Steering wheel bio-sensing driving risk intervention system and method

Abstract

The application discloses a steering wheel biological perception driving risk intervention system and method, and belongs to the field of vehicle active safety technology. The system collects signals related to the contact area between the driver and the steering wheel and vehicle behavior signals to form original multi-modal time series data; determines the contact effectiveness coefficient and performs weight reduction, interpolation or shielding processing on distorted samples to obtain effective multi-modal time series data; establishes individual baseline parameters according to the calm driving stage, performs baseline correction and multi-modal fusion on the multi-modal characteristics, and outputs emotion recognition results; determines the comprehensive risk score in combination with the driving behavior abnormality and the scene risk degree, and outputs the monitoring state, the early warning state or the active intervention state control instruction according to the risk threshold, and the active intervention state control instruction is verified and released step by step when the recovery condition is met. The application can improve the accuracy, stability and safety of the driver state recognition and risk assessment and the active intervention control.

Description

Technical Field

[0001] This invention relates to the field of active vehicle safety technology, and in particular to a steering wheel-based bio-sensing driving risk intervention system and method. Background Technology

[0002] With the development of intelligent driving assistance technology and vehicle active safety technology, research on driver state perception, risk identification, and intervention control is gradually increasing. Existing technologies already exist that detect driver state based on driver physiological state information, driving behavior information, or in-vehicle environment information, and trigger alerts, warnings, or vehicle control accordingly.

[0003] However, existing technologies still have shortcomings in practical applications. During driving, driver state-related information exhibits significant dynamic changes. The data is easily affected by factors such as changes in driving operations, physical contact states, and driving scenarios during collection and processing, leading to insufficient stability and reliability of state information and impacting the accuracy of driver state identification and risk assessment results. Different drivers differ in physiological response characteristics, behavioral habits, and risk tolerance. Without adaptive processing to address individual differences, identification bias and inaccurate assessments are likely to occur. When the system needs to comprehensively utilize multi-source state information for judgment, fluctuations in the validity and reliability of different information sources also reduce the consistency and stability of the overall judgment results.

[0004] On the other hand, existing technologies typically focus more on trigger control after an abnormal state occurs, while neglecting the matching relationship between intervention intensity and risk level, as well as the exit control mechanism after risk reduction. If intervention triggers are too singular or lack hierarchy, it may affect driving comfort and system acceptability; if several pre-release mechanisms lack sufficient verification and reasonable exit procedures, the risk may rise again before it has stabilized and subsided, affecting vehicle operation safety.

[0005] Therefore, existing technologies urgently need to address the following technical issues: how to improve the accuracy and stability of driver state recognition and risk assessment during driving, under conditions where driver state information is susceptible to dynamic disturbances, individual differences are significant, and the reliability of multi-source state information is inconsistent, and on this basis, to achieve graded intervention that matches the level of risk and safe and reasonable intervention exit control, thereby improving the level of active vehicle safety. Summary of the Invention

[0006] The purpose of this invention is to provide a steering wheel bio-sensing driving risk intervention system and method to address the problems in existing technologies, such as insufficient data reliability due to the susceptibility of driver state information to changes in contact, manipulation, and scene during driving; individual differences among different drivers leading to biased identification results; and inconsistent reliability of multi-source state information resulting in insufficient stability of comprehensive judgment. Furthermore, this invention addresses the problems in existing technologies, such as poor matching between intervention intensity and risk level, unreasonable timing of intervention exit, and lack of stable recovery mechanisms when risks escalate. This improves the accuracy, stability, and safety of driver state identification, risk assessment, and proactive intervention control.

[0007] To address the aforementioned technical problems, this invention provides a steering wheel-based bio-sensing driving risk intervention system, comprising: The steering wheel biosensing module is used to collect skin conductance signals, electromyographic correlation signals, grip force distribution signals, contact temperature signals and vehicle behavior signals related to the driver's contact area with the steering wheel, forming raw multimodal time-series data; The contact validity determination module is used to determine the contact validity coefficient based on the grip force distribution signal and vehicle behavior signal, and to perform weight reduction, interpolation or masking on the distorted samples in the original multimodal time series data based on the contact validity coefficient, and output valid multimodal time series data. The emotion state recognition module is used to determine the calm driving stage and establish individual baseline parameters based on the continuous time period after driving starts and the preset stable conditions are met. Based on the contact effectiveness coefficient and individual baseline parameters, the module performs baseline correction on the multimodal feature vectors corresponding to the effective multimodal time series data to form a standardized multimodal feature sequence. After time series feature extraction and multimodal fusion, the module outputs the emotion recognition result. The driving risk assessment module is used to determine a comprehensive risk score based on emotion recognition results, abnormality of driving behavior, and danger level of the scenario. The graded active intervention module is used to output monitoring control commands, early warning control commands, or active intervention control commands based on the relationship between the comprehensive risk score and the first risk threshold and the second risk threshold. The active intervention control commands include vehicle control commands, cabin adjustment commands, or a combination of the two. When the comprehensive risk score is continuously lower than the recovery threshold and continues to reach the preset recovery time, the module verifies the release conditions of the active intervention control commands and releases the active intervention control commands level by level after the verification is passed.

[0008] In some embodiments, the steering wheel biosensor module includes: A skin conductivity acquisition unit is installed on the surface of the steering wheel grip area to collect skin conductivity signals from the area in contact with the driver's palm. An electromyography (EMG) correlation acquisition unit is located inside the steering wheel grip area to acquire EMG signals related to the driver's palm and forearm. The grip force distribution acquisition unit is located in the steering wheel grip area and is used to acquire grip force distribution signals and pressure center location information; The contact temperature acquisition unit is located in the steering wheel grip area and establishes a corresponding relationship with the grip force distribution acquisition unit in the sampling area to collect the driver's palm contact temperature signal. The vehicle behavior acquisition unit is connected to the vehicle bus and is used to collect steering wheel angle, vehicle speed, throttle opening, braking intensity, vehicle distance, lane center offset, road curvature and lateral acceleration. The steering wheel biosensing module is also used to align skin conductance signals, electromyography-related signals, grip force distribution signals, contact temperature signals, and vehicle behavior signals according to a unified time reference, and to correlate them according to the corresponding relationship of the sampling areas to form original multimodal time series data.

[0009] In some implementations, the contact effectiveness determination module determines the steering wheel contact coverage, the mean value of the pressure matrix, and the pressure center offset based on the grip force distribution signal; determines the steering wheel angular velocity based on the vehicle behavior signal; and determines the contact effectiveness coefficient based on the steering wheel contact coverage, the mean value of the pressure matrix, the steering wheel angular velocity, and the pressure center offset; the contact effectiveness coefficient is calculated according to the following formula: ; In the formula, Indicates time The contact effectiveness coefficient, Indicates steering wheel contact coverage. This represents the mean of the pressure matrix. Indicates the angular velocity of the steering wheel. Indicates the offset of the pressure center. , , and Represents the weight parameters. Represents the Sigmoid mapping function; The contact validity determination module determines sampling points with contact validity coefficients lower than a preset threshold as distorted samples, and performs weight reduction processing, interpolation processing, or masking processing on the distorted samples.

[0010] In some implementations, the emotion state recognition module determines a calm driving phase based on a continuous time period after driving begins that meets preset stability conditions, and establishes individual baseline parameters based on the statistical results of multimodal feature vectors corresponding to multiple consecutive analysis windows within the calm driving phase. These individual baseline parameters include the baseline mean and baseline standard deviation of the multimodal feature vectors. The emotion state recognition module extracts multimodal feature vectors from valid multimodal time-series data and performs baseline correction according to the following formula to obtain standardized multimodal feature vectors: ; In the formula, Indicates time The multimodal feature vectors, Indicates time The standardized multimodal feature vector, Indicates the baseline mean. Indicates the baseline standard deviation. Represents a very small positive number. Indicates time The contact effectiveness coefficient; the emotion state recognition module arranges the standardized multimodal feature vectors corresponding to multiple consecutive moments in chronological order to form a standardized multimodal feature sequence.

[0011] In some implementations, the emotion state recognition module uses a sliding window to continuously update the standardized multimodal feature sequence and extracts temporal features from the standardized multimodal feature sequence within the current sliding window; The emotion state recognition module determines modality credibility based on the contact validity coefficient, effective sample ratio, and missing rate corresponding to different input source data within the current sliding window, and performs weighted fusion of each modality feature according to the modality credibility to output the probability of emotion category. The emotion categories include anger, anxiety, and fatigue; When the probability corresponding to the same emotion category is higher than the preset recognition threshold within multiple consecutive sliding windows, an emotion state change is determined to have occurred.

[0012] In some implementations, the driving risk assessment module determines the degree of abnormality of driving behavior based on the steering wheel angle sequence, throttle opening sequence, braking input sequence and vehicle speed change, determines the degree of danger of the scene based on the distance between vehicles, lane center deviation, road curvature and lateral acceleration, and determines a comprehensive risk score based on the emotion recognition results, the degree of abnormality of driving behavior and the degree of danger of the scene.

[0013] In some implementations, the abnormality of driving behavior, the hazard level of the scenario, and the overall risk score are determined according to the following formula: ; In the formula, Indicates time abnormality of driving behavior Indicates length is The sequence of steering wheel angles within the analysis window. Indicates standard deviation, Indicates the throttle opening sequence. Indicates the braking input sequence. Represents sample entropy. Indicates the change in vehicle speed. Indicates time The level of danger in the scene , , ,and These represent the normalized oncoming distance, lane center offset, road curvature, and lateral acceleration, respectively. , , and This refers to the pre-calibrated and stored weight parameters in the driving risk assessment module. Indicates time The overall risk score, , and Representing time respectively The probability of anger, anxiety, and fatigue. , , , , , , , and Represents the weight parameters. Indicates the preset analysis window length.

[0014] In some implementations, the graded active intervention module outputs a monitoring state control command, an early warning state control command, or an active intervention state control command based on the relationship between the comprehensive risk score and the first risk threshold and the second risk threshold, wherein the first risk threshold is less than the second risk threshold; The active intervention control commands include vehicle control commands, cabin adjustment commands, or a combination of both. The graded active intervention module outputs torque limiting commands and vehicle warning commands when the emotional state corresponding to the probability of anger is dominant, outputs target following distance adjustment commands and cabin adjustment commands when the emotional state corresponding to the probability of anxiety is dominant, and outputs lane centering enhancement commands and steering wheel vibration commands when the emotional state corresponding to the probability of fatigue is dominant.

[0015] In some implementations, the hierarchical active intervention module verifies the release conditions of the active intervention state control command when the system is in an active intervention state; The conditions for lifting the restrictions include: the comprehensive risk score is continuously lower than the recovery threshold and continues to reach the preset recovery time, and the recovery threshold is lower than the second risk threshold; After the release condition verification is passed, the hierarchical active intervention module releases the corresponding control commands level by level in a preset order; During the step-by-step de-escalation process, if the comprehensive risk score reaches or exceeds the second risk threshold again, the active intervention control command will be resumed.

[0016] The present invention also provides a steering wheel bio-sensing driving risk intervention method, applied to a steering wheel bio-sensing driving risk intervention system as described above, comprising the following steps: S1. Collect skin conductance signals, electromyographic correlation signals, grip force distribution signals, contact temperature signals, and vehicle behavior signals related to the driver's contact area with the steering wheel to form raw multimodal time series data; S2. Determine the contact effectiveness coefficient based on the grip force distribution signal and vehicle behavior signal, and perform weight reduction, interpolation or masking on the distorted samples in the original multimodal time series data based on the contact effectiveness coefficient, and output effective multimodal time series data; S3. Determine the calm driving phase based on the continuous time period after driving begins that meets the preset stability conditions, and establish individual baseline parameters. S4. Based on the contact effectiveness coefficient and individual baseline parameters, the baseline correction is performed on the multimodal feature vectors corresponding to the effective multimodal time series data to form a standardized multimodal feature sequence. Then, the time series feature extraction and multimodal fusion are performed on the standardized multimodal feature sequence to output the emotion recognition result. S5. Determine the comprehensive risk score based on the emotion recognition results, the degree of abnormality in driving behavior, and the degree of danger in the scenario; S6. Output a monitoring control command, an early warning control command, or an active intervention control command based on the relationship between the comprehensive risk score and the first risk threshold and the second risk threshold. The active intervention control command includes a vehicle control command, a cabin adjustment command, or a combination of the two. S7. When the comprehensive risk score is continuously lower than the recovery threshold and the preset recovery time is reached, the conditions for releasing the active intervention control command are verified, and the active intervention control command is released step by step after the verification is passed.

[0017] Compared with existing technologies, this invention has the following advantages: By collecting skin conductance signals, electromyography signals, grip force distribution signals, contact temperature signals, and vehicle behavior signals related to the driver's contact area with the steering wheel, this invention forms raw multimodal time-series data, providing a more complete and continuous data foundation for subsequent driver state recognition and risk assessment. By determining the contact effectiveness coefficient based on the grip force distribution signal and vehicle behavior signal, and by performing weight reduction, interpolation, or masking on distorted samples in the raw multimodal time-series data, the impact of contact instability, short-term disengagement, local slippage, and drastic changes in handling on data quality can be reduced, improving the effectiveness and reliability of the input data. By establishing individual baseline parameters based on the calm driving phase, and performing baseline correction on the multimodal feature vectors corresponding to the effective multimodal time-series data based on the contact effectiveness coefficient and individual baseline parameters, a standardized multimodal feature sequence is formed, which is then processed... Temporal feature extraction and multimodal fusion output of emotion recognition results can reduce the impact of individual driver differences on recognition results, improving the accuracy and stability of driver state recognition. By determining a comprehensive risk score based on emotion recognition results, abnormality of driving behavior, and scenario hazard, a comprehensive characterization of the driver's current state, degree of operational abnormality, and driving scenario risk can be achieved, improving the comprehensiveness and rationality of risk assessment results. By outputting monitoring-state control commands, early warning-state control commands, or active intervention-state control commands based on the relationship between the comprehensive risk score and a first risk threshold and a second risk threshold, and verifying the release conditions of the active intervention-state control command when the comprehensive risk score continuously falls below the recovery threshold and reaches a preset recovery time, the active intervention-state control command is released level by level after successful verification, ensuring that the intervention intensity is adapted to the risk level and improving the smoothness and safety of the active intervention exit process. This invention is beneficial for improving the accuracy of driver state recognition, the stability of risk assessment, and the safety of active intervention control. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the modular structure of the system of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0020] like Figure 1 As shown in the illustration, this is an embodiment of the present invention, which provides a steering wheel bio-sensing driving risk intervention system, including: a steering wheel bio-sensing module, a contact effectiveness determination module, an emotion state recognition module, a driving risk assessment module, and a graded active intervention module. The steering wheel bio-sensing module outputs raw multimodal time-series data to the contact effectiveness determination module; the contact effectiveness determination module outputs valid multimodal time-series data and contact effectiveness coefficients to the emotion state recognition module; the emotion state recognition module outputs the probability of anger, the probability of anxiety, the probability of fatigue, and the confirmation result of emotion state changes to the driving risk assessment module; the driving risk assessment module outputs a comprehensive risk score to the graded active intervention module; the graded active intervention module outputs corresponding control results based on the comprehensive risk score and the dominant emotion state, and performs maintenance, deactivation, or restoration control according to recovery conditions.

[0021] In this application, the original multimodal time-series data refers to the data collected by the steering wheel biosensing module and formed after being aligned with a unified time reference and associated with the corresponding sampling areas; the effective multimodal time-series data refers to the data after the contact validity determination module has processed the distorted samples by weight reduction, interpolation, or masking; the multimodal feature vector refers to the feature combination extracted from the effective multimodal time-series data; the standardized multimodal feature vector refers to the multimodal feature vector corrected by individual baseline parameters and contact validity coefficients; the standardized multimodal feature sequence refers to the sequence formed by arranging the standardized multimodal feature vectors in chronological order; the fused feature vector refers to the result formed by weighting and fusing the temporal features of each modality according to modality credibility within the current sliding window; and the comprehensive risk score refers to the unified risk quantity jointly determined by the emotion recognition result, the abnormality of driving behavior, and the danger level of the scene.

[0022] (1) Steering wheel biosensing module In this embodiment, the steering wheel biosensing module is used to collect skin conductance signals, electromyographic correlation signals, grip force distribution signals, contact temperature signals and vehicle behavior signals related to the area where the driver contacts the steering wheel, forming raw multimodal time-series data.

[0023] The steering wheel biosensing module includes: a skin conductivity acquisition unit, disposed on the surface of the steering wheel grip area, for acquiring skin conductivity signals from the area in contact with the driver's palm; an electromyography (EMG) correlation acquisition unit, disposed on the inner side of the steering wheel grip area, for acquiring EMG correlation signals from the driver's palm and forearm; a grip force distribution acquisition unit, disposed in the steering wheel grip area, for acquiring grip force distribution signals and pressure center location information; a contact temperature acquisition unit, disposed in the steering wheel grip area and corresponding to the grip force distribution acquisition unit in the sampling area, for acquiring the driver's palm contact temperature signal; and a vehicle behavior acquisition unit, connected to the vehicle bus, for acquiring steering wheel angle, vehicle speed, throttle opening, braking intensity, vehicle distance, lane center offset, road curvature, and lateral acceleration. The steering wheel biosensing module also aligns the skin conductivity signals, EMG correlation signals, grip force distribution signals, contact temperature signals, and vehicle behavior signals according to a unified time reference and correlates them based on the corresponding sampling areas to form raw multimodal time-series data.

[0024] Specifically, the steering wheel biosensing module is installed in the vehicle steering wheel assembly and its associated control circuitry, serving as the system's data acquisition front-end. The skin conductance signal is used to characterize changes in the driver's autonomic nervous activity; the electromyographic correlation signal is used to characterize the electrophysiological changes related to the driver's hand gripping movements and forearm muscle activity; the grip force distribution signal and pressure center location information are used to characterize the contact range, contact intensity, and changes in the contact center of gravity between the driver and the steering wheel; the contact temperature signal is used to characterize changes in the thermal state of the contact area; and the vehicle behavior signal is used to characterize the vehicle's current handling and driving states.

[0025] In this embodiment, each acquisition unit is connected to a data acquisition control board. The data acquisition control board includes a signal conditioning circuit, an analog-to-digital conversion circuit, a time synchronization circuit, a buffer circuit, and a communication interface circuit. It receives the raw signals output by each acquisition unit and outputs data frames in a unified format. For signals with different sampling frequencies, the steering wheel biosensing module maps them to the same time-series grid using a unified timestamp, resampling, interpolation, or synchronous buffering. For contact signals within the steering wheel grip area, the steering wheel biosensing module performs correlation processing based on the corresponding sampling areas, enabling the skin conductance signal, grip force distribution signal, and contact temperature signal corresponding to the same contact area at the same sampling time to form joint data. After the above processing, the steering wheel biosensing module outputs the raw multimodal time-series data to the contact validity determination module.

[0026] (2) Contact validity determination module In this embodiment, the contact validity determination module is used to determine the contact validity coefficient based on the grip force distribution signal and the vehicle behavior signal, and to perform weight reduction, interpolation or masking processing on the distorted samples in the original multimodal time series data based on the contact validity coefficient, and output valid multimodal time series data.

[0027] Specifically, the contact validity determination module is connected to the steering wheel biosensing module. It receives grip force distribution signals, pressure center location information, and vehicle behavior signals from the raw multimodal time-series data and performs contact state analysis point by point according to the sampling time. The inputs of the contact validity determination module include grip force distribution signals, pressure center location information, and vehicle behavior signals, and the outputs include the contact validity coefficients corresponding to each sampling time and the processed valid multimodal time-series data.

[0028] In this embodiment, the grip force distribution signal is organized into a pressure matrix according to the sampling areas of the steering wheel grip area. Let the steering wheel grip area be divided into [number] sections. Each pressure sampling area, at time... No. The pressure value of each sampling area is recorded as follows: .when When the pressure exceeds the contact detection threshold, the sampling area is considered a valid contact area; when... If the pressure does not exceed the pressure contact threshold, the sampling area is classified as a non-contact area. If at time... The number of areas determined to be valid contact areas is Then the steering wheel contact coverage Determine using the following formula: ; In the formula, the steering wheel contact coverage rate It is used to characterize the actual contact range of the driver with the steering wheel grip area at the current sampling time.

[0029] In this embodiment, the contact validity determination module determines the time... The average pressure values ​​of all sampled regions in the pressure matrix are averaged to obtain the mean of the pressure matrix. It is determined by the following formula: ; In the formula, the mean of the pressure matrix It is used to characterize the overall grip pressure level of the driver on the steering wheel at the current sampling moment.

[0030] In this embodiment, the contact effectiveness determination module calculates the time based on the pressure values ​​of each sampling area in the pressure matrix and their corresponding positions. The coordinates of the pressure center. Let the first... The coordinates of each sampling region are Then at time Pressure center coordinates Determine using the following formula: ; Pressure center coordinates at the previous sampling time As a reference position, then at time Pressure center offset Determine using the following formula: ; In the formula, the offset of the pressure center It is used to characterize the degree of change in the center of gravity of contact between adjacent sampling times.

[0031] In this embodiment, the contact validity determination module determines the steering wheel angular velocity based on the steering wheel angle in the vehicle behavior signal. Let the time be... Steering wheel angle ,time Steering wheel angle The time interval between two adjacent sampling times is Then the steering wheel angular velocity Determine using the following formula: ; Among them, steering wheel angular velocity Used to characterize the degree of change in the current steering operation.

[0032] In this embodiment, the contact effectiveness coefficient is calculated according to the following formula: ; In the formula, Indicates time The contact effectiveness coefficient, Indicates steering wheel contact coverage. This represents the mean of the pressure matrix. Indicates the angular velocity of the steering wheel. Indicates the offset of the pressure center. , , and Represents the weight parameters. This represents the Sigmoid mapping function; the steering wheel contact coverage and the mean of the pressure matrix have a positive effect on the contact effectiveness coefficient, while the steering wheel angular velocity and the pressure center offset have a negative effect on the contact effectiveness coefficient.

[0033] In this embodiment, the contact validity determination module classifies sampling points with contact validity coefficients lower than the contact validity threshold as distorted samples, and retains sampling points with contact validity coefficients not lower than the contact validity threshold as valid sampling points. The contact validity threshold is the preset threshold in the claims. For distorted samples, the contact validity determination module performs weight reduction processing, interpolation processing, or masking processing. Weight reduction processing means retaining the original data of the distorted sample and reducing the contribution weight of that sampling point in subsequent feature extraction and multimodal fusion; interpolation processing means using the previous and next sampling points to complete the distorted sample when both are valid sampling points; masking processing means marking the distorted sample as invalid data and not participating in subsequent feature extraction and state recognition. After the above processing, the contact validity determination module outputs valid multimodal time-series data and contact validity coefficients to the emotion state recognition module.

[0034] (3) Emotional state recognition module In this embodiment, the emotion state recognition module is used to determine the calm driving stage based on the continuous time period after driving starts and meet the preset stable conditions, and to establish individual baseline parameters. Based on the contact effectiveness coefficient and individual baseline parameters, the module performs baseline correction on the multimodal feature vectors corresponding to the effective multimodal time series data to form a standardized multimodal feature sequence. After time series feature extraction and multimodal fusion, the module outputs the emotion recognition result.

[0035] Specifically, the emotion state recognition module is connected to the contact validity determination module to receive valid multimodal time-series data and contact validity coefficients corresponding to each sampling time. The inputs of the emotion state recognition module include valid multimodal time-series data and contact validity coefficients, and the outputs include the probability of anger, the probability of anxiety, the probability of fatigue, and the confirmation result of emotion state changes.

[0036] In this embodiment, the emotion state recognition module first determines a calm driving phase based on a continuous time period after driving begins that meets preset stability conditions. These preset stability conditions include: the absolute value of the steering wheel angular velocity is less than a first stability threshold, the absolute value of the vehicle speed change is less than a second stability threshold, the change in braking intensity is less than a third stability threshold, and the absolute value of the lane center deviation is less than a fourth stability threshold, all continuously reaching a preset stability duration. The continuous time period meeting these conditions is defined as the calm driving phase. The first stability threshold, second stability threshold, third stability threshold, fourth stability threshold, and preset stability duration are all pre-set judgment parameters stored in the emotion state recognition module. An analysis window within the calm driving phase is used to establish individual baseline parameters. In the subsequent online recognition phase, a sliding window is used to continuously update and recognize the standardized multimodal feature sequence.

[0037] In this embodiment, the emotion state recognition module divides the effective multimodal time-series data during the calm driving phase into multiple consecutive analysis windows according to time order, and extracts a multimodal feature vector for each analysis window. Let the first... The original sequence corresponding to each modality within the current analysis window is: Then, the mean, standard deviation, slope of change, and peak-to-peak value are extracted from the original sequence to form the window feature vector of that mode. The window feature vector corresponding to the skin conductance signal is denoted as... The window feature vector corresponding to the electromyographic associated signal is denoted as The window feature vector corresponding to the grip force distribution signal is denoted as The window feature vector corresponding to the contact temperature signal is denoted as The feature vectors of the above modal windows are concatenated in a fixed order to obtain the multimodal feature vector corresponding to the current analysis window. .

[0038] In this embodiment, the emotion state recognition module establishes individual baseline parameters based on the statistical results of multimodal feature vectors corresponding to multiple consecutive analysis windows during the calm driving phase. Let's assume that a total of [number missing] emotion state data were obtained during the calm driving phase. Each analysis window corresponds to a multimodal feature vector. Then the baseline mean The baseline standard deviation is obtained by averaging the above multimodal eigenvectors dimension by dimension. The standard deviation is obtained by calculating the standard deviation of the above multimodal feature vectors dimension by dimension. The individual baseline parameters include the baseline mean and the baseline standard deviation.

[0039] In this embodiment, the emotion state recognition module extracts time from effective multimodal time series data. Multimodal feature vectors The time mentioned Multimodal feature vectors Because of time The standardized multimodal feature vector is obtained by extracting effective multimodal time series data within the preset feature window at the current time and performing baseline correction according to the following formula: ; In the formula, Indicates time The multimodal feature vectors, Indicates time The standardized multimodal feature vector, Indicates the baseline mean. Indicates the baseline standard deviation. Represents a very small positive number. Indicates time The contact effectiveness coefficient; the minimum positive number To avoid a denominator of zero; the contact effectiveness coefficient Outputted by the contact validity determination module, it is used to reduce the impact of the corresponding sampling point on the subsequent recognition results when the contact reliability is insufficient.

[0040] In this embodiment, the emotion state recognition module arranges the standardized multimodal feature vectors corresponding to multiple consecutive moments in chronological order to form a standardized multimodal feature sequence, and continuously updates the standardized multimodal feature sequence using a sliding window. Let the current sliding window contain the first... The standardized feature subsequences corresponding to each modality are: Then for Extract the window mean, window standard deviation, window slope, and window peak value to form the first... Temporal feature vectors of each modality within the current sliding window .

[0041] In this embodiment, the emotion state recognition module determines modality reliability based on the contact validity coefficient, effective sample ratio, and missing rate corresponding to different input source data within the current sliding window. Let the first... The average contact effectiveness coefficient of each mode within the current sliding window is: The percentage of valid samples is The missing rate is Then the modal reliability of this mode. Determine using the following formula: ; In the formula, , and This refers to the credibility weight parameter pre-set and stored in the emotion state recognition module. The average contact effectiveness coefficient... The effective sample ratio is obtained by averaging the contact effectiveness coefficients of the sampling points corresponding to this mode within the current sliding window; The missing rate is the proportion of sampling points with a contact validity coefficient not lower than the contact validity threshold within the current sliding window to the total number of sampling points; This represents the proportion of the number of masked sampling points within the current sliding window to the total number of sampling points.

[0042] In this embodiment, the emotion state recognition module performs weighted fusion of the temporal feature vectors of each modality according to modality confidence. Assume that the current sliding window includes a total of... If there are multiple modalities, then the feature vectors are fused. Determine using the following formula: ; In the formula, Indicates time The fused feature vector corresponding to the current sliding window, Indicates the first Modal reliability of each modality Indicates the first The temporal feature vector of each modality within the current sliding window.

[0043] In this embodiment, the emotion state recognition module is based on fused feature vectors. Calculate anger, anxiety, and fatigue scores. Let the rating parameters for anger be... and r, the rating parameter corresponding to anxiety is and The scoring parameters corresponding to fatigue are: and The state scores for the three emotion categories are determined by the following formulas: ; In the formula, , , The scoring weight vector is pre-set and stored in the emotion state recognition module. , , These are the corresponding rating bias parameters. Subsequently, the state scores for the three emotion categories are normalized to obtain the probability of anger. Anxiety probability and fatigue probability It is determined by the following formula: The sum of the three is 1.

[0044] In this embodiment, when the probability corresponding to the same emotion category exceeds a preset recognition threshold within multiple consecutive sliding windows, the emotion state recognition module determines that an emotion state change has occurred and outputs an emotion state change confirmation result. Let the threshold for the number of consecutive sliding windows be... The identification threshold is Then, when the probability corresponding to a certain emotion category is continuous Within each sliding window, all are greater than When the emotion category is reached, the corresponding emotion state change confirmation result is output. The threshold for the number of consecutive sliding windows and the recognition threshold are both pre-set judgment parameters stored in the emotion state recognition module. After the above processing, the emotion state recognition module outputs the probability of anger, the probability of anxiety, the probability of fatigue, and the emotion state change confirmation result to the driving risk assessment module.

[0045] (4) Driving risk assessment module

[0046] In this embodiment, the driving risk assessment module is used to determine a comprehensive risk score based on emotion recognition results, abnormality of driving behavior, and danger level of the scenario.

[0047] Specifically, the driving risk assessment module is connected to the emotion state recognition module to receive the probabilities of anger, anxiety, and fatigue. Simultaneously, the driving risk assessment module also receives vehicle behavior signals such as steering wheel angle, throttle opening, brake input, vehicle speed, distance to other vehicles, lane center deviation, road curvature, and lateral acceleration, either output by the steering wheel biosensor module or provided by the vehicle behavior acquisition unit. The inputs to the driving risk assessment module include the emotion recognition results and vehicle behavior signals, and the output is the comprehensive risk score corresponding to each sampling time.

[0048] In this embodiment, the driving risk assessment module determines the degree of abnormality in driving behavior based on the steering wheel angle sequence, throttle opening sequence, braking input sequence, and vehicle speed change; determines the scene hazard level based on vehicle distance, lane center deviation, road curvature, and lateral acceleration; and determines a comprehensive risk score based on emotion recognition results, driving behavior abnormality, and scene hazard level. To ensure the temporal continuity of the risk assessment results at each moment, the driving risk assessment module uses a length of [length missing]. The analysis window performs windowing processing on vehicle behavior signals. Let time... The corresponding analysis window is from time 1000. At that time If the continuous sampling interval is defined, then the steering wheel angle sequence within this analysis window is denoted as... The throttle opening sequence is denoted as The braking input sequence is denoted as .

[0049] In this embodiment, the degree of abnormality in driving behavior Determine using the following formula: ; In the formula, Indicates time abnormality of driving behavior Indicates length is The sequence of steering wheel angles within the analysis window. Indicates standard deviation, Indicates the throttle opening sequence. Indicates the braking input sequence. Represents sample entropy. Indicates the change in vehicle speed. , , and This refers to the weight parameters that are pre-calibrated and stored in the driving risk assessment module. The sample entropy is calculated on the corresponding sequence using a preset embedding dimension and similarity tolerance, which are calculation parameters that are pre-set and stored in the driving risk assessment module.

[0050] The change in vehicle speed From the current vehicle speed Vehicle speed at the previous sampling time The difference is determined, that is: ; In this embodiment, the scene hazard level Determine using the following formula: ; In the formula, Indicates time The level of danger in the scene , , and These represent the normalized oncoming distance, lane center offset, road curvature, and lateral acceleration, respectively. , , and This refers to the weight parameters that are pre-calibrated and stored in the driving risk assessment module.

[0051] Among them, time The distance between vehicles is recorded as Then the reverse distance can be expressed as ,in To prevent extremely small positive numbers with a denominator of zero, linear normalization based on a preset range is applied to the oncoming distance, lane center offset, road curvature, and lateral acceleration, uniformly mapping them to the same numerical range. Let any quantity to be normalized be... Then its normalization result Determine using the following formula: ; In the formula, and These represent the preset minimum and maximum values ​​of the quantity, respectively. Therefore, the normalized reverse distance is denoted as... The normalized lane center offset is denoted as The normalized road curvature is denoted as The normalized lateral acceleration is denoted as .

[0052] In this embodiment, the comprehensive risk score Determine using the following formula: ; In the formula, Indicates time The overall risk score, , and Representing time respectively The probability of anger, anxiety, and fatigue. , , , and This represents the pre-calibrated weight parameters stored in the driving risk assessment module. Analysis window length. The preset analysis window length is used to determine the time range corresponding to the calculation of driving behavior anomaly degree.

[0053] In this embodiment, the driving risk assessment module outputs the comprehensive risk score corresponding to the current moment at each sampling time or each analysis window update. The comprehensive risk score is output to the hierarchical active intervention module, which is used to subsequently determine the monitoring state control command, the early warning state control command, or the active intervention state control command based on the first risk threshold and the second risk threshold.

[0054] (5) Tiered active intervention module In this embodiment, the graded active intervention module is used to output monitoring control commands, early warning control commands, or active intervention control commands based on the relationship between the comprehensive risk score and the first risk threshold and the second risk threshold. The active intervention control commands include vehicle control commands, cabin adjustment commands, or a combination of the two. When the comprehensive risk score is continuously lower than the recovery threshold and continues to reach the preset recovery time, the release conditions of the active intervention control commands are verified, and the active intervention control commands are released level by level after the verification is passed.

[0055] Specifically, the tiered proactive intervention module is connected to the driving risk assessment module to receive the comprehensive risk score corresponding to each sampling time. It also receives the probability of anger output by the emotion state recognition module. Anxiety probability and fatigue probability The inputs to the tiered proactive intervention module include a comprehensive risk score, probability of anger, probability of anxiety, and probability of fatigue. The outputs include monitoring control commands, early warning control commands, proactive intervention control commands, and the release results of proactive intervention control commands.

[0056] In this embodiment, the first risk threshold and the second risk threshold are pre-set risk grading parameters stored in the tiered active intervention module, and the first risk threshold is less than the second risk threshold. When the comprehensive risk score... When the risk score is below the first risk threshold, the tiered active intervention module outputs a monitoring-state control command; when the comprehensive risk score is below the first risk threshold, the module outputs a monitoring-state control command. When the risk score is greater than or equal to the first risk threshold and less than the second risk threshold, the tiered proactive intervention module outputs a warning control command; when the comprehensive risk score is... When the risk level is greater than or equal to the second risk threshold, the graded active intervention module outputs an active intervention control command. The active intervention control command includes vehicle control commands, cabin adjustment commands, or a combination of both.

[0057] In this embodiment, the graded proactive intervention module determines the level of anger based on the probability of anger. Anxiety probability and fatigue probability The dominant emotional state is determined. If the maximum value among the three corresponds to the probability of anger, then the emotional state of anger is determined to be dominant; if the maximum value corresponds to the probability of anxiety, then the emotional state of anxiety is determined to be dominant; if the maximum value corresponds to the probability of fatigue, then the emotional state of fatigue is determined to be dominant. Thus, the dominant emotional state is directly linked to the probability of the emotional category output by the emotional state recognition module.

[0058] In active intervention mode, when anger is the dominant emotion, the tiered active intervention module outputs torque limiting and vehicle warning commands; when anxiety is the dominant emotion, it outputs target following distance adjustment and cabin adjustment commands; and when fatigue is the dominant emotion, it outputs lane centering enhancement and steering wheel vibration commands. The torque limiting command limits vehicle power output or torque response; the vehicle warning command alerts the driver to current risks via a warning device; the target following distance adjustment command increases the longitudinal control target distance; the cabin adjustment command changes current cabin environmental parameters; the lane centering enhancement command strengthens lateral control; and the steering wheel vibration command provides a tactile reminder to the driver. Therefore, each active intervention control command is linked to both the overall risk score and the dominant emotional state.

[0059] In this embodiment, the recovery threshold and preset recovery duration are pre-set and stored in the hierarchical active intervention module as release judgment parameters, and the recovery threshold is lower than the second risk threshold. After outputting the active intervention state control command, the hierarchical active intervention module continuously monitors whether the comprehensive risk score is continuously lower than the recovery threshold and counts the duration of continuous lower than the recovery threshold. When the comprehensive risk score is continuously lower than the recovery threshold and continuously reaches the preset recovery duration, it serves as the basic condition for release condition verification. If the comprehensive risk score reaches or exceeds the recovery threshold again during the statistical process, the continuous timing restarts.

[0060] In this embodiment, the release condition verification is based at least on the aforementioned basic conditions. That is, when the comprehensive risk score is continuously lower than the recovery threshold and persists for a preset recovery time, the tiered active intervention module verifies the release condition of the active intervention control command. In some implementations, the release condition verification may further combine the emotional probability corresponding to the current dominant emotional state and the abnormality of the current driving behavior for additional verification. Specifically, a preset release probability threshold and a preset behavior recovery threshold can be set, and after the basic conditions are met, it is further determined whether the emotional probability corresponding to the current dominant emotional state is lower than the preset release probability threshold, and whether the abnormality of the current driving behavior is lower than the preset behavior recovery threshold. The preset release probability threshold and the preset behavior recovery threshold are both preset verification parameters stored in the tiered active intervention module.

[0061] In this embodiment, after the release condition verification is passed, the tiered active intervention module releases the active intervention control commands level by level according to a preset order. The preset order is a release order parameter pre-set and stored in the tiered active intervention module. If the current active intervention control command includes multiple control items, the control items with higher vehicle control intensity can be released first, followed by warning and reminder control items, and finally cabin adjustment control items. A preset release interval time can also be set between each release level to avoid sudden changes in control state; this preset release interval time is a time parameter pre-set and stored in the tiered active intervention module.

[0062] In this embodiment, during the step-by-step deactivation process, if the comprehensive risk score reaches or exceeds the second risk threshold again, the active intervention control command is resumed. The comprehensive risk score reaching or exceeding the second risk threshold again constitutes the basic recovery condition for resuming the output of the active intervention control command. In some implementations, the emotional probability corresponding to the dominant emotional state and the degree of abnormal driving behavior can be further combined as additional recovery conditions to improve the safety and targeting of the recovery control. Specifically, when the comprehensive risk score reaches or exceeds the second risk threshold again, it can be simultaneously determined whether the emotional probability corresponding to the dominant emotional state and the degree of abnormal driving behavior have reached the corresponding recovery thresholds; when the additional recovery conditions are met, the active intervention control command is resumed.

[0063] After the above processing, the tiered proactive intervention module outputs control results corresponding to the current comprehensive risk level and dominant emotional state, and executes the proactive intervention state control instructions to maintain, release, or restore them based on the comprehensive risk score and the verification results of the release conditions.

[0064] like Figure 2 As shown, another embodiment of the present invention also provides a steering wheel bio-sensing driving risk intervention method, applied to a steering wheel bio-sensing driving risk intervention system as described above, comprising the following steps: S1. Collect skin conductance signals, electromyographic correlation signals, grip force distribution signals, contact temperature signals, and vehicle behavior signals related to the driver's contact area with the steering wheel to form raw multimodal time series data; S2. Determine the contact effectiveness coefficient based on the grip force distribution signal and vehicle behavior signal, and perform weight reduction, interpolation or masking on the distorted samples in the original multimodal time series data based on the contact effectiveness coefficient, and output effective multimodal time series data; S3. Determine the calm driving phase based on the continuous time period after driving begins that meets the preset stability conditions, and establish individual baseline parameters. S4. Based on the contact effectiveness coefficient and individual baseline parameters, the baseline correction is performed on the multimodal feature vectors corresponding to the effective multimodal time series data to form a standardized multimodal feature sequence. Then, the time series feature extraction and multimodal fusion are performed on the standardized multimodal feature sequence to output the emotion recognition result. S5. Determine the comprehensive risk score based on the emotion recognition results, the degree of abnormality in driving behavior, and the degree of danger in the scenario; S6. Output a monitoring control command, an early warning control command, or an active intervention control command based on the relationship between the comprehensive risk score and the first risk threshold and the second risk threshold. The active intervention control command includes a vehicle control command, a cabin adjustment command, or a combination of the two. S7. When the comprehensive risk score is continuously lower than the recovery threshold and the preset recovery time is reached, the conditions for releasing the active intervention control command are verified, and the active intervention control command is released step by step after the verification is passed.

[0065] In this embodiment, the above method steps correspond to the processing procedures of the aforementioned steering wheel biosensing module, contact validity determination module, emotional state recognition module, driving risk assessment module, and graded active intervention module. The specific implementation methods of signal acquisition, parameter determination, data processing, state recognition, risk assessment, and graded intervention control involved in each step can be found in the relevant content of the aforementioned system embodiments, and will not be repeated here. Through the above method, continuous perception, recognition, assessment, and graded intervention of the driver's state can be achieved, which helps to improve the accuracy, stability, and safety of driver state recognition, risk assessment, and active intervention control.

[0066] In summary, this invention, through the collaborative design of collecting multi-source information related to the driver-steering wheel contact area, processing contact effectiveness, recognizing emotional states, assessing risks, and implementing tiered active intervention control, forms a complete, coherent, and implementable technical solution. This solution can improve the accuracy and stability of driver state recognition and risk assessment, and enhance the safety and rationality of active intervention and exit control processes, thus possessing significant engineering application value.

[0067] The present invention has been further described above with reference to specific embodiments, but the present invention is not limited to the above embodiments. The various embodiments and technical features described in this specification can be combined with each other without contradiction. Without departing from the concept of the present invention, those skilled in the art can make equivalent substitutions or modifications to the technical solutions of the present invention, and these should all fall within the protection scope of the present invention. The protection scope of the present invention is defined by the claims.

Claims

1. A steering wheel-based bio-sensing driving risk intervention system, characterized in that, include: The steering wheel biosensing module is used to collect skin conductance signals, electromyographic correlation signals, grip force distribution signals, contact temperature signals and vehicle behavior signals related to the driver's contact area with the steering wheel, forming raw multimodal time-series data; The contact validity determination module is used to determine the contact validity coefficient based on the grip force distribution signal and vehicle behavior signal, and to perform weight reduction, interpolation or masking on the distorted samples in the original multimodal time series data based on the contact validity coefficient, and output valid multimodal time series data. The emotion state recognition module is used to determine the calm driving stage based on a continuous time period after driving begins and meets preset stable conditions, and to establish individual baseline parameters. Based on the contact effectiveness coefficient and individual baseline parameters, the module performs baseline correction on the multimodal feature vectors corresponding to the effective multimodal time series data, forming a standardized multimodal feature sequence. After time series feature extraction and multimodal fusion, the module outputs the emotion recognition result. The driving risk assessment module is used to determine a comprehensive risk score based on the emotion recognition result, the abnormality of driving behavior, and the hazard level of the scenario. The graded active intervention module is used to output monitoring control commands, early warning control commands, or active intervention control commands based on the relationship between the comprehensive risk score and the first risk threshold and the second risk threshold. The active intervention control commands include vehicle control commands, cabin adjustment commands, or a combination of the two. When the comprehensive risk score is continuously lower than the recovery threshold and continues to reach the preset recovery time, the module verifies the release conditions of the active intervention control commands and releases the active intervention control commands level by level after the verification is passed.

2. The steering wheel bio-sensing driving risk intervention system according to claim 1, characterized in that, The steering wheel biosensing module includes: The system includes a skin conductance acquisition unit, located on the surface of the steering wheel grip area, for acquiring skin conductance signals from the driver's palm contact area; an electromyography (EMG) correlation acquisition unit, located inside the steering wheel grip area, for acquiring EMG correlation signals from the driver's palm and forearm; a grip force distribution acquisition unit, located in the steering wheel grip area, for acquiring grip force distribution signals and pressure center location information; a contact temperature acquisition unit, located in the steering wheel grip area and corresponding to the grip force distribution acquisition unit in the sampling area, for acquiring the driver's palm contact temperature signal; and a vehicle behavior acquisition unit, connected to the vehicle bus, for acquiring steering wheel angle, vehicle speed, throttle opening, braking intensity, vehicle distance, lane center offset, road curvature, and lateral acceleration. The steering wheel biosensing module also aligns the skin conductance signals, EMG correlation signals, grip force distribution signals, contact temperature signals, and vehicle behavior signals according to a unified time reference and correlates them based on the corresponding sampling areas to form raw multimodal time-series data.

3. The steering wheel bio-sensing driving risk intervention system according to claim 1, characterized in that, The contact effectiveness determination module determines the steering wheel contact coverage, average pressure matrix, and pressure center offset based on the grip force distribution signal, determines the steering wheel angular velocity based on the vehicle behavior signal, and determines the contact effectiveness coefficient based on the steering wheel contact coverage, average pressure matrix, steering wheel angular velocity, and pressure center offset; the contact effectiveness coefficient is calculated according to the following formula: ； In the formula, Indicates time The contact effectiveness coefficient, Indicates steering wheel contact coverage. This represents the mean of the pressure matrix. Indicates the angular velocity of the steering wheel. Indicates the offset of the pressure center. , , and Represents the weight parameters. Represents the Sigmoid mapping function; The contact validity determination module determines sampling points with contact validity coefficients lower than a preset threshold as distorted samples, and performs weight reduction processing, interpolation processing, or masking processing on the distorted samples.

4. The steering wheel bio-sensing driving risk intervention system according to claim 1, characterized in that, The emotion state recognition module determines the calm driving phase based on a continuous time period after driving begins that meets preset stability conditions. It then establishes individual baseline parameters based on the statistical results of multimodal feature vectors corresponding to multiple consecutive analysis windows within the calm driving phase. These individual baseline parameters include the baseline mean and baseline standard deviation of the multimodal feature vectors. The emotion state recognition module extracts multimodal feature vectors from valid multimodal time-series data and performs baseline correction according to the following formula to obtain standardized multimodal feature vectors: ； In the formula, Indicates time The multimodal feature vectors, Indicates time The standardized multimodal feature vector, Indicates the baseline mean. Indicates the baseline standard deviation. Represents a very small positive number. Indicates time The contact effectiveness coefficient; the emotion state recognition module arranges the standardized multimodal feature vectors corresponding to multiple consecutive moments in chronological order to form a standardized multimodal feature sequence.

5. A steering wheel bio-sensing driving risk intervention system according to claim 4, characterized in that, The emotion state recognition module uses a sliding window to continuously update the standardized multimodal feature sequence and extracts temporal features from the standardized multimodal feature sequence within the current sliding window. The emotion state recognition module determines modality credibility based on the contact validity coefficient, effective sample ratio, and missing rate of data from different input sources within the current sliding window, and performs weighted fusion of each modality feature according to the modality credibility to output the probability of the emotion category; the emotion categories include anger, anxiety, and fatigue; when the probability corresponding to the same emotion category is higher than the preset recognition threshold in multiple consecutive sliding windows, it is determined that an emotion state change has occurred.

6. The steering wheel bio-sensing driving risk intervention system according to claim 1, characterized in that, The driving risk assessment module determines the degree of abnormality of driving behavior based on the steering wheel angle sequence, throttle opening sequence, braking input sequence and vehicle speed change, determines the degree of danger of the scene based on the distance between vehicles, lane center deviation, road curvature and lateral acceleration, and determines the comprehensive risk score based on the emotion recognition results, the degree of abnormality of driving behavior and the degree of danger of the scene.

7. The steering wheel bio-sensing driving risk intervention system according to claim 1, characterized in that, The abnormality of driving behavior, the danger level of the scenario, and the comprehensive risk score are determined according to the following formula: ； In the formula, Indicates time abnormality of driving behavior This indicates that the preset analysis window length is... The sequence of steering wheel angles within the analysis window. Indicates standard deviation, Indicates the throttle opening sequence. Indicates the braking input sequence. Represents sample entropy. Indicates the change in vehicle speed. Indicates time The level of danger in the scene , , and These represent the normalized oncoming distance, lane center offset, road curvature, and lateral acceleration, respectively. , , and This refers to the pre-calibrated and stored weight parameters in the driving risk assessment module. Indicates time The overall risk score, , and Representing time respectively The probability of anger, anxiety, and fatigue. , , , , , , , and Represents the weight parameters. Indicates the preset analysis window length.

8. The steering wheel bio-sensing driving risk intervention system according to claim 1, characterized in that, The graded active intervention module outputs monitoring-state control commands, warning-state control commands, or active intervention-state control commands based on the relationship between the comprehensive risk score and the first risk threshold and the second risk threshold, wherein the first risk threshold is less than the second risk threshold; the active intervention-state control commands include vehicle control commands, cabin adjustment commands, or a combination of both; the graded active intervention module outputs torque limiting commands and vehicle warning commands when the emotional state corresponding to the anger probability is dominant, outputs target following distance adjustment commands and cabin adjustment commands when the emotional state corresponding to the anxiety probability is dominant, and outputs lane centering enhancement commands and steering wheel vibration commands when the emotional state corresponding to the fatigue probability is dominant.

9. A steering wheel bio-sensing driving risk intervention system according to claim 8, characterized in that, When the system is in an active intervention state, the hierarchical active intervention module verifies the release conditions of the active intervention state control command. The release conditions include: the comprehensive risk score is continuously lower than the recovery threshold and continues to reach the preset recovery time, and the recovery threshold is lower than the second risk threshold; after the release conditions are verified, the hierarchical active intervention module releases the corresponding control commands level by level in a preset order; during the step-by-step release process, if the comprehensive risk score reaches or exceeds the second risk threshold again, the active intervention state control command is restored.

10. A steering wheel bio-sensing driving risk intervention method, applied to a steering wheel bio-sensing driving risk intervention system as described in any one of claims 1-9, characterized in that, Includes the following steps: S1. Collect skin conductance signals, electromyographic correlation signals, grip force distribution signals, contact temperature signals, and vehicle behavior signals related to the driver's contact area with the steering wheel to form raw multimodal time series data; S2. Determine the contact effectiveness coefficient based on the grip force distribution signal and vehicle behavior signal, and perform weight reduction, interpolation or masking on the distorted samples in the original multimodal time series data based on the contact effectiveness coefficient, and output effective multimodal time series data; S3. Determine the calm driving phase based on the continuous time period after driving begins that meets the preset stability conditions, and establish individual baseline parameters. S4. Based on the contact effectiveness coefficient and individual baseline parameters, the baseline correction is performed on the multimodal feature vectors corresponding to the effective multimodal time series data to form a standardized multimodal feature sequence. Then, the time series feature extraction and multimodal fusion are performed on the standardized multimodal feature sequence to output the emotion recognition result. S5. Determine the comprehensive risk score based on the emotion recognition results, the degree of abnormality in driving behavior, and the degree of danger in the scenario; S6. Output a monitoring control command, an early warning control command, or an active intervention control command based on the relationship between the comprehensive risk score and the first risk threshold and the second risk threshold. The active intervention control command includes a vehicle control command, a cabin adjustment command, or a combination of the two. S7. When the comprehensive risk score is continuously lower than the recovery threshold and the preset recovery time is reached, the conditions for releasing the active intervention control command are verified, and the active intervention control command is released step by step after the verification is passed.

Images