A millimeter wave radar-based driver state monitoring and safety warning method

Abstract

This invention discloses a driver state monitoring and safety early warning method based on millimeter-wave radar, comprising: acquiring driver physiological micro-motion signals and driver macro-behavioral signals respectively through millimeter-wave radar to obtain driving environment information in real time; denoising the driver physiological micro-motion signals and extracting features to obtain driver physiological feature signals; constructing a time-frequency spectrum using driver macro-behavioral signals; determining the driver physiological arousal index and driver emotional state risk value based on the physiological feature signals; obtaining the driver dynamic behavior risk value based on the time-frequency spectrum; calculating an environmental complexity index based on driving environment information; constructing a driver state risk field model based on the driver physiological arousal index, driver emotional state risk value, and driver dynamic behavior risk value; calculating the driver state risk value using the driver state risk field model and environmental complexity index; and providing graded early warning based on the driver state risk value.

Description

Technical Field

[0001] This invention belongs to the field of intelligent vehicle and human-computer interaction technology, and specifically relates to a method for driver status monitoring and safety early warning based on millimeter-wave radar. Background Technology

[0002] As autonomous driving technology advances to Level 3 and above, the driver's role is gradually shifting from continuous operator to system supervisor and, when necessary, takeover operator. In this process, the driver's physical and mental state not only affects takeover performance but also directly impacts the safety and comfort of the entire driving experience. Research shows that physiological abnormalities (such as sudden heart rate arrhythmias), intense emotional fluctuations (such as anger or anxiety), severe distraction, or fatigue, even when the autonomous driving system is functioning normally, can pose safety hazards due to sudden intervention needs or system boundary scenarios. Therefore, continuous, non-invasive monitoring and assessment of the driver's state has become a crucial component of intelligent vehicle safety systems.

[0003] Current driver condition monitoring technologies are mainly divided into three categories:

[0004] Visual sensor-based methods use cameras to capture information such as the driver's facial expressions, eye movements, and head posture to determine their fatigue, distraction, and emotional state. While intuitive, these methods are significantly limited by changes in lighting, occlusion, and camera angle. Furthermore, continuous video monitoring raises privacy concerns, posing challenges to acceptance and compliance in automotive-grade applications.

[0005] Wearable device-based methods: These methods collect physiological signals such as heart rate, skin conductance, and brain waves using devices like smart bracelets, heart rate belts, and EEG caps. While these methods offer high data accuracy, they require users to actively wear the devices, leading to intrusiveness and a burden. Furthermore, device battery life, comfort, and long-term user compliance are difficult to guarantee, making them unsuitable for seamlessly integrated in-vehicle environments.

[0006] Indirect inference methods based on other in-vehicle sensors: such as inferring the driver's state indirectly through vehicle interaction signals like steering wheel grip force and pedal operation characteristics. These methods are reactive or ad-hoc judgments, lacking the ability to perceive the driver's state during non-control phases, and are easily confused with driving style, resulting in limited timeliness and accuracy of warnings.

[0007] Millimeter-wave radar technology has emerged in the field of in-vehicle sensing in recent years due to its unique advantages such as non-contact operation, strong penetration, high environmental robustness, and good privacy protection. This technology transmits electromagnetic waves in the millimeter-wave band and receives their echoes, enabling precise detection of microscopic movements in the chest cavity (micrometer level) caused by breathing and heartbeat, as well as macroscopic movements such as body posture and hand gestures. Compared to visual solutions, millimeter-wave radar is unaffected by lighting conditions, does not acquire visual images, and is privacy-friendly; compared to wearable devices, it does not require wearing, enabling imperceptible and continuous monitoring; and compared to indirect inference from vehicle interaction signals, it provides direct and rich physiological and behavioral information.

[0008] Existing technical solutions mostly focus on fragmented single-point functions such as occupant presence detection, vital sign monitoring, or simple gesture recognition. They have not yet formed a panoramic state monitoring system that can cover multiple dimensions such as driver physiology, emotions, and behavior, and lack dynamic risk assessment indicators that are compatible with it. Therefore, they cannot provide a systematic decision-making basis for judging takeover capabilities and safety warnings in high-level autonomous driving scenarios. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide a driver status monitoring and safety early warning method based on millimeter-wave radar. This method collects driver physiological data and behavioral information in a non-contact manner and constructs a dynamic driver status risk field model in combination with environmental risks, which can achieve accurate identification and graded early warning of the driver's real-time status.

[0010] The technical solution provided by this invention is as follows:

[0011] A method for driver status monitoring and safety early warning based on millimeter-wave radar, comprising:

[0012] The system collects driver physiological micro-motion signals and driver macro-behavioral signals using millimeter-wave radar, and also acquires real-time driving environment information.

[0013] After denoising the driver's physiological micro-motion signals, feature extraction is performed to obtain the driver's physiological feature signals; and a time-frequency spectrum is constructed using the driver's macroscopic behavioral signals.

[0014] The driver's physiological arousal index and driver's emotional state risk value are determined based on the physiological characteristic signals; and the driver's dynamic behavior risk value is obtained based on the time-frequency spectrum.

[0015] Calculate the environmental complexity index based on the driving environment information;

[0016] A driver state risk field model is constructed based on the driver's physiological arousal index, driver's emotional state risk value, and driver's dynamic behavior risk value. The driver's state risk value is calculated using the driver state risk field model and the environmental complexity index.

[0017] Early warnings are issued in tiers based on the driver's risk level.

[0018] Preferably, after denoising the driver's physiological micro-motion signal, feature extraction is performed to obtain the normalized deviation value of heart rate and the normalized deviation value of respiratory rate, and the driver's physiological arousal index is calculated based on the normalized deviation value of heart rate and the normalized deviation value of respiratory rate.

[0019] The formula for calculating the heart rate normalization deviation is as follows:

[0020] ;

[0021] The formula for calculating the normalized deviation of the respiratory rate is:

[0022] ;

[0023] In the formula, The average heart rate of the driver. The average breathing rate of the driver. Baseline heart rate This is the baseline respiratory rate.

[0024] Preferably, the driver's physiological arousal index is:

[0025] ;

[0026] in, and These are the weighting coefficients for heart rate and respiratory rate; and These are the standard deviations of the normalized deviations of heart rate and respiratory rate, respectively. This is the offset parameter.

[0027] Preferably, the driver state risk field model is as follows:

[0028] ;

[0029] in, Driver status risk value; The risk value is the driver's emotional state. This represents the driver's dynamic behavioral risk value. , and These are the driver's physiological arousal weight coefficient, driver's emotional state risk weight coefficient, and driver's behavioral risk weight coefficient, respectively, determined based on the complexity of the driving environment.

[0030] Preferably, the formula for calculating the complexity of the driving environment is:

[0031] ;

[0032] in, , , These are the normalized values ​​of traffic density, road curvature, and weather factors, respectively. For the collision risk function based on TTC, , , and These are the weight coefficients for the corresponding items.

[0033] Preferably, the formula for calculating the driver's emotional state risk value is as follows:

[0034] ;

[0035] in, This represents the baseline emotional risk coefficient at the current moment. This represents the confidence level corresponding to the current emotion category.

[0036] Preferably, the formula for calculating the driver's dynamic behavior risk value is as follows:

[0037] ;

[0038] Where τ is the time decay constant, This represents the basic risk coefficient of the behavior at the current moment. Indicates the duration of the action at the current moment.

[0039] Preferably, the collision risk function based on TTC is:

[0040] ;

[0041] in, As a safety threshold, This is the emergency threshold.

[0042] Preferably, the driver status monitoring and safety warning method based on millimeter-wave radar further includes:

[0043] Set up an emergency TTC handling mechanism: When At that time, the highest level of warning will be activated.

[0044] The beneficial effects of this invention are:

[0045] 1. Enables all-weather, contactless, and highly privacy-protected driver status monitoring: Utilizing millimeter-wave radar technology, it is unaffected by changes in light, does not collect visual images, and requires no equipment worn by the driver. While ensuring privacy, it achieves seamless and continuous physiological and behavioral monitoring, solving the problems of occlusion and privacy associated with traditional visual solutions, as well as the intrusiveness and compliance issues of wearable devices.

[0046] 2. Multi-dimensional state fusion assessment to improve the comprehensiveness and accuracy of state judgment: Simultaneously collect physiological micro-motion signals and macro-behavioral signals, and combine them with multi-dimensional indicators such as emotion, arousal level, and behavioral posture to construct a comprehensive driver state profile, avoiding the limitations of a single signal source and significantly improving the robustness and reliability of state recognition.

[0047] 3. Environmentally Adaptive Dynamic Risk Assessment Mechanism: Environmental information (traffic, road conditions, weather, etc.) is used as the basis for weight adjustment, rather than direct risk factors. Different state weights are preset according to low, medium, and high environmental complexity, so that the system has differentiated risk perception and response strategies in different driving scenarios, thereby enhancing the system's scenario adaptability.

[0048] 4. Tiered early warning and intelligent decision support: Based on the driver's state risk field, the system classifies risks into low, medium and high levels and matches them with multimodal early warning and system intervention strategies. In autonomous driving takeover scenarios, it provides real-time driver capability assessment and environmental risk information, supports personalized and progressive takeover guidance and safety intervention, and improves the safety and smoothness of human-machine collaboration.

[0049] 5. Enhance the overall safety and user experience of intelligent vehicles: The system is not only used in emergency takeover scenarios, but can also identify and warn of driver fatigue, distraction, and abnormal emotions in the early stages, which helps prevent accidents, improve driving comfort, and promote the safe implementation and user acceptance of high-level autonomous driving systems. Attached Figure Description

[0050] Figure 1 This is a block diagram of the overall structure of the driver status monitoring and safety early warning system based on millimeter-wave radar described in this invention.

[0051] Figure 2 This is a schematic diagram of a millimeter-wave radar arrangement in one embodiment of the present invention.

[0052] Figure 3 This is a flowchart of the driver status monitoring and safety early warning method based on millimeter-wave radar described in this invention. Detailed Implementation

[0053] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0054] This invention provides a driver status monitoring and safety warning method based on millimeter-wave radar, which is implemented through a driver status monitoring and safety warning system based on millimeter-wave radar.

[0055] like Figure 1 As shown, the driver state monitoring and safety early warning system based on millimeter-wave radar includes: a data acquisition module 110, a time synchronization and data processing module 120, a driver state analysis module 130, a dynamic driver risk field construction module 140, and a hierarchical early warning and decision support module 150. The data acquisition module 110 includes a radar sensing unit and an environmental information unit. The radar sensing unit includes at least one millimeter-wave radar unit for high-precision physiological signal acquisition and at least one millimeter-wave radar unit for wide-range behavioral signal acquisition, enabling non-contact synchronous data acquisition of the driver. The environmental information unit acquires environmental data in real time. The time synchronization and data fusion processing module 120 performs high-precision timestamp marking, clock synchronization, and time axis alignment on the data streams from multiple radar units. It also preprocesses and extracts features from the aligned physiological and behavioral signals to provide a consistent and reliable data foundation for subsequent analysis. The driver state analysis module 130 includes an emotion... Along with the arousal recognition unit and the behavior posture analysis unit, the emotion and arousal recognition unit combines physiological signal characteristics with specific behavioral patterns to identify the driver's emotional state and physiological arousal level through a machine learning model. The behavior posture analysis unit uses a deep learning model to analyze behavioral signals, identify the driver's body posture and behavioral actions, and determine whether the driver has the ability to cope with emergencies and the relevant risk level. The dynamic driver state risk field construction module 140 receives multi-dimensional state indicators from the driver state analysis module 130 and integrates real-time information from the vehicle bus (such as vehicle speed, steering angle, and acceleration) and the external environment perception system (such as traffic participant information, road curvature, and weather). Based on predefined coupling rules and risk assessment algorithms, a spatially and temporally distributed "driver state risk field" is dynamically calculated and generated. The dynamic driver state risk field quantitatively reflects the comprehensive risk level that may be introduced due to the driver's own condition under the current vehicle and environmental conditions. The graded early warning and decision support module 150 executes corresponding early warning strategies and provides decision-making basis for response strategies based on the output of the dynamic driver state risk field and preset risk thresholds at each level.

[0056] In one embodiment, the data acquisition module 110 is equipped with a physiological signal radar, a behavioral signal radar, and other necessary onboard sensors or vehicle-to-everything (V2X) communication systems to acquire information about other traffic-participating vehicles and the external environment. The physiological signal radar uses a Texas Instruments AWR1642 evaluation board, operating at a frequency of 77 GHz. Figure 2 As shown, the physiological signal radar 210 is installed at the upper edge of the windshield directly above the steering wheel, pointing towards the driver's chest area, for high-precision acquisition of micro-motion signals caused by breathing and heartbeat; the behavioral signal radar uses Infineon's BGT60LTR11AIP radar chip, operating at a frequency of 60GHz, and the behavioral signal radar 220 is installed on the A-pillar interior panel on the driver's side, with the beam pointing downwards to cover the driver's upper body and arm movement area, for monitoring macroscopic behaviors such as posture and gestures; the information on other traffic-related vehicles and the external environment includes, but is not limited to, comprehensive environmental information such as traffic flow status, road conditions, weather conditions, and potential hazardous events around the vehicle.

[0057] The time synchronization and data processing module 120 uses a host computer equipped with a high-performance, high-computing-power system-on-a-chip as its core controller. The host computer has a built-in high-precision system clock and connects to the physiological signal radar, behavioral signal radar, and other auxiliary sensors via multiple independent hardware serial ports (UARTs). Each UART port is configured with a high baud rate to support high-speed transmission of the radar's raw data stream. At the beginning of each processing cycle, the host computer generates an absolute timestamp based on its high-precision clock and sends a lightweight synchronization command packet containing the absolute timestamp to all radar units via the serial ports. Upon receiving the synchronization command, each radar unit immediately uses this timestamp as the zero-point reference for its local data acquisition timeline and adds a relative timestamp based on this reference to each subsequent frame of data. The host computer asynchronously receives data packets from different radars via the serial ports, aligns the data according to the relative timestamps, performs real-time data parsing on the aligned radar data, and outputs the processed data and environmental information from other sensors.

[0058] After receiving the preprocessed data, the driver state analysis module 130 inputs the physiological signal data to the emotion and arousal recognition unit, calculates the physiological arousal index, and outputs the emotion label using the LightGBM machine learning classification model; it also transmits the behavioral signal data to the behavioral posture analysis unit, which uses a convolutional neural network (CNN) to input 10 consecutive frames of time-frequency graphs and outputs the behavioral label.

[0059] The driver state risk field construction module 140 receives emotion tags, physiological arousal levels, behavioral tags, and environmental information, comprehensively calculates the driver state risk value, and outputs it.

[0060] The graded early warning and decision support module 150 executes corresponding early warning strategies and provides decision-making basis for response strategies based on the driver's status risk value, environmental complexity information, and preset risk thresholds at each level.

[0061] like Figure 3 As shown, the specific implementation process of the driver status monitoring and safety early warning method based on millimeter-wave radar provided by the present invention is as follows.

[0062] S310: Utilizes a multimodal millimeter-wave radar system deployed inside the vehicle to simultaneously and non-contactly collect the driver's physiological micro-motion signals and macro-behavioral signals, and obtains information about other traffic participants and the external environment through other onboard sensors or vehicle-to-everything (V2X) communication systems.

[0063] S320: Performs preprocessing on the multi-source radar data acquired by S310, including time synchronization, timestamp alignment, filtering and denoising, and extracts the corresponding time domain, frequency domain and time-frequency domain features.

[0064] S330: Based on the feature signals processed by S320, assess the driver's physiological arousal level and behavioral risk value;

[0065] S331: Assess environmental complexity based on information obtained from S310 regarding other traffic-participating vehicles and the external environment;

[0066] S340: The driver state index obtained in step S330 is fused with the environmental complexity obtained in S331, and the driver state risk field is dynamically constructed based on the risk assessment model.

[0067] S350: Based on the driver status risk field and preset strategies quantified by S340, and combined with environmental complexity as a classification reference, execute corresponding safety warnings, system interaction adjustments, or takeover decision support operations.

[0068] As a preferred embodiment, in S310, the driver's physiological micro-motion signals acquired by the radar sensing unit in the data acquisition module include, but are not limited to, heart rate and respiratory rate signals. Both the physiological micro-motion signals and macroscopic behavioral signals are represented in the form of radar echoes. Preferably, a millimeter-wave radar with a frequency-modulated continuous wave (FMCW) system is used, the basic principle of which is as follows:

[0069] Transmit signal For linear frequency modulated continuous wave:

[0070] ;

[0071] in, For the amplitude of the transmitted signal, For carrier frequency, For signal bandwidth, For frequency modulation period, For the initial phase, This indicates the time elapsed since the radar started transmitting.

[0072] The received echo signal after being reflected by the target There is a delay :

[0073] ;

[0074] in, For the received signal amplitude, The phase change introduced for target reflection.

[0075] After the echo signal and the transmitted signal are mixed and low-pass filtered, the intermediate frequency signal (beat frequency signal) is obtained.

[0076] ;

[0077] Among them, the shooting frequency Distance from target Proportional:

[0078] ;

[0079] In the formula, It is the speed of light.

[0080] The target distance can then be calculated:

[0081] ;

[0082] For micro-motion sensing, physiological activities such as breathing and heartbeat of the driver will cause periodic micro-displacements on the surface of the chest cavity. This displacement will modulate the echo phase. :

[0083] ;

[0084] in, This is the radar wavelength.

[0085] The environmental information collected by the environmental information unit includes traffic-participating vehicle operation status information, external road and traffic environment information, and interaction and risk target information.

[0086] The vehicles participating in the traffic include the vehicle itself and all vehicles in the lane 150 meters in front of and 70 meters behind it in the direction of travel, as well as all vehicles in adjacent lanes.

[0087] The traffic-participating vehicle operation status information includes, but is not limited to:

[0088] Vehicle speed: The instantaneous speed of vehicles participating in traffic;

[0089] Acceleration: The instantaneous acceleration of vehicles participating in traffic;

[0090] Heading angle: The angle of relative motion of vehicles participating in traffic, with the positive direction of travel axis of the vehicle coordinate system as the reference;

[0091] The external road and traffic environment information includes, but is not limited to:

[0092] Traffic flow density: The number of vehicles participating in traffic per unit length of road segment, quantifying the degree of traffic congestion and the risk of following other vehicles;

[0093] Road curvature: The radius or curvature of the current road, which affects driving load and vehicle control difficulty;

[0094] Lane information: The type, width, and clarity of the lane where the vehicle is located, used to determine the drivable area and the risk of deviation;

[0095] Weather condition factors: Quantitative levels of weather such as rain, fog, and hail (e.g., rainfall intensity, visibility) affect the road surface adhesion coefficient and perception range;

[0096] Lighting conditions: Changes in lighting conditions such as day and night, tunnel entrances and exits, and glare may indirectly affect the driver's visual state and the system's perception capabilities;

[0097] The interaction and risk target information includes, but is not limited to:

[0098] Estimated time of collision: The estimated time of collision between the vehicle and the nearest obstacle (vehicle, pedestrian, etc.) at the current relative speed is a direct indicator for quantifying the risk of a forward collision.

[0099] Relative distance and speed: The relative motion state of the vehicle and key traffic participants in the surrounding area (such as the vehicle in front and cutting in);

[0100] Traffic incident information: Location and type of beyond-line-of-sight risk events such as accidents, construction, and congestion endpoints obtained through V2X or cloud platforms;

[0101] Special road area markings: such as school zones, construction zones, slippery road sections, and other areas requiring special attention or speed reduction;

[0102] Preferably, in S320, the raw echo signals acquired by the physiological signal radar are processed sequentially as follows:

[0103] a. Static clutter removal to eliminate reflections from fixed objects such as seats and steering wheels; preferably, a background subtraction algorithm is used for filtering and noise reduction, with the following formula:

[0104] ;

[0105] in, Representing discrete time points, The original (radar) signal of physiological micro-motions after time alignment. This is a pure background signal collected during system initialization or when the driver is not seated. To suppress physiological micro-motion signals that are disturbed by static background;

[0106] b. Perform phase demodulation and phase unrolling on the signal to obtain continuous physiological micro-motion signals;

[0107] c. Micro-displacement of the thoracic cavity (micro-motion signal) Bandpass filtering was performed to separate the respiratory frequency band signal (0.1–0.5 Hz) and the heart rate band signal (0.8–2.0 Hz).

[0108] d. Perform spectral analysis or autocorrelation analysis on the filtered signal, extract the dominant frequency peak, and calculate the driver's average heart rate (HR), driver's average respiratory rate (RR), and heart rate variability characteristics. Preferably, to reduce individual differences, a baseline heart rate is introduced. Compared with baseline respiratory rate Define the normalization bias:

[0109] ;

[0110] ;

[0111] in, This is due to heart rate normalization bias. This represents the normalization bias of respiratory rate.

[0112] Preferably, the heart rate variability features include SDNN (standard deviation of all sinus intervals), RMSSD (root mean square of the difference between adjacent RR intervals), and LF / HF (ratio of low-frequency power to high-frequency power), which are calculated using the following formula:

[0113] ;

[0114] ;

[0115] ;

[0116] in, For the first A period of time between heartbeats For all The arithmetic mean, The total number of RR intervals involved in the calculation; the power spectral density can be obtained by performing spectral analysis on the RR interval sequence, the low-frequency power LF is obtained by integrating the power in the 0.04-0.15Hz frequency band, and the high-frequency power HF is obtained by integrating the power in the 0.15-0.4Hz frequency band.

[0117] Short-time Fourier transform (STFT) is performed on the echo signals acquired by the behavior signal radar to construct a 64x64 time-frequency spectrum, which is used to describe the upper body movement characteristics of the driver and serves as the input to the behavior recognition model.

[0118] As a preferred embodiment, the specific steps in S330 are as follows:

[0119] a. Analyze the driver's physiological arousal level and construct a physiological arousal index based on normalized bias. The Sigmoid function is used to map the output to the [0,1] interval. A higher value indicates a higher level of physiological arousal. The calculation formula is as follows:

[0120] ;

[0121] in, and As a weighting coefficient, it determines the relative contribution of heart rate and respiratory changes to arousal. and θ represents the standard deviation of the corresponding indicator within the normal fluctuation range; θ is the offset parameter used to adjust the center point of the function; the value range of θ is [-1, 1]. When θ decreases, the curve shifts to the left and becomes more sensitive to physiological changes; when θ increases, the curve shifts to the right and becomes less sensitive to physiological changes.

[0122] b. Analyze the driver's emotional state. Calculate the driver's average heart rate (HR), average respiratory rate (RR), heart rate variability, and physiological arousal index to form a feature vector. Input this vector into a pre-trained LightGBM machine learning classification model to obtain the driver's most likely current emotional state label. and the corresponding confidence level, for each emotion category A static baseline emotional risk coefficient with a value range of [0,1] is preset. This output After determining the emotion category at the current moment, we can obtain the corresponding result. The baseline emotional risk coefficient at any given time; and the emotional state risk value is calculated. The calculation formula is as follows:

[0123] ;

[0124] in, The confidence level for each emotion category; This represents the baseline emotional risk coefficient at the current moment.

[0125] In one embodiment, the emotion state label The basic emotional risk coefficients for the emotions include calmness, joy, sadness, anxiety, anger, and fear, with corresponding coefficients of 0.1, 0.2, 0.5, 0.6, 0.8, and 1.0, respectively.

[0126] c. A lightweight temporal convolutional neural network (CNN) deep learning model is used to identify driver behavior and posture. The model receives a short-sequence time-frequency graph stack as input and outputs the most likely action category at the current moment. For each identifiable action category Based on the severity of its impact on driving task interruption and the prior statistical probability of causing an accident, a static basic risk coefficient with a value range of [0,1] is pre-defined. Furthermore, the system tracks the duration of each identified action in real time. Calculate the current time Dynamic behavioral risk value The calculation formula is as follows:

[0127] ;

[0128] Where τ is the time decay constant, This represents the basic risk coefficient of the behavior at the current moment.

[0129] In one embodiment, action category The basic risk factors include focusing on observing driving conditions, talking to passengers, drinking water / eating, operating the central control screen, making phone calls, turning one's head, using a mobile phone / sending messages, and large-scale physical movements (bending over, turning to retrieve objects, etc.). The values ​​are: 0.1, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, and 0.9.

[0130] As a preferred option, in S331, the environmental complexity index is calculated based on the environmental and vehicle status information collected by the system. The calculation formula is as follows:

[0131] ;

[0132] in, , , These are normalized traffic density, road curvature, and weather factors, respectively. For the collision risk function based on TTC, Weights for each environmental factor;

[0133] As a preferred embodiment, the collision risk function formula for the TTC is as follows:

[0134] ;

[0135] in, As a safety threshold, Emergency threshold;

[0136] Preferably, based on the environmental complexity index The driving environment is divided into three levels: low, medium, and high. Three sets of weight coefficients are preset for the three levels of environmental complexity. As the environmental complexity increases, the behavioral risk weight gradually increases, reflecting the higher requirements for the driver's operating ability. The total weight is always 1, ensuring that the risk field value is within the range of [0,1].

[0137] Preferably, the classification of driving environment complexity levels is as follows: when When, it has low complexity; when When, it is of medium complexity; when At this time, it is considered to have high complexity;

[0138] As a preferred option, in S340, the corresponding weighted reorganization is selected based on the current environmental complexity level to calculate the driver's state risk field. The calculation formula is as follows:

[0139] ;

[0140] in, For the driver's state risk field; , , Preset weights are selected based on the level of environmental complexity.

[0141] As a preferred option, , , For low complexity, set them to 0.3, 0.2, and 0.5 respectively; for medium complexity, set them to 0.25, 0.15, and 0.6 respectively; and for high complexity, set them to 0.2, 0.1, and 0.7 respectively.

[0142] As a preferred option, in the S350, a graded early warning and response strategy is set based on the quantified driver state risk field output (driver state risk value):

[0143] when At that time, it was a low-risk level, and only the status was recorded without any warnings;

[0144] when At that time, the risk level was medium, and a mild visual and auditory dual-mode warning was adopted;

[0145] when When the risk level is high, a three-mode warning system (visual, auditory, and tactile) is used, and the system intervenes to reduce the vehicle speed. In extreme cases, the system executes a minimum risk strategy, which means that, under the premise of ensuring safety, the system slows down, turns on the hazard lights, and smoothly pulls the vehicle to the side of the road and stops it.

[0146] To ensure safety redundancy, an emergency TTC (Total Traffic Control) mechanism is set up. When the TTC falls below the emergency threshold, the highest level of warning is immediately triggered, and the system automatically executes emergency braking or avoidance strategies.

[0147] In one embodiment, a humanized takeover request mechanism is also included, which determines the advance time of the takeover request based on the driver's state risk value when the autonomous driving system requests takeover.

[0148] When the driver's condition is at a low-risk level, use gentle prompts and extend the preparation time;

[0149] When the driver's condition is at a medium risk level, a tiered prompting system is used, with an appropriate preparation time.

[0150] When a driver is in a high-risk state, an emergency alert is used, and the preparation time is shortened.

[0151] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A millimeter wave radar based driver state monitoring and safety warning method, characterized in that, include: The system collects driver physiological micro-motion signals and driver macro-behavioral signals using millimeter-wave radar, and also acquires real-time driving environment information. After denoising the driver's physiological micro-motion signal, feature extraction is performed to obtain the driver's physiological feature signal; And constructing a time-frequency spectrum using the driver's macroscopic behavioral signals; The driver's physiological arousal index and driver's emotional state risk value are determined based on the aforementioned physiological characteristic signals; And obtain the driver's dynamic behavior risk value based on the aforementioned time-frequency spectrum; Calculate the environmental complexity index based on the driving environment information; A driver state risk field model is constructed based on the driver's physiological arousal index, driver's emotional state risk value, and driver's dynamic behavior risk value. The driver's state risk value is calculated using the driver state risk field model and the environmental complexity index. Early warnings are issued in tiers based on the driver's risk level.

2. The millimeter wave radar based driver state monitoring and safety warning method of claim 1, wherein, After denoising the driver's physiological micro-motion signal, feature extraction is performed to obtain the normalized deviation of heart rate and the normalized deviation of respiratory rate. The driver's physiological arousal index is then calculated based on the normalized deviation of heart rate and the normalized deviation of respiratory rate. The formula for calculating the heart rate normalization deviation is as follows: ； The formula for calculating the normalized deviation of the respiratory rate is: ； wherein is the average heart rate of the driver, is the average respiration rate of the driver, is the baseline heart rate, is the baseline respiration rate.

3. The millimeter-wave radar-based driver state monitoring and safety warning method according to claim 2, characterized in that, The driver's physiological arousal index is: ； wherein, and are a heart rate weight coefficient and a respiration rate weight coefficient; and are a standard deviation of the heart rate normalized deviation and a standard deviation of the respiration rate normalized deviation, respectively; is an offset parameter.

4. The millimeter-wave radar-based driver state monitoring and safety warning method according to any one of claims 1-3, characterized in that, The driver state risk field model is as follows: ； wherein, is a driver state risk value; is a driver emotional state risk value, is a driver dynamic behavior risk value; , and are respectively a driver physiological arousal weight coefficient, a driver emotional state risk weight coefficient and a driver behavior risk weight coefficient, determined according to the driving environment complexity.

5. The millimeter-wave radar-based driver state monitoring and safety warning method according to claim 4, characterized in that, The formula for calculating the complexity of the driving environment is: ； wherein, , , are a normalized value of traffic density, a normalized value of road curvature and a normalized value of weather factor, respectively, is a TTC-based collision risk function, , , and are weight coefficients for the corresponding terms.

6. The millimeter-wave radar-based driver state monitoring and safety warning method according to claim 5, characterized in that, The formula for calculating the risk value of the driver's emotional state is as follows: ； in, This represents the baseline emotional risk coefficient at the current moment. This represents the confidence level corresponding to the current emotion category.

7. The driver status monitoring and safety early warning method based on millimeter-wave radar according to claim 6, characterized in that, The formula for calculating the driver's dynamic behavior risk value is as follows: ； Where τ is the time decay constant, This represents the basic risk coefficient of the behavior at the current moment. Indicates the duration of the action at the current moment.

8. The driver status monitoring and safety early warning method based on millimeter-wave radar according to claim 7, characterized in that, The collision risk function based on TTC is: ； in, As a safety threshold, This is the emergency threshold.

9. The driver status monitoring and safety early warning method based on millimeter-wave radar according to claim 8, characterized in that, Also includes: Set up an emergency TTC handling mechanism: When At that time, the highest level of warning will be activated.

Images