Vital sign monitoring equipment for vehicle and use method

Abstract

The invention relates to the technical field of intelligent automobile safety monitoring, and discloses a vital sign monitoring device for a vehicle and a use method.The vital sign monitoring device comprises a vehicle body and is characterized in that an instrument desk and a plurality of seats are arranged in the vehicle body, and a control panel is arranged on the instrument desk; a plurality of sensing assemblies are arranged on the instrument desk and the seats close to the instrument desk; the control panel is provided with a touch display screen, is internally provided with a central processing unit, a signal processing module, an algorithm analysis module and a communication module, and is used for processing data uploaded by the sensing assembly and outputting early warning; the system effectively solves the problems that a single sensor is liable to be interfered, the traditional monitoring precision is insufficient, the multimode cooperative capability is lacked, and the actual deployment is limited, and improves the reliability, accuracy and practicability of the life safety monitoring of people in a vehicle.

Description

Technical Field

[0001] This invention relates to the field of intelligent vehicle safety monitoring technology, specifically to a vital signs monitoring device for vehicles and its usage method. Background Technology

[0002] With the rapid development of intelligent vehicle technology, vehicle safety has received increasing attention, especially the health and safety of passengers. Traditional vehicle safety systems mainly focus on collision protection and anti-theft alarms, while their ability to monitor and warn of the vital signs of occupants in real time is relatively weak. In situations such as high temperatures, lack of oxygen, or sudden illness, occupants may face life-threatening dangers if they are not detected in time.

[0003] Currently, existing technologies mainly achieve vital sign recognition through a single sensor. For example, infrared thermal imagers or cameras are used for body temperature or facial recognition, or ordinary millimeter-wave radar is used to monitor breathing and heartbeat. These sensors are usually deployed on the dashboard or seat back and rely on image or signal processing algorithms to determine whether a person is in an abnormal state and push warning information through the vehicle display screen or mobile terminal.

[0004] However, the aforementioned existing technologies have the following obvious shortcomings: single sensors are susceptible to environmental interference, such as decreased recognition accuracy in low light, obstruction, or large temperature changes; traditional millimeter-wave radar has low resolution and is difficult to accurately capture weak heartbeat and breathing signals; the systems often lack multi-mode fusion and intelligent collaboration capabilities, making it impossible to achieve stable and reliable vital sign monitoring in complex scenarios; in addition, existing solutions often do not fully consider the installation layout and multi-seat coverage issues in actual vehicle environments, limiting their effectiveness and universality in practical applications. Summary of the Invention

[0005] This invention provides a vital signs monitoring device and method for vehicles, which has the advantages of improving monitoring reliability in complex environments through multimodal intelligent fusion perception, achieving high-precision vital signs extraction and risk assessment using advanced algorithms, and optimizing system layout to enhance the universality of practical applications. It solves the problems mentioned in the background art, such as the susceptibility of single sensors to environmental interference, the difficulty of traditional radar to capture weak vital signs signals, the lack of multimodal fusion and coordination capabilities of the system, and the insufficient consideration of actual vehicle installation layout and multi-seat coverage in existing solutions.

[0006] The present invention provides the following technical solution: a vital signs monitoring device for a vehicle, comprising a vehicle body, characterized in that: an instrument panel and several seats are provided inside the vehicle body, a control panel is provided on the instrument panel, and several sensing components are provided on the instrument panel and the seats near the instrument panel; The control panel has a touch screen and a built-in central processing unit, signal processing module, algorithm analysis module and communication module, which are used to process the data uploaded by the sensing components and output early warnings. The sensing components specifically include millimeter-wave radar, infrared camera, and temperature sensor.

[0007] In a preferred embodiment, the millimeter-wave radar is a 60GHz millimeter-wave radar used to collect respiratory and heartbeat signals of passengers within a target area inside the vehicle.

[0008] In a preferred embodiment, the infrared camera supports dual-mode recognition of thermal imaging and visible light, and works in conjunction with an AI acceleration module.

[0009] In a preferred embodiment, the temperature sensor includes a clinical-grade temperature sensor for measuring body surface temperature and an environmental sensor for measuring ambient temperature and humidity; it also includes a GPS / BeiDou dual-mode positioning chip for positioning.

[0010] In a preferred embodiment, the point cloud data acquired by the millimeter-wave radar is used to separate human body parts using an improved DBSCAN clustering algorithm, and respiratory and heartbeat signals are extracted using a variational mode decomposition algorithm to calculate real-time respiratory rate and heart rate.

[0011] In a preferred embodiment, the thermal imaging data collected by the infrared camera is used to identify key body temperature areas using a deep learning model, and combined with a thermal conduction correction model to predict core body temperature.

[0012] In a preferred embodiment, the central processing unit of the control panel is equipped with a multimodal spatiotemporal feature fusion network based on an attention mechanism, which is used to dynamically weight and fuse heterogeneous data from different sensing components to generate a fused spatiotemporal feature vector of vital signs.

[0013] In a preferred embodiment, the multimodal spatiotemporal feature fusion network includes a three-layer fusion mechanism of sensor-level attention, feature-level spatiotemporal attention, and decision-level ensemble learning; the decision-level ensemble learning uses a lightweight gradient booster to integrate the outputs of multiple neural network branches and calculates the interaction effects between physiological indicators to output a comprehensive risk score of vital signs.

[0014] In a preferred embodiment, the health status assessment and decision-making layer of the control panel uses a model combining a temporal convolutional network and a Transformer encoder to perform pattern recognition on the fused features and execute a graded early warning mechanism including primary alerts, intermediate alarms, advanced warnings, and emergency responses based on the risk score.

[0015] The present invention also provides a method for monitoring vital signs in a vehicle, comprising the following steps: S1: Multimodal data synchronous acquisition, activate millimeter-wave radar, infrared thermal imaging camera, temperature sensor and environmental sensor to synchronously collect data on passengers' breathing, heartbeat, body surface temperature distribution, ambient temperature and humidity and carbon dioxide concentration in the target area inside the vehicle; S2: Signal preprocessing and feature extraction. For point cloud data acquired by millimeter-wave radar, an improved DBSCAN clustering algorithm is used to separate human body parts. Respiratory and heartbeat signals are extracted using variational mode decomposition algorithm, and real-time respiratory rate and heart rate are calculated. For infrared thermal imaging data, a deep network is used to identify key body temperature areas, and a heat conduction correction model is used to predict body temperature. For environmental data, an in-vehicle thermodynamic model is constructed to predict temperature change trends. S3: Multimodal spatiotemporal feature fusion. The respiratory, heart rate, body temperature and environmental risk features extracted by S2 are input into the multimodal spatiotemporal feature fusion network based on the attention mechanism. Through the three-layer attention mechanism of sensor level, feature level and decision level, heterogeneous data are dynamically weighted and fused to generate the fused vital sign spatiotemporal feature vector. S4: Intelligent health status assessment, which combines a feature input temporal convolutional network with a Transformer encoder to identify the combined risk pattern of sleep apnea, arrhythmia and heatstroke precursors, and outputs a comprehensive risk score of vital signs in the range of 0-1 based on a lightweight gradient booster integrating physiological indicators and their interaction effects. S5: Tiered early warning and proactive response. A four-level early warning response mechanism is implemented based on the risk score: When the score < 0.2, it is considered a normal state, and no warning is output; when 0.2 ≤ score < 0.5, a primary alert is triggered, displaying a mild warning message on the control panel; when 0.5 ≤ score < 0.8, a medium-level alarm is triggered, activating audio-visual prompts and highlighting abnormal information; when the score ≥ 0.8, a high-level warning is triggered, executing a combined audio-visual alarm and seat vibration; if the dangerous state persists beyond the set threshold, an emergency response is automatically executed, including dialing a preset emergency number, sending real-time location information, and unlocking the vehicle. S6: Based on the federated learning framework, it anonymizes and stores users' historical vital signs data locally, and periodically uploads encrypted features to the cloud for model aggregation and updates, enabling the system to adaptively optimize for different users, vehicle models and environments, thereby improving long-term monitoring accuracy and personalized service capabilities.

[0016] The present invention has the following beneficial effects: 1. This invention employs a multimodal sensing component consisting of millimeter-wave radar, infrared thermal imaging, and environmental sensors. It designs an attention-based multimodal spatiotemporal feature fusion network (AMM-STFN) to dynamically weight heterogeneous data, effectively overcoming the shortcomings of traditional single sensors that are susceptible to environmental interference (such as low light, obstruction, and temperature and humidity changes). The 60GHz millimeter-wave radar accurately identifies heartbeat and respiratory micro-movements at the 0.1mm level. Combined with the infrared thermal imaging dual-light system and thermodynamic model, it can maintain medical-grade monitoring accuracy even in complex in-vehicle environments, significantly improving the system's environmental adaptability and data reliability.

[0017] 2. This invention uses an architecture that combines a temporal convolutional network (TCN) with a Transformer encoder to simultaneously capture local temporal patterns and long-range cross-modal dependencies. It accurately identifies complex risks such as sleep apnea, arrhythmia, and heatstroke precursors, calculates the independent risk of each physiological indicator, introduces interaction terms to quantify the synergistic effect of multiple indicators, outputs a precise vital sign risk score, and triggers a four-level early warning system from screen prompts and audible and visual alarms to automatic emergency contact, thus achieving a leap from passive monitoring to proactive safety protection.

[0018] 3. This invention constructs a complete intelligent closed loop of "perception-fusion-decision-interaction," combining engineering practicality with the potential for personalized services. It employs a high-performance in-vehicle computing platform and AI acceleration chip to ensure real-time algorithm operation. On the software side, it forms an interpretable and adaptive fusion decision-making process through a multi-layered attention mechanism at the sensor, feature, and decision levels. It provides real-time health status and trend reports, and continuously optimizes the model while protecting privacy through a federated learning framework, adapting to different users, vehicle models, and environments. This effectively solves the shortcomings of existing technologies in terms of multi-seat coverage, actual installation layout, and long-term stability, providing a highly reliable and implementable system-level solution for intelligent vehicle health and safety monitoring. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a top view of the present invention; Figure 3 This is a schematic diagram of the control panel architecture of the present invention; Figure 4 This is a schematic diagram of the overall method of the present invention.

[0020] In the picture: 1. Vehicle body; 2. Dashboard; 3. Sensing components; 4. Control panel; 5. Seat. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The vital signs monitoring device and method for vehicles involved in the present invention are not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figures 1-3 The device shown is a vital signs monitoring device for a vehicle, including a vehicle body 1, an instrument panel 2 and several seats 5 inside the vehicle body 1, a control panel 4 on the instrument panel 2, and several sensing components 3 on the instrument panel 2 and the seats 5 near the instrument panel 2. The control panel 4 has a touch screen for displaying monitoring information and setting parameters and receiving user commands; The control panel 4 has been completely reconstructed based on the traditional in-vehicle display screen, forming a vital sign monitoring center with intelligent perception, multimodal data fusion processing, and proactive safety decision-making capabilities. The control panel adopts a multi-layer embedded system architecture, such as... Figure 3 As shown, it includes: a multi-source data acquisition and preprocessing layer, a vital sign feature extraction and fusion layer, a health status assessment and decision-making layer, and a human-computer interaction and early warning output layer. The core technology lies in the construction of an attention-based multimodal spatiotemporal feature fusion network (AMM-STFN), which can dynamically weight and process heterogeneous data from different sensors and adaptively adjust the feature fusion strategy according to changes in the in-vehicle environment.

[0023] In one specific embodiment, the control panel hardware is equipped with a Renesas R-Car H3 high-performance automotive SoC as the main processor, along with 8GB of LPDDR5 memory and 256GB of UFS 3.1 storage, ensuring the real-time operation of complex algorithms. The touchscreen display is a 12.3-inch OLED curved screen with a resolution of 2560×1600, supporting multi-touch and glove operation mode. A Huawei Ascend 310 AI acceleration chip is integrated as a coprocessor for accelerating neural network inference, reducing the latency of vital sign recognition to less than 200ms.

[0024] The multi-source data acquisition and preprocessing layer includes a millimeter-wave radar data processing module. The millimeter-wave radar uses a Texas Instruments AWR1843 60GHz FMCW radar, configured in a multiple-input multiple-output (MIMO) array mode, forming 16 virtual antenna channels. It can perform fine scanning of passengers in three-dimensional space. The radar acquires 100 frames of point cloud data per second, and each frame includes four-dimensional information such as azimuth, elevation, range, and Doppler velocity.

[0025] The specific data processing flow is as follows: An improved DBSCAN clustering algorithm is used, incorporating velocity consistency constraints, to separate the point clouds of different parts of the passenger's body. The algorithm first calculates the adaptive distance threshold between the point clouds: ε=α·σv+β·(1−SNR / SNRmax), Where σv is the velocity variance, SNR is the signal-to-noise ratio, SNRmax is the maximum signal-to-noise ratio, and α and β are empirical coefficients, with α taking the value of 0.3 and β taking the value of 0.7.

[0026] For point clouds in the chest region, the variational mode decomposition (VMD) algorithm is applied to separate respiratory and heartbeat signals. VMD decomposes the signal into K eigenmode functions by constructing a variational problem: , Σkuk(t)=f(t), Where uk is the k-th eigenmode function, ωk is the center frequency of the k-th eigenmode, f(t) is the original input signal, and δ(t) is the unit impulse function, satisfying: δ(t) = +∞, t = 0, t ≠ 0 , Combined with the Hilbert transform kernel j / πt, it forms an analytic signal construction operator; j / πt is the kernel function of the Hilbert transform. To obtain the partial derivative with respect to time, It is a complex exponential function used to shift the modal spectrum to the baseband.

[0027] Instantaneous frequencies of respiratory rate (BR) and heart rate (HR) are extracted using Hilbert transform: BR = 12π·ddt[arg(Hi(sb))], HR=12π·ddt[arg(Hi(sh))]×60, Where sb is the respiratory signal, sh is the heartbeat signal, Hi is the Hilbert transform operator, arg is the argument function of the complex number, 60 is the bpm conversion coefficient, and 12π is the unit conversion coefficient.

[0028] The multi-source data acquisition and preprocessing layer also includes an infrared thermal imaging data processing module. The infrared camera uses a FLIR Lepton 3.5 thermal imaging sensor with a resolution of 160×120 and a thermal sensitivity of <50mK. It forms a dual-light system with a visible light camera (Sony IMX585). A deep learning-based multispectral feature fusion method is proposed, specifically including: Thermal anomaly detection: Using an improved YOLOv5-Thermal model, ROI extraction was performed in the neck, armpit, and temples. The model was trained on a self-built thermal imaging dataset and achieved an mAP of 0.92.

[0029] Core body temperature estimation: A heat conduction correction model is established to estimate the core body temperature Tc using multiple surface temperature points. Tc=Σiwi·Ts,i+ΔTa+ΔTm, The weights wi are obtained through neural network learning. wi is the i-th temperature weight coefficient, with a value of 0.40 for the carotid artery area, 0.30 for the axillary area, 0.15-0.25 for the temples / forehead, and 0.10 for other areas. Ts,i is the i-th individual surface temperature, ΔTa is the ambient temperature correction term with a value of 0.01-0.05, and ΔTm is the metabolic rate correction term with a value of 0.005-0.02.

[0030] Micro-expression thermal feature analysis: 3D-CNN is used to extract time-series features of facial thermal distribution to identify abnormal states (such as pain, syncope precursors); feature dimensions include: temperature gradient, thermal asymmetry, and thermal diffusion rate.

[0031] The multi-source data acquisition and preprocessing layer also includes an environmental and positioning data processing module. The temperature sensor used is the MAX30205 clinical-grade human body temperature sensor with an accuracy of ±0.1℃, combined with an SHT40 environmental temperature and humidity sensor. The positioning module uses the ATGM336H dual-mode positioning chip and establishes a thermodynamic model of the in-vehicle environment, as detailed below: Hazard index calculation: Constructing a vehicle heat accumulation model to predict the trend of temperature changes inside the vehicle: , Where Ti is the interior temperature, α is the solar radiation absorption coefficient (0.6-0.9), Qs is the solar radiation flux, β is the exterior ambient temperature, β is the vehicle body thermal conductivity (2.0-5.0), γ is the air convection heat transfer coefficient (1200-2000), Av is the ventilation area, va is the air velocity, and Ca is the air heat capacity.

[0032] Positioning-assisted monitoring: Combining GPS / BeiDou positioning and vehicle status information (such as door and window status) to determine whether the vehicle is in a stationary and closed state.

[0033] The vital sign feature extraction and fusion layer includes a multimodal feature fusion network and a hierarchical attention fusion mechanism with three levels: Sensor-level attention: Assigning dynamic weights to each type of sensor data. wr=σ(MLP([SNR,mv,cr])), wt=σ(MLP([tg,fd,ct])), we = σ(MLP([td,h,CO2l])), Where σ is the sigmoid function, wr dynamically reflects the reliability of the millimeter-wave radar at the current moment, wt evaluates the availability of infrared thermal imaging data, we reflects the credibility and importance of environmental sensor data, and SNR, mv, cr, tg, fd, ct, td, h, CO2l are the signal-to-noise ratio, motion variance, radar confidence, temperature gradient, face detection confidence, thermal imaging confidence, internal and external temperature difference, vehicle interior humidity, and carbon dioxide concentration, respectively.

[0034] Feature-level spatiotemporal attention: using a Transformer encoder to capture cross-modal spatiotemporal dependencies. Attention(Q,K,V)=softmax(QKT / dk)V, Where Q, K, and V are feature maps from different modalities, namely the query matrix, key matrix, and value matrix, respectively; dk is the key-query dimension; QKT is the query-key dot product; and softmax is the normalized exponential function.

[0035] Decision-level ensemble learning: A lightweight gradient boosting machine (LightGBM) is used to integrate the outputs of multiple neural network branches, ultimately outputting a vital sign risk score. Rs=Σiλi·fi(x)+Σi <jμij·fi(x)fj(x), Where x is the input feature vector, fi(x) / fj(x) are the outputs of the i-th and j-th base predictors respectively, λi is the linear weight of the i-th base predictor, the respiratory abnormality detector has a value of 0.3, the heart rate abnormality detector has a value of 0.25, the body temperature abnormality detector has a value of 0.20, the environmental risk detector has a value of 0.18, the motion state detector has a value of 0.07, μij is the interaction weight between the i-th and j-th predictors, respiratory × heart rate has a value of 0.1, respiratory × body temperature has a value of 0.06, heart rate × body temperature has a value of 0.08, respiratory × environment has a value of 0.05, heart rate × environment has a value of 0.05, body temperature × environment has a value of 0.15, and motion × respiratory has a value of -0.01, and i,j are predictor indices.

[0036] The health status assessment and decision-making layer includes a physiological state classification model, which establishes a four-dimensional health status classification system based on fusion features: Normal state (risk score < 0.2); Attention state (0.2 ≤ score < 0.5): such as mild respiratory abnormalities; Warning state (0.5 ≤ score < 0.8): such as abnormal body temperature or arrhythmia; Dangerous state (score ≥ 0.8): such as apnea or heatstroke precursors.

[0037] Temporal Convolutional Networks (TCNs) are used to perform pattern recognition on continuous monitoring data. The dilated convolution formula for TCNs is: , Where t is the current output time index, k is the convolution kernel size, f(k) is the weight of the k-th convolution kernel, which is learned through training and has no preset fixed value. The initial value is randomly sampled from the standard normal distribution with a standard deviation of 0.01. d is the dilation coefficient. Layer 1: d=1 (capturing second-level changes), Layer 2: d=2 (5-10 second mode), Layer 3: d=4 (15-30 second mode), Layer 4: d=8 (1-2 minute mode), Layer 5: d=16 (3-5 minute mode), and d·k is the dilation offset.

[0038] The human-computer interaction and early warning output layer provides four levels of early warning response on the control panel: primary reminder: a gentle prompt on the screen, suitable for a state of attention; intermediate alarm: an audible prompt + a bright screen display, suitable for a warning state; advanced warning: a combined audible and visual alarm + seat vibration, suitable for a dangerous state; emergency response: automatically dialing emergency services + sending location + unlocking doors, suitable for a life-threatening situation.

[0039] In addition, the system also provides health trend reports, which visually display the changes in vital signs over the past 24 hours / 7 days / 30 days and provide health advice.

[0040] The touch screen is equipped with a central processing unit, which includes a signal processing module, an algorithm analysis module, and a communication module, for processing data uploaded by the sensing components. The sensing component 3 specifically includes a millimeter-wave radar, an infrared camera, and a temperature sensor; The millimeter-wave radar uses the 60GHz millimeter-wave radar sensor R60ABD1 as its core sensor. Compared with the traditional 24GHz radar, its resolution is improved by 4 times (up to 4cm), and it can accurately identify heartbeats and respiratory waves at the 0.1mm level. The infrared camera can work in conjunction with the K210 AI acceleration module and supports dual-mode recognition of thermal imaging and visible light, which can compensate for the shortcomings of radar in image detail recognition. The temperature sensor features automatic day / night mode switching, a temperature measurement range of -55-125℃, and an accuracy of ±0.1℃. It employs an ATGM336H positioning module that supports "GPS + Beidou" multi-mode positioning.

[0041] A method for monitoring vital signs in vehicles, such as Figure 4 As shown, it includes S1: Multimodal data synchronous acquisition, which activates millimeter-wave radar, infrared thermal imaging camera, temperature sensor and environmental sensor to synchronously collect data on passengers' breathing, heartbeat, body surface temperature distribution, ambient temperature and humidity and carbon dioxide concentration in the target area inside the vehicle.

[0042] S2: Signal preprocessing and feature extraction. For point cloud data acquired by millimeter-wave radar, an improved DBSCAN clustering algorithm is used to separate human body parts. Respiratory and heartbeat signals are extracted through variational mode decomposition algorithm, and real-time respiratory rate and heart rate are calculated. For infrared thermal imaging data, a deep network is used to identify key body temperature areas, and a heat conduction correction model is used to predict body temperature. For environmental data, an in-vehicle thermodynamic model is constructed to predict temperature change trends.

[0043] S3: Multimodal spatiotemporal feature fusion. The respiratory, heart rate, body temperature and environmental risk features extracted by S2 are input into the multimodal spatiotemporal feature fusion network based on the attention mechanism. Through the three-layer attention mechanism of sensor level, feature level and decision level, heterogeneous data are dynamically weighted and fused to generate the fused vital sign spatiotemporal feature vector.

[0044] S4: Intelligent health status assessment combines a temporal convolutional network with a Transformer encoder to identify a composite risk pattern of sleep apnea, arrhythmia, and heatstroke precursors. Based on a lightweight gradient booster, it integrates physiological indicators and their interaction effects to output a comprehensive risk score of vital signs in the range of 0-1.

[0045] S5: Tiered early warning and proactive response. A four-level early warning response mechanism is implemented based on the risk score: When the score < 0.2, it is considered a normal state, and no warning is output; when 0.2 ≤ score < 0.5, a primary alert is triggered, displaying a mild warning message on the control panel; when 0.5 ≤ score < 0.8, an intermediate alarm is triggered, activating audio-visual prompts and highlighting abnormal information; when the score ≥ 0.8, a high-level warning is triggered, executing a combined audio-visual alarm and seat vibration; if the dangerous state persists beyond the set threshold, an emergency response is automatically executed, including dialing a preset emergency number, sending real-time location information, and unlocking the vehicle.

[0046] S6: Based on the federated learning framework, it anonymizes and stores users' historical vital signs data locally, and periodically uploads encrypted features to the cloud for model aggregation and updates, enabling the system to adaptively optimize for different users, vehicle models and environments, thereby improving long-term monitoring accuracy and personalized service capabilities.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A vital signs monitoring device for a vehicle, comprising a vehicle body (1), characterized in that: The vehicle body (1) is provided with an instrument panel (2) and several seats (5). The instrument panel (2) is provided with a control panel (4). Several sensing components (3) are provided on the instrument panel (2) and the seats (5) near the instrument panel (2). The control panel (4) has a touch screen and a built-in central processing unit, signal processing module, algorithm analysis module and communication module, which are used to process the data uploaded by the sensing component (3) and output warnings; The sensing component (3) specifically includes a millimeter-wave radar, an infrared camera, and a temperature sensor.

2. The vital signs monitoring device for a vehicle according to claim 1, characterized in that: The millimeter-wave radar is a 60GHz millimeter-wave radar used to collect the breathing and heartbeat signals of passengers in the target area inside the vehicle.

3. The vital signs monitoring device for a vehicle according to claim 1, characterized in that: The infrared camera supports dual-mode recognition of thermal imaging and visible light, and works in conjunction with the AI ​​acceleration module.

4. A vital signs monitoring device for a vehicle according to claim 1, characterized in that: The temperature sensor includes a clinical-grade temperature sensor for measuring body surface temperature and an environmental sensor for measuring ambient temperature and humidity; it also includes a GPS / BeiDou dual-mode positioning chip for positioning.

5. A vital signs monitoring device for a vehicle according to claim 2, characterized in that: The point cloud data acquired by the millimeter-wave radar is used to separate human body parts using an improved DBSCAN clustering algorithm, and respiratory and heartbeat signals are extracted using a variational mode decomposition algorithm to calculate real-time respiratory rate and heart rate.

6. A vital signs monitoring device for a vehicle according to claim 3, characterized in that: The thermal imaging data collected by the infrared camera is used to identify key body temperature areas using a deep learning model, and combined with a thermal conduction correction model to predict core body temperature.

7. A vital signs monitoring device for a vehicle according to claim 1, characterized in that: The central processing unit of the control panel (4) is equipped with a multimodal spatiotemporal feature fusion network based on an attention mechanism, which is used to dynamically weight and fuse heterogeneous data from different sensing components (3) to generate a fused spatiotemporal feature vector of vital signs.

8. A vital signs monitoring device for a vehicle according to claim 7, characterized in that: The multimodal spatiotemporal feature fusion network includes a three-layer fusion mechanism: sensor-level attention, feature-level spatiotemporal attention, and decision-level ensemble learning. The decision-level ensemble learning uses a lightweight gradient booster to integrate the outputs of multiple neural network branches and calculates the interaction effects between physiological indicators to output a comprehensive risk score of vital signs.

9. A vital signs monitoring device for a vehicle according to claim 1, characterized in that: The health status assessment and decision-making layer of the control panel (4) uses a model combining a temporal convolutional network and a Transformer encoder to perform pattern recognition on the fused features and execute a graded early warning mechanism including primary alerts, intermediate alarms, advanced warnings and emergency responses based on the risk score.

10. A method for monitoring vital signs in a vehicle, used to implement the vital signs monitoring device for a vehicle as described in any one of claims 1-9, characterized in that, include: S1: Multimodal data synchronous acquisition, activate millimeter-wave radar, infrared thermal imaging camera, temperature sensor and environmental sensor to synchronously collect data on passengers' breathing, heartbeat, body surface temperature distribution, ambient temperature and humidity and carbon dioxide concentration in the target area inside the vehicle; S2: Signal preprocessing and feature extraction. For point cloud data acquired by millimeter-wave radar, the improved DBSCAN clustering algorithm is used to separate human body parts. The variational mode decomposition algorithm is used to extract respiratory and heartbeat signals and calculate real-time respiratory rate and heart rate. For infrared thermal imaging data, a deep network is used to identify key body temperature areas, and a heat conduction correction model is used to predict body temperature. Based on environmental data, construct an in-vehicle thermodynamic model to predict temperature change trends; S3: Multimodal spatiotemporal feature fusion. The respiratory, heart rate, body temperature and environmental risk features extracted by S2 are input into the multimodal spatiotemporal feature fusion network based on the attention mechanism. Through the three-layer attention mechanism of sensor level, feature level and decision level, heterogeneous data are dynamically weighted and fused to generate the fused vital sign spatiotemporal feature vector. S4: Intelligent health status assessment, which combines a feature input temporal convolutional network with a Transformer encoder to identify the combined risk pattern of sleep apnea, arrhythmia and heatstroke precursors, and outputs a comprehensive risk score of vital signs in the range of 0-1 based on a lightweight gradient booster integrating physiological indicators and their interaction effects. S5: Tiered early warning and proactive response. A four-level early warning response mechanism is implemented based on the risk score: when the score is <0.2, it is judged as a normal state and no early warning is output; when 0.2≤score<0.5, a primary reminder is triggered and a mild prompt message is displayed on the control panel; when 0.5≤score<0.8, an intermediate alarm is triggered, an audible and visual prompt is activated and the abnormal information is highlighted. When the score is ≥0.8, an advanced warning is triggered, and a combined sound and light alarm and seat vibration are activated; If the vehicle remains in a dangerous state for more than a set threshold, an emergency response will be automatically executed, including dialing a preset emergency number, sending real-time location information, and unlocking the vehicle. S6: Based on the federated learning framework, it anonymizes and stores users' historical vital signs data locally, and periodically uploads encrypted features to the cloud for model aggregation and updates, enabling the system to adaptively optimize for different users, vehicle models and environments, thereby improving long-term monitoring accuracy and personalized service capabilities.

Images