the steering wheel steering angle and the energy of the beta wave in the motor cortex (M_drive[SA][MCBP] = 0.72), which reflects that the driver's motion control system is actively preparing to execute the steering operation. At the same time, there is also a strong positive correlation between the lane keeping ability and the coherence of the theta wave in the prefrontal-parietal region (M_drive[LKA][FPTC] = 0.85), indicating that the driver's spatial cognitive system is making full efforts to maintain the correct steering trajectory. In terms of vehicle speed control, it is found that there is a moderate negative correlation between the vehicle speed control accuracy and the energy of the alpha wave in the parietal lobe (M_drive[SCP][PAP] = -0.63), which may reflect the impact of increased visual spatial processing load on speed control. From the feedback of EEG activities, when the driver performs a large steering action, the energy of the beta wave in the motor cortex increases significantly with the increase of the steering angle (M_eeg[MCBP][SA] = 0.68). When keeping the vehicle driving stably, there is a high positive correlation between the coherence of the theta wave in the prefrontal-parietal region and the lane keeping ability (M_eeg[FPTC][LKA] = 0.79), indicating that accurate lane keeping requires close cooperation of different regions of the brain. During the acceleration process, it is observed that there is a moderate negative correlation between the energy of the alpha wave in the parietal lobe and the position of the accelerator pedal (M_eeg[PAP][AP] = -0.57), which reflects that the acceleration operation may reduce the processing intensity of visual spatial information.

[0073] Step S104, according to the EEG interaction parameters and driving operation interaction parameters, generate the feature codes of non-homogeneous features based on the feature coding generation model, and the feature codes of all non-homogeneous features are combined into a feature code set.

[0074] Specifically, the core idea of the feature coding generation model is to convert non-homogeneous features into numerical vectors of fixed length while retaining the key information and interaction relationships of the features. A deep learning model based on an autoencoder is used to achieve this goal. The autoencoder model structure includes an input layer, an encoder, a bottleneck layer, a decoder, and an output layer. In the continuous steering scenario, when inputting a non-homogeneous feature (such as the theta wave energy time series feature reflecting attention allocation), the model first compresses it to the bottleneck layer through the encoder to form a feature code, and then tries to reconstruct the original input through the decoder.

[0075] The mathematical expressions of the model are: encoding process \(E = f_e(X; \theta_e)\), decoding process \(X' = f_d(E; \theta_d)\), where \(X\) is the input heterogeneous feature, \(E\) is the generated feature encoding, \(X'\) is the reconstructed feature, \(f_e\) and \(f_d\) are the encoder and decoder functions respectively, and \(\theta_e\) and \(\theta_d\) are the corresponding parameters. To incorporate the interaction parameters obtained in the above steps into the feature encoding generation process, a regularization term is introduced into the loss function of the autoencoder: \(L = MSE(X, X')+\lambda*R(E, M_drive, M_eeg)\), where \(MSE(X, X')\) is the reconstruction error, \(R(E, M_drive, M_eeg)\) is the regularization term based on the interaction parameters, and \(\lambda\) is the balancing parameter.

[0076] The regularization term \(R\) is designed as follows: \(R(E, M_drive, M_eeg)=\sum|corr(E_i, E_j)-(M_drive[i][j]+M_eeg[j][i]) / 2|\), where \(corr(E_i, E_j)\) is the correlation coefficient between the \(i\)-th and \(j\)-th dimensions of the feature encoding \(E\), and \(M_drive[i][j]\) and \(M_eeg[j][i]\) are the corresponding driving operation interaction parameter and EEG interaction parameter respectively. The design of this regularization term ensures that the correlation between the dimensions of the feature encoding is consistent with the interaction parameter matrix obtained in the above steps. For example, in the steering scenario, if \(M_drive[SA][MCBP]=0.72\) indicates a strong correlation between the steering wheel steering angle and the energy of the beta wave in the motor cortex, then the correlation between the corresponding dimensions in the feature encoding should also be close to this value.

[0077] Taking the heterogeneous feature of the dynamic ratio of theta wave energy to beta wave energy as an example, it has 100 sampling points in the time dimension and 64 EEG electrodes in the spatial dimension. The original feature representation is a \(100\times64\) matrix. The specific structure of the autoencoder is: input layer with 6400 nodes (\(100 * 64\)), the encoder compresses through layers of \(3200\rightarrow1600\rightarrow800\rightarrow400\rightarrow200\) nodes, bottleneck layer with 128 nodes (feature encoding), the decoder reconstructs through layers of \(200\rightarrow400\rightarrow800\rightarrow1600\rightarrow3200\) nodes, and output layer with 6400 nodes. In the steering scenario, when the driver performs a sharp turning maneuver, this feature matrix is encoded into a 128-dimensional vector: \(E = [0.75, -0.23, 0.91,\cdots,0.45, -0.62]\). This encoding not only preserves the spatio-temporal pattern of the original feature, but also through the constraint of the regularization term, makes the correlation between its dimensions reflect the interaction relationship between driving operations and EEG activities.

[0078] Repeat this encoding process for all heterogeneous features. For example, for the spatial distribution features of β-wave energy, θ-wave coherence network features, sample entropy change rate features, and θ-γ phase-amplitude coupling features, their corresponding 128-dimensional feature encodings are generated respectively. Finally, these feature encodings are combined into a feature encoding set S = {E_1, E_2, ..., E_10}, where each E_i is a 128-dimensional vector. This feature encoding set S not only realizes the dimensionality reduction representation of the original heterogeneous features, but more importantly, by integrating the interaction parameter information of the above steps, it retains the complex interaction relationship between driving operations and EEG activities.

[0079] Step S105, according to the predefined driving operation descriptions and EEG data descriptions in the interaction mapping table, split the feature encoding and insert it into the data units of the interaction mapping table in sequence to obtain the interaction mapping table filled with data.

[0080] Specifically, first, the structure of the interaction mapping table needs to be clarified. It contains predefined driving operation descriptions and EEG data descriptions, forming a 5x5 structure, corresponding to 5 driving operation indicators (steering wheel steering angle, accelerator pedal position, brake pedal force, vehicle speed control accuracy, and lane keeping ability) and 5 EEG data indicators (prefrontal θ-wave energy, motor cortex β-wave energy, parietal α-wave energy, prefrontal-parietal θ-wave coherence, and prefrontal θ-γ phase amplitude coupling) respectively.

[0081] Uniformly divide each 128-dimensional feature encoding vector in the feature encoding set S obtained in the above steps into 25 sub-vectors, and each sub-vector contains 5 elements. This division method ensures that each sub-vector can correspond to a cell in the mapping table. Calculate 5 key statistical features for each sub-vector: mean, standard deviation, maximum value, minimum value, and median. These statistical features can comprehensively describe the characteristics of the sub-vector, reflecting both the overall trend and capturing the extreme values and distribution. In a continuous steering scenario, taking the encoding E_1 as an example, its first sub-vector [0.75, -0.23, 0.91, 0.45, -0.62] after statistical feature calculation gives: mean 0.252, standard deviation 0.654, maximum value 0.91, minimum value -0.62, median 0.45.

[0082] After determining the statistical characteristics of the sub-vectors, it is necessary to decide how to map this information into the interaction mapping table. This step utilizes the interaction parameter matrices (M_drive and M_eeg) obtained in the above steps. The correlation between each sub-vector and the corresponding elements of the interaction parameter matrix is calculated, and this process helps to determine where each sub-vector should be mapped in the interaction mapping table. The Pearson correlation coefficient is used for the calculation of the correlation, and this method can effectively capture linear relationships and is insensitive to scale changes.

[0083] Based on the results of the correlation, the interaction mapping table starts to be filled. The statistical characteristics of each sub-vector are inserted into the corresponding positions. However, during this process, there may be a situation where multiple sub-vectors compete for the same mapping position. To solve this problem, a weighted average method based on the intensity of the interaction parameters is adopted. For example, when multiple sub-vectors are all related to the mapping unit of the steering wheel steering angle and the energy of the beta wave in the motor cortex, weights are assigned according to their correlation intensity with the interaction parameter (M_drive[SA][MCBP] = 0.72) corresponding to this position, and then the weighted average is calculated. This method ensures that the final filling result can reflect the contributions of all relevant features and also maintains consistency with the original interaction parameters.

[0084] During the filling process, it is also necessary to consider the temporal information that the feature encoding may carry. Although the feature encoding itself is a static vector, it may encode information about certain dynamic processes. To retain this information, the position of the sub-vector in the original 128-dimensional vector is considered during filling. The sub-vectors at the front may represent earlier information, while the sub-vectors at the back may represent more recent information. This consideration makes the filled mapping table not only reflect the spatial relationships but also retain the sequential information in time to a certain extent.

[0085] After the initial filling is completed, the entire mapping table needs to be normalized. This step uses the Z-score normalization method to transform each statistical characteristic into a standard normal distribution. The purpose of normalization is to ensure the comparability of the numerical values between different units and eliminate the biases that may be caused by different feature scales. In the normalized mapping table, each element represents the degree of deviation from the average level, making it easier for this application to identify abnormal or significant patterns.

[0086] After filling and normalization are completed, consistency checks are also required. The purpose of this step is to ensure that the filled mapping table is consistent with the interaction parameter matrix obtained in the above steps in terms of the overall trend. Calculate the correlation between the average value of each cell in the filled mapping table and the corresponding interaction parameter. If significant deviations are found, it is necessary to trace back and adjust the mapping process. Possible adjustments include reassigning the mapping positions of sub-vectors or adjusting the weight distribution of weighted averages.

[0087] Through this complex process, the abstract feature encoding is successfully transformed into a specific and interpretable mapping relationship. The finally obtained filled interaction mapping table is a three-dimensional structure of 5x5x5, containing 25 cells, and each cell contains 5 statistical features. This structure comprehensively describes the complex relationship between driving operations and EEG data, considering both their direct effects and the indirect associations reflected through heterogeneous features.

[0088] Step S106, extract EEG interaction parameters and driving operation interaction parameters from the interaction mapping table according to predefined driving operation categories and EEG types.

[0089] Specifically, perform the first round of parameter extraction on the 5x5x5 three-dimensional mapping table obtained in the above steps. Each cell of this mapping table contains 5 statistical features (mean, standard deviation, maximum value, minimum value, and median). The focus of the analysis is to determine the relationship between each driving operation and all EEG indicators. This process is actually "slicing" the three-dimensional mapping table, and each slice represents the influence of a driving operation on all EEG indicators. The parameter extraction uses the weighted average method, considering the importance of each statistical feature. The calculation formula is: P_EEG(i,j) = Σ(w_k * M[i,j,k]) / Σw_k, where P_EEG(i,j) represents the interaction parameter of driving operation i on EEG indicator j, M[i,j,k] is the k-th statistical feature at the corresponding position in the mapping table, and w_k is the weight of this statistical feature. Taking the sharp turn scenario as an example, when taking the "slice" in the mapping table that reflects the steering operation, it is found that the energy of the prefrontal theta wave shows an obvious change trend before the start of the turn, which corresponds to a relatively high interaction parameter value, reflecting the early mobilization of the driver's attention.

[0090] In some embodiments, the extracting of EEG interaction parameters and driving operation interaction parameters from the interaction mapping table according to predefined driving operation categories and EEG types includes: performing the first round of locking on the interaction mapping table according to the predefined driving operation categories to obtain EEG interaction parameters; performing the second round of locking on the interaction mapping table according to the predefined EEG types to obtain the corresponding driving operation interaction parameters; and obtaining optimized EEG interaction parameters and driving operation interaction parameters through cross-validation.

[0091] The determination of weights is a crucial step, adopting an assignment method based on importance ranking: the mean and median obtain relatively high weights (0.3 and 0.25) because they reflect the overall trend; the standard deviation obtains a medium weight (0.2) because it reflects the degree of variation; the maximum and minimum values obtain relatively low weights (0.125 each) because they may be affected by extreme values. This weight allocation shows good stability in practical applications. For example, in the continuous curve driving scenario, it can effectively capture the continuous impact of driving operations on EEG activities.

[0092] Through this process, a 5x5 EEG interaction parameter matrix P_EEG is obtained. Each element in the matrix quantifies the influence intensity of a certain driving operation on a specific EEG index. This matrix provides a perspective to observe EEG activities from the perspective of driving operations, revealing the EEG response patterns that may be triggered by different driving behaviors. For example, in the sharp turn operation, it is found that the steering operation has a strong positive correlation with the energy of theta waves in the prefrontal lobe (correlation coefficient 0.82), while having a slight negative correlation with the energy of alpha waves in the parietal lobe (correlation coefficient -0.35).

[0093] In the second round of parameter extraction, change the perspective and mainly determine the driving operation interaction parameters under each EEG index. Lock the mapping table for the second time and focus on analyzing the influence of each type of brain wave on all driving operation indexes. This process is to "slice" the three-dimensional mapping table in another direction. Each slice represents the influence of an EEG index on all driving operations. The parameter extraction also uses the weighted average method, and the calculation formula is: P_DRIVE(j,i) = Σ(w_k * M[i,j,k]) / Σw_k. Here, P_DRIVE(j,i) represents the interaction parameter of EEG index j on driving operation i, and the meanings of other parameters are the same as those in the first round. Taking the cornering scenario as an example, when analyzing the "slice" of beta wave activity in the mapping table, it is found that the change in beta wave energy in the motor cortex area is highly correlated with the accuracy of the steering operation, which reflects the predictive value of EEG activities for driving operations.

[0094] Through the verification of the results of the two rounds of parameter extraction, theoretically, P_EEG and P_DRIVE should be transpose matrices of each other. By calculating the differences between the corresponding elements of the two matrices, the reliability of parameter extraction is evaluated. If significant deviations (differences exceeding 0.1) are found, the original data and the extraction process need to be rechecked. For example, in a typical mountain road driving scenario, when the driver continuously passes through multiple sharp turns, this two-way parameter extraction process clearly demonstrates the mutual influence between driving operations and EEG activities. If significant differences are found between the parameters representing the influence of steering operations on beta waves in P_EEG and the parameters representing the influence of beta waves on steering operations in P_DRIVE, the weights need to be adjusted or the data at the corresponding positions in the mapping table need to be reexamined. The finally obtained optimized EEG interaction parameter matrix P_EEG and driving operation interaction parameter matrix P_DRIVE not only accurately quantify this two-way relationship but also ensure the reliability of the parameters through verification. For example, when the driver is about to enter a sharp turn, the P_EEG matrix shows a strong positive correlation between the steering operation and the energy of theta waves in the prefrontal lobe, indicating that the driver will significantly improve the level of spatial attention before turning; at the same time, a slight negative correlation with the energy of alpha waves in the parietal lobe is observed, reflecting the change in the load of visual spatial processing. From the perspective of the P_DRIVE matrix, when an increase in the energy of beta waves in the motor cortex area is detected, it often indicates an improvement in the accuracy of steering operations, which provides a reliable basis for predicting the driver's operation intention. Such a verified and optimized parameter matrix provides accurate quantitative indicators for subsequent driving state monitoring and auxiliary decision-making.

[0095] Step S107, obtain the current condition of the driver according to the EEG interaction parameters and driving operation interaction parameters, so as to generate a vehicle auxiliary control instruction according to the current condition of the driver.

[0096] Specifically, first, a real-time EEG data acquisition system is established. This system continuously acquires the driver's EEG signals at a sampling rate of 500 Hz, covering 32 EEG channels, including key brain regions such as the prefrontal lobe, parietal lobe, and occipital lobe. To ensure data quality, the system integrates a real-time signal quality detection mechanism and uses the threshold method to identify and mark possible artifacts. When the signal amplitude exceeds 100 μV, the system will mark it as a possible blink artifact; when the high-frequency component of the signal exceeds 5 times the standard deviation, the system will mark it as a possible myoelectric interference.

[0097] In some embodiments, the obtaining the current condition of the driver according to the EEG interaction parameters and driving operation interaction parameters includes: obtaining the driver's EEG data collected in real time; using a pattern recognition algorithm to extract features from the driver's EEG data, and combining the optimized EEG interaction parameters and driving operation interaction parameters to obtain the current condition of the driver.

[0098] The collected raw EEG data is immediately preprocessed. The first step of preprocessing is band-pass filtering. A 4th-order Butterworth filter is used with a frequency band set to 0.5 - 45 Hz. This step effectively removes high-frequency noise and baseline drift, improving the signal-to-noise ratio of the signal. Next, the common spatial pattern (CSP) method is applied for spatial filtering. The CSP method can maximize the variance difference between different category signals, thereby enhancing the separability of the signals. In this system, the first 5 spatial filters with the largest eigenvalues are selected, which reduces the dimension of the data while retaining the key information. Finally, the continuous EEG signals are divided into 2-second time windows with an overlap rate of 50% between windows. This division method ensures both time resolution and provides enough data points for feature extraction.

[0099] After preprocessing, multi-level feature extraction is carried out. In the time domain, the statistical features of each channel are calculated, including mean, standard deviation, skewness, and kurtosis. These features reflect the basic statistical properties of the signal, and a total of 32×4 = 128 features are obtained. In the frequency domain, the fast Fourier transform (FFT) is used to calculate the relative energy of each frequency band. The frequency bands of concern include delta waves (0.5 - 4 Hz), theta waves (4 - 8 Hz), alpha waves (8 - 13 Hz), beta waves (13 - 30 Hz), and gamma waves (30 - 45 Hz). This step obtains 32×5 = 160 features, which reflect the activity levels of different brain regions in each frequency band. In the time-frequency domain, the continuous wavelet transform (CWT) is used to calculate the time-frequency joint distribution features. Five scales are selected, and 32×5 = 160 features are obtained. These features capture the synchronous changes of the signal in time and frequency. In the spatial domain, the coherence between 10 pairs of key brain region combinations is calculated, and 10 features are obtained, which reflect the functional connection strength between different brain regions.

[0100] Next, the extracted feature vectors are combined with the optimized interaction parameter matrices P_EEG and P_DRIVE in the above steps for feature mapping. This process can be represented by two mathematical formulas: Y_EEG = F(X) * P_EEG and Y_DRIVE = F(X) * P_DRIVE. Where F(X) is the feature extraction function, X is the preprocessed EEG data, Y_EEG represents the prediction from EEG features to driving operations, and Y_DRIVE represents the prediction from EEG features to driving behaviors. This mapping converts the high-dimensional EEG features into low-dimensional representations related to driving, providing a good basis for subsequent state judgment.

[0101] To accurately determine the current condition of the driver, an ensemble learning method based on decision trees is adopted. This method combines the advantages of multiple decision tree models, can handle high-dimensional features, resist overfitting, and provide good generalization ability. Specifically, the Random Forest algorithm is used. This algorithm constructs 100 decision trees, with the maximum depth of each tree set to 10 and the minimum number of samples in leaf nodes set to 5. Randomness is introduced during the construction of each tree (such as randomly selecting a subset of features), thus generating a powerful ensemble model. These parameter settings achieve a good balance between model complexity and generalization ability.

[0102] To further improve the accuracy of the judgment, the temporal characteristics of electroencephalogram (EEG) signals are also considered. Using the sliding window technique, not only the EEG features at the current moment are analyzed, but also the trend of feature changes in the past 10 seconds is incorporated. This method can capture the dynamic changes of the driver's state and improve the stability and reliability of the judgment. Specifically, a series of statistics, such as mean, variance, slope, etc., are calculated for the feature sequence within 10 seconds, and these statistics are used as additional features and input into the Random Forest model.

[0103] Finally, the state judgment results of the system are divided into four dimensions: fatigue level, attention level, cognitive load, and emotional state. The first three dimensions are represented by continuous values from 0 to 1, where 0 represents the lowest level and 1 represents the highest level. For example, in the fatigue level dimension, 0 means fully awake and 1 means extremely fatigued. The emotional state is divided into four discrete categories: calm, mildly tense, moderately tense, and highly tense, represented by 0, 1, 2, and 3 respectively. The judgment result of each dimension is accompanied by a confidence index, which is calculated based on the prediction consistency of all decision trees in the Random Forest model. The higher the confidence, the more reliable the judgment result.

[0104] In some embodiments, generating the vehicle auxiliary control instruction according to the current condition of the driver includes: obtaining the real-time driving environment information of the vehicle; generating the vehicle auxiliary control instruction according to the current condition of the driver and the real-time driving environment information of the vehicle, where the vehicle auxiliary control instruction includes speed adjustment, direction control, and braking intensity, and is used to implement intelligent driving assistance functions, and the intelligent driving assistance functions at least include adaptive cruise and automatic emergency braking.

[0105] Based on the determination result of the driver's condition in the above steps, combined with the real-time driving environment information of the vehicle, the driving assistance plan is converted into specific vehicle control instructions, including but not limited to parameters such as speed adjustment, direction control, braking intensity, etc., and the on-vehicle execution system receives and implements the corresponding control instructions to achieve intelligent driving assistance functions such as adaptive cruise and automatic emergency braking.

[0106] First, integrate information from multiple sources. This includes the driver condition determination results (fatigue level, attention level, cognitive load, and emotional state, each dimension accompanied by a confidence index) output from the above steps, as well as the real-time driving environment information of the vehicle. The driving environment information is provided by various sensors of the vehicle, mainly including the Global Positioning System, millimeter-wave radar, camera system, ultrasonic sensor, and on-board diagnostic system. The Global Positioning System provides the precise position, speed, and direction of the vehicle, with a positioning accuracy better than 0.5 meters; the millimeter-wave radar detects the distance and relative speed of surrounding vehicles and obstacles, with a detection range of 200 meters; the camera system identifies lane lines, traffic signs, and other road users, with a field of view of 120 degrees; the ultrasonic sensor is used for close-range obstacle detection, especially when driving at low speeds, with a detection range of 3 meters; the on-board diagnostic system provides information on the internal state of the vehicle, such as engine speed, throttle position, etc. The data from these sensors is transmitted to the central processing unit in real time through the in-vehicle network at a sampling frequency of 100Hz, forming a comprehensive environmental perception model.

[0107] This information is input into a complex decision-making system. The system adopts a hybrid approach that combines a rule-based method and a machine learning model. The rule-based part handles clearly defined situations, such as automatically increasing the safety distance from the vehicle ahead when the detected driver fatigue level exceeds 0.8. The machine learning model deals with more complex scenarios, such as predicting the behavior of other vehicles. The core of the decision-making system is a multi-objective optimization algorithm that balances the three main objectives of safety, comfort, and efficiency. This algorithm can be expressed as a mathematical optimization problem, where the decision variables include speed v, direction angle θ, acceleration a, etc., and the objective function is expressed as: where S, C, and E represent the safety, comfort, and efficiency indicators respectively, are the weight coefficients. The constraint conditions include vehicle performance limitations (such as maximum steering angle, maximum acceleration) and traffic rules (such as speed limit requirements). This optimization problem is solved by an improved particle swarm optimization algorithm, which uses 100 particles, sets the maximum number of iterations to 200, and the convergence threshold to 0.001.

[0108] The result of the optimization is a set of optimal control parameters, which are then converted into specific vehicle control instructions. The main control instructions include speed regulation, direction control, braking intensity, lane keeping, and following distance. Speed regulation is achieved by controlling the engine output power and the braking system. The system dynamically adjusts the target speed according to the driver's fatigue level and attention level. For example, when the fatigue level is greater than 0.6, the cruise speed is reduced by 10%. Direction control adjusts the steering assist and feedback through the electric power steering system. When it detects that the driver's attention is not concentrated (attention level less than 0.4), the system increases the steering feedback force to improve the driver's sense of direction. Braking intensity is controlled by adjusting the anti-lock braking system and the electronic brake force distribution system. The system pre-adjusts the braking force according to the distance and relative speed of the vehicle ahead to handle possible emergencies. Lane keeping keeps the vehicle driving within the lane by fine-tuning the steering angle. When the driver's cognitive load is high (greater than 0.8), the system increases the intervention degree of the lane keeping assist. The following distance is achieved by adjusting the target following time of the adaptive cruise control system. The system dynamically adjusts the safety distance according to the driver's state and the current traffic flow.

[0109] These control instructions are transmitted to each execution unit through the in-vehicle bus system. Each execution unit is equipped with a dedicated microcontroller, which is responsible for converting the high-level control instructions into specific mechanical or electronic operations. For example, the speed regulation instruction is received by the engine management system and then converted into specific parameters such as throttle opening and fuel injection volume. To ensure the smoothness and safety of the control, the system adopts a progressive control strategy. The change of control parameters is not instantaneous but smoothly transitions within a certain time. This transition process can be described by a mathematical function: where P0 is the initial parameter value, P1 is the target parameter value, τ is the time constant, and t is the time variable. This smooth transition not only improves driving comfort but also avoids the safety risks that sudden control changes may bring.

[0110] In this way, the system implements a variety of intelligent driving assistance functions. The adaptive cruise function can automatically adjust the vehicle speed according to the driver's state and the distance to the vehicle in front to maintain a safe following distance. When the system detects that the driver is fatigued (greater than 0.7), it will automatically increase the safe distance from the vehicle in front and may reduce the cruising speed. The automatic emergency braking function automatically performs emergency braking when a collision risk is detected and the driver does not respond in time. This function takes into account the driver's attention level and reaction time. The lane departure warning function warns the driver through sound or seat vibration when the vehicle tends to deviate from the lane. The intensity and frequency of the warning will be adjusted according to the driver's cognitive load and attention level. The fatigue driving warning function will issue a warning and recommend a rest when it detects that the driver's fatigue level exceeds the threshold. The system will choose the most suitable reminder method according to the driver's emotional state. The intelligent speed limit function automatically adjusts the maximum speed according to the road speed limit information and the driver's state. When the driver is in poor condition, the system may set the speed limit lower.

[0111] In order to execute the intelligent vehicle-mounted driving assistance method based on EEG signals corresponding to the above method embodiment, so as to achieve the corresponding functions and technical effects. Figure 2 , Figure 2 The structure block diagram of an intelligent vehicle-mounted driving assistance device 200 provided in an embodiment of the present application is shown. For the convenience of explanation, only the parts related to the present embodiment are shown. The intelligent vehicle-mounted driving assistance device 200 provided in an embodiment of the present application includes:

[0112] The operation analysis module 201 is used to obtain the target driving operation and the corresponding target EEG data within a preset time period as a training set, and perform a benchmark comparison analysis on the target driving operation and the target EEG data;

[0113] A result acquisition module 202, for determining the EEG interaction range of each target driving operation according to the analysis results corresponding to the benchmark control analysis, and generating corresponding heterogeneous features;

[0114] A feature input module 203, for inputting the feature code corresponding to the inhomogeneous feature into a preset interactive mapping table to determine the EEG interaction parameters under the preset driving operation index and the driving operation interaction parameters under the preset EEG index;

[0115] A parameter acquisition module 204, for generating feature codes of heterogeneous features based on the feature code generation model according to the EEG interaction parameters and the driving operation interaction parameters, and combining the feature codes of all heterogeneous features into a feature code set;

[0116] A feature encoding module 205, configured to perform encoding splitting on the feature encoding according to predefined driving operation descriptions and electroencephalogram data descriptions in the interaction mapping table, and sequentially insert the split results into data units of the interaction mapping table, so as to obtain an interaction mapping table filled with data;

[0117] An operation locking module 206, configured to extract electroencephalogram interaction parameters and driving operation interaction parameters from the interaction mapping table according to predefined driving operation categories and electroencephalogram types;

[0118] An instruction generation module 207, configured to obtain the current status of the driver according to the electroencephalogram interaction parameters and driving operation interaction parameters, and generate a vehicle auxiliary control instruction according to the current status of the driver.

[0119] The above intelligent vehicle driving assistance device 200 can implement the intelligent vehicle driving assistance method based on electroencephalogram signals in the above method embodiment. The optional items in the above method embodiment are also applicable to this embodiment, and will not be elaborated here. The remaining content of this application embodiment can refer to the content of the above method embodiment, and will not be repeated in this embodiment.

[0120] Figure 3 It is a schematic structural diagram of a computer device provided in an embodiment of the present application. As Figure 3 shown, the computer device 3 in this embodiment includes: at least one processor 30 ( Figure 3 only one is shown in the figure), a memory 31, and a computer program 32 stored in the memory 31 and executable on the at least one processor 30. When the processor 30 executes the computer program 32, the steps in any of the above method embodiments are implemented.

[0121] The computer device 3 may be a computing device such as a smart phone, a tablet computer, a desktop computer, and a cloud server. The computer device may include but is not limited to the processor 30 and the memory 31. Those skilled in the art can understand that Figure 3 merely examples of the computer device 3, and do not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine some components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0122] The so-called processor 30 may be a Central Processing Unit (CPU), and the processor 30 may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0123] In some embodiments, the memory 31 may be an internal storage unit of the computer device 3, such as the hard disk or memory of the computer device 3. In some other embodiments, the memory 31 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., equipped on the computer device 3. Further, the memory 31 may also include both the internal storage unit and the external storage device of the computer device 3. The memory 31 is used to store an operating system, application programs, a BootLoader, data, and other programs, such as the program code of the computer program, etc. The memory 31 may also be used to temporarily store data that has been output or is to be output.

[0124] In addition, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0125] An embodiment of the present application provides a computer program product, and when the computer program product runs on a computer device, the computer device is caused to implement the steps in each of the above method embodiments when executed.

[0126] In several embodiments provided by the present application, it can be understood that each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and the module, the program segment, or the part of code includes one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved.

[0127] If the described functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs and other various media that can store program codes.

[0128] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present application. It should be understood that the above description is only specific embodiments of the present application and is not used to limit the protection scope of the present application. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An intelligent vehicle driving assistance method based on electroencephalogram signals, characterized in that, Including: Obtain the target driving operations and corresponding target EEG data within a preset time period as a training set, and conduct a benchmark comparison analysis on the target driving operations and target EEG data, including: obtaining the driving events corresponding to the driving operation data; the driving events at least include straight driving, acceleration, deceleration, and emergency braking; extracting time-domain features, frequency-domain features, and time-frequency features from the EEG data segments corresponding to the driving events in the target EEG data; conducting a benchmark comparison analysis on the driving events and the corresponding time-domain features, frequency-domain features, and time-frequency features, including: constructing EEG features based on the time-domain features, frequency-domain features, and time-frequency features; calculating the average EEG feature vector and covariance matrix of the EEG features corresponding to the driving events; conducting variance analysis based on the average EEG feature vector and covariance matrix to determine abnormal EEG features; the abnormal EEG features are significantly different among multiple driving events; using principal component analysis or linear discriminant analysis to reduce the feature dimension of the abnormal EEG features and generate the analysis result corresponding to the benchmark comparison analysis; According to the analysis result corresponding to the benchmark comparison analysis, determine the EEG interaction range of each target driving operation and generate corresponding inhomogeneous features; Input the feature codes corresponding to the inhomogeneous features into a preset interaction mapping table to determine the EEG interaction parameters under the preset driving operation indicators and the driving operation interaction parameters under the preset EEG indicators; Based on the EEG interaction parameters and driving operation interaction parameters, generate the feature codes of the inhomogeneous features based on the feature coding generation model, and combine the feature codes of all inhomogeneous features into a feature code set; According to the predefined driving operation descriptions and EEG data descriptions in the interaction mapping table, split the feature codes and insert them into the data units of the interaction mapping table in sequence to obtain the interaction mapping table filled with data; Extract the EEG interaction parameters and driving operation interaction parameters from the interaction mapping table according to the predefined driving operation categories and EEG wave types; Obtain the current condition of the driver according to the EEG interaction parameters and driving operation interaction parameters, and generate a vehicle auxiliary control instruction according to the current condition of the driver.

2. The method according to claim 1, wherein Before inputting the feature codes corresponding to the inhomogeneous features into the preset interaction mapping table, it further includes: Construct a deviation distribution map of the target driving operation according to the EEG interaction range, including distribution columns corresponding to the target EEG data and driving deviation types; Compare the deviation distribution map with the standard driving operation distribution map, determine the amplitude change amount of each distribution column, and screen the minimum amplitude change amount from the amplitude change amounts to correct the remaining amplitude change amounts; Redetermine the EEG interaction range based on the corrected deviation distribution map; Obtain the feature expression of the corrected EEG interaction range and input the feature expression into a preset inhomogeneous analysis model to obtain optimized inhomogeneous features.

3. The method according to claim 1, wherein Before conducting the benchmark comparison analysis on the target driving operations and target EEG data, it further includes: For the driving operation data, identify and remove outliers based on the 3σ criterion, perform smoothing using moving average filtering, and perform Z-score normalization; For the target EEG data, use notch filtering to remove power frequency interference, use independent component analysis method to remove artifacts of electrooculogram and electromyogram, and perform band-pass filtering to retain the main EEG activity frequency band to ensure that the signal-to-noise ratio of the target EEG data is greater than the preset signal-to-noise ratio; Use high-precision timestamps to accurately align the driving operation data and the target EEG data in time.

4. The method according to claim 1, wherein According to the predefined driving operation categories and EEG types, extract EEG interaction parameters and driving operation interaction parameters from the interaction mapping table, including: Perform the first round of locking on the interaction mapping table according to the predefined driving operation categories to obtain EEG interaction parameters; Perform the second round of locking on the interaction mapping table according to the predefined EEG types to obtain the corresponding driving operation interaction parameters; Through cross-validation, obtain the optimized EEG interaction parameters and driving operation interaction parameters.

5. The method according to claim 1, characterized in that, According to the EEG interaction parameters and driving operation interaction parameters, obtain the current condition of the driver, including: Obtain the real-time collected EEG data of the driver; Use a pattern recognition algorithm to extract features from the EEG data of the driver, and combine the optimized EEG interaction parameters and driving operation interaction parameters to obtain the current condition of the driver.

6. The method according to claim 1, wherein According to the current condition of the driver, generate vehicle auxiliary control instructions, including: Obtain the real-time driving environment information of the vehicle; According to the current condition of the driver and the real-time driving environment information of the vehicle, generate the vehicle auxiliary control instructions, and the vehicle auxiliary control instructions include speed adjustment, direction control and braking intensity, and are used to implement intelligent driving assistance functions, and the intelligent driving assistance functions at least include adaptive cruise and automatic emergency braking.

7. An intelligent in-vehicle driving assistance device based on electroencephalogram signals, characterized in that, Include: An operation analysis module, which is used to obtain the target driving operation and the corresponding target EEG data within a preset time period as a training set, and perform a benchmark comparison analysis on the target driving operation and the target EEG data, including: obtaining the driving events corresponding to the driving operation data; the driving events at least include straight driving, acceleration, deceleration and emergency braking; according to the EEG data segments corresponding to the driving events in the target EEG data, extract time domain features, frequency domain features and time-frequency features; perform a benchmark comparison analysis on the driving events and the corresponding time domain features, frequency domain features and time-frequency features, including: constructing EEG features according to the time domain features, frequency domain features and time-frequency features; calculating the average EEG feature vector and covariance matrix of the EEG features corresponding to the driving events; performing variance analysis according to the average EEG feature vector and covariance matrix to determine abnormal EEG features; the abnormal EEG features are significantly different among multiple driving events; use principal component analysis or linear discriminant analysis to reduce the feature dimension of the abnormal EEG features, and generate the analysis result corresponding to the benchmark comparison analysis; A result acquisition module, which is used to determine the EEG interaction range of each target driving operation according to the analysis result corresponding to the benchmark comparison analysis, and generate the corresponding heterogeneous features; A feature input module, configured to input the feature encoding corresponding to the inhomogeneous feature into a preset interaction mapping table to determine the electroencephalogram (EEG) interaction parameters under a preset driving operation index and the driving operation interaction parameters under a preset EEG index; A parameter acquisition module, configured to generate the feature encoding of the inhomogeneous feature based on a feature encoding generation model according to the EEG interaction parameters and the driving operation interaction parameters, and the feature encodings of all the inhomogeneous features are combined into a feature encoding set; A feature encoding module, configured to perform encoding splitting on the feature encoding according to the predefined driving operation description and EEG data description in the interaction mapping table and sequentially insert the split encoding into the data unit of the interaction mapping table to obtain the interaction mapping table filled with data; An operation locking module, configured to extract the EEG interaction parameters and the driving operation interaction parameters from the interaction mapping table according to the predefined driving operation category and EEG wave type; An instruction generation module, configured to obtain the current condition of the driver according to the EEG interaction parameters and the driving operation interaction parameters, and generate a vehicle auxiliary control instruction according to the current condition of the driver.

8. A computer device, characterized in that, It includes a memory and a processor; the memory is used for storing a computer program; the processor is used for executing the computer program and implementing the method according to any one of claims 1 to 6 when executing the computer program.

Images