Vehicle control method, electronic device, and computer-readable storage medium

Abstract

This application provides a vehicle control method, electronic device, and computer-readable storage medium, relating to the field of vehicle control technology. The method includes: acquiring the driver's height, weight, and facial data, wherein the facial data is acquired through a data acquisition device within the vehicle's cockpit; determining the driver's spinal posture angle based on the height and facial data; inputting the height, weight, and spinal posture angle into a pre-trained target spinal stress prediction model to obtain a prediction result, wherein the target spinal stress prediction model is a regression mapping model trained based on biomechanical simulation data; determining a driver health intervention strategy based on the prediction result and spinal biomechanical load constraints; and generating vehicle control commands based on the driver health intervention strategy. This application solves the technical problem in related technologies where real-time prediction of driver spinal stress is impossible, leading to a lack of precise biological basis for driver health interventions.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, and more specifically, to a vehicle control method, electronic device, and computer-readable storage medium. Background Technology

[0002] The stress state of the driver's spine is a core biomechanical parameter for assessing driver fatigue and optimizing seat ergonomics. During driving, the spine bears complex loads from the combined effects of body weight, seat posture, pedal and steering wheel reactions. The accumulation of these loads can lead to intervertebral disc degeneration. Therefore, achieving real-time and accurate prediction of spinal stress is a crucial prerequisite for building an intelligent and healthy driving system.

[0003] However, current methods for assessing spinal stress rely on high-fidelity biomechanical simulation software. While these methods can provide highly accurate stress distribution results, a single simulation can take tens of minutes to several hours, which is completely insufficient to meet the real-time requirements of vehicle systems for millisecond-level response. Consequently, it is impossible to implement health interventions for drivers based on real-time spinal stress in existing vehicles.

[0004] There is currently no good solution to the above problems. Summary of the Invention

[0005] This application provides a vehicle control method, electronic device, and computer-readable storage medium to at least solve the technical problem in the related art that the driver's spinal force cannot be predicted in real time, resulting in a lack of accurate biological basis for health interventions for the driver.

[0006] According to one aspect of the embodiments of this application, a vehicle control method is provided, comprising: acquiring the driver's height, weight, and facial data, wherein the facial data is acquired through a data acquisition device in the vehicle cabin; determining the driver's spinal posture angle based on the height and facial data; inputting the height, weight, and spinal posture angle into a pre-trained target spinal stress prediction model to obtain a prediction result, wherein the target spinal stress prediction model is a regression mapping model trained based on biomechanical simulation data; determining a driver health intervention strategy based on the prediction result and spinal biomechanical load constraints; and generating vehicle control commands based on the driver health intervention strategy.

[0007] Furthermore, the facial data includes: initial facial data and current facial data. Based on height and facial data, the driver's spinal posture angle is determined, including: determining a reference facial depth based on the initial facial data, wherein the initial facial data is used to characterize the driver's facial data when maintaining an upright sitting posture in the driver's seat, and the reference facial depth is used to characterize the reference distance of the driver's face relative to the data acquisition device; determining the current facial depth based on the current facial data; and determining the spinal posture angle based on the reference facial depth, the current facial depth, and height.

[0008] Furthermore, based on the reference facial depth, the current facial depth, and the height, the spinal posture angle is determined, including: determining the depth change based on the reference facial depth and the current facial depth; determining the target length based on the height, the target length being used to characterize the distance from the driver's hip joint rotation center to the eye point; and determining the spinal posture angle based on the depth change and the target length.

[0009] Furthermore, the driver health intervention strategy includes: a seat adjustment strategy and a fatigue warning strategy. Based on the prediction results and spinal biomechanical load constraints, the driver health intervention strategy is determined, including: determining a seat adjustment strategy based on the prediction results and spinal biomechanical load constraints; or, determining a fatigue warning strategy based on the prediction results and spinal biomechanical load constraints. Based on the driver health intervention strategy, vehicle control commands are generated, including: generating driver seat adjustment commands based on the seat adjustment strategy; or, generating driver fatigue warning commands based on the fatigue warning strategy.

[0010] Furthermore, the vehicle control method also includes: acquiring multiple sets of sample data corresponding to multiple drivers, wherein any one set of sample data includes: driver height, driver weight, and driver spinal posture angle; analyzing the multiple sets of sample data to construct multiple skeletal muscle models corresponding to multiple drivers; setting simulation conditions for the multiple skeletal muscle models, wherein the simulation conditions are used to describe the force application of the multiple skeletal muscle models; under the simulation conditions, using biomechanical simulation software to perform force simulation calculations on the multiple skeletal muscle models respectively, obtaining multiple simulation results; and constructing biomechanical simulation data based on the multiple sets of sample data and multiple simulation results.

[0011] Furthermore, the vehicle control method also includes: training an initial regression mapping model based on biomechanical simulation data to obtain an initial spinal force prediction model, wherein the initial spinal force prediction model includes an initial first sub-network, an initial second sub-network, and an initial third sub-network. The initial first sub-network is used to predict the normal pressure along the spinal axis and perpendicular to the intervertebral disc plane, the initial second sub-network is used to predict the tangential force along the anterior-posterior direction of the spine and parallel to the intervertebral disc plane, and the initial third sub-network is used to predict the transverse tangential force along the lateral direction of the spine and parallel to the intervertebral disc plane; and model compression is performed on the initial spinal force prediction model to obtain the target spinal force prediction model.

[0012] Furthermore, the initial spinal stress prediction model is compressed to obtain the target spinal stress prediction model, including: performing network pruning on the initial first subnetwork, initial second subnetwork, and initial third subnetwork respectively to obtain the pruned first subnetwork, pruned second subnetwork, and pruned third subnetwork; and performing knowledge distillation on the pruned first subnetwork, pruned second subnetwork, and pruned third subnetwork respectively to obtain the target spinal stress prediction model.

[0013] According to another aspect of the embodiments of this application, a vehicle control system is also provided, including: an acquisition module for acquiring the driver's height, weight, and facial data, wherein the facial data is acquired through a data acquisition device in the vehicle cabin; a first determination module for determining the driver's spinal posture angle based on the height and facial data; a prediction module for inputting the height, weight, and spinal posture angle into a pre-trained target spinal stress prediction model to obtain a prediction result, wherein the target spinal stress prediction model is a regression mapping model trained based on biomechanical simulation data; a second determination module for determining a driver health intervention strategy based on the prediction result and spinal biomechanical load constraints; and a generation module for generating vehicle control commands based on the driver health intervention strategy.

[0014] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0015] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a memory storing an executable program; and a processor for running the executable program, wherein the executable program executes the methods in various embodiments of this application when it runs.

[0016] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0017] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0018] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the methods in various embodiments of this application.

[0019] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0020] In this embodiment, the driver's height, weight, and facial data are acquired, with the facial data collected by a data acquisition device in the vehicle cabin. Based on the height and facial data, the driver's spinal posture angle is determined. The height, weight, and spinal posture angle are input into a pre-trained target spinal force prediction model to obtain prediction results. The target spinal force prediction model is a regression mapping model trained based on biomechanical simulation data. Based on the prediction results and spinal biomechanical load constraints, a driver health intervention strategy is determined. Based on the driver health intervention strategy, vehicle control commands are generated. This application first obtains the driver's height, weight, and facial data collected by the in-cabin data acquisition device as the basis for subsequent spinal stress prediction. Based on the height and facial data, the spinal posture angle is calculated, realizing real-time posture estimation and solving the technical bottleneck of posture measurement relying on high-cost hardware. The height, weight, and spinal posture angle are input into a regression mapping model (i.e., the target spinal stress prediction model) trained with biomechanical simulation data, compressing the originally time-consuming simulation calculation to a millisecond-level response, realizing real-time and accurate mapping from the driver's current sitting posture to spinal stress, breaking through the computational barrier that force prediction cannot run online in existing technologies. Based on the predicted spinal stress value (i.e., the prediction result), a health intervention strategy is generated in combination with spinal biomechanical load constraints, so that the vehicle's health intervention behavior (such as seat adjustment, fatigue warning, etc.) has scientificity and pertinence based on real biomechanical load, rather than relying on fuzzy judgments based on experience or indirect pressure signals. Finally, vehicle control commands are generated according to the health intervention strategy, realizing closed-loop control from real-time stress prediction to active health intervention. In summary, the technical solution disclosed in this application effectively solves the technical problem in related technologies that it is impossible to predict the driver's spinal stress in real time, which leads to a lack of accurate biological basis for health interventions for drivers. It achieves the technical effect of predicting the driver's spinal stress in real time and then using the spinal stress as a basis for accurate health interventions for drivers. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0022] Figure 1 This is a flowchart of a vehicle control method according to an embodiment of this application;

[0023] Figure 2 This is a flowchart of a training method for a spinal stress prediction model according to an embodiment of this application;

[0024] Figure 3 This is a schematic diagram illustrating the principle of spinal posture angle estimation according to an embodiment of this application;

[0025] Figure 4 This is a structural block diagram of a vehicle control system according to an embodiment of this application;

[0026] Figure 5 This is a schematic diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] According to an embodiment of this application, an embodiment of a vehicle control method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] This embodiment provides a vehicle control method. Figure 1 This is a flowchart of a vehicle control method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0031] Step S10: Obtain the driver's height, weight, and facial data, wherein the facial data is obtained through a data acquisition device inside the vehicle's cabin.

[0032] Step S11: Determine the driver's spinal posture angle based on height and facial data;

[0033] Step S12: Input the height, weight and spinal posture angle into the pre-trained target spinal force prediction model to obtain the prediction result. The target spinal force prediction model is a regression mapping model trained based on biomechanical simulation data.

[0034] Step S13: Based on the prediction results and spinal biomechanical load constraints, determine the driver's health intervention strategy for the vehicle.

[0035] Step S14: Generate vehicle control commands based on driver health intervention strategies.

[0036] The driver's height and weight were pre-entered through the vehicle's infotainment system or manually entered by the driver upon first use of the vehicle (unit: cm) and weight (unit: kg). Height and weight are used as basic input parameters for individual biomechanical characteristics.

[0037] The aforementioned facial data refers to pixel-level features extracted from images captured by the data acquisition device that are directly related to the three-dimensional spatial pose of the head, including but not limited to the two-dimensional coordinates of key anatomical landmarks such as the center of the pupils, the tip of the nose, the corners of the mouth, and the brow bone. These key points provide the geometric constraint basis for subsequent three-dimensional pose calculation.

[0038] Optionally, a driver monitoring system (DMS) camera already deployed in the vehicle's cabin can be used to capture real-time visible light image sequences of the driver's facial area. The DMS camera is typically a low-power complementary metal-oxide-semiconductor image sensor in the infrared or visible light band, mounted above the steering wheel or dashboard, and features autofocus and auto-exposure capabilities, enabling stable imaging at night or in bright light conditions.

[0039] In one optional embodiment, after the driver starts the vehicle, voice prompts guide the driver to complete an initial action of assuming a standard upright sitting posture. Simultaneously, facial images are captured to establish an individual baseline, improving the accuracy of subsequent posture estimation. Height and weight parameters can be stored in the vehicle's local user profile and can be remotely updated via a mobile application or vehicle networking platform to ensure the continuous validity of the input.

[0040] Determining the driver's spinal posture angle based on height and facial data refers to using prior knowledge of anthropometry and geometric projection relationships to invert height information and head posture information in two-dimensional facial images into an estimated value of the overall spinal posture.

[0041] In one optional embodiment, based on statistical laws of human anatomy, the driver's height is multiplied by an empirical proportional coefficient to calculate the upper body length. This length is defined as the straight-line distance from the hip joint rotation center to the eye point (or the center of the bridge of the nose), serving as a geometric proxy for the spinal midline. Combining this with the spatial coordinates of facial key points in the three-dimensional world coordinate system calculated by the DMS camera using the Perspective-n-Point (PnP) algorithm, the depth value of the current head center point is obtained. Based on the baseline depth recorded during the initialization phase when the driver is in an upright sitting posture, the degree of torso tilt relative to the vertical direction, i.e., the spinal posture angle, can be inferred by calculating the depth change.

[0042] The aforementioned spinal posture angle is directly used for subsequent force prediction. Its advantage lies in its complete reliance on the vehicle-mounted vision system, eliminating the need for additional inertial sensors or markers, thus achieving truly contactless and non-invasive posture monitoring.

[0043] The aforementioned target spinal stress prediction model is a mathematical mapper built in a data-driven manner. Its essence is a machine learning model trained on a large number of high-fidelity simulation samples (such as multiple sets of individualized spinal stress data generated by simulation software). Its training objective is to learn the nonlinear functional relationship from "human body parameters + posture angle" to "intervertebral disc compression force, anterior and posterior shear force, and lateral shear force".

[0044] Optionally, after training, the target spinal force prediction model undergoes lightweight processing such as model pruning, parameter quantization, and knowledge distillation, compressing the model size to less than 15KB and the inference time to less than 1ms, which can be directly embedded in the vehicle controller.

[0045] The above predictions refer to the biomechanical load on key segments of the spine, i.e., the spinal force values. Optionally, the spinal force values ​​include: intervertebral disc compressive force, anterior-posterior shear force, and lateral shear force.

[0046] In one optional embodiment, the target spinal stress prediction model takes normalized height, weight, and posture angles as input and outputs the biomechanical loads on key spinal segments. During the inference phase, after receiving real-time input, the target spinal stress prediction model can output predicted values ​​through a single forward calculation, without any iterative or numerical integration process, completely eliminating the high computational overhead of traditional simulation software.

[0047] Optionally, during the training of the target spinal force prediction model, algorithms such as support vector regression or Gaussian process regression are used, with radial basis functions as kernel functions, so that the model can effectively capture the nonlinear force response caused by posture changes, and can maintain prediction stability, especially when the human posture is close to the limit angle, thus achieving accurate translation from "posture perception" to "internal mechanical state".

[0048] The aforementioned spinal biomechanical load constraints refer to the acceptable stress range set to ensure the driver's spinal health (i.e., the physiological safety boundaries set based on ergonomic research and clinical medical data). For example: the intervertebral disc compressive force threshold is set at 1000N (exceeding this value will significantly increase the risk of intervertebral disc degeneration), the anterior-posterior shear force threshold is set at 150N (exceeding this value can easily induce facet joint slippage), and the lateral shear force threshold is set at 120N (excessive levels may cause lateral ligament strain). When any force component exceeds the threshold limit, the current driving posture is determined to pose a potential health threat.

[0049] In one optional embodiment, determining a driver health intervention strategy based on prediction results and spinal biomechanical load constraints involves comparing the predicted spinal stress values ​​with physiological safety thresholds recognized in the medical and engineering fields to dynamically generate health intervention decisions.

[0050] Optionally, the above health intervention strategy is a risk-level-based tiered intervention strategy.

[0051] For example, if the spinal stress value is within the warning range (e.g., intervertebral disc compression force of 800 to 1000 N), mild intervention is triggered, such as displaying a "suggest adjusting seating posture" prompt on the central control screen or automatically activating the seat vibration reminder. If the stress enters the danger range (e.g., intervertebral disc compression force > 1000 N), active intervention is initiated, such as automatically adjusting the seat back angle by 5°, increasing the pressure of the lumbar support airbag, or activating the seat heating and massage functions to promote local blood circulation. In addition, health intervention strategies should consider the cumulative effect over time. If the high-load state continues for more than 15 minutes, even if the instantaneous value does not exceed the limit, a fatigue accumulation warning will be triggered, and the driver will be advised to rest.

[0052] In one alternative embodiment, the development of health intervention strategies incorporates medical thresholds, engineering feasibility, and human-computer interaction psychology to ensure that the response is both biomechanically grounded and avoids frequent false triggers that could disrupt driving safety.

[0053] The aforementioned generation of vehicle control commands based on driver health intervention strategies refers to converting the aforementioned intervention decisions into specific control signals that can be executed by the vehicle system and sending them to the corresponding execution unit via CAN bus or Ethernet.

[0054] Optionally, vehicle control commands are encapsulated by the control logic unit in the vehicle's infotainment system and communicate directly with subsystems such as the seat control module, air conditioning system, and voice prompt module. During the generation of vehicle control commands, a gradual adjustment strategy is prioritized to avoid sudden actions that could cause driver discomfort.

[0055] For example, the seat back angle adjustment moves smoothly at a rate of 1° per second, rather than abruptly; the airbag pressure increase uses a stepped increment to ensure a comfortable transition.

[0056] Meanwhile, to prevent command conflicts, a priority mechanism is set up: if the driver manually operates the seat adjustment knob, all automatic interventions will be automatically suspended, and monitoring will resume 5 seconds after the manual operation ends, ensuring the initiative and safety of human-machine collaboration.

[0057] In addition, the execution results of vehicle control commands can form a closed loop through sensor feedback. For example, the seat angle sensor confirms that the angle has been adjusted to the correct position, and the airbag pressure sensor verifies that the pressure has reached the standard. Based on this, the status is updated and an intervention log is recorded for subsequent health trend analysis and model adaptive optimization.

[0058] In this embodiment, the driver's height, weight, and facial data are acquired, with the facial data collected by a data acquisition device in the vehicle cabin. Based on the height and facial data, the driver's spinal posture angle is determined. The height, weight, and spinal posture angle are input into a pre-trained target spinal force prediction model to obtain prediction results. The target spinal force prediction model is a regression mapping model trained based on biomechanical simulation data. Based on the prediction results and spinal biomechanical load constraints, a driver health intervention strategy is determined. Based on the driver health intervention strategy, vehicle control commands are generated. This application first obtains the driver's height, weight, and facial data collected by the in-cabin data acquisition device as the basis for subsequent spinal stress prediction. Based on the height and facial data, the spinal posture angle is calculated, realizing real-time posture estimation and solving the technical bottleneck of posture measurement relying on high-cost hardware. The height, weight, and spinal posture angle are input into a regression mapping model (i.e., the target spinal stress prediction model) trained with biomechanical simulation data, compressing the originally time-consuming simulation calculation to a millisecond-level response, realizing real-time and accurate mapping from the driver's current sitting posture to spinal stress, breaking through the computational barrier that force prediction cannot run online in existing technologies. Based on the predicted spinal stress value (i.e., the prediction result), a health intervention strategy is generated in combination with spinal biomechanical load constraints, so that the vehicle's health intervention behavior (such as seat adjustment, fatigue warning, etc.) has scientificity and pertinence based on real biomechanical load, rather than relying on fuzzy judgments based on experience or indirect pressure signals. Finally, vehicle control commands are generated according to the health intervention strategy, realizing closed-loop control from real-time stress prediction to active health intervention. In summary, the technical solution disclosed in this application effectively solves the technical problem in related technologies that it is impossible to predict the driver's spinal stress in real time, which leads to a lack of accurate biological basis for health interventions for drivers. It achieves the technical effect of predicting the driver's spinal stress in real time and then using the spinal stress as a basis for accurate health interventions for drivers.

[0059] The vehicle control method in the embodiments of this application will be further described below.

[0060] Optionally, the facial data includes: initial facial data and current facial data, and based on height and facial data, determining the driver's spinal posture angle, including:

[0061] Step S111: Determine the reference facial depth based on the initial facial data, wherein the initial facial data is used to characterize the driver's facial data when the driver is in an upright sitting position in the driver's seat, and the reference facial depth is used to characterize the reference distance of the driver's face relative to the data acquisition device.

[0062] Step S112: Determine the current facial depth based on the current facial data;

[0063] Step S113: Determine the spinal posture angle based on the baseline facial depth, the current facial depth, and the height.

[0064] The aforementioned initial facial data refers to the sequence of facial images of the driver in a preset upright sitting posture, collected by the onboard driver monitoring system (DMS) camera during the system initialization phase.

[0065] The above-mentioned upright sitting posture is as follows: the angle between the torso and thighs is close to 90°, the head is kept in a neutral position, the eyes look straight ahead, and the back is lightly against the back of the chair. This posture corresponds to the "neutral spinal posture" defined by ergonomics, which is the benchmark state with the least spinal load.

[0066] The data acquisition device is usually an infrared or visible light camera integrated above the steering wheel or inside the dashboard, and its optical axis has a fixed geometric relationship with the driver's line of sight.

[0067] The aforementioned baseline facial depth refers to the three-dimensional spatial depth coordinates of key facial points (such as the tip of the nose, the center of the eyebrows, or the midpoint of the line connecting the two eyes) of the driver in the camera coordinate system, calculated using a perspective-n-point algorithm under the upright sitting posture. In other words, it is the vertical distance from that point to the camera's image sensor plane. This depth value is individual-specific, does not depend on absolute distance, and serves as a relative reference for subsequent posture changes.

[0068] In one optional embodiment, the system prompts the driver to maintain an upright sitting posture for approximately 3 to 5 seconds. The DMS camera continuously captures multiple frames of facial images at a frequency of 20 to 30 Hz. A facial landmark detection algorithm extracts at least 68 facial feature points. Then, using a perspective-n-point algorithm combined with a known 3D facial template model, the coordinates of these landmarks in 3D space are deduced. Finally, the median of the nose tip depth values ​​across 10 consecutive frames is taken as the baseline facial depth to suppress transient jitter and measurement noise. This step does not rely on any external calibration equipment and is entirely based on the in-vehicle camera and a pre-set facial geometry model, ensuring automatic calibration in mass-produced vehicles without manual intervention.

[0069] It is easy to understand that determining the baseline facial depth is about establishing an individualized spatial reference baseline for the driver in order to eliminate posture measurement deviations caused by differences in individual height and seat position.

[0070] The aforementioned current facial data refers to the driver's facial image stream continuously collected by the DMS camera at a frequency of 30Hz or higher during driving. Its content includes real-time changes in the driver's head position and posture caused by actions such as adjusting seating posture, drowsiness, turning, or reaching out to operate.

[0071] In one optional embodiment, the current frame image is processed by a perspective-n-point algorithm to calculate the real-time three-dimensional depth coordinates of facial key points (such as the tip of the nose) in the camera coordinate system, which is denoted as the current facial depth.

[0072] The current facial depth changes continuously with the driver's head movement. When the driver leans forward, the face moves closer to the camera, and the current facial depth decreases; when the driver leans back, the face moves away from the camera, and the current facial depth increases. To improve measurement robustness, an optical flow-based facial feature point tracking mechanism is introduced during the calculation process to avoid the loss of key points due to blinking, occlusion, or changes in lighting.

[0073] Meanwhile, to reduce the noise impact of false detections in a single frame, a sliding window averaging method is used to smooth the current facial depth of five consecutive frames, outputting a stable and reliable estimate of the current facial depth. This process is completed on the vehicle's embedded processor with a latency of less than 5ms, meeting real-time requirements.

[0074] It should be noted that facial depth measurement does not rely on absolute distance calibration, but only on the relative geometric relationship between the camera and the face. Therefore, it has good generalization ability across different car models, different seat positions, and different drivers, without the need to calibrate camera parameters separately for each car model.

[0075] The aforementioned spinal posture angle is defined as the angle between the center of the driver's first lumbar vertebra (L1) and the horizontal plane. This angle directly reflects the degree of inclination of the trunk relative to the pelvis and is the core variable affecting the stress on the intervertebral discs in the L4-L5 segment.

[0076] In the above steps, by establishing individualized reference depth, continuously tracking facial spatial displacement, and combining height parameters for geometric conversion, for the first time, high-precision, low-latency, and vehicle-deployable real-time perception of the driver's spinal posture angle was achieved without any wearable devices, external calibration, or dedicated sensors. This provides an indispensable input foundation for the subsequent lightweight prediction of spinal stress.

[0077] Optionally, the spinal posture angle is determined based on the reference facial depth, the current facial depth, and the height, including:

[0078] Step S1131: Determine the depth change based on the reference facial depth and the current facial depth;

[0079] Step S1132: Based on height, determine the target length, which is used to characterize the distance from the driver's hip joint rotation center to the eye point;

[0080] Step S1133: Determine the spine posture angle based on the depth change and target length.

[0081] The aforementioned depth change reflects the displacement of the driver's head relative to the reference posture along the camera's optical axis.

[0082] In one alternative embodiment, the depth change is equal to the difference between the reference facial depth and the current facial depth.

[0083] A positive value for the depth change indicates that the driver's head is moving forward, meaning the upper body is leaning forward. A negative value for the depth change indicates that the head is moving backward, meaning the upper body is leaning backward.

[0084] The aforementioned target length is a key geometric parameter in human biomechanical modeling. Its physical essence is the equivalent rigid rod length of the driver's upper body, which is the straight-line distance from the center of rotation of the hip joint (located near the pelvis, roughly above the navel and at the midpoint of the line connecting the anterior superior pelvic spines) to the eye point (the midpoint of the line connecting the pupils of both eyes). This distance determines the geometric mapping relationship between the driver's head movement and trunk tilt.

[0085] Since direct measurement of the internal anatomical structure of the human body is not feasible in a vehicle environment, an empirical formula based on the statistical laws of anthropometry is used to estimate the target length. The empirical formula is shown below:

[0086] Formula (1)

[0087] in, The target length (i.e., upper body length). Height, Proportion Coefficient The range of values ​​is Preferred .

[0088] The target length serves as a rigid geometric constraint, so that subsequent attitude angle calculations no longer depend on the absolute accuracy of the absolute depth, but only on the ratio of depth change to this fixed length, thereby significantly improving the generalization ability and cross-individual applicability of attitude estimation.

[0089] The aforementioned spinal posture angle refers to the angle of inclination of the trunk relative to the vertical direction.

[0090] When the depth change is positive (i.e., the driver leans forward), the driver's head moves forward along the optical axis of the camera. The formula for calculating the spine posture angle at this time is:

[0091] Formula (2)

[0092] in, For the spinal posture angle, For the target length, This represents the depth change.

[0093] When the depth change is negative (i.e., the driver leans back), the head moves backward, and the angle between the upper body and the vertical direction exceeds 90°. The formula for calculating the spinal posture angle at this point is:

[0094] Formula (3)

[0095] in, For the spinal posture angle, For the target length, This represents the depth change.

[0096] In addition, to improve stability, a first-order Kalman filter is applied to the spine posture angles calculated in consecutive frames to suppress instantaneous jumps caused by camera exposure fluctuations, occlusion, or blinking, ultimately outputting smooth spine posture angles as the core input variables for subsequent spine stress prediction models.

[0097] In the above steps, without relying on any dedicated hardware, high-precision, low-latency, and individualized estimation of spinal posture angles can be achieved using only the existing DMS system, providing a reliable and real-time dynamic input basis for subsequent lightweight spinal stress prediction models.

[0098] Optionally, driver health intervention strategies include: seat adjustment strategies and fatigue warning strategies. Based on prediction results and spinal biomechanical load constraints, driver health intervention strategies are determined, including:

[0099] Step S131: Based on the prediction results and spinal biomechanical load constraints, determine the seat adjustment strategy;

[0100] Step S132: Based on the prediction results and the spinal biomechanical load constraints, determine the fatigue early warning strategy.

[0101] Optionally, after completing the real-time prediction of the driver's spinal stress, based on the predicted intervertebral disc compression force, anterior-posterior shear force, and lateral-lateral shear force (i.e., prediction results), and combined with the preset spinal safety load threshold (i.e., spinal biomechanical load constraint conditions), a seat adjustment strategy is determined to reduce local spinal stress and delay fatigue accumulation.

[0102] The aforementioned spinal biomechanical load constraints refer to the acceptable stress range set to ensure the spinal health of drivers. For example, the compressive force of the L4-L5 intervertebral disc should not exceed 1000N, the anterior-posterior shear force should not exceed 150N, and the lateral shear force should not exceed 120N. These thresholds are derived from clinical medical literature and human biomechanical experimental data, representing the engineering safety boundary that will not cause chronic damage under long-term exposure.

[0103] In one optional embodiment, the real-time predicted force value is first compared with each constraint threshold. If any index exceeds the upper limit, the seat adaptive adjustment process is triggered.

[0104] Optionally, the generation of the seat adjustment strategy relies on multi-objective optimization logic. Based on the current force distribution characteristics, it is determined whether leaning forward causes excessive compressive force or leaning backward causes increased shear force, and then decides to adjust parameters such as seat back angle, seat cushion depth, lumbar support height, or seat cushion firmness.

[0105] For example, when the intervertebral disc compression force exceeds the threshold limit and the spinal posture angle is less than 85°, it is determined that excessive forward tilting has led to a decrease in lumbar lordosis. At this time, the lumbar support height is automatically increased and the backrest tilt angle is appropriately increased to 105° to restore the normal physiological curvature of the lumbar spine. If the anterior and posterior shear forces exceed the standard and the posture angle is greater than 100°, it is determined that excessive trunk posterior tilting has led to pelvic posterior rotation. At this time, the height of the front edge of the seat cushion is reduced and the seat position is moved forward to make the hip joint in a more neutral flexion angle, thereby reducing the lumbar shear load.

[0106] In one alternative embodiment, the seat adjustment strategy is determined based on a preset "force-posture-seat parameter" mapping table, which is constructed by simulation optimization and expert experience to ensure that each abnormal force mode corresponds to a unique, safe, effective and ergonomic seat response action.

[0107] Furthermore, when the predicted value of spinal stress exceeds the safety threshold or shows a cumulative increasing trend, the system automatically determines that the driver is in a state of spinal fatigue and generates targeted visual, auditory, or tactile warning information to remind the driver to rest or adjust their posture in time, thereby preventing the occurrence of occupational low back pain.

[0108] In one alternative embodiment, based on a dynamic fatigue accumulation model, the integral value of the force on the spine per unit time is calculated to form a "load-time" accumulation, which reflects the potential for continuous strain damage to muscles and intervertebral discs.

[0109] Optionally, the fatigue warning strategy is generated based on a tiered alarm mechanism. Level 1 (mild fatigue) is triggered when the load approaches 70% of the threshold, displaying a gentle text message on the central control screen: "It is recommended to slightly adjust your posture to relieve lumbar pressure," accompanied by slight seat vibrations (such as a 200ms localized pulse vibration of the backrest). Level 2 (moderate fatigue) is triggered when the load exceeds 90% of the threshold, initiating a voice announcement: "You have been driving continuously for more than 2 hours; it is recommended to stop and rest for 10 minutes." Simultaneously, a yellow lumbar spine icon illuminates on the instrument panel, and the air conditioning fan speed is automatically reduced to decrease muscle tension caused by cold stimulation. Level 3 (severe fatigue) is triggered when the cumulative load exceeds the limit or a sudden peak force (such as a single compression force exceeding 1500N). This forces the vehicle into silent mode, flashes a bright red warning light, and forcibly wakes the driver through a combination of steering wheel vibration and full-width seat vibration, while automatically navigating to the nearest service area and playing a safety reminder video.

[0110] The above steps together constitute a closed loop of proactive health intervention based on real-time spinal stress prediction. The two work together to achieve a technological leap from "passive monitoring" to "proactive protection".

[0111] Based on driver health intervention strategies, vehicle control commands are generated, including:

[0112] Step S141: Generate driver's seat adjustment command based on seat adjustment strategy;

[0113] Step S142: Based on the fatigue warning strategy, generate a driver fatigue warning instruction.

[0114] It should be noted that the driver's seat adjustment command is not a simple qualitative command such as "move forward" or "recline backward," but rather includes continuous control signals in multiple dimensions, specifically: backrest tilt adjustment amount (unit: °), seat cushion fore-and-aft sliding distance (unit: mm), seat cushion height adjustment value (unit: mm), lumbar support pressure intensity (unit: kPa), and headrest longitudinal / tilt fine-tuning value. These parameters are all generated in the form of standardized digital control signals, which can be directly mapped to the CAN bus protocol message of the vehicle's electric seat controller to drive the corresponding motor to perform actions.

[0115] Furthermore, the driver fatigue warning command is not a single audio prompt, but a complex set of control signals integrating visual, acoustic, and tactile feedback. During the generation of the driver fatigue warning command, an adaptive response mechanism is employed: if the driver actively adjusts their posture within 5 seconds (posture angle returns to a safe range), the current warning is automatically canceled; if the same level of warning is triggered three times consecutively without response, the next warning level is automatically increased by one level.

[0116] In addition, driver fatigue warning commands are distributed through the unified event bus of the in-vehicle infotainment system to ensure coordinated operation with subsystems such as navigation, audio, and air conditioning, and to avoid resource conflicts.

[0117] The above steps, without adding any external hardware, complete the whole-cycle, non-invasive protection of the driver's lumbar spine health, significantly improving the biomechanical safety and comfort experience during driving.

[0118] Optionally, the vehicle control method further includes:

[0119] Step S151: Obtain multiple sets of sample data corresponding to multiple drivers, wherein any set of sample data includes: driver height, driver weight, and driver spinal posture angle.

[0120] Step S152: Analyze multiple sets of sample data to construct multiple skeletal muscle models corresponding to multiple drivers;

[0121] Step S153: Set simulation conditions for multiple skeletal muscle models, wherein the simulation conditions are used to describe the force application of multiple skeletal muscle models.

[0122] Step S154: Under simulation conditions, use biomechanical simulation software to perform force simulation calculations on multiple skeletal muscle models to obtain multiple simulation results.

[0123] Step S155: Construct biomechanical simulation data based on multiple sets of sample data and multiple simulation results.

[0124] In one optional embodiment, multiple sets of sample data corresponding to multiple drivers are obtained, and each set of sample data contains three key input features: driver height, driver weight, and driver spinal posture angle.

[0125] Among them, driver height refers to the vertical distance from the soles of the feet to the top of the head, in centimeters, reflecting the overall size of the human body; driver weight refers to the mass of the particle corresponding to the gravity acting on the driver in a static state, in kilograms, used to characterize the level of body load; driver spinal posture angle refers to the degree of inclination of the torso relative to the vertical direction, used to quantify the degree of forward or backward tilting of the driver's sitting posture.

[0126] Furthermore, based on each driver's height and weight information, biomechanical modeling software was used to perform personalized geometric scaling and mass distribution adjustments on the standard human body model, constructing a multibody dynamic musculoskeletal model that conforms to individual anatomical characteristics.

[0127] The musculoskeletal model consists of rigid skeletons (including the spine, pelvis, and lower limbs), muscle units (such as the erector spinae, rectus abdominis, quadratus lumborum, and other core trunk muscles), joint connections, and neural control modules, which can realistically simulate the processes of muscle activation, force transmission, and joint torque generation.

[0128] During the modeling process, based on each driver's height and weight information, parameters such as bone length, muscle origin and insertion points, and muscle cross-sectional area are linearly scaled to ensure that the model is consistent with the real driver in terms of geometry and mechanics. For example, for every 10cm increase in height, the spine length is extended proportionally, and the physiological cross-sectional area of ​​the muscles is adjusted according to the square root of body weight to maintain the physiological correspondence between muscle strength and body weight.

[0129] The above-mentioned simulation conditions for multiple skeletal muscle models refer to defining the external load environment for each personalized model under driving conditions in order to simulate the biomechanical response under real driving situations.

[0130] The simulation conditions include: a gravitational load, vertically downward, with a magnitude equal to the driver's weight multiplied by gravitational acceleration; a pedal reaction force, applied to the right foot pedal contact point, with a magnitude of 50N and a direction perpendicular to the pedal plane, representing the force exerted by the driver's foot during constant speed cruising or slight acceleration; and a steering wheel reaction force, applied to the hands' grip points, with a magnitude of 20N and a direction horizontally backward, simulating the grip force required by the driver to maintain directional control.

[0131] In addition, the simulation system sets the seat support force to a rigid contact constraint to simulate the support effect of a real seat on the buttocks and back, but does not apply an active adjustment force.

[0132] Under simulation conditions, multiple musculoskeletal models are subjected to stress simulation calculations using biomechanical simulation software to obtain multiple simulation results. This refers to using a professional biomechanical simulation platform to perform dynamic mechanical solutions on each personalized musculoskeletal model and output the stress response of the target vertebral segment.

[0133] Optionally, the simulation employs a forward dynamics method, calculating the internal force distribution among the vertebrae of the spine at each time step by solving the muscle co-activation equation and the joint force balance equation. The simulation targets are the L4-L5 and L5-S1 segments, which are the key areas of the lumbar spine where stress is most concentrated and strain is most likely to occur.

[0134] The simulation output force data (i.e., simulation results) includes: intervertebral disc compressive force, i.e., axial pressure perpendicular to the vertebral endplate, reflecting the axial load borne by the intervertebral disc; anterior-posterior shear force, the horizontal shear component along the long axis of the spine, reflecting the tensile stress on the posterior annulus fibrosus of the intervertebral disc when the trunk leans forward; and lateral shear force, the transverse shear component perpendicular to the sagittal plane of the spine, reflecting the torsional shear force borne by the spine when turning or the vehicle rolls. The simulation runs at a frequency of 100Hz, with each posture condition lasting for 5 seconds to ensure that the steady-state force state is fully captured.

[0135] Constructing biomechanical simulation data based on multiple sets of sample data and simulation results involves using the acquired driver's height, weight, and spinal posture angles as input variables, and the simulated intervertebral disc compression force, anterior-posterior shear force, and lateral shear force as output labels to build a structured training dataset. This dataset is a multidimensional regression sample library from three-dimensional input (height, weight, and spinal posture angles) to three-channel output (forces in three directions). Each data point is uniquely identified by its sample number, thus forming complete high-fidelity simulation data (i.e., biomechanical simulation data).

[0136] In the above steps, biomechanical simulation data is constructed to provide a high-precision, high-coverage "knowledge base" for subsequent lightweight model training.

[0137] Optionally, the vehicle control method further includes:

[0138] Step S156: Based on biomechanical simulation data, the initial regression mapping model is trained to obtain the initial spinal force prediction model. The initial spinal force prediction model includes an initial first sub-network, an initial second sub-network, and an initial third sub-network. The initial first sub-network is used to predict the normal pressure along the spinal axis and perpendicular to the intervertebral disc plane. The initial second sub-network is used to predict the tangential force along the anterior-posterior direction of the spine and parallel to the intervertebral disc plane. The initial third sub-network is used to predict the transverse tangential force along the lateral direction of the spine and parallel to the intervertebral disc plane.

[0139] Step S157: Compress the initial spinal force prediction model to obtain the target spinal force prediction model.

[0140] It is easy to understand that the forces exerted on the spine during changes in driving posture are not a resultant force in a single direction, but a three-dimensional force vector composed of normal pressure (i.e., vertical compressive force of the intervertebral disc), anterior-posterior tangential force (i.e., shear force caused by flexion / extension), and lateral tangential force (i.e., shear force caused by lateral bending or torsion). These three forces are not independent in terms of biomechanical mechanisms, but their driving factors and the nonlinear coupling relationship with the posture angle differ significantly. Therefore, a "component decoupling modeling" strategy is adopted to decompose the originally complex multi-output regression problem into three independent single-output regression tasks, handled by the initial first sub-network, initial second sub-network, and initial third sub-network, respectively, thereby improving the prediction accuracy of each component and the model convergence stability.

[0141] The initial input to the first subnetwork is the driver's height, weight, and spinal posture angle, and the output is the normal compressive force on the intervertebral disc in the L4-L5 segment.

[0142] The initial second subnetwork predicts the shear force in the anterior-posterior direction of the intervertebral disc. This force mainly originates from the torque imbalance caused by gravity when the trunk leans forward, as well as the forward pulling effect transmitted through the pelvic-spinal chain by the reaction force of the steering wheel and pedal. Its trend tends to flatten out after the posture angle exceeds 100°, and there is a clear inflection point nonlinearity.

[0143] The initial third subnetwork is used to predict the lateral shear force in the left-right direction. This force is usually small and is mainly caused by slight tilting, steering operation or asymmetrical seat support during long-term driving. Its variation is highly random and individual variability.

[0144] In one optional embodiment, all three sub-networks employ the same deep neural network architecture: the input layer receives normalized human body parameters (height, weight) and spinal posture angles, passing through two to three fully connected hidden layers, each containing 128 to 256 neurons, with modified linear units as the activation function to enhance the model's ability to fit nonlinear relationships; the output layer is a single neuron that directly outputs the predicted values ​​of the target force components, without an activation function to ensure continuous real-valued outputs. During training, mean squared error is used as the loss function, adaptive moment estimation is selected as the optimization algorithm, the learning rate is set to 0.001, the batch size is 32, the training epochs are 500, and an early stopping mechanism is used to prevent overfitting. The three sub-networks are trained independently and do not share parameters. Furthermore, to improve the model's generalization ability, Gaussian noise perturbation (standard deviation of 3% of the feature mean) and data augmentation strategies are introduced during the training phase, such as randomly perturbing the posture angle within ±5° and interpolating to generate new samples, thereby simulating the small fluctuations in posture during real driving and making the model robust to measurement errors. After training, when the three sub-networks meet the preset prediction accuracy on the validation set, it indicates that the model has fully learned the complex mapping relationship from human body parameters and posture angles to the three-dimensional force components of the spine.

[0145] Furthermore, without significantly sacrificing prediction accuracy, the initial spinal force prediction model is compressed to meet the storage resource and computing power constraints of the vehicle-mounted embedded controller.

[0146] In the above steps, a target spinal force prediction model is obtained through training, forming a software-based, embeddable, high-precision, and low-latency real-time spinal force prediction technology solution, providing a practical core algorithm engine for intelligent seat adaptive adjustment, driver fatigue warning, and occupational health protection.

[0147] Optionally, the initial spinal stress prediction model is compressed to obtain the target spinal stress prediction model, including:

[0148] Step S1571: Perform network pruning on the initial first subnetwork, initial second subnetwork and initial third subnetwork respectively to obtain the pruned first subnetwork, pruned second subnetwork and pruned third subnetwork.

[0149] Step S1572: Perform knowledge distillation on the first, second, and third pruned subnetworks respectively to obtain the target spinal force prediction model.

[0150] The initial spinal stress prediction model has a large number of parameters and a dense structure, containing a large number of connection weights and neurons that contribute little to the output.

[0151] The aforementioned network pruning refers to the technical process of removing network components that have minimal impact on prediction results without significantly impairing model performance.

[0152] In one optional embodiment, the network pruning operation employs a sensitivity analysis strategy based on weight magnitude: first, the absolute value of each connection weight is calculated, and its gradient response on the validation set is statistically analyzed to identify the low-magnitude parameters least sensitive to output perturbations; then, these parameters are set to zero, and their updates are frozen, forming sparse subnetworks. The pruning process is performed layer by layer, prioritizing the removal of redundant neurons in fully connected layers while preserving key feature mapping paths. Optionally, the pruning ratio is dynamically determined based on the precision-complexity trade-off curve of each subnetwork, with the compression rate controlled within the 60%–80% range to ensure that the final model can maintain stable inference on embedded processors. After pruning, the storage space of each subnetwork is reduced from hundreds of KB to less than tens of KB, and the number of floating-point operations required for inference is reduced by more than 70%, laying the foundation for subsequent lightweight deployment.

[0153] Although the three types of subnetworks after pruning have sparse structures, they still have some redundant computational paths. Knowledge distillation is a model compression method. Its core idea is to use a high-precision, high-capacity teacher model (i.e., a subnetwork that retains complete predictive ability after pruning) to guide a student model with fewer parameters and a simpler structure to learn its output distribution and internal decision-making logic.

[0154] In one optional embodiment, each pruned sub-network acts as an independent teacher model, and its output is treated as a "soft label" used to train the corresponding student model. The student model employs a minimal feedforward neural network structure, containing only 1–2 fully connected layers, with no more than 16 neurons per layer. During training, the loss function consists of two parts: first, the mean squared error between the student model output and the true label (from the simulation dataset), used to ensure basic regression ability; second, the mean squared error between the student model output and the teacher model output, used to guide the student to imitate the teacher's predicted distribution, thereby learning the implicit relationship between the teacher and the nonlinear combination of input features.

[0155] Optionally, distillation training employs a temperature scaling technique, smoothing the outputs of both teachers and students at a temperature scale of T=3 using a Softmax function. This enhances information transmission in low-probability regions, making it easier for students to capture the relative confidence structure of the teacher's output, rather than focusing solely on the maximum value. The training data utilizes the aforementioned biomechanical simulation data, incorporating random noise perturbation features to enhance generalization. After distillation, the model size of each sub-network is further compressed to 2-5KB, with the total number of parameters controlled below 1000, and the inference latency consistently below 0.6ms. Finally, the three distilled mini-models are encapsulated into independent prediction modules, which are then combined to obtain a unified target spinal force prediction model. Using height, weight, and spinal posture angle as inputs to the target spinal force prediction model, the model outputs three types of force components in parallel, achieving end-to-end millisecond-level spinal force prediction.

[0156] In the above steps, the two-stage lightweight processing of network pruning and knowledge distillation achieves the ultimate optimization of model size and computational efficiency, providing core technical support for contactless, low-power, and high-response driver spinal health monitoring in intelligent cockpit systems.

[0157] Optionally, Figure 2 This is a flowchart of a training method for a spinal stress prediction model according to an embodiment of this application, as shown below. Figure 2 As shown, the training of the spinal stress prediction model includes: sample data acquisition; simulation dataset construction; lightweight mapping model training; model simplification and compression; model verification and solidification; and real-time deployment and application in vehicles.

[0158] In one optional embodiment, 80 sample drivers, aged 20-65 years, representing the 5th, 50th, and 95th percentiles of male and female anthropometric dimensions, were selected. Height H (cm) and weight W (kg) were measured for each driver. An inertial motion capture system was used to record the spinal posture angles of each driver in 10 typical driving postures. These 10 typical driving postures included seat back angles of 90°, 100°, 110°, and 120°, each combined with different seat cushion tilt angles. The spinal posture angle α was defined as the angle between the first lumbar vertebra and the horizontal line. The spinal posture angle was approximately 90° when sitting upright, less than 90° when leaning forward, and greater than 90° when leaning back.

[0159] Optionally, a personalized skeletal muscle model was created for each sample driver using biomechanical simulation software. Model parameters were scaled according to height and weight to ensure the model conformed to real human biomechanical characteristics. Simulation boundary conditions were set for each sample driver's posture: gravity direction vertically downwards; 50N pedal force applied to the right foot; 20N steering wheel grip force applied to both hands. Simulation calculations were run to extract the intervertebral disc compression force F_comp, anterior-posterior shear force F_shear_ap, and lateral shear force F_shear_ml for the L4-L5 segments. Through this process, a total of 800 sets of sample data were obtained, each set including input features (H, W, α) and output labels (F_comp, F_shear_ap, F_shear_ml). 650 sets were used as the training set, and 150 sets as the validation set.

[0160] Optionally, the training data is normalized to unify the dimensions of each feature. A support vector regression algorithm is used, with the radial basis function as the kernel function, to train three independent regression models to predict F_comp, F_shear_ap, and F_shear_ml, respectively.

[0161] Optionally, the model training employs five-fold cross-validation to optimize hyperparameters and ultimately determine the penalty coefficient. Kernel function parameters The core prediction function formula is as follows:

[0162] Formula (4)

[0163] Input features: , after average and standard deviation Normalization H and W determine the basic magnitude of the force. Kernel function parameters act as dynamic attitude variables, dominating nonlinear changes. The value is 0.1, located in the exponent term. In this context, the width of the Gaussian kernel is controlled to balance the sensitivity to local attitude changes with the smoothness of the overall curve. Penalty coefficient. The value is 10, which, although not directly appearing in the prediction formula, constrains the trade-off between fitting error and model complexity during the training phase, ensuring high accuracy of the base model. Output results This represents the predicted spinal force value, which needs to be inversely normalized back to Newton units.

[0164] Considering the deployment efficiency of support vector regression models in embedded environments, a support vector reduction algorithm is used to compress the number of support vectors from 80% of the original training set to 20% (i.e., ...). Significantly reduces model storage space and computation time. The compressed model retains the key support vector set. and bias The compressed model takes approximately 0.8ms for a single prediction on the test platform, meeting the millisecond-level real-time computing requirements (i.e., on the target embedded hardware, the time for a single complete prediction is <10ms, which is preferred). ).

[0165] Alternatively, principal component analysis can be used for dimensionality reduction, pruning algorithms can be used to remove redundant neurons, model quantization techniques can be used to convert floating-point parameters into low-bit parameters, or knowledge distillation techniques can be used to simplify complex models into lightweight models.

[0166] In addition, the compressed model was accuracy verified using 150 sets of reserved validation data. The verified models (e.g., mean absolute percentage error) were then used. or coefficient of determination Three models were solidified to obtain a real-time prediction model of the driver's spinal force. This model is stored in binary file format and internally encapsulates normalized parameters. Hyperparameters And a compressed support vector library, which will be used for subsequent deployment and application of vehicle terminals to achieve millisecond-level calculation of spinal force based on real-time attitude.

[0167] In one alternative embodiment, a method for real-time prediction of driver spinal stress in a vehicle environment is provided, which does not require wearing any motion capture equipment and can achieve attitude estimation and stress prediction using only an existing DMS camera.

[0168] First, obtain the driver's height H, and calculate the upper body length using an empirical formula. The upper body length is defined as the distance from the center of hip rotation to the center of the head. Secondly, when the driver uses the system for the first time, they are prompted to maintain an upright sitting posture (the angle between the torso and thighs is approximately 90°, at which point the α angle is approximately 90°). Facial images are captured by the DMS camera, and the 3D coordinates of the head center are estimated using a perspective-n-point algorithm. The average depth value over 3 seconds is taken as the baseline depth d0. During driving, the DMS camera continuously captures facial images of the driver at a frame rate of 30fps, and the current facial depth value is calculated for each frame using the perspective-n-point algorithm. Calculate the depth change According to geometric relationships, When it is a positive value (indicating that the driver is leaning forward), α is calculated using formula (2). When the value is negative (indicating the driver is leaning back), α is calculated using formula (3). Kalman filtering is applied to the attitude angles calculated over multiple consecutive frames to obtain smoothed attitude angles. The smoothed attitude angles, combined with the driver's height H and weight W, are input into a pre-built rapid prediction model of spinal stress (i.e., the target spinal stress prediction model), and the current attitude angle is output in real time. Segmental intervertebral disc compressive force Front and rear shear forces Left and right shear forces F_shear_ml.

[0169] Alternatively, motion capture equipment can be used to accurately measure the driver's spinal posture angles in different sitting positions, serving as training benchmark data for high-fidelity simulation models.

[0170] Through comparative experiments, although the method in this embodiment introduces attitude angle measurement error, the final force prediction error can be controlled within a certain range due to the robustness of spinal forces to changes in attitude angle. Within this range, it meets the needs of engineering applications and achieves completely contactless real-time vehicle-mounted monitoring. The above comparative tests refer to the use of motion capture equipment, DMS method (single frame) measurement, and DMS method (filtered) measurement to predict attitude angles and forces. Among them, the attitude angle measurement error corresponding to the motion capture equipment measurement is ±0.5°, but it cannot predict forces and is difficult to deploy in vehicles. The attitude angle measurement error corresponding to the DMS method (single frame) measurement is ±3.5°, and the force error is <10%, which can be deployed in vehicles. The attitude angle measurement error corresponding to the DMS method (filtered) measurement is ±2.8°, and the force error is <8%, which can also be deployed in vehicles.

[0171] Figure 3 This is a schematic diagram illustrating the principle of spinal posture angle estimation according to an embodiment of this application, as shown below. Figure 3 As shown, the perspective-n-point algorithm includes: determining camera intrinsic parameters; determining world coordinates (standard 3D face model); determining image coordinates (2D pixel position); estimating head center depth: reference depth and current facial depth value. When the driver leans forward, the current facial depth value is less than the reference depth, and α is calculated using formula (2). When the driver leans backward, the current facial depth value is greater than the reference depth, and α is calculated using formula (3).

[0172] In summary, this application simplifies the complex biomechanical simulation process into a single forward computation of a mathematical model by constructing a lightweight mapping model. Testing shows that a single prediction takes less than 1ms, representing a 3-4 order of magnitude improvement compared to the computation time of traditional simulation software, which can take tens of minutes to hours. This also marks the first time that real-time onboard monitoring of spinal stress has been achieved. Furthermore, this application is the first to use spinal posture angles as core input variables, establishing an explicit mapping relationship from (height, weight, posture angles) to (intervertebral disc compressive force, shear force). Further, this application compresses the model using techniques such as feature selection, model pruning, and knowledge distillation, ultimately controlling the model size to within a certain range. Within its scope, it can be directly programmed into embedded devices such as vehicle controllers and seat control units without relying on a high-performance computing platform, demonstrating good engineering practicality. Finally, by using the spinal posture angle as an input variable, this application can respond in real time to changes in driver posture (such as adjusting the seat, leaning forward, leaning back, etc.) and dynamically output the corresponding spinal force values, providing a technical basis for monitoring the cumulative spinal load during long-term driving.

[0173] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0174] According to an embodiment of this application, a system embodiment of a vehicle control system is provided. It should be noted that this system can be used to execute the above-described vehicle control method.

[0175] According to another aspect of the embodiments of this application, a vehicle control system is also provided. Figure 4 This is a structural block diagram of a vehicle control system according to an embodiment of this application, such as... Figure 4 As shown, the vehicle control system 400 includes: an acquisition module 401 for acquiring the driver's height, weight, and facial data, wherein the facial data is acquired through a data acquisition device in the vehicle's cockpit; a first determination module 402 for determining the driver's spinal posture angle based on the height and facial data; a prediction module 403 for inputting the height, weight, and spinal posture angle into a pre-trained target spinal force prediction model to obtain a prediction result, wherein the target spinal force prediction model is a regression mapping model trained based on biomechanical simulation data; a second determination module 404 for determining the vehicle's driver health intervention strategy based on the prediction result and spinal biomechanical load constraints; and a generation module 405 for generating vehicle control commands based on the driver health intervention strategy.

[0176] Optionally, the facial data includes: initial facial data and current facial data. The first determining module 402 is further configured to: determine a reference facial depth based on the initial facial data, wherein the initial facial data is used to characterize the driver's facial data when maintaining an upright sitting posture in the driver's seat, and the reference facial depth is used to characterize the reference distance of the driver's face relative to the data acquisition device; determine a current facial depth based on the current facial data; and determine a spinal posture angle based on the reference facial depth, the current facial depth, and the height.

[0177] Optionally, the first determining module 402 is further configured to: determine the depth change based on the reference facial depth and the current facial depth; determine the target length based on the height, the target length being used to characterize the distance from the driver's hip joint rotation center to the eye point; and determine the spinal posture angle based on the depth change and the target length.

[0178] Optionally, the driver health intervention strategy includes a seat adjustment strategy and a fatigue warning strategy. The second determining module 404 is further configured to: determine the seat adjustment strategy based on the prediction results and spinal biomechanical load constraints; or, determine the fatigue warning strategy based on the prediction results and spinal biomechanical load constraints. The generating module 405 is further configured to: generate a driver seat adjustment command based on the seat adjustment strategy; or, generate a driver fatigue warning command based on the fatigue warning strategy.

[0179] Optionally, the vehicle control system 400 also includes a training module (not shown in the figure). The training module is used to: acquire multiple sets of sample data corresponding to multiple drivers, wherein any set of sample data includes: driver height, driver weight, and driver spinal posture angle; analyze the multiple sets of sample data to construct multiple skeletal muscle models corresponding to multiple drivers; set simulation conditions for the multiple skeletal muscle models, wherein the simulation conditions are used to describe the force application of the multiple skeletal muscle models; under the simulation conditions, use biomechanical simulation software to perform force simulation calculations on the multiple skeletal muscle models respectively to obtain multiple simulation results; and construct biomechanical simulation data based on the multiple sets of sample data and multiple simulation results.

[0180] Optionally, the training module is also used to: train the initial regression mapping model based on biomechanical simulation data to obtain an initial spinal stress prediction model, wherein the initial spinal stress prediction model includes an initial first sub-network, an initial second sub-network, and an initial third sub-network. The initial first sub-network is used to predict the normal pressure along the spinal axis and perpendicular to the intervertebral disc plane, the initial second sub-network is used to predict the tangential force along the anterior-posterior direction of the spine and parallel to the intervertebral disc plane, and the initial third sub-network is used to predict the transverse tangential force along the lateral direction of the spine and parallel to the intervertebral disc plane; and to compress the initial spinal stress prediction model to obtain the target spinal stress prediction model.

[0181] Optionally, the training module is also used to: perform network pruning on the initial first subnetwork, the initial second subnetwork, and the initial third subnetwork respectively to obtain the pruned first subnetwork, the pruned second subnetwork, and the pruned third subnetwork; and perform knowledge distillation on the pruned first subnetwork, the pruned second subnetwork, and the pruned third subnetwork respectively to obtain the target spinal force prediction model.

[0182] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0183] Optionally, Figure 5 This is a schematic diagram of an electronic device according to an embodiment of this application, such as... Figure 5 As shown, the electronic device 500 may include a memory 510 and a processor 520, wherein the memory 510 is used to store an executable program; and the processor 520 is used to run the program stored in the memory 510, and the program executes the methods in various embodiments of this application when it runs.

[0184] Optionally, in this embodiment, the executable program performs the following steps when it runs:

[0185] Step S10: Obtain the driver's height, weight, and facial data, wherein the facial data is obtained through a data acquisition device inside the vehicle's cabin.

[0186] Step S11: Determine the driver's spinal posture angle based on height and facial data;

[0187] Step S12: Input the height, weight and spinal posture angle into the pre-trained target spinal force prediction model to obtain the prediction result. The target spinal force prediction model is a regression mapping model trained based on biomechanical simulation data.

[0188] Step S13: Based on the prediction results and spinal biomechanical load constraints, determine the driver's health intervention strategy for the vehicle.

[0189] Step S14: Generate vehicle control commands based on driver health intervention strategies.

[0190] Optionally, the facial data includes: initial facial data and current facial data. When the executable program is run, it performs the following steps: determining a reference facial depth based on the initial facial data, wherein the initial facial data is used to characterize the driver's facial data when maintaining an upright sitting posture in the driver's seat, and the reference facial depth is used to characterize the reference distance of the driver's face relative to the data acquisition device; determining a current facial depth based on the current facial data; and determining the spinal posture angle based on the reference facial depth, the current facial depth, and the height.

[0191] Optionally, the executable program performs the following steps when it runs: determining the depth change based on the reference facial depth and the current facial depth; determining the target length based on the height, the target length being used to characterize the distance from the driver's hip joint rotation center to the eye point; and determining the spinal posture angle based on the depth change and the target length.

[0192] Optionally, the driver health intervention strategy includes a seat adjustment strategy and a fatigue warning strategy. When the above executable program is running, it performs the following steps: determining the seat adjustment strategy based on the prediction results and spinal biomechanical load constraints; or, determining the fatigue warning strategy based on the prediction results and spinal biomechanical load constraints. When the above executable program is running, it performs the following steps: generating a driver seat adjustment command based on the seat adjustment strategy; or, generating a driver fatigue warning command based on the fatigue warning strategy.

[0193] Optionally, the executable program executes the following steps during runtime: acquiring multiple sets of sample data corresponding to multiple drivers, wherein any set of sample data includes: driver height, driver weight, and driver spinal posture angle; analyzing the multiple sets of sample data to construct multiple skeletal muscle models corresponding to multiple drivers; setting simulation conditions for the multiple skeletal muscle models, wherein the simulation conditions are used to describe the force application of the multiple skeletal muscle models; under the simulation conditions, using biomechanical simulation software to perform force simulation calculations on the multiple skeletal muscle models respectively, obtaining multiple simulation results; and constructing biomechanical simulation data based on the multiple sets of sample data and multiple simulation results.

[0194] Optionally, the executable program performs the following steps during runtime: Based on biomechanical simulation data, an initial regression mapping model is trained to obtain an initial spinal stress prediction model. This initial spinal stress prediction model includes an initial first sub-network, an initial second sub-network, and an initial third sub-network. The initial first sub-network is used to predict the normal pressure along the spinal axis and perpendicular to the intervertebral disc plane; the initial second sub-network is used to predict the tangential force along the anterior-posterior direction of the spine and parallel to the intervertebral disc plane; and the initial third sub-network is used to predict the transverse tangential force along the lateral direction of the spine and parallel to the intervertebral disc plane. The initial spinal stress prediction model is then compressed to obtain a target spinal stress prediction model.

[0195] Optionally, the executable program performs the following steps during runtime: performing network pruning on the initial first subnetwork, initial second subnetwork, and initial third subnetwork respectively to obtain pruned first subnetwork, pruned second subnetwork, and pruned third subnetwork; and performing knowledge distillation on the pruned first subnetwork, pruned second subnetwork, and pruned third subnetwork respectively to obtain the target spinal force prediction model.

[0196] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0197] Optionally, in this embodiment, the executable program can be configured to store an executable program for performing the following steps:

[0198] Step S10: Obtain the driver's height, weight, and facial data, wherein the facial data is obtained through a data acquisition device inside the vehicle's cabin.

[0199] Step S11: Determine the driver's spinal posture angle based on height and facial data;

[0200] Step S12: Input the height, weight and spinal posture angle into the pre-trained target spinal force prediction model to obtain the prediction result. The target spinal force prediction model is a regression mapping model trained based on biomechanical simulation data.

[0201] Step S13: Based on the prediction results and spinal biomechanical load constraints, determine the driver's health intervention strategy for the vehicle.

[0202] Step S14: Generate vehicle control commands based on driver health intervention strategies.

[0203] Optionally, the facial data includes: initial facial data and current facial data. The executable program can be configured to store an executable program for performing the following steps: determining a reference facial depth based on the initial facial data, wherein the initial facial data is used to characterize the driver's facial data when maintaining an upright sitting posture in the driver's seat, and the reference facial depth is used to characterize the reference distance of the driver's face relative to the data acquisition device; determining a current facial depth based on the current facial data; and determining a spinal posture angle based on the reference facial depth, the current facial depth, and the height.

[0204] Optionally, the executable program described above can be configured to store an executable program for performing the following steps: determining the depth change based on a reference facial depth and the current facial depth; determining a target length based on height, the target length being used to characterize the distance from the driver's hip joint rotation center to the eye point; and determining the spinal posture angle based on the depth change and the target length.

[0205] Optionally, the driver health intervention strategy includes a seat adjustment strategy and a fatigue warning strategy. The executable program can be configured to store executable programs for performing the following steps: determining a seat adjustment strategy based on prediction results and spinal biomechanical load constraints; or determining a fatigue warning strategy based on prediction results and spinal biomechanical load constraints. The executable program can also be configured to store executable programs for performing the following steps: generating a driver seat adjustment command based on the seat adjustment strategy; or generating a driver fatigue warning command based on the fatigue warning strategy.

[0206] Optionally, the executable program can be configured to store an executable program for performing the following steps: acquiring multiple sets of sample data corresponding to multiple drivers, wherein any set of sample data includes: driver height, driver weight, and driver spinal posture angle; analyzing the multiple sets of sample data to construct multiple skeletal muscle models corresponding to multiple drivers; setting simulation conditions for the multiple skeletal muscle models, wherein the simulation conditions are used to describe the force application of the multiple skeletal muscle models; under the simulation conditions, using biomechanical simulation software to perform force simulation calculations on the multiple skeletal muscle models respectively, obtaining multiple simulation results; and constructing biomechanical simulation data based on the multiple sets of sample data and multiple simulation results.

[0207] Optionally, the executable program can be configured to store an executable program for performing the following steps: training an initial regression mapping model based on biomechanical simulation data to obtain an initial spinal stress prediction model, wherein the initial spinal stress prediction model includes an initial first sub-network, an initial second sub-network, and an initial third sub-network. The initial first sub-network is used to predict the normal pressure along the spinal axis and perpendicular to the intervertebral disc plane, the initial second sub-network is used to predict the tangential force along the anterior-posterior direction of the spine and parallel to the intervertebral disc plane, and the initial third sub-network is used to predict the transverse tangential force along the lateral direction of the spine and parallel to the intervertebral disc plane; and model compression is performed on the initial spinal stress prediction model to obtain a target spinal stress prediction model.

[0208] Optionally, the executable program can be configured to store an executable program for performing the following steps: performing network pruning on the initial first subnetwork, the initial second subnetwork, and the initial third subnetwork respectively to obtain the pruned first subnetwork, the pruned second subnetwork, and the pruned third subnetwork; and performing knowledge distillation on the pruned first subnetwork, the pruned second subnetwork, and the pruned third subnetwork respectively to obtain a target spinal force prediction model.

[0209] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0210] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0211] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the methods in various embodiments of this application.

[0212] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods described in the various embodiments of this application.

[0213] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0214] In this application, "multiple" refers to two or more.

[0215] In this application, unless otherwise expressly defined, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0216] The terms “first,” “second,” “third,” “fourth,” etc., in this application (if present) are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0217] In this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, in this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0218] In the embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0219] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0220] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0221] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0222] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A vehicle control method, characterized in that, include: The driver's height, weight, and facial data are acquired, wherein the facial data is acquired through a data acquisition device inside the vehicle's cockpit. Based on the height and facial data, the driver's spinal posture angle is determined; The height, weight, and spinal posture angle are input into a pre-trained target spinal force prediction model to obtain the prediction result. The target spinal force prediction model is a regression mapping model trained based on biomechanical simulation data. Based on the prediction results and the spinal biomechanical load constraints, a health intervention strategy for the vehicle driver is determined. Based on the driver health intervention strategy, vehicle control commands are generated.

2. The vehicle control method according to claim 1, characterized in that, The facial data includes: initial facial data and current facial data. Based on the height and the facial data, the driver's spinal posture angle is determined, including: Based on the initial facial data, a reference facial depth is determined, wherein the initial facial data is used to characterize the driver's facial data when the driver is in an upright sitting position in the driver's seat, and the reference facial depth is used to characterize the reference distance of the driver's face relative to the data acquisition device; Based on the current facial data, determine the current facial depth; The spinal posture angle is determined based on the reference facial depth, the current facial depth, and the height.

3. The vehicle control method according to claim 2, characterized in that, Determining the spinal posture angle based on the reference facial depth, the current facial depth, and the height includes: The depth change is determined based on the reference facial depth and the current facial depth; Based on the height, a target length is determined, which is used to characterize the distance from the driver's hip joint rotation center to the eye point; The spinal posture angle is determined based on the depth change and the target length.

4. The vehicle control method according to claim 1, characterized in that, The driver health intervention strategy includes: a seat adjustment strategy and a fatigue warning strategy. Based on the prediction results and the spinal biomechanical load constraints, the driver health intervention strategy is determined, including: Based on the prediction results and the spinal biomechanical load constraints, determine the seat adjustment strategy; or, Based on the prediction results and the spinal biomechanical load constraints, the fatigue early warning strategy is determined. Based on the driver health intervention strategy, the vehicle control commands are generated, including: Based on the aforementioned seat adjustment strategy, generate a driver's seat adjustment command; or... Based on the fatigue warning strategy, a driver fatigue warning instruction is generated.

5. The vehicle control method according to claim 1, characterized in that, The vehicle control method further includes: Obtain multiple sets of sample data corresponding to multiple drivers, wherein any one set of sample data includes: driver height, driver weight, and driver spinal posture angle; The multiple sets of sample data are analyzed to construct multiple skeletal muscle models corresponding to the multiple drivers; Simulation conditions are set for the plurality of skeletal muscle models, wherein the simulation conditions are used to describe the force application of the plurality of skeletal muscle models; Under the simulation conditions, the stress simulation calculations were performed on the multiple skeletal muscle models using biomechanical simulation software, and multiple simulation results were obtained. Based on the multiple sets of sample data and the multiple simulation results, the biomechanical simulation data is constructed.

6. The vehicle control method according to claim 5, characterized in that, The vehicle control method further includes: Based on the biomechanical simulation data, the initial regression mapping model is trained to obtain an initial spinal force prediction model. The initial spinal force prediction model includes an initial first sub-network, an initial second sub-network, and an initial third sub-network. The initial first sub-network is used to predict the normal pressure along the spinal axis and perpendicular to the intervertebral disc plane. The initial second sub-network is used to predict the tangential force along the anterior-posterior direction of the spine and parallel to the intervertebral disc plane. The initial third sub-network is used to predict the transverse tangential force along the lateral direction of the spine and parallel to the intervertebral disc plane. The initial spinal force prediction model is compressed to obtain the target spinal force prediction model.

7. The vehicle control method according to claim 6, characterized in that, The initial spinal stress prediction model is compressed to obtain the target spinal stress prediction model, including: The initial first sub-network, the initial second sub-network, and the initial third sub-network are pruned respectively to obtain the pruned first sub-network, the pruned second sub-network, and the pruned third sub-network; Knowledge distillation is performed on the first pruned subnetwork, the second pruned subnetwork, and the third pruned subnetwork respectively to obtain the target spinal force prediction model.

8. A vehicle control system, characterized in that, include: The acquisition module is used to acquire the driver's height, weight, and facial data, wherein the facial data is acquired through a data acquisition device in the vehicle's cockpit. The first determining module is used to determine the driver's spinal posture angle based on the height and the facial data; The prediction module is used to input the height, weight and spinal posture angle into a pre-trained target spinal force prediction model to obtain the prediction result, wherein the target spinal force prediction model is a regression mapping model trained based on biomechanical simulation data. The second determining module is used to determine the driver's health intervention strategy based on the prediction results and the spinal biomechanical load constraints. The generation module is used to generate vehicle control commands based on the driver health intervention strategy.

9. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the executable program, wherein the executable program, when running, performs the vehicle control method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the computer-readable storage medium is located to perform the vehicle control method according to any one of claims 1 to 7.

Images