A driver fatigue early warning system and method based on an intelligent seat of a vehicle

Abstract

This invention relates to the field of automotive intelligent seats and driver state monitoring technology, specifically a driver fatigue warning system and method based on an automotive intelligent seat. The system includes a first sensing node, a second sensing node, and an alarm control unit. The first sensing node calculates the variance of a first modality of physiological data in real time. When the variance exceeds a preset motion threshold, it broadcasts a channel silence token. The second sensing node discards the currently collected second modality of physiological data and stops wireless transmission. When the variance is less than a preset steady-state threshold, it broadcasts a steady-state synchronization token. The second sensing node restarts data collection using the timestamp of the steady-state synchronization token arrival as a time reference, extracts physiological features, and transmits them to the alarm control unit to generate a driver fatigue warning signal. When the variance is between the two thresholds, the system continues to operate according to the current configuration. This invention can suppress invalid transmissions and improve the reliability of fatigue warnings.

Description

Technical Field

[0001] This invention relates to the field of automotive intelligent seats and driver condition monitoring technology, specifically to a driver fatigue early warning system and method based on automotive intelligent seats. Background Technology

[0002] With the development of automotive intelligence and active safety technologies, alarm systems targeting abnormal driver conditions play a crucial role in preventing traffic accidents. Current intelligent alarm systems typically employ distributed sensor networks, continuously collecting physiological data such as the driver's heart rate by embedding sensors in the car's intelligent seats. This data is then transmitted to the alarm control unit for analysis, triggering an abnormal condition alarm signal when fatigue is detected.

[0003] Existing abnormal state alarm systems still have technical shortcomings in practical applications. The reliability of alarm decisions is limited by data transmission quality, and the system is prone to generating false alarms or missing alarms. Traditional alarm systems typically collect physiological waveforms continuously and send them to the alarm control unit. In real driving environments, drivers frequently adjust their posture, shift their position, or straighten their backs, and these postural disturbances introduce significant motion artifacts into the electrophysiological signals. The alarm control unit performs feature extraction and fatigue determination based on the interfered data, which easily leads to false alarms or missed alarms, reducing the overall reliability of the alarm system.

[0004] The alarm system's internal data flow and signaling transmission lack physical state-based control logic. During periods of driver physical disturbance, the front-end nodes continue to transmit invalid raw high-frequency waveforms to the alarm control unit. This not only fails to provide accurate fatigue assessment but also severely consumes the wireless channel bandwidth of the in-vehicle alarm control network, significantly increasing the power consumption of the sensor nodes. The continuous transmission of invalid data leads to communication delays in subsequent, actual alarm signaling.

[0005] Existing alarm systems fail to effectively establish steady-state synchronization verification logic for cross-modal data. When a driver returns from a motion state to a stable sitting posture, the alarm front end lacks a unified time reference to align with and verify the validity of physiological characteristics. This makes it difficult for the system to ensure that the subsequently extracted physiological characteristics actually come from the stable sitting period, thus affecting the accuracy of the warning assessment.

[0006] The current urgent problem to be solved is how to use changes in mechanical posture to impose a priori constraints on the validity of electrophysiological signals from the perspective of abnormal state alarm system architecture design, block invalid data transmission at the transmission end, and establish a highly reliable alarm triggering mechanism with a unified time reference when the system stabilizes again. Summary of the Invention

[0007] The purpose of this invention is to provide a driver fatigue warning system and method based on a smart car seat, and to solve the following technical problems:

[0008] By leveraging the prior constraints of mechanical state changes on the validity of electrophysiological signals, invalid transmissions are suppressed at the data source, and a unified time reference can be established upon re-stabilization. This enables a fatigue early warning processing flow that is low-power, low-collision, and more reliable.

[0009] The objective of this invention can be achieved through the following technical solutions:

[0010] A driver fatigue warning system based on a smart car seat includes: a first sensing node adapted to collect first modal physiological data of a target object; and a second sensing node adapted to collect second modal physiological data of the target object.

[0011] An alarm control unit is configured to receive physiological features extracted from the second modality of physiological data by the second sensing node, and generate an abnormal state alarm signal based on the physiological features; a wireless communication connection is established between the first sensing node, the second sensing node, and the alarm control unit; the first sensing node calculates the variance of the first modality of physiological data in real time;

[0012] When the variance is greater than a preset motion threshold, the first sensor node broadcasts a channel silence token, and the second sensor node responds to the channel silence token by discarding the currently collected second modality physiological data and stopping wireless transmission.

[0013] When the variance is less than the preset steady-state threshold, the first sensing node broadcasts a steady-state synchronization token, and the second sensing node responds to the steady-state synchronization token by restarting the collection of the second modal physiological data based on the timestamp of the arrival of the steady-state synchronization token, extracting physiological features from the second modal physiological data and transmitting them to the alarm control unit.

[0014] When the variance is between the preset steady-state threshold and the preset motion threshold and includes boundary values, the second sensing node continues to operate according to the current acquisition and transmission configuration.

[0015] Optionally, the first sensing node includes: a pressure sensor array adapted to acquire mechanical stress signals as the first modal physiological data;

[0016] A first wireless microcontroller is adapted to calculate the variance and broadcast the channel silence token and the steady-state synchronization token.

[0017] Optionally, the second sensing node includes: an electrophysiological sensor, which is adapted to acquire electromagnetic electrocardiogram signals as the second modal physiological data;

[0018] A second wireless microcontroller is adapted to receive the channel silence token and the steady-state synchronization token, and to control the transmission status of the second modality physiological data.

[0019] Optionally, in response to the channel silence token, the second sensing node discards the currently acquired second modality physiological data and stops wireless transmission, including:

[0020] The second wireless microcontroller is internally configured with an analog-to-digital converter and a wireless transmission buffer. The second wireless microcontroller blocks the data transfer operation from the analog-to-digital converter to the wireless transmission buffer.

[0021] The second wireless microcontroller controls the second sensor node to enter a low-power sleep mode in order to release wireless channel resources.

[0022] Optionally, extracting physiological features from the second modality physiological data and transmitting them to the alarm control unit includes: the second sensing node extracting heart rate variability features from the restarted second modality physiological data based on the time reference, and generating feature data as the physiological features;

[0023] The second sensing node compresses the feature data and sends the compressed feature data to the alarm control unit.

[0024] Optionally, the first sensing node calculates the variance of the first modal physiological data in real time, including: the first sensing node extracts the numerical distribution characteristics of the first modal physiological data within the sliding window based on a preset sliding window in real time.

[0025] The first sensing node uses a variance statistics algorithm to calculate the sliding window variance of the distribution characteristics, which is then used as the variance.

[0026] Optional, for use in smart car seats;

[0027] The first sensing node is deployed in the seat cushion area of ​​the smart car seat; the second sensing node is deployed in the backrest area of ​​the smart car seat; the abnormal state alarm signal is a driver fatigue warning signal.

[0028] A driver fatigue warning method based on a smart car seat includes the following steps: a first sensing node acquires first modal physiological data of a target object, a second sensing node acquires second modal physiological data of the target object; the first sensing node calculates the variance of the first modal physiological data in real time;

[0029] When the variance is greater than a preset motion threshold, the first sensor node broadcasts a channel silence token, and the second sensor node responds to the channel silence token by discarding the currently collected second modality physiological data and stopping wireless transmission.

[0030] When the variance is less than the preset steady-state threshold, the first sensing node broadcasts a steady-state synchronization token, and the second sensing node responds to the steady-state synchronization token by restarting the collection of the second modality physiological data based on the timestamp of the arrival of the steady-state synchronization token.

[0031] When the variance is between the preset steady-state threshold and the preset motion threshold and includes the boundary value, the second sensing node continues to operate according to the current acquisition and transmission configuration; the second sensing node extracts physiological features from the second modal physiological data and transmits them to the alarm control unit, so that the alarm control unit can generate an abnormal state alarm signal based on the physiological features.

[0032] Optionally, the second sensing node, in response to the channel silence token, discards the currently collected second modality physiological data and stops wireless transmission, including: blocking the data transfer operation from the analog-to-digital converter configured inside the second sensing node to its wireless transmission buffer.

[0033] The currently collected second modality physiological data is discarded at the source; the second sensing node is controlled to enter a low-power sleep mode.

[0034] Optionally, the second sensing node extracts physiological features and transmits them to the alarm control unit, including: the second sensing node aligns the time axis of the second modality physiological data with the steady-state synchronization token based on the time reference;

[0035] The second sensing node extracts heart rate variability features from the second modality physiological data as the physiological features; the second sensing node compresses the physiological features and sends the compressed data to the alarm control unit for status assessment.

[0036] The beneficial effects of this invention are:

[0037] 1. This invention evaluates body perturbation by calculating the variance of modal data in real time through the first sensing node. When the variance exceeds the motion threshold, a silence token is broadcast to control the second sensing node to discard data contaminated by motion artifacts and stop transmission, effectively preventing interference from low-reliability data and significantly improving the accuracy of fatigue determination.

[0038] 2. When the variance is less than the steady-state threshold, the present invention broadcasts a steady-state synchronization token, which enables the second sensing node to restart the acquisition based on the timestamp of the token. By using the mechanical steady state as a priori constraint on the validity of the electrophysiological signal, the invention ensures that the re-acquired data is in the sitting posture stable period, thus achieving highly reliable feature extraction.

[0039] 3. In a silent state, this invention directly blocks the data transfer from the analog-to-digital converter to the wireless transmission buffer, discards the currently disturbed data at the source, and controls the node to enter a low-power sleep mode, effectively avoiding the accumulation of invalid buffers, significantly reducing node power consumption, and releasing the limited wireless bandwidth in the vehicle.

[0040] 4. Based on time-referenced multimodal data, this invention extracts heart rate variability features locally, compresses them, and then transmits them to the alarm control unit. This transforms the high-frequency raw waveform into low-bandwidth feature data for fatigue assessment, reducing the amount of wireless communication data in the system and improving the real-time performance of fatigue analysis.

[0041] 5. This invention uses a preset sliding window and variance statistics algorithm to extract distribution features in real time, and maps changes in the original data into stable gating decision indicators in an online manner with controllable computational load. This avoids frequent state switching errors and takes into account the real-time response of the system and the feasibility of vehicle-mounted embedded devices. Attached Figure Description

[0042] The invention will now be further described with reference to the accompanying drawings.

[0043] Figure 1 A schematic diagram of a driver fatigue warning system based on an intelligent car seat is provided for an embodiment of this application;

[0044] Figure 2 This is a flowchart illustrating a driver fatigue warning method based on a smart car seat provided in an embodiment of this application. Detailed Implementation

[0045] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 A driver fatigue warning system based on a smart car seat includes: a first sensing node adapted to collect first modal physiological data of a target object; and a second sensing node adapted to collect second modal physiological data of the target object.

[0047] An alarm control unit is adapted to receive physiological features extracted from the second modality physiological data by the second sensing node, and generate an abnormal state alarm signal indicating driver fatigue based on the physiological features.

[0048] A wireless communication connection is established between the first sensing node, the second sensing node, and the alarm control unit; the first sensing node calculates the variance of the first modality physiological data in real time;

[0049] When the variance is greater than the preset motion threshold, the first sensor node broadcasts a channel silence token, and the second sensor node responds to the channel silence token by discarding the currently collected second modality physiological data and stopping wireless transmission.

[0050] When the variance is less than the preset steady-state threshold, the first sensor node broadcasts a steady-state synchronization token, and the second sensor node responds to the steady-state synchronization token by restarting the acquisition of second modal physiological data based on the timestamp of the arrival of the steady-state synchronization token, extracting physiological features from the second modal physiological data and transmitting them to the alarm control unit.

[0051] When the variance is between the preset steady-state threshold and the preset motion threshold and includes boundary values, the second sensing node continues to operate according to the current acquisition and transmission configuration.

[0052] This embodiment provides an implementation mechanism for a driver fatigue warning system based on a smart car seat. Specifically, the system is applied to a high-speed trunk transport vehicle at night, and the monitoring object is the driver performing long-distance transport tasks. A first sensing node is arranged in the seat cushion area to sense changes in the load distribution of the driver's lower limbs and buttocks. A second sensing node is arranged in the seat back area to collect second modal physiological data related to cardiac activity.

[0053] An alarm control unit is installed in the center console or seat controller to receive the extracted physiological characteristics and output fatigue warning signals; the three are connected by a short-range wireless connection, which can be a low-power Bluetooth broadcast link, a ZigBee link, or an equivalent low-power in-vehicle wireless link.

[0054] To avoid ambiguity caused by the same object being referred to by multiple names throughout the text, in this embodiment and subsequent embodiments, unless otherwise specified, the silent token refers to the channel silent token, the synchronization token refers to the steady-state synchronization token, and the arrival timestamp, token arrival timestamp, and arrival time all refer to the local time stamp formed when the second sensing node actually receives the steady-state synchronization token and completes local time latching, and not just the timestamp field value carried in the message.

[0055] The timestamp field carried in the message is uniformly used for token identification, sequence verification, window number association, or drift correction; therefore, when abbreviations are used in the following text without changing the technical meaning, they should all be understood according to the above-mentioned unique meaning.

[0056] Specifically, the first sensing node does not directly provide a fatigue conclusion, but rather serves as the arbiter of the physical state of the entire system; it continuously receives physiological data of the first modality and calculates the variance within a continuous time window to determine whether the driver is currently in a stage of significant physical disturbance; here, a sliding window with a length of 1 second can be used, and the variance result is updated every 100 milliseconds;

[0057] To facilitate understanding, the following specific implementation scenario can be used as an example: Assume that the total pressure sequence within a certain window is 100, 101, 99, 100, 100, and its dimensions are the dimensionless quantized values ​​of the analog-to-digital converter. Then, according to the variance formula... The calculated variance of the sequence is 0.5, which is less than the preset steady-state threshold of 4. If the next window becomes 100, 130, 85, 120, 90, the variance calculated according to the above formula is 375, which is greater than the preset motion threshold of 25. This represents the total number of data items in the sequence within this window. For the first in the sequence One data point, This is the average value of the sequence;

[0058] The system presets two thresholds, for example, a steady-state threshold of 4 and a motion threshold of 25. When the variance is greater than 25, it indicates that the driver is moving, straightening their back, turning around, or adjusting their sitting posture. At this time, the electrophysiological data collected from the backrest side usually contains motion artifacts with amplitudes exceeding the preset interference threshold. Continuing wireless transmission will not only occupy the channel but also send low-reliability data to the alarm control unit. Therefore, the first sensor node immediately broadcasts a channel silence token. After receiving the token, the second sensor node discards the data that has not yet been uploaded in the current sampling period and stops wireless transmission.

[0059] When the variance drops below 4, it indicates that the driver has re-entered a relatively stable sitting posture; at this time, the first sensor node broadcasts a steady-state synchronization token, which carries at least one timestamp field that can be recognized by the second sensor node.

[0060] After receiving the data, the second sensor node no longer uses the old time reference before the interruption during the movement. Instead, it uses the arrival time when the token is received by the second sensor node and completes local time latching as the new time zero point to restart the second modality physiological data acquisition. The aforementioned timestamp field is mainly used for token identification, sequential verification and window number association. The direct time reference used when the node restarts acquisition is still the actual arrival timestamp of the steady-state synchronization token.

[0061] Physiological features, such as R-wave interval sequences, heart rate variability statistics, or respiratory rhythm parameters, are extracted from the recovered data and uploaded to the alarm control unit. The latter then compares the time-domain or frequency-domain indicators of the physiological features with a preset health baseline threshold according to preset fatigue judgment rules and outputs an abnormal state alarm signal.

[0062] In the transition range between the two thresholds, the system does not forcibly change the working mode of the second sensing node. The reason for this setting is that slight load changes may be normal driving fine-tuning or a transition precursor to significant body disturbances. If silence or wake-up commands are frequently issued in this range, it will cause the node to switch repeatedly, which will increase the system control burden. Therefore, maintaining the current acquisition and transmission configuration at this time is beneficial to suppressing jitter.

[0063] Furthermore, to avoid threshold jumps, a minimum hold time can be set in this embodiment; for example, a silent token is issued only when the variance is greater than the motion threshold for three consecutive update cycles; a synchronization token is issued only when the variance is less than the steady-state threshold for three consecutive update cycles; if the second sensing node fails to receive a synchronization token, it remains in sleep mode or in the previous state until the next valid synchronization token arrives; if the wireless link is interrupted for a short time, the alarm control unit can record the feature absence status instead of directly outputting a fatigue alarm to avoid false alarms caused by communication abnormalities.

[0064] For example, in a scenario where a heavy truck is traveling on a highway at 2 a.m., the driver becomes fatigued due to continuous driving and first makes a noticeable back straightening and shifting motion; the pressure distribution in the seat area changes drastically within 1 second, and the first sensor node calculates that the variance rises to 36, exceeding the motion threshold, so it broadcasts a silent token.

[0065] The second sensor node in the backrest area stops uploading ECG data for this stage to avoid sending waveforms contaminated by artifacts to the central control unit. After about 2 seconds, the driver's posture stabilizes again, the pressure variance drops to 2, and the first sensor node broadcasts a steady-state synchronization token. The second sensor node resamples with the arrival time of the token as the zero point, extracts the heart rate variability features for the next 30 seconds, and uploads them. The alarm control unit detects the decrease in heart rate variability and the homogenization of the rhythm, triggers a fatigue warning signal, and activates the in-vehicle buzzer and seat vibration to provide a reminder.

[0066] The purpose of this mechanism is to use the prior constraints of mechanical state changes on the effectiveness of electrophysiological signals to suppress invalid transmissions at the data source and establish a unified time reference when the system re-stabilizes, thereby achieving a fatigue early warning process with low power consumption, low collisions and higher reliability.

[0067] In a preferred embodiment of the present invention, the first sensing node includes: a pressure sensor array adapted to collect mechanical stress signals as first modal physiological data; and a first wireless microcontroller adapted to calculate variance and broadcast channel silence tokens and steady-state synchronization tokens.

[0068] This embodiment provides a specific structure of a first sensing node; specifically, a pressure sensor array and a first wireless microcontroller are provided in the seat cushion area, which can be integrated on a flexible circuit board and embedded between the seat foam layer and the cover to withstand vibration and repetitive loads over a long period of time.

[0069] Specifically, the pressure sensor array can adopt a 4×4, 6×6, or other distributed measuring point structure suitable for the seat area; each measuring point outputs a local mechanical stress signal, the first wireless microcontroller periodically reads the values ​​of each measuring point and forms a pressure distribution vector, and by calculating the spatial weighted average of the values ​​of each measuring point in the pressure distribution vector, the center of gravity coordinates at the current moment can be obtained, and the center of gravity offset sequence is formed by the center of gravity coordinates of multiple consecutive moments; taking the specific implementation architecture of 4 measuring points as an example, the distribution vector at a certain moment can be written as [20, 25, 18, 22], representing the relative pressure of the left front, right front, left rear, and right rear regions;

[0070] If the next moment becomes [10, 35, 12, 30], it indicates that the driver's center of gravity has shifted significantly. The first wireless microcontroller can first sum the total pressure sequence at each measuring point, or it can calculate the variance of the derived quantities such as the center of gravity position, left-right difference, and front-back difference. As long as it can reflect the degree of body posture disturbance, it can be used as the basis for subsequent arbitration.

[0071] Relying solely on total pressure can sometimes mask local shifts in the center of gravity; for example, if the total remains constant but the left side decreases while the right side increases, it still constitutes a significant posture change. Therefore, in this embodiment, it is preferable for the first wireless microcontroller to simultaneously calculate at least one spatial distribution feature. For example, consider the following specific implementation scenario: if the left-right distribution difference is 2 at time T1, 1 at time T2, and 0 at time T3, then the variance of the left-right distribution difference calculated based on the above sequence is less than a preset motion threshold. If T1 is 2, T2 is 15, and T3 is -12, then the variance of the calculated left-right distribution difference is greater than the preset motion threshold, and even if the total pressure remains unchanged, it is still determined to meet the motion stage conditions.

[0072] In terms of broadcast control, the first wireless microcontroller can encapsulate the silence token and the synchronization token into short messages; the silence token can at least include the node identifier, token type, and generation time; the synchronization token can also carry a synchronization timestamp; the token prefers the broadcast mode rather than establishing a long connection because the transmission latency of the broadcast mode is lower than the establishment latency of establishing a long connection, which is suitable for quickly triggering the back-side node to enter the silence or restart state.

[0073] Furthermore, if some pressure measurement points fail, for example, if a certain measurement point outputs a fixed value for a long time, the first wireless microcontroller can shield that measurement point from the current calculation and continue to calculate the variance based on the remaining measurement points. It can continue to work as long as the number of effective measurement points reaches the minimum threshold. If the number of effective measurement points is lower than the threshold, it enters the degradation mode, only sends the untrusted status packet of the seat cushion, and no longer triggers the silence and synchronization control to prevent erroneous arbitration.

[0074] For example, in the aforementioned nighttime high-speed transportation scenario, the driver briefly adjusts his sitting posture before approaching the service area. Among the 16 measuring points on the seat cushion, the right rear region shows a significant increase for three consecutive frames, while the left front region shows a synchronous decrease. The first wireless microcontroller calculates the variance exceeding the limit based on the center of gravity offset sequence and quickly issues a silence token. After the driver sits back down, the measuring points return to a gradual change, and the microcontroller issues a synchronization token to drive the backrest node to resume effective data acquisition.

[0075] The purpose of this step is to obtain mechanical state information that is more sensitive to changes in driving posture through the pressure array, and to have the first wireless microcontroller directly perform local calculations and token broadcasts, thereby reducing the central processing burden and improving control response speed.

[0076] In a preferred embodiment of the present invention, the second sensing node includes: an electrophysiological sensor adapted to collect electrocardiogram signals in an electromagnetic state as second modal physiological data; and a second wireless microcontroller adapted to receive a channel silence token and a steady-state synchronization token, and to control the transmission state of the second modal physiological data.

[0077] This embodiment provides a specific structure for a second sensing node; specifically, an electrophysiological sensor and a second wireless microcontroller are arranged in the backrest area. The electrophysiological sensor can be a capacitively coupled ECG electrode, a dry electrode, or a similar device that can collect cardiac electrical activity through clothing; the second wireless microcontroller is responsible for sampling control, token response, feature extraction, and wireless transmission management.

[0078] Specifically, the data output rate of the electrophysiological sensors in the backrest area is greater than that of the pressure sensor array in the seat cushion, and the amplitude of motion artifact interference is also greater than the preset level. If the original waveform is continuously transmitted, it will not only occupy the wireless bandwidth, but also easily generate a lot of meaningless data when the driver turns or the road is vibrating. Therefore, the second wireless microcontroller does not simply forward the sampling results, but listens for the control token from the seat cushion area, and then decides to keep, stop or restart the acquisition and transmission process based on the token.

[0079] To avoid ambiguity in notation, in this embodiment, individual ECG digital samples arranged in the order of sampling are denoted as E1, E2, E3, ..., where Ek represents the k-th ECG sample after analog-to-digital conversion; the synchronization arrival time when the second wireless microcontroller actually receives the steady-state synchronization token and completes local time latching is denoted as t0; the above notation is only used to illustrate the timing relationship in the sampling and restart process and does not represent any additional independent physical quantity type;

[0080] To facilitate understanding, the following specific implementation scenario is used as an example; assuming that the second wireless microcontroller samples ECG data every 4 milliseconds to form a sequence E1, E2, E3... When no control token is received, the node continues to work according to the current configuration;

[0081] If a silence token is received at time E125, then even if samples E126 and thereafter have entered the local buffer, they will not enter the wireless transmission process. If a synchronization token is received later, and its arrival time is recorded as t0, then the node will reorganize the sampling batch from t0. For example, a new ECG segment can be formed between t0 and t0+8 seconds, and then subsequent features can be extracted from this segment. The technical effect of this setting is to ensure that the re-acquired data segment is during the period when the target object is in a stable sitting posture, rather than completely retaining the entire sampling waveform.

[0082] Simply stopping transmission without stopping data acquisition may still lead to local cache buildup and increased power consumption. Therefore, in this embodiment, the second wireless microcontroller can synchronously reduce the sampling frequency during the silent phase, or temporarily shut down the high-power operating stage of the front-end amplifier to reduce unnecessary power consumption. After receiving the synchronization token again, it can then restore the target sampling rate required for fatigue monitoring.

[0083] Furthermore, if the second wireless microcontroller receives multiple silence tokens consecutively, it will use the most recent silence token as the standard and maintain the silence state without switching repeatedly; if it does not receive a synchronization token within the predetermined maximum silence duration, it can perform periodic self-tests and report the status byte waiting for steady-state recovery, but will still not upload the original high-frequency data; if the token verification fails, it will be regarded as an invalid packet and ignored to prevent other wireless signals in the vehicle from being triggered by mistake.

[0084] For example, during nighttime trunk line transportation, when the vehicle passes through a section of road with dense road joints, the interference to the ECG sensor in the backrest area increases. At this time, without mechanical gating, the node will continuously upload waveforms with distorted amplitude. In this embodiment, the second wireless microcontroller prioritizes responding to the silence token from the seat cushion area and suspends the high-frequency transmission of that section. After the vehicle leaves the section with significant vibration and the driver's posture stabilizes again, effective data acquisition is resumed based on the synchronization token.

[0085] The purpose of this step is to enable the back-end node to respond quickly to changes in external physical conditions, and to change the electrophysiological data acquisition and wireless transmission from a continuous stream mode to a controlled event mode, thereby achieving a higher effective feature output ratio.

[0086] In a preferred embodiment of the present invention, the second sensing node, in response to a channel silence token, discards the currently collected second modality physiological data and stops wireless transmission, including: the second wireless microcontroller is internally configured with an analog-to-digital converter and a wireless transmission buffer; the second wireless microcontroller blocks the data transfer operation from the analog-to-digital converter to the wireless transmission buffer; the second wireless microcontroller controls the second sensing node to enter a low-power sleep mode to release wireless channel resources;

[0087] This embodiment provides a source-end discarding mechanism during the silent phase. Specifically, in the previous layer scheme, although the overall logic for discarding and stopping transmission has been given, if the data segment is only marked as invalid at the application layer, and the analog-to-digital converter continues to send data to the wireless buffer, the buffer will still occupy storage and bus resources, and the node power consumption will decrease only slightly. Therefore, this embodiment further pushes the silent control down to the data transport path.

[0088] Specifically, the second wireless microcontroller can internally include a sampling front-end, an analog-to-digital converter, a direct memory access channel or equivalent transfer channel, a wireless transmission buffer, and a transmitter. During normal operation, the electrophysiological analog signal is conditioned by the front-end and then enters the analog-to-digital converter to generate digital samples, which then enter the wireless transmission buffer and are finally transmitted in the form of data frames. Upon receiving a silence token, the second wireless microcontroller performs two actions: first, it blocks data transfer from the analog-to-digital converter to the wireless transmission buffer; second, the control node enters a low-power sleep mode or a partial module sleep mode. In this way, even if a small amount of sampling remains in the front-end for a short period, no new data will continue to be injected into the transmission path.

[0089] To maintain consistency with the description of discarding the currently collected second modality physiological data in the aforementioned embodiments, the term "current collection" in this embodiment uniformly refers to the current sampling batch, the current cache batch, or the set of pending data to be generated from the batch when the silence token takes effect;

[0090] Data that has been transmitted over the air and received normally by the other end before the silence token arrives is not included in this discarding; however, data that has entered the wireless transmission buffer but is still part of the current disturbed sampling batch and has not yet been effectively transmitted can be discarded as part of the currently collected data. With this limitation, the scope of data discarded by the source end is consistent with the silence semantics in the embodiment, and does not mistakenly include previously successfully transmitted steady-state valid data in the deletion scope.

[0091] As a specific data processing implementation scenario; assuming that the wireless transmission buffer can hold a maximum of 8 frames, each frame contains 20 sampling points; under normal conditions, the analog-to-digital converter fills 1 frame every 80 milliseconds and triggers transmission; if a silence token is received when the 3rd frame is half-filled, this embodiment does not wait for the frame to be filled, but immediately stops subsequent processing and regards the incomplete data of the 3rd frame as a source-end discard object;

[0092] If the first and second frames have not yet been transmitted, it is preferable to clear them directly and prohibit them from entering the over-the-air transmission process while their content still belongs to the current disturbed sampling batch, so as to ensure that the currently collected second modality physiological data is discarded at the source and the wireless transmission stops immediately; if a frame has been effectively transmitted before the silence token arrives, no further processing is required; in order to maintain the integrity of the underlying communication protocol, if the transmitter is already in the end-of-transmission phase of a physical layer bit at the moment the silence token arrives, only the shortest protocol termination sequence required for the end of the frame can be retained, and no new payload data will be transmitted; this situation is not considered as continuing to transmit the currently collected second modality physiological data;

[0093] For low-power sleep mode, a layered sleep mode can be adopted. If the expected silence duration is less than the preset duration threshold, only the wireless transmission module and the high-frequency processing clock can be turned off, while the token receiving capability is retained. If the silence duration is greater than or equal to the preset duration threshold, the front-end high-power amplifier stage is further turned off, while only the low-power wake-up receiving unit is retained. This ensures that the synchronization token can be received in the future, while avoiding the additional delay introduced by recalibration after a complete power outage.

[0094] Furthermore, if a frame has already started transmitting in the air when the silence token arrives, the second wireless microcontroller will preferentially prevent that frame from carrying a new physiological payload into the subsequent transmission sequence, and will immediately stop subsequent handling and transmission after the minimum protocol is completed, in order to avoid protocol abnormalities.

[0095] If the buffer is empty, it will directly enter sleep mode; if the analog-to-digital converter is calibrating or the front-end is self-testing, it will first complete the minimum protection actions and then switch to silent mode to prevent abnormal bias caused by sudden disconnection of the analog front-end; if the node fails to receive the synchronization token within the set time in low-power sleep mode, it can periodically wake up briefly to listen, ensuring that it will not be permanently disconnected.

[0096] For example, in the aforementioned truck continuous driving scenario, if the driver suddenly leans forward and adjusts the backrest angle due to lower back discomfort, the seat cushion area immediately determines that the movement phase has begun; when the backrest area receives the silence token, the analog-to-digital converter has just completed half of the sampling buffer; at this time, the second wireless microcontroller directly cuts off the transport path, clears the transmission buffer, shuts down the transmitter, and retains only the low-power receiving channel; in the following few seconds, the in-vehicle wireless channel is no longer occupied by invalid ECG data, reserving resources for subsequent steady-state reconstruction;

[0097] The purpose of this mechanism is to implement the cessation of transmission as hardware-level data blocking and sleep control, thereby reducing the accumulation of invalid buffers, reducing node power consumption, and significantly freeing up the limited wireless channel resources in the vehicle.

[0098] In a preferred embodiment of the present invention, extracting physiological features from the second modality physiological data and transmitting them to the alarm control unit includes: the second sensing node extracting heart rate variability features from the restarted second modality physiological data based on a time reference, generating feature data as physiological features; the second sensing node compressing the feature data and sending the compressed feature data to the alarm control unit.

[0099] This embodiment provides a feature extraction and compression mechanism after steady-state recovery. Specifically, in the previous layer scheme, if the complete waveform is continuously uploaded after synchronization, although the artifacts have been reduced, the wireless link still suffers from high bandwidth pressure. Therefore, this embodiment further extracts heart rate variability features locally at the back-to-back node and only sends the compressed feature data.

[0100] Specifically, after receiving the synchronization token, the second sensing node uses the arrival time of the token as the new time reference to fragment the collected ECG data. Preferably, the R-wave position is extracted within a fixed-length window, and then the RR interval sequence is calculated. A specific implementation example can be used: assuming the R-wave is detected at 1.00 seconds, 1.82 seconds, 2.65 seconds, and 3.48 seconds within a steady-state window, the adjacent RR intervals are 0.82 seconds, 0.83 seconds, and 0.83 seconds, respectively. From this, time-domain indicators such as the average RR interval, standard deviation, and root mean square of adjacent differences can be calculated, as well as frequency-domain indicators such as the energy ratio of low and high frequencies can be estimated. These indicators constitute physiological characteristics.

[0101] Since the alarm control unit ultimately needs features that reflect fatigue trends, rather than all sampling points, this embodiment compresses the feature data. Compression can be achieved through fixed-point quantization, differential coding, threshold pruning, or lightweight packetization. Taking the example above, if the average RR interval is 0.826 seconds, the standard deviation is 0.005 seconds, and the root mean square of adjacent differences is 0.004 seconds, it can be multiplied by a uniform scaling factor to convert it into integers 826, 5, and 4, and then packaged into a short feature frame for transmission. Compared to the original hundreds or even thousands of sampling points, only a small number of bytes are needed to express the information required for the current fatigue state. After receiving the above feature data, the alarm control unit compares it with a preset health baseline threshold. For example, when the average RR interval becomes longer, or the root mean square of adjacent differences is lower than the preset fatigue baseline threshold, it is determined that the driver is in a fatigued state.

[0102] Uploading only single-window features may be too sensitive to short-term anomalies. Therefore, in this embodiment, multi-window smoothing can also be performed on the node side. For example, feature groups F1, F2, and F3 are obtained from three consecutive steady-state windows, and the average result or the trend of change is sent. This reduces the burden on the central end and improves the stability of the judgment.

[0103] Furthermore, if the number of valid R waves within the synchronized window is insufficient, for example, due to the driver's slight movement causing the available interval to be less than the preset threshold, then the window will not generate a formal feature, but will be marked as insufficient features and wait for the next window; if extreme values ​​exceeding the coding range occur during compression, saturation truncation can be used with an abnormal flag bit so that the alarm control unit can distinguish between physiological abnormalities and coding overflow; if valid features cannot be generated continuously for a period of time, the alarm control unit can make a conservative assessment by combining the results of the previous period and the vehicle running time, instead of directly issuing a strong alarm based on this.

[0104] For example, in the fourth hour of a long-distance night transport, after the driver adjusts his sitting posture and stabilizes again, the second sensor node starts to count the subsequent 60-second ECG segment with the synchronization token time as zero, extracts the average heart rate, the standard deviation of the RR interval, and the high-low frequency ratio, and sends these indicators to the central control unit after quantification and compression; the central control unit does not need to analyze the entire waveform, but can determine whether the fatigue level continues to deepen based on the compression characteristics alone;

[0105] The purpose of this step is to transform the high-frequency raw waveform into low-bandwidth, high-diagnostic feature data, thereby reducing the wireless load and improving the real-time performance of fatigue analysis.

[0106] In a preferred embodiment of the present invention, the first sensing node calculates the variance of the first modality physiological data in real time, including: the first sensing node extracts the numerical distribution characteristics of the first modality physiological data within a preset sliding window in real time; the first sensing node calculates the sliding window variance of the distribution characteristics using a variance statistics algorithm, and uses it as the variance.

[0107] This embodiment provides a specific step for real-time variance calculation. Specifically, when the mechanical signal of the seat area changes continuously, if the variance is calculated based on all historical data each time, not only will the amount of calculation be large, but it will also be difficult to reflect the current posture change in a timely manner. Therefore, this embodiment adopts a sliding window and an online variance statistics algorithm.

[0108] Specifically, the first sensing node extracts distribution features from the pressure array; these distribution features are not limited to a single measurement point value, but can be total pressure, center of gravity coordinates, left-right load difference, front-back load difference, local gradient, or a combination thereof; and the variance is calculated within a preset sliding window.

[0109] To facilitate understanding, let's take a specific implementation scenario with a window length of 5 as an example. Assume that the left and right load differences at 5 consecutive moments are 1, 2, 1, 2, 2, then the mean is about 1.6 and the variance is small. When the window slides to the next group of 3, 8, -4, 7, -2, although the mean is close to 2.4, each value deviates from the mean by a large margin, and the variance increases significantly. Based on this, the system judges that the driver's posture is unstable.

[0110] When implemented online, an incremental statistical algorithm suitable for embedded devices can be used; that is, the first wireless microcontroller does not need to store all the original data for a long time, but only maintains the limited state quantities required for the window, and updates the variance when each new sample enters and an old sample is removed; this approach is particularly suitable for automotive-grade low-power controllers.

[0111] If only a single distribution feature is selected, it may be affected by the overall vibration of the seat under certain road conditions; for example, when passing through a rough road surface, the total pressure will fluctuate synchronously, but the driver's posture may not actually change; therefore, in this embodiment, multiple distribution features can be optionally combined to generate a comprehensive variance index; for example, the variances of the lateral change of the center of gravity and the difference between the left and right loads are calculated separately, and then summed according to preset weights as the final index used for threshold comparison; this can improve the ability to distinguish real posture changes.

[0112] Furthermore, if the sliding window length is less than the first preset length, it will reduce the variance's ability to resist instantaneous noise; if the sliding window length is greater than the second preset length, it will increase the system's response delay. Therefore, calibration can be performed in different vehicle models based on seat material, vehicle vibration characteristics, and target driving time. If the number of valid samples in a certain window is insufficient, for example, due to sampling interruption or multiple measurement points failing simultaneously, the variance of that window will not participate in the decision-making process, and the previous valid result will be used or a conservative state will be entered. If the overall variance is exactly at the boundary value, the current configuration will remain unchanged according to the aforementioned rules to avoid frequent switching.

[0113] For example, during the endurance road test of the same heavy truck, the engineers found that the total pressure fluctuated periodically when passing through the bridge joint, but the change in the lateral center of gravity was small. Therefore, the variance of the lateral center of gravity and the variance of the difference between the left and right loads were used as the main criteria. In actual operation, when the driver was only slightly affected by the road surface, the overall variance remained within the steady state range. However, when the driver actively turned to the side to get water or adjusted his leg posture, the overall variance increased rapidly, successfully triggering silent control.

[0114] The purpose of this step is to map the changes in the original pressure distribution into a stable index that can be used for gating decisions using an online statistical method with controllable computational load, thereby balancing real-time performance, reliability, and embedded feasibility.

[0115] In a preferred embodiment of the present invention, it is applied to a smart car seat; a first sensing node is deployed in the seat cushion area of ​​the smart car seat; a second sensing node is deployed in the backrest area of ​​the smart car seat; and the abnormal state alarm signal is a driver fatigue warning signal.

[0116] This embodiment provides a deployment method for intelligent car seats; specifically, the system is centrally embedded in the seat cushion area and backrest area of ​​the intelligent car seat; the technical effect of its physical deployment is that the seat cushion area is suitable for stably acquiring mechanical load distribution information, and the backrest area is suitable for acquiring relevant electrophysiological information of the chest and back without additional operating load.

[0117] Specifically, the first sensing node in the seat cushion area can be attached to the upper surface of the foam layer or the inner neutral layer, which can sense pressure changes without affecting seat comfort; the second sensing node in the backrest area can be arranged in the effective sensing area from the upper back to the lower back, preferably close to the position where electrocardiogram signals are more easily coupled.

[0118] The alarm control unit can be integrated with the seat controller, body domain controller or driver status monitoring module, and output a driver fatigue warning signal when fatigue trend is detected; this signal can drive audible and visual reminders, seat vibration, air conditioning ventilation, instrument display, or be uploaded to the fleet management platform.

[0119] In this deployment method, the seat cushion and backrest play different roles: the former focuses on posture and motion gating, while the latter focuses on steady-state physiological feature extraction; their physical division of labor is clear, which is conducive to improving their respective signal quality; compared with stacking all functions in one place, this partitioned deployment can reduce mutual interference; for example, if the ECG sensor is placed in the seat cushion area, it will be more significantly affected by large hip displacement; if the pressure array is placed in the backrest area, the long-term fit and load stability are often not as good as those in the seat cushion area.

[0120] Furthermore, different vehicle models have different preset characteristics in their seat structures. For example, the thickness of the buffer layer of a commercial vehicle seat is greater than the first thickness threshold, heavy truck seats have air suspension, and ride-hailing vehicle seats have a wear rate greater than the preset standard. Therefore, the embedding depth and fixing method of the sensing nodes can be changed according to the vehicle model.

[0121] If the signal quality in the backrest area deteriorates due to thick winter clothing, the system can raise the steady-state judgment threshold, extend the feature extraction window, or temporarily reduce its dependence on frequency domain features while still maintaining the fatigue warning function. If the seat is unoccupied, the cushion area can first close the entire monitoring process through occupancy judgment to avoid malfunction when there is no driver.

[0122] For example, in a long-haul logistics tractor equipped with an air-damped seat, the sensor node in the seat cushion area monitors the driver's center of gravity shift and posture stability, while the sensor node in the backrest area acquires electrocardiogram data when the driver is stably leaning against the seat.

[0123] The central control system detected a continuous decrease in heart rate variability after 4 hours of continuous driving. Heart rate variability is quantified and extracted using the root mean square of adjacent RR intervals, and its calculation formula is as follows: ,in, This represents the total number of extracted RR intervals. For the first The duration of each RR interval For the first The driver is alerted by the duration of the RR interval and the fact that the road ahead still requires a long drive, so a fatigue warning signal is output, triggering seat vibration and instrument text reminders to prompt the driver to enter a service area for rest as soon as possible.

[0124] The purpose of this deployment method is to leverage the physical characteristics of smart seats, which directly contact the target object and require no additional wearable devices, to achieve integrated implementation of fatigue monitoring, wireless door control, and alarm output.

[0125] Please see Figure 2 A driver fatigue warning method based on a smart car seat includes the following steps: a first sensing node acquires first modal physiological data of a target object, a second sensing node acquires second modal physiological data of the target object; the first sensing node calculates the variance of the first modal physiological data in real time;

[0126] When the variance is greater than a preset motion threshold, the first sensor node broadcasts a channel silence token, and the second sensor node responds to the channel silence token by discarding the currently collected second modality physiological data and stopping wireless transmission.

[0127] When the variance is less than the preset steady-state threshold, the first sensing node broadcasts a steady-state synchronization token, and the second sensing node responds to the steady-state synchronization token by restarting the collection of the second modality physiological data based on the timestamp of the arrival of the steady-state synchronization token.

[0128] When the variance is between the preset steady-state threshold and the preset motion threshold and includes the boundary value, the second sensing node continues to operate according to the current acquisition and transmission configuration; the second sensing node extracts physiological features from the second modal physiological data and transmits them to the alarm control unit, so that the alarm control unit can generate an abnormal state alarm signal based on the physiological features.

[0129] This embodiment provides a flowchart of a driver fatigue warning method based on a smart car seat; specifically, the method can run in the aforementioned system and perform gating collection of second modality physiological data according to the actual posture changes during vehicle driving.

[0130] Specifically, the method can be executed in a continuous loop; the seat area continuously acquires the first modality of physiological data, and the backrest area continuously acquires the second modality of physiological data; the seat area calculates the variance based on the real-time updated window and divides the current state into the motion phase, steady-state phase, or transition phase;

[0131] The data strategy for the backrest area is controlled based on the division results: if it is in the motion phase, a silence token is broadcast, the backrest area discards the data currently being generated and stops wireless transmission; if it is in the steady state phase, a synchronization token is broadcast, and the backrest area restarts acquisition using the timestamp of the token as the new time reference; if it is in the intermediate transition phase, the current working configuration of the backrest area is not changed; the backrest area extracts physiological features from the steady state acquisition segment and uploads them, and the alarm control unit performs abnormal state assessment.

[0132] For ease of understanding, specific data examples are as follows: Assuming that the period from 0 to 10 seconds is the stable cruising phase of the vehicle, the variance of the seat cushion area is 2, which is less than the steady-state threshold of 4. Therefore, a synchronization token is broadcast once near 0 seconds, and the backrest area collects data based on this. During the period from 10 to 13 seconds, the driver rotates his waist and back to adjust his posture, and the variance rises to 28, which is greater than the motion threshold of 25. The seat cushion area broadcasts a silence token, and the backrest area stops sending data.

[0133] Between 13 and 15 seconds, the variance drops back to 10, falling into the middle range, and the current silent state remains unchanged; after 15 seconds, the variance drops to 3, and the synchronization token is broadcast again, and the backrest area resumes effective collection from 15 seconds; then, physiological characteristics are extracted and transmitted between 15 and 45 seconds; throughout the process, the control switching is driven by mechanical state changes, rather than by random backoff determined by the communication layer.

[0134] The technical advantages of the above timing configuration are as follows: by introducing mechanical steady-state determination before acquiring and transmitting second-mode data, the alarm control unit avoids processing high-frequency data contaminated by motion artifacts, thereby reducing the computational load and improving alarm reliability; in this embodiment, by sequentially executing the steps of mechanical arbitration, re-acquisition and feature extraction, the data completes physical validity screening before being transmitted to the alarm control unit.

[0135] Furthermore, if a silence token is received again during steady-state acquisition, the current feature extraction window will terminate immediately, and data within the window that has not yet met the minimum length requirement can be discarded directly; if the minimum length requirement has been met, it can also be used as a valid window that ends early and sent to subsequent calculations.

[0136] If the backrest area fails to form a valid feature after synchronization, a valid data insufficiency flag can be sent to the alarm control unit. If the alarm control unit only receives this flag in multiple consecutive cycles, it can issue a lower-level prompt based on driving duration, time period, and historical feature changes, rather than the highest-level alarm.

[0137] For example, during long-haul freight missions in the early morning hours, the system updates the seat pressure variance every 100 milliseconds; if the driver moves frequently after driving continuously for 3 hours, the process automatically enters a closed loop of silence-waiting-synchronization-feature extraction; the heart rate variability features extracted in the subsequent stabilization phase show a decline in autonomic nervous system regulation ability, and the alarm control unit outputs a fatigue warning accordingly, prompting the driver to stop and rest as soon as possible.

[0138] The purpose of this method is to link cross-modal gating, synchronous reconstruction, and feature transmission into an executable data processing chain, thereby improving the practicality and engineering deployability of fatigue early warning.

[0139] In a preferred embodiment of the present invention, the second sensing node, in response to a channel silence token, discards the currently collected second modality physiological data and stops wireless transmission, including: blocking the data transfer operation from the analog-to-digital converter configured inside the second sensing node to its wireless transmission buffer; performing source-end discard processing on the currently collected second modality physiological data; and controlling the second sensing node to enter a low-power sleep mode.

[0140] This embodiment provides a method-level silent execution step; specifically, although the silent token trigger to stop transmission has been defined in the previous process, if the execution granularity is not fine enough, problems such as the external transmission of remaining data in the buffer, invalid power consumption of the transmitter, or redundant operation of the sampling link may still occur; therefore, this embodiment refines the silent response into three consecutive actions: blocking the transfer, discarding at the source end, and low-power sleep.

[0141] Specifically, when the backrest area receives a silence token, it blocks the data transfer from the analog-to-digital converter to the wireless transmission buffer; this action directly cuts off the entry point of the sampled value into the transmission link; and discards the currently collected but not yet processed second-modality physiological data at the source end.

[0142] For example, if an ECG segment is originally planned to be accumulated to 2 seconds before R-wave detection, but a silence token is received when it is accumulated to 1.2 seconds, then the 1.2-second segment is directly discarded and will no longer participate in subsequent feature calculations; the control node enters a low-power sleep mode, shuts down the transmission-related modules, and retains only the minimum functional units required for listening to the synchronization token;

[0143] As a further specific example of the working process; assume that the buffer currently has 4 slots, denoted as C1, C2, C3, and C4 respectively; where C1 has been sent, C2 is waiting to be sent, C3 is being written, and C4 is empty; when the silent token arrives, this embodiment can terminate the writing of C3 and clear it, C2 can be cleared directly or cleared after sending according to the policy, and C4 remains empty, stopping new transfer requests; in this way, from the time the silent token takes effect, no more complete new data frames will enter the transmission path;

[0144] If only source-end discarding is performed without entering sleep mode, the node may still maintain a high clock frequency to listen to local processes, resulting in limited power saving. Therefore, the third step in this embodiment is equally important. Sleep mode can be set to light sleep or deep sleep. Light sleep is suitable for situations where the expected silence duration is less than the first preset duration, and its wake-up latency is less than the preset latency threshold. Deep sleep is suitable for situations where the attitude change amplitude is greater than the preset amplitude or communication congestion occurs, and its power consumption is lower than that of light sleep mode. The mode can be dynamically selected based on recent silence duration statistics.

[0145] Furthermore, if a node's battery level is detected to be low during the execution of a silent action, it will enter a deeper sleep state first to extend its available time. If a node cannot completely block the transfer due to hardware reasons, an invalid flag can be added to the cache entry point to ensure that this data is not sent by mistake. If a duplicate silent token is received by mistake during the silent period, only the silent timer will be refreshed, and the entire set of switches will not be executed repeatedly to reduce state machine jitter.

[0146] For example, on a long downhill section of a highway, the driver frequently changes his body posture due to nervous operation; the backrest node receives a silence token twice in a row. The method flow first cuts off the transfer between the analog-to-digital converter and the transmission buffer, and then discards the sampled but not fully featured data fragments on the spot, causing the node to enter a light sleep and wait for the next synchronization token sent by the seat area.

[0147] The purpose of this step is to implement silent control at the underlying level, thereby ensuring that invalid data is truly stopped at the source and achieving quantifiable energy-saving and channel-clearing effects.

[0148] In a preferred embodiment of the present invention, the second sensing node extracts physiological features and transmits them to the alarm control unit, including: the second sensing node aligning the time axis of the second modality physiological data with the steady-state synchronization token based on a time reference; the second sensing node extracting heart rate variability features from the second modality physiological data as physiological features; the second sensing node compressing the physiological features and sending the compressed data to the alarm control unit for status assessment.

[0149] This embodiment provides a method and steps for feature extraction and time alignment. Specifically, in the case of only restarting the acquisition after synchronization, although the backrest area has obtained a new time zero point, if the two types of modal data are not further aligned on the same time axis, the alarm control unit still finds it difficult to determine which sitting posture stable interval a certain heart rate variability feature corresponds to. Therefore, this embodiment adds a time axis alignment step before feature extraction.

[0150] Specifically, after receiving the synchronization token, the second sensing node records the arrival time of the token as t0, and uses this as the numbering of subsequent local sampling segments;

[0151] In this embodiment, the time axis of the first modal physiological data and the second modal physiological data is aligned based on a time reference. Preferably, the second sensing node uses a common time reference established by the steady-state synchronization token from the first sensing node to make the steady-state interval identifier corresponding to the first modal data consistent with the sampling segment corresponding to the second modal data in terms of time attribution. Rather, it does not require the second sensing node to hold all the original waveforms of the first modal physiological data.

[0152] The cushion area can generate a lightweight time summary corresponding to the first modality physiological data around t0, such as steady-state start marker, steady-state duration, window number, or steady-state interval label; the second sensing node uses this to associate the local second modality sampling segment with the same time axis identifier, thereby completing the time alignment of the two types of modal data at the method level;

[0153] To avoid confusion of notations, in this embodiment, the steady-state interval labels corresponding to the first modality of physiological data can be denoted as S1 and S2, the sampling window labels corresponding to the second modality of physiological data can be denoted as W1 and W2, and the feature packets extracted from W1 and W2 respectively can be denoted as F1 and F2; the above labels are only used to indicate the time attribution relationship and are used in this sense throughout the text.

[0154] The following specific implementation scenario is used as an example: Assume that the synchronization token arrives at 100 seconds. The first sensor node confirms that it has entered the stable interval S1 starting from 100 seconds. The second sensor node marks the ECG segments acquired between 100 and 130 seconds as window W1 and establishes the correspondence between S1 and W1. After synchronizing again at 150 seconds, the first sensor node confirms the stable interval S2. The second sensor node generates window W2 between 150 and 180 seconds and establishes the correspondence between S2 and W2.

[0155] The second sensing node extracts feature packets F1 and F2 from W1 and W2 respectively; the alarm control unit sees that F1 corresponds to the window [100, 130] and is associated with the stable interval S1, and F2 corresponds to the window [150, 180] and is associated with the stable interval S2, and can associate them with the steady-state information of the corresponding time period of the cushion area, without mistakenly thinking that F1 comes from the section in the middle where there is a drastic change in body posture.

[0156] In terms of heart rate variability extraction, a lightweight approach combining time and frequency domains can be adopted. In the time domain, the mean RR interval, standard deviation, and root mean square difference between adjacent intervals can be extracted. In the frequency domain, low-frequency power, high-frequency power, or the ratio of the two can be estimated while ensuring that the computing power is manageable.

[0157] After extraction, the data is compressed. The compressed package can include at least a window identifier, a set of feature values, and a validity flag. If it is necessary to further reduce the amount of data transmitted, only the trend of change can be sent, such as whether the data has increased, decreased, or remained unchanged compared to the previous window.

[0158] If only features are sent without time information, the central end may not be able to distinguish the sequential relationship between two consecutive stable windows; if only time is aligned without feature compression, there will still be a problem of high bandwidth usage; therefore, this embodiment designs time alignment-feature extraction-compression transmission as a continuous operation.

[0159] Furthermore, if a small time drift occurs between the first modal data and the second modal data after a certain synchronization, such as a millisecond-level deviation in the local clock of the back node, the local cumulative offset can be attached to the feature packet for correction by the alarm control unit;

[0160] If the complete frequency domain index cannot be extracted within a certain window, at least the time domain index shall be retained and a missing index shall be set; if the compressed feature packet is lost in wireless transmission, the alarm control unit can detect the missing packet according to the window number and wait for the next window data, without making excessive inferences about the missing window.

[0161] For example, in the same night freight mission, the system receives a synchronization token at 02:15:30. The backrest node uses this token as a basis to generate a stable window feature package, which includes the window number, average RR interval, standard deviation, and quantized low-frequency ratio. At the same time, the feature package carries a time label corresponding to the steady-state interval of the seat area. After synchronizing again at 02:17:10, the next feature package is generated.

[0162] The alarm control unit evaluated the data in the order of window time and found that the characteristics showed a continuous fatigue trend. Therefore, it raised the warning level and further uploaded the event record to the fleet backend when the driver failed to respond for a continuous period of time.

[0163] The purpose of this step is to ensure that the extracted fatigue-related physiological features are not only small in size but also have a clear temporal attribution, thereby enabling a state assessment process that allows for cross-modal information alignment and traceability.

[0164] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A driver fatigue warning system based on a smart car seat, characterized in that, include: The first sensing node is adapted to collect the first modality of physiological data of the target object; The second sensing node is adapted to collect the second modality physiological data of the target object; An alarm control unit is configured to receive physiological features extracted from the second modality of physiological data by the second sensing node, and generate an abnormal state alarm signal based on the physiological features; a wireless communication connection is established between the first sensing node, the second sensing node, and the alarm control unit; the first sensing node calculates the variance of the first modality of physiological data in real time; When the variance is greater than a preset motion threshold, the first sensor node broadcasts a channel silence token, and the second sensor node responds to the channel silence token by discarding the currently collected second modality physiological data and stopping wireless transmission. When the variance is less than the preset steady-state threshold, the first sensing node broadcasts a steady-state synchronization token, and the second sensing node responds to the steady-state synchronization token by restarting the collection of the second modal physiological data based on the timestamp of the arrival of the steady-state synchronization token, extracting physiological features from the second modal physiological data and transmitting them to the alarm control unit. When the variance is between the preset steady-state threshold and the preset motion threshold and includes boundary values, the second sensing node continues to operate according to the current acquisition and transmission configuration. The first sensing node includes: a pressure sensor array, which is adapted to acquire mechanical stress signals as the first modal physiological data; A first wireless microcontroller, the first wireless microcontroller being adapted to calculate the variance and broadcast the channel silence token and the steady-state synchronization token; The second sensing node includes: an electrophysiological sensor, which is adapted to acquire electromagnetic electrocardiogram signals as the second modal physiological data; A second wireless microcontroller is adapted to receive the channel silence token and the steady-state synchronization token, and to control the transmission status of the second modality physiological data.

2. The driver fatigue warning system based on an intelligent car seat according to claim 1, characterized in that, In response to the channel silence token, the second sensing node discards the currently acquired second modality physiological data and stops wireless transmission, including: The second wireless microcontroller is internally configured with an analog-to-digital converter and a wireless transmission buffer. The second wireless microcontroller blocks the data transfer operation from the analog-to-digital converter to the wireless transmission buffer. The second wireless microcontroller controls the second sensor node to enter a low-power sleep mode in order to release wireless channel resources.

3. The driver fatigue warning system based on an intelligent car seat according to claim 1, characterized in that, The step of extracting physiological features from the second modality physiological data and transmitting them to the alarm control unit includes: the second sensing node extracting heart rate variability features from the restarted second modality physiological data based on the time reference, and generating feature data as the physiological features; The second sensing node compresses the feature data and sends the compressed feature data to the alarm control unit.

4. The driver fatigue warning system based on an intelligent car seat according to claim 1, characterized in that, The first sensing node calculates the variance of the first modal physiological data in real time, including: the first sensing node extracts the numerical distribution characteristics of the first modal physiological data within the sliding window based on a preset sliding window in real time; The first sensing node uses a variance statistics algorithm to calculate the sliding window variance of the distribution characteristics, which is then used as the variance.

5. The driver fatigue warning system based on an intelligent car seat according to claim 1, characterized in that, Applications in smart car seats; The first sensing node is deployed in the seat cushion area of ​​the smart car seat; the second sensing node is deployed in the backrest area of ​​the smart car seat; the abnormal state alarm signal is a driver fatigue warning signal.

6. A driver fatigue warning method based on a smart car seat, employing the driver fatigue warning system based on a smart car seat according to any one of claims 1-5, characterized in that, Includes the following steps: The first sensing node acquires the first modal physiological data of the target object, and the second sensing node acquires the second modal physiological data of the target object; the first sensing node calculates the variance of the first modal physiological data in real time. When the variance is greater than a preset motion threshold, the first sensor node broadcasts a channel silence token, and the second sensor node responds to the channel silence token by discarding the currently collected second modality physiological data and stopping wireless transmission. When the variance is less than the preset steady-state threshold, the first sensing node broadcasts a steady-state synchronization token, and the second sensing node responds to the steady-state synchronization token by restarting the collection of the second modality physiological data based on the timestamp of the arrival of the steady-state synchronization token. When the variance is between the preset steady-state threshold and the preset motion threshold and includes boundary values, the second sensing node continues to operate according to the current acquisition and transmission configuration. The second sensing node extracts physiological features from the second modality physiological data and transmits them to the alarm control unit, so that the alarm control unit can generate an abnormal state alarm signal based on the physiological features.

7. The driver fatigue warning method based on a smart car seat according to claim 6, characterized in that, The second sensing node, in response to the channel silence token, discards the currently collected second modality physiological data and stops wireless transmission, including: blocking the data transfer operation from the analog-to-digital converter configured inside the second sensing node to its wireless transmission buffer; The currently collected second modality physiological data is discarded at the source; the second sensing node is controlled to enter a low-power sleep mode.

8. The driver fatigue early warning method based on a smart car seat according to claim 6, characterized in that, The second sensing node extracts physiological features from the second modal physiological data and transmits them to the alarm control unit, including: the second sensing node aligns the time axis of the second modal physiological data with the steady-state synchronization token based on the time reference; The second sensing node extracts heart rate variability features from the second modality physiological data as the physiological features; the second sensing node compresses the physiological features and sends the compressed data to the alarm control unit for status assessment.

Images