Integrated identification method and monitoring platform for dangerous driving behavior of road transport vehicle

Abstract

This invention discloses an integrated identification method and monitoring platform for dangerous driving behaviors of road transport vehicles, relating to the field of vehicle safety alarm system technology. It addresses the problems of existing alarm systems, which are prone to false alarms and have delayed alarms due to their reliance on single thresholds. First, by analyzing the frequency domain characteristics of line of sight and steering torque, the phase lag of the driver's intended action and the low-frequency coupling coefficient are extracted. Then, a dynamic state space matrix is ​​constructed by combining the vehicle's underlying mechanical response lag, and a cognitive-mechanical decoupling index is extracted to quantify the trend of human-machine collaborative loss of control. This index is then fused with the forward continuous collision time margin to construct an alarm modulation transfer function, outputting a time-varying risk warning gradient. Finally, the warning gradient is converted into a trigger signal envelope, triggering multimodal audio-visual and tactile vibration graded alarms and remote telemetry based on high and low frequency components, thus constructing an interference-resistant active warning closed loop, greatly improving driving safety.

Description

Technical Field

[0001] This invention relates to the field of vehicle safety alarm system technology, specifically to an integrated identification method and monitoring platform for dangerous driving behavior of road transport vehicles. Background Technology

[0002] With the rapid development of modern transportation, road transport vehicles play a vital role in logistics operations, and their operational safety is directly related to the safety of public life and property. Due to the typically long distances and complex road conditions of road transport, drivers are highly susceptible to fatigue, distraction, and other physiological and psychological factors during long drives, leading to dangerous driving behaviors such as visual deviation and slow operation. These dangerous driving behaviors are a core contributing factor to major traffic accidents and pose a serious threat to vehicle operational safety. Therefore, real-time monitoring of dangerous driving behaviors of road transport vehicles and timely issuance of effective warnings have become the primary task in the field of proactive safety monitoring and early warning systems.

[0003] Existing vehicle-mounted dangerous driving warning systems typically trigger alarms based on a single-dimensional threshold judgment mechanism. For example, they may rely solely on facial images to extract the duration of eye closure or solely on vehicle trajectory deviation to trigger a buzzer. In complex and ever-changing real-world transportation scenarios, such a single judgment logic is highly susceptible to interference from sudden changes in lighting, drivers wearing obstructions, or worn road markings, leading to frequent false alarms or missed alarms. Furthermore, existing alarm strategies often employ static and uniform triggering logic, severing the inherent connection between the driver's physiological cognitive delay and the mechanical physical lag of the vehicle chassis. They also fail to consider the dynamic collision risk margin between the vehicle and the external environment. This rigid recognition and alarm mechanism results in a severe disconnect between the timing of the warning signal triggering and the actual evolution of the danger. This not only easily causes unnecessary fright and interference to drivers in safe road sections but also leads to missed opportunities for correction and rescue in high-risk emergency situations due to delayed alarm actions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an integrated method and monitoring platform for identifying dangerous driving behaviors of road transport vehicles, thus solving the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an integrated method for identifying dangerous driving behaviors of road transport vehicles, comprising the following steps: S1, acquiring the spatial coordinate sequence of the driver's gaze point, the time series sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis; estimating the cross-power spectral density of the spatial coordinate sequence of the gaze point and the time series sequence of the steering wheel fine-tuning torque, and extracting the intentional action phase hysteresis and low-frequency coupling coefficient; S2, constructing an autoregressive moving average model between the steering wheel fine-tuning torque time series sequence and the vehicle chassis yaw rate sequence, analyzing the principal poles of the autoregressive moving average model to extract the vehicle mechanical response hysteresis; constructing the intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis into a dynamic state space matrix and performing eigenvalue decomposition, extracting the maximum Lyapunov exponent as the recognition... S3. Collect the relative distance and relative speed sequences between the vehicle and the target obstacle ahead. Based on the time decay characteristics of the relative distance and relative speed sequences, and combined with the vehicle's maximum longitudinal braking deceleration, fit the longitudinal collision avoidance safety boundary to obtain the continuous collision time margin. Construct an alarm modulation transfer function using the cognitive-mechanical decoupling index as a state variable and the continuous collision time margin as a boundary constraint, and solve for the output time-varying risk warning gradient. S4. Convert the time-varying risk warning gradient into an alarm trigger signal envelope according to a nonlinear mapping rule and input it into the vehicle's multi-channel alarm controller. By analyzing the amplitude of the high and low frequency components of the alarm trigger signal envelope, trigger the multimodal audio-visual alarm device and the vehicle's seat tactile vibration device respectively. Simultaneously, encode the cognitive-mechanical decoupling index and the time-varying risk warning gradient into an alarm telemetry data packet and send it to the remote monitoring center.

[0006] Further, the specific process of obtaining the driver's gaze point spatial coordinate sequence, the steering wheel fine-tuning torque timing sequence, and the vehicle chassis yaw rate sequence is as follows: The driver's facial feature image is captured by the cockpit infrared binocular vision sensor. The facial feature image is transformed into a spatial coordinate system to extract the three-dimensional coordinates of the gaze point. The image is then resampled at a fixed sampling frequency to generate the gaze point spatial coordinate sequence. Initial torque electrical signals are collected by the steering column torque sensor. These signals are then denoised and filtered to extract effective steering intervention signals, generating the steering wheel fine-tuning torque timing sequence. Six-degree-of-freedom motion data of the chassis is collected by the onboard inertial measurement unit. The longitudinal and roll components are removed, and the yaw rate data is extracted to generate the vehicle chassis yaw rate sequence. The gaze point spatial coordinate sequence, the steering wheel fine-tuning torque timing sequence, and the vehicle chassis yaw rate sequence are then time-domain aligned using a hardware clock hard synchronization mechanism.

[0007] Furthermore, the specific process of estimating the cross-power spectral density of the gaze point spatial coordinate sequence and the steering wheel fine-tuning torque time sequence, and extracting the intentional action phase hysteresis and low-frequency coupling coefficient is as follows: A sliding time window is applied to the time-domain aligned gaze point spatial coordinate sequence and steering wheel fine-tuning torque time sequence to truncate the data, eliminating spectral leakage at the sequence edges. The truncated gaze point spatial coordinate sequence and steering wheel fine-tuning torque time sequence are then transformed to the frequency domain, and their cross-period graph is analyzed to extract the cross-power spectral density function. The real and imaginary parts of the cross-power spectral density function are decomposed. Within the preset low-frequency band of human-machine interaction, the principal values ​​of the arctangent of the real and imaginary components are extracted as the intentional action phase hysteresis. Based on the respective power spectral density functions of the gaze point spatial coordinate sequence and the steering wheel fine-tuning torque time sequence, the cross-power spectral density function is analyzed using spectral coherence normalization to extract the low-frequency coupling coefficient.

[0008] Furthermore, an autoregressive moving average model is constructed between the steering wheel fine-tuning torque time series and the vehicle chassis yaw rate sequence. The specific process of extracting the vehicle mechanical response hysteresis from the principal poles of the autoregressive moving average model is as follows: The steering wheel fine-tuning torque time series is used as the forward excitation input, and the vehicle chassis yaw rate sequence is used as the state feedback output to establish a discrete difference equation. The augmented least squares algorithm is used to identify the parameters of the discrete difference equation, determine the autoregression order and the moving average order to generate the autoregressive moving average model; the zero-pole distribution of the autoregressive moving average model is mapped, and the eigenvalue closest to the unit circle in the complex plane is located as the principal pole of the model. The decay time constant corresponding to the principal pole of the model is extracted, and the decay time constant is transformed according to the sampling period of the chassis yaw rate sequence to obtain the vehicle mechanical response hysteresis.

[0009] Furthermore, the specific process of constructing a dynamic state space matrix by integrating the intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis, and performing eigenvalue decomposition to extract the maximum Lyapunov exponent as the cognitive-mechanical decoupling exponent is as follows: The intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis are embedded as independent state variables into a multi-dimensional state observation space to reconstruct the human-machine interaction state evolution trajectory and construct a dynamic state space matrix; the Jacobian matrix sequence of the dynamic state space matrix is ​​deduced along the evolution direction of the human-machine interaction state evolution trajectory; the Jacobian matrix sequence is orthogonally decomposed and updated; the exponential divergence rate of neighboring trajectories within the observation space is tracked to obtain the Lyapunov exponent spectrum of the dynamic state space matrix; the scalar extremum characterizing the maximum divergence trend of human-machine collaborative instability is extracted from the Lyapunov exponent spectrum of the dynamic state space matrix as the cognitive-mechanical decoupling exponent.

[0010] Furthermore, the specific process of acquiring the relative distance and relative velocity sequences between the vehicle and the target obstacle ahead, and fitting the longitudinal collision avoidance safety boundary based on the temporal decay characteristics of the relative distance and relative velocity sequences, combined with the vehicle's maximum longitudinal braking deceleration, to obtain the continuous collision time margin is as follows: The target point cloud data acquired by the vehicle-mounted millimeter-wave radar and the visual feature points acquired by the vehicle-mounted forward-facing camera are fused together to extract the motion vector of the target obstacle ahead and calculate the relative distance and relative velocity sequences between the vehicle and the target obstacle ahead; First-order difference mapping is performed on the relative distance sequence using the relative velocity sequence to extract the temporal decay characteristics of the relative distance and relative velocity sequences; the vehicle's maximum longitudinal braking deceleration and the vehicle's mechanical response hysteresis are input into the kinematic prediction model, and the longitudinal collision avoidance safety boundary is projected to generate the longitudinal collision avoidance safety boundary; By solving the intersection of the relative distance sequence and the longitudinal collision avoidance safety boundary in the spatiotemporal dimension, the remaining time length before the vehicle reaches the collision critical state is extracted as the continuous collision time margin.

[0011] Furthermore, the specific process of constructing an alarm modulation transfer function using the cognitive electromechanical decoupling index as a state variable and the continuous collision time margin as a boundary constraint, and solving the output time-varying risk warning gradient is as follows: The cognitive electromechanical decoupling index, as a state variable, is converted into a warning dynamic response coefficient through a nonlinear gain function; the continuous collision time margin is mapped to a time-varying risk probability distribution field, and the boundary threshold constraint for the evolution of dangerous states is defined; the warning dynamic response coefficient is used as the numerator and the continuous collision time margin is used as the denominator to construct the alarm modulation transfer function; the amplitude-frequency response characteristics of the alarm modulation transfer function in the complex frequency domain are analyzed; the directional derivative is calculated on the output envelope of the alarm modulation transfer function to extract the time-varying risk warning gradient.

[0012] Furthermore, the time-varying risk warning gradient is converted into an alarm trigger signal envelope according to a nonlinear mapping rule and input into the vehicle multi-channel alarm controller. The specific process of triggering the multimodal audio-visual alarm device and the vehicle seat tactile vibration device by analyzing the amplitude of the high and low frequency components of the alarm trigger signal envelope is as follows: The time-varying risk warning gradient is mapped to a preset oscillation waveform generator to generate an alarm trigger signal envelope whose amplitude changes with the gradient. The alarm trigger signal envelope is then spectrum-stripped using a bandpass filter with specific frequency response characteristics to separate the high-frequency and low-frequency characteristic components. The high-frequency characteristic components are encapsulated into audio-visual driving commands and sent to the multimodal audio-visual alarm device, while the low-frequency characteristic components are encapsulated into tactile sequence pulse commands and sent to the vehicle seat tactile vibration device.

[0013] Furthermore, the specific process of simultaneously encoding the cognitive electromechanical decoupling index and the time-varying risk warning gradient into an alarm telemetry data packet and sending it to the remote monitoring center is as follows: The cognitive electromechanical decoupling index, the time-varying risk warning gradient, and the current global positioning coordinates are asynchronously aligned using multi-source data. The aligned data is serialized into a bitstream using a preset industrial communication protocol, and a vehicle identification code and alarm level identifier are inserted. Error detection is performed on the serialized data using a cyclic redundancy check algorithm. The data packet is then encapsulated to generate an alarm telemetry data packet. The priority scheduling algorithm of the vehicle wireless communication link is called to load the alarm telemetry data packet into a high-priority transmission queue and upload it to the remote monitoring center.

[0014] An integrated monitoring platform for dangerous driving behavior of road transport vehicles is used to execute the aforementioned integrated identification method for dangerous driving behavior of road transport vehicles. It includes: a synchronous acquisition module, used to capture the spatial coordinate sequence of the driver's gaze point, the timing sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis through an onboard sensor network, and to perform time-domain alignment processing through a hardware clock hard synchronization mechanism; a state calculation module, used to estimate the cross-power spectral density and identify parameters of the time-domain aligned sequences, extract the intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis, and extract the cognitive-mechanical decoupling index through state-space feature decomposition; a warning generation module, used to fuse onboard radar and visual data to fit the longitudinal collision avoidance safety boundary to obtain the continuous collision time margin, and input the cognitive-mechanical decoupling index and the continuous collision time margin into the alarm modulation transfer function to solve and generate a time-varying risk warning gradient; and an alarm telemetry module, used to separate the high and low frequency components of the signal envelope corresponding to the time-varying risk warning gradient through a filter, independently triggering the multimodal audio-visual alarm device and the onboard seat tactile vibration device, and simultaneously encapsulating the warning data into an alarm telemetry data packet and uploading it to the remote monitoring center.

[0015] The present invention has the following beneficial effects:

[0016] (1) An integrated identification method for dangerous driving behavior of road transport vehicles, by acquiring the spatial coordinate sequence of gaze point, the time sequence of steering wheel fine-tuning torque, and the yaw rate sequence of vehicle chassis, and performing cross-power spectral density estimation and autoregressive moving average model analysis, successfully establishes a deep correlation and deconstruction between the driver's physiological cognitive intention and the physical mechanical response of the vehicle chassis in the spatiotemporal domain. This process effectively extracts the phase hysteresis of the intentional action, the low-frequency coupling coefficient, and the hysteresis of the vehicle's mechanical response, and then constructs a dynamic state space matrix and extracts the maximum Lyapunov exponent as the cognitive-mechanical decoupling exponent. It accurately quantifies the actual danger of human-machine interaction disconnection from the perspective of underlying system dynamics, and overcomes the technical problem that traditional alarm systems are easily misjudged due to external interference by relying on a single superficial feature.

[0017] (2) An integrated monitoring platform for dangerous driving behavior of road transport vehicles fits the longitudinal collision avoidance safety boundary by introducing the temporal decay characteristics of the relative distance and relative speed between the vehicle and the target obstacle ahead. The above-mentioned cognitive electromechanical decoupling index and continuous collision time margin are integrated to construct an alarm modulation transfer function and output a time-varying risk warning gradient. On this basis, the time-varying risk warning gradient is converted into an alarm trigger signal envelope. By analyzing the amplitude of high and low frequency components, multimodal audio-visual alarm devices and vehicle seat tactile vibration devices are triggered respectively, and alarm telemetry data packets are uploaded simultaneously. This mechanism realizes dynamic hierarchical warning response under the game of internal and external risks. It performs gentle tactile awakening in low-risk conditions and implements strong audio-visual intervention in high-risk conditions, effectively solving the technical pain points of frequent false interference caused by static fixed thresholds in existing alarm systems and alarm lag in emergency situations.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] Figure 1 This is a flowchart of the integrated identification method for dangerous driving behavior of road transport vehicles according to the present invention.

[0020] Figure 2 A schematic diagram of the electromechanical response timing and mechanical hysteresis of a human-machine vehicle.

[0021] Figure 3 This is a schematic diagram illustrating the divergence of the spatial trajectory of cognitive electromechanical states.

[0022] Figure 4 This is a schematic diagram illustrating the nonlinear relationship between collision time margin and risk warning gradient.

[0023] Figure 5 This is a flowchart of the integrated monitoring platform for dangerous driving behavior of road transport vehicles according to the present invention. Detailed Implementation

[0024] This application's embodiments solve the problems of frequent false alarms caused by single-modal judgment in existing vehicle alarm systems and the disconnect between alarm triggering timing and actual danger evolution by using an integrated identification method and monitoring platform for dangerous driving behavior of road transport vehicles.

[0025] The overall approach of the scheme in this application embodiment is as follows: First, frequency domain cross-analysis is performed on the driver's gaze wandering state and steering wheel fine-tuning actions to extract the primary intention hysteresis features of human-machine interaction; second, a system model between steering wheel input and vehicle chassis yaw output is established, the mechanical response hysteresis at the vehicle's bottom layer is analyzed and extracted, and the above-mentioned multi-dimensional hysteresis features are projected into the dynamic state space to calculate the cognitive-mechanical decoupling index characterizing the trend of loss of control in human-vehicle collaboration; then, the forward dynamic collision avoidance time margin outside the vehicle and the cognitive-mechanical decoupling index inside are integrated to construct an alarm modulation model with boundary constraints to generate a time-varying risk warning gradient that changes dynamically with risk; finally, high and low frequency split control is performed on the in-vehicle alarm actuator according to the frequency component characteristics of the warning gradient to drive different modal hierarchical alarm devices respectively, and remote data encoding and telemetry uploading of the warning state are completed.

[0026] Please see Figure 1 This invention provides a technical solution: an integrated method for identifying dangerous driving behavior of road transport vehicles, comprising the following steps: S1, acquiring the spatial coordinate sequence of the driver's gaze point, the time sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis; estimating the cross-power spectral density of the spatial coordinate sequence of the gaze point and the time sequence of the steering wheel fine-tuning torque, and extracting the intentional action phase hysteresis and low-frequency coupling coefficient; S2, constructing an autoregressive moving average model between the steering wheel fine-tuning torque time sequence and the vehicle chassis yaw rate sequence, analyzing the principal poles of the autoregressive moving average model to extract the vehicle mechanical response hysteresis; constructing the intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis into a dynamic state space matrix and performing eigenvalue decomposition, extracting the maximum Lyapunov exponent as the cognitive machine electrolysis. Coupling index; S3, Collect the relative distance sequence and relative speed sequence between the vehicle and the target obstacle in front. Based on the time decay characteristics of the relative distance and relative speed sequences, and combined with the maximum longitudinal braking deceleration of the vehicle, fit the longitudinal collision avoidance safety boundary to obtain the continuous collision time margin; Construct the alarm modulation transfer function with the cognitive electromechanical decoupling index as the state variable and the continuous collision time margin as the boundary constraint, and solve the output time-varying risk warning gradient; S4, Convert the time-varying risk warning gradient into an alarm trigger signal envelope according to the nonlinear mapping rule and input it into the vehicle multi-channel alarm controller. By analyzing the high and low frequency component amplitudes of the alarm trigger signal envelope, trigger the multimodal sound and light alarm device and the vehicle seat tactile vibration device respectively. Simultaneously encode the cognitive electromechanical decoupling index and the time-varying risk warning gradient into an alarm telemetry data packet and send it to the remote monitoring center.

[0027] In this implementation scheme, step S1 is used to acquire the driver's basic physiological action signals and analyze their cognitive reaction delay. Cross-power spectral density estimation refers to analyzing the frequency coherence and phase relationship between the driver's line-of-sight coordinate sequence and the steering wheel action sequence in the frequency domain. The intention action phase hysteresis represents the time difference between the driver's eyes observing the road conditions ahead and the hands actually performing fine-tuning of the steering wheel. The low-frequency coupling coefficient reflects the consistency between the driver's intention and actual physical operation at normal driving frequencies. Through this step, the system can accurately capture the phenomenon of cognitive perception and physical action disconnect that occurs in the early stages of driver's slight fatigue or distraction, thereby providing a deeper physiological behavior judgment basis for the generation of subsequent alarm signals.

[0028] Step S2 is used to calculate the mechanical and physical delay of the vehicle chassis and comprehensively evaluate the runaway trend of human-vehicle cooperative control. The autoregressive moving average model is a dynamic time series mathematical model that uses past steering input sequences to predict the current yaw output state of the vehicle. Analyzing the principal poles of this model aims to extract the objective mechanical delay time of the vehicle chassis overcoming its own inertia and generating an actual steering response. The maximum Lyapunov exponent is used in system dynamics to measure the rate at which the system state diverges over time. Here, it is extracted as the cognitive-mechanical decoupling exponent, which is used to quantify the degree of human-vehicle cooperative control failure caused by the superposition of the driver's physiological reaction delay and the vehicle's physical and mechanical delay. This step effectively overcomes the shortcomings of traditional alarm systems that rely only on single superficial features such as the driver closing their eyes and yawning, ensuring that the alarm assessment mechanism can truly reflect the overall dynamic instability risk of the human-vehicle system.

[0029] Step S3 is used to dynamically generate the source signal for alarm control by combining the risks of the vehicle's external environment. The continuous collision time margin refers to the limit safety reaction time remaining before a rear-end collision occurs, under the premise of comprehensively considering the maximum braking deceleration performance of the vehicle. The alarm modulation transfer function is a signal transformation relationship with the driver's internal loss of control state as the independent variable and the external collision time as the limiting boundary condition. The time-varying risk warning gradient represents the rate slope of the rapid deterioration of the current comprehensive danger situation. This step deeply binds the degree of danger inside the driver with the fault tolerance space of the external road environment, so that the alarm system can adaptively suppress invalid false alarms in open and safe road sections, and can rapidly increase the intensity of the warning signal when the following distance is extremely compressed.

[0030] Step S4 is used to perform multimodal hierarchical physical alarm interaction and remote status telemetry. The alarm trigger signal envelope refers to the outer contour of the comprehensive electrical signal waveform containing risk gradient information. The system separates the high and low frequencies of the signal envelope. The low-frequency component represents latent risks and is used to drive the seat internal device to perform directional tactile vibration to achieve subconscious physical awakening of the driver. The high-frequency component represents emergency danger and is used to directly activate the high-decibel sound and bright flashing light device in the cockpit for forced visual and auditory intervention. This step realizes a closed loop from signal evaluation to terminal physical alarm. The hierarchical multimodal alarm method effectively avoids the shock and interference caused to the driver by a single abrupt alarm. At the same time, the quantified risk status data is packaged and transmitted back to the remote monitoring center in real time, which meets the fleet's background dynamic supervision needs for dangerous driving behavior of transport vehicles.

[0031] Specifically, the process of obtaining the spatial coordinate sequence of the driver's gaze point, the timing sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis is as follows: The driver's facial feature image is captured by an infrared binocular vision sensor in the cockpit. The facial feature image is transformed into a spatial coordinate system to extract the three-dimensional coordinates of the gaze point. The gaze point spatial coordinate sequence is generated by resampling at a fixed sampling frequency. Initial torque electrical signals are collected by a steering column torque sensor. The initial torque electrical signals are denoised and filtered to extract effective steering intervention signals, generating the steering wheel fine-tuning torque timing sequence. Six-degree-of-freedom motion data of the chassis is collected by an onboard inertial measurement unit. The longitudinal and roll components are removed, and the yaw rate data is extracted to generate the vehicle chassis yaw rate sequence. The gaze point spatial coordinate sequence, the steering wheel fine-tuning torque timing sequence, and the vehicle chassis yaw rate sequence are time-domain aligned using a hardware clock hard synchronization mechanism.

[0032] In this implementation scheme, when capturing facial features using the cockpit infrared binocular vision sensor, the system utilizes an intrinsic and extrinsic parameter calibration matrix to perform spatial depth mapping and three-dimensional reconstruction of two-dimensional facial pixel feature points. The aim is to eliminate recognition biases caused by the driver's natural head rotation and changes in cabin lighting, thereby obtaining the three-dimensional coordinates of the gaze point that accurately reflects the driver's visual perception focus. For the initial torque electrical signal collected by the steering column torque sensor, considering the high-frequency road vibrations of the commercial vehicle chassis and electromagnetic crosstalk from the vehicle's electrical system, an adaptive median filter combined with Gaussian smoothing denoising algorithm is used to remove high-frequency random glitches. This operation can completely preserve the low-frequency effective steering intervention signal that accurately reflects the driver's subjective fine-tuning intentions. In the onboard inertial measurement unit... In the data extraction stage, to prevent interference from the three-axis acceleration components generated by long downhill terrain or high-gravity side-tilt curves, the longitudinal and roll components in the six-degree-of-freedom motion data of the chassis are selectively stripped, and the pure yaw rate data reflecting the actual lateral offset trajectory of the vehicle is accurately extracted. Finally, the hardware clock hard synchronization mechanism is triggered by the same timing pulse of the underlying bus controller to forcibly unify the data output timestamps of the above three types of multi-source heterogeneous sensors, completely eliminating the physical time difference between data transmission and sampling cycles, and ensuring the absolute alignment of the spatial coordinate sequence of the gaze point, the timing sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis in the time domain dimension. This provides an unbiased underlying data foundation for the subsequent accurate calculation of the time delay between driver cognition and vehicle response.

[0033] Specifically, the process of estimating the cross-power spectral density of the gaze point spatial coordinate sequence and the steering wheel fine-tuning torque time sequence, and extracting the intention action phase hysteresis and low-frequency coupling coefficient is as follows: A sliding time window is applied to the time-domain aligned gaze point spatial coordinate sequence and steering wheel fine-tuning torque time sequence to truncate the data, eliminating spectral leakage at the sequence edges. The truncated gaze point spatial coordinate sequence and steering wheel fine-tuning torque time sequence are then transformed to the frequency domain, and their cross-period graph is analyzed to extract the cross-power spectral density function. The real and imaginary parts of the cross-power spectral density function are decomposed. Within the preset low-frequency band of human-machine interaction, the principal values ​​of the arctangent of the real and imaginary components are extracted as the intention action phase hysteresis. Based on the respective power spectral density functions of the gaze point spatial coordinate sequence and the steering wheel fine-tuning torque time sequence, the cross-power spectral density function is analyzed using spectral coherence normalization to extract the low-frequency coupling coefficient.

[0034] In this implementation, a sliding time window is applied before frequency domain conversion to overlap and truncate the time-domain aligned sequence. This smooths the signal edge amplitudes and suppresses spectral leakage caused by non-periodic truncation during Fourier transform, thereby ensuring the energy purity of the output frequency domain characteristic components. After converting the truncated line-of-sight and torque sequences to the frequency domain, the cross-power spectral density function is analyzed by calculating their corresponding cross-period diagrams. This step aims to quantify the dynamic interaction between the driver's perceived line of sight and hand correction actions at different frequency components. To further explore the driver's physiological reaction delay, the system performs complex domain decomposition on the cross-power spectral density function to extract its real and imaginary solutions. Based on a preset arctangent integral model in the low-frequency band of human-computer interaction, the phase hysteresis of the intended action is calculated. The specific calculation formula is as follows: In the formula, This indicates the phase hysteresis of the intended action; This indicates the upper frequency limit of the preset low-frequency band for human-computer interaction; The lower limit of the preset low-frequency band for human-computer interaction is indicated by the frequency upper and lower limits. The frequency upper and lower limits are obtained by calibrating the maximum likelihood estimation method based on the concentrated distribution range of the spectrum energy of a large number of pre-collected visual and torque-related operations under normal driving conditions without fatigue. This indicates the angular frequency at the current integration point; The cross power spectral density function at angular frequency is The imaginary component at the location; The cross power spectral density function at angular frequency is The real component at the given point, by obtaining the principal value of the arctangent and equalizing the integral within the low-frequency band that reflects the regular human-computer interaction patterns, can objectively and stably reflect the intrinsic neural transmission time difference from the driver's visual acquisition of road condition deviations to the actual execution of steering correction by the hand muscles. Furthermore, to assess the synchronization and coupling tightness of the line of sight and steering action in the energy distribution spectrum, the cross-power spectral density function is analyzed using spectral coherence normalization based on its respective self-power spectral density function to extract the low-frequency coupling coefficient. The specific formula is as follows: In the formula, Indicates the low-frequency coupling coefficient; The cross-power spectral density function representing the spatial coordinate sequence of the gaze point and the temporal sequence of steering wheel fine-tuning torque; The power spectral density function representing the spatial coordinate sequence of the gaze point; This represents the self-power spectral density function of the steering wheel fine-tuning torque time sequence. The analytical process normalizes the signal by using the ratio of the square of the cross-power spectral density function amplitude to the product of their self-power spectral density functions, cleverly eliminating interference from the absolute amplitude of a single signal. This precisely quantifies the degree of attenuation in the consistency between the driver's cognitive intention sequence and the actual mechanical control sequence under slight fatigue or distraction, thus providing core physical parameters for subsequently constructing a dynamic state space matrix characterizing the trend of instability and divergence in human-vehicle coordination.

[0035] Please see Figure 2 Specifically, the process of constructing an autoregressive moving average model between the steering wheel fine-tuning torque time series and the vehicle chassis yaw rate sequence, and extracting the vehicle mechanical response hysteresis from the principal poles of the autoregressive moving average model is as follows: A discrete difference equation is established using the steering wheel fine-tuning torque time series as the forward excitation input and the vehicle chassis yaw rate sequence as the state feedback output. The augmented least squares algorithm is used to identify the parameters of the discrete difference equation, determining the autoregressive order and the moving average order to generate the autoregressive moving average model. Zero-pole distribution mapping is performed on the autoregressive moving average model, and the eigenvalue closest to the unit circle in the complex plane is located as the principal pole of the model. The decay time constant corresponding to the principal pole is extracted, and the decay time constant is transformed according to the sampling period of the chassis yaw rate sequence to obtain the vehicle mechanical response hysteresis.

[0036] In this implementation scheme, a discrete difference equation is established using the steering wheel fine-tuning torque time sequence as the forward excitation input and the vehicle chassis yaw rate sequence as the state feedback output. This is to directly extract the mechanical inertial characteristics of the vehicle chassis itself from the input-output mapping of dynamic data, eliminate road condition interference, and use the augmented least squares algorithm to identify the parameters of the difference equation. This effectively overcomes the problem of biased parameter estimation caused by complex road noise, thereby accurately determining the autoregressive order and the moving average order and generating an autoregressive moving average model. Subsequently, the generated autoregressive moving average model is mapped to a zero-pole distribution, and the eigenvalue closest to the unit circle in the complex plane is located as the principal pole of the model. The principal pole represents the dynamic mode with the slowest decay and the dominant role in the transient response of vehicle dynamics. In order to transform the pole characteristics of this complex frequency domain into an intuitive time hysteresis index, the system extracts the decay time constant corresponding to the principal pole of the model and transforms it according to the sampling period of the chassis yaw rate sequence to obtain the vehicle mechanical response hysteresis. The specific calculation formula is as follows: In the formula, This indicates the amount of mechanical lag in the vehicle's response; This indicates the sampling period of the chassis yaw rate sequence; This indicates the autoregressive order of the autoregressive moving average model; m represents the index number of the model's poles. Let represent the real part of the m-th pole in the complex plane; Let m represent the imaginary part of the m-th pole in the complex plane; This indicates the maximum value operation, which involves iterating through all poles to find the maximum value of the modulus to lock the principal pole of the model closest to the unit circle; This represents the vehicle load inertia compensation weight coefficient. The method for determining this weight coefficient is as follows: The actual total load weight of the vehicle is obtained through the vehicle chassis air suspension pressure sensor. The difference between the actual total load weight and the vehicle's standard unloaded weight is calculated. This difference is divided by the standard unloaded weight to obtain the load increase ratio. Finally, this load increase ratio is input into a preset natural exponential growth function for mapping and calculation. By introducing weight compensation through this step, the alarm system can accurately grasp the physical hysteresis of the vehicle chassis to the driver's steering operation under different full-load or unloaded states. This provides a core decoupling parameter for subsequently distinguishing whether dangerous driving behavior stems from a slow driver reaction or a slow response due to heavy load.

[0037] Please see Figure 3 Specifically, the process of constructing a dynamic state space matrix by integrating the intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis, and performing eigenvalue decomposition to extract the maximum Lyapunov exponent as the cognitive-mechanical decoupling exponent is as follows: The intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis are embedded as independent state variables into a multi-dimensional state observation space to reconstruct the human-machine interaction state evolution trajectory and construct a dynamic state space matrix. The Jacobian matrix sequence of the dynamic state space matrix is ​​deduced along the evolution direction of the human-machine interaction state evolution trajectory. The Jacobian matrix sequence is then orthogonally decomposed and updated. The exponential divergence rate of neighboring trajectories within the observation space is tracked to obtain the Lyapunov exponent spectrum of the dynamic state space matrix. Finally, the scalar extremum characterizing the maximum divergence trend of human-machine collaborative instability is extracted from the Lyapunov exponent spectrum of the dynamic state space matrix as the cognitive-mechanical decoupling exponent.

[0038] In this implementation scheme, the intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis are embedded as independent state variables into the multi-dimensional state observation space. This is to overcome the limitations of the single-dimensional linear threshold judgment in traditional alarm systems. By reconstructing the human-machine interaction state evolution trajectory, which includes the driver's physiological hysteresis and the vehicle's physical hysteresis, a dynamic state space matrix representing the depth of human-vehicle coupling is established within the topological dynamic space. The Jacobian matrix sequence of the dynamic state space matrix is ​​derived along the evolution direction of the human-machine interaction state trajectory, which can accurately capture the local linear evolution law of the human-vehicle system state when faced with minor road disturbances or operational errors. Furthermore, the Jacobian matrix sequence is orthogonally decomposed and updated, continuously tracking the exponential divergence rate of neighboring trajectories within the state observation space, thereby obtaining the Lyapunov exponent spectrum of the dynamic state space matrix reflecting the global chaotic divergence characteristics of the system. Finally, the scalar extremum representing the maximum divergence trend of human-machine cooperative instability is extracted from the Lyapunov exponent spectrum as the cognitive-electromechanical decoupling index, the specific calculation formula of which is: In the formula, Indicates the cognitive-mechanical decoupling index; This represents the total number of tracking iterations for orthogonal decomposition and updating along the human-computer interaction state evolution trajectory; This represents the discrete time step of a single tracking evolution; n represents the index number of the evolution steps. It represents the maximum element value on the main diagonal of the upper triangular matrix obtained after orthogonal decomposition of the Jacobian matrix of the dynamic state space matrix at the nth evolution step. Its physical meaning is the maximum local divergence ratio of the system state trajectory at this evolution step. This represents the adaptive speed compensation coefficient. The method for determining the adaptive speed compensation coefficient is as follows: real-time acquisition of the actual driving speed of the vehicle, calculation of the difference between the actual driving speed and the system's preset benchmark safe cruise speed, nonlinear normalization of the difference through the hyperbolic tangent function, and addition of the basic bias constant. Through the above steps, the system can accurately characterize the degree of loss of control caused by the superposition of the driver's fatigue-induced sluggish control intention and the vehicle chassis response lag at the physical level, quantifying the latent period of extreme speed danger hidden under the appearance of normal driving into an intuitive underlying warning trigger basis.

[0039] Specifically, the process of acquiring the relative distance and relative velocity sequences between the vehicle and the target obstacle ahead, and fitting the longitudinal collision avoidance safety boundary based on the temporal decay characteristics of the relative distance and relative velocity sequences, combined with the vehicle's maximum longitudinal braking deceleration, to obtain the continuous collision time margin is as follows: The target point cloud data acquired by the vehicle-mounted millimeter-wave radar and the visual feature points acquired by the vehicle-mounted forward-facing camera are fused together to extract the motion vector of the target obstacle ahead and calculate the relative distance and relative velocity sequences between the vehicle and the target obstacle ahead; First-order difference mapping is performed on the relative distance sequence using the relative velocity sequence to extract the temporal decay characteristics of the relative distance and relative velocity sequences; the vehicle's maximum longitudinal braking deceleration and the vehicle's mechanical response hysteresis are input into the kinematic prediction model, and the longitudinal collision avoidance safety boundary is projected to generate the longitudinal collision avoidance safety boundary; By solving the intersection of the relative distance sequence and the longitudinal collision avoidance safety boundary in the spatiotemporal dimension, the remaining time length before the vehicle reaches the collision critical state is extracted as the continuous collision time margin.

[0040] In this implementation scheme, target point cloud data collected by vehicle-mounted millimeter-wave radar and visual feature points collected by vehicle-mounted forward-looking camera are integrated. This aims to overcome the limitations of a single radar lacking target semantic recognition capabilities and a single camera being susceptible to complex lighting conditions such as backlighting at night. Through multi-sensor physical pixel-level and low-level data-level front-end cross-fusion, the motion vector of the target obstacle in front is accurately extracted under a globally unified coordinate system. Then, the high-frequency and high-precision relative distance sequence and relative velocity sequence between the vehicle and the target obstacle in front are calculated. Using the relative velocity sequence to perform a first-order difference mapping on the relative distance sequence, the nonlinear reduction trend of the physical distance between the two vehicles within a very short time window can be dynamically captured, thereby effectively extracting the temporal decay characteristics of the relative distance sequence and the relative velocity sequence. Subsequently, the maximum longitudinal braking deceleration of the vehicle and the aforementioned extracted vehicle mechanical response hysteresis are input into the kinematic prediction model. A longitudinal collision avoidance safety boundary that dynamically changes with vehicle speed is generated by projecting along the longitudinal tangent direction of the vehicle's current driving trajectory. This boundary represents the limit of stopping required for road transport vehicles to take emergency physical braking under heavy load inertia. By solving for the intersection of the relative distance sequence and the longitudinal collision avoidance safety boundary in the spatiotemporal dimension, the critical state with a truly imminent rear-end collision threat can be identified. Then, the remaining time before the vehicle reaches the critical collision state can be extracted as the continuous collision time margin. The specific calculation formula is as follows: In the formula, Indicates the time margin for consecutive collisions; Indicates the relative distance between this vehicle and the target obstacle ahead; This indicates the current absolute longitudinal speed of the vehicle; This represents the vehicle mechanical response hysteresis extracted in the preceding steps. This indicates the maximum longitudinal braking deceleration output by the chassis brake chamber of this vehicle; This represents the dynamic weighting coefficient of road surface adhesion friction. This indicates the relative speed between the vehicle and the target obstacle ahead; This represents a small positive compensation constant to prevent calculation overflow. The method for determining the dynamic weight coefficient of the road surface adhesion friction is as follows: the instantaneous wheel speed of each wheel is obtained in real time through the vehicle chassis anti-lock braking control unit, the relative slip ratio between the instantaneous wheel speed and the absolute reference speed of the vehicle center is calculated, and the relative slip ratio and the rainfall level output by the current vehicle rain sensor are input into the preset nonlinear friction mapping surface. The value of the corresponding coordinate point is directly extracted as the friction weight of the current road using the surface fitting interpolation method. This calculation step does not use the standard braking distance under ideal conditions, but incorporates the mechanical lag of the driver's operation of the vehicle chassis and the depth of road friction attenuation in rainy and snowy weather into the model, so that the final output continuous collision time margin becomes the extreme fault tolerance boundary that closely matches the real complex transportation road conditions, laying an extremely reliable physical constraint foundation for the accurate triggering of subsequent anti-interference alarms.

[0041] Please see Figure 4 Specifically, the process of constructing an alarm modulation transfer function using the cognitive electromechanical decoupling index as a state variable and the continuous collision time margin as a boundary constraint, and solving for the output time-varying risk warning gradient is as follows: The cognitive electromechanical decoupling index, as a state variable, is converted into a warning dynamic response coefficient through a nonlinear gain function; the continuous collision time margin is mapped to a time-varying risk probability distribution field, and boundary threshold constraints for the evolution of dangerous states are defined; the warning dynamic response coefficient is used as the numerator and the continuous collision time margin is used as the denominator to construct the alarm modulation transfer function; the amplitude-frequency response characteristics of the alarm modulation transfer function in the complex frequency domain are analyzed; the directional derivative of the output envelope of the alarm modulation transfer function is calculated, and the time-varying risk warning gradient is extracted.

[0042] In this implementation scheme, the cognitive electromechanical decoupling index, which is a state variable, is converted into a warning dynamic response coefficient through a nonlinear gain function. The core purpose is to amplify and adjust the alarm sensitivity for the driver's hidden physiological danger state. When the decoupling index increases sharply, indicating that the driver is in severe fatigue or limb stiffness, the warning dynamic response coefficient will increase exponentially, thereby forcibly increasing the response priority of the alarm system at the algorithm level. Subsequently, the continuous collision time margin is mapped to the time-varying collision risk model, and the boundary threshold constraint of the evolution of the danger state is defined at the physical safety space level to ensure that when an external physical collision is imminent, the alarm logic can be forcibly intervened without being disturbed by fluctuations in the internal physiological state assessment. Furthermore, in order to comprehensively evaluate the superimposed excitation effect of internal and external dual risks in the frequency domain, the warning dynamic response coefficient is used as the numerator of the transfer function and the continuous collision time margin is used as the denominator of the transfer function to construct an alarm modulation transfer function, and the amplitude-frequency response characteristics of the transfer function are fully analyzed in the complex frequency domain. Finally, following the risk evolution direction in the multidimensional state space, the directional derivative of the alarm envelope output by the alarm modulation transfer function is calculated to accurately extract the time-varying risk warning gradient that reflects the slope of the rate of risk deterioration at this moment. The specific calculation formula is as follows: In the formula, Indicates the time-varying risk warning gradient; This represents the direction vector of risk state evolution within the multidimensional state observation space; Indicates the dynamic response coefficient for early warning; Indicates the time margin for consecutive collisions; This represents the cutoff angular frequency used in complex frequency domain analysis to filter out common road vibration interference. This refers to the alarm intensity amplification compensation weight coefficient. The method for determining the alarm intensity amplification compensation weight coefficient is as follows: The electronic freight manifest bound to the vehicle terminal of the current road transport vehicle is parsed through the vehicle communication gateway, and the dangerous goods classification label code of the transported goods is extracted. When the dangerous goods classification label code belongs to the catalog of flammable liquids, explosives, or highly toxic chemicals, the highest risk penalty multiple constant preset in the system memory is retrieved and assigned to the weight coefficient; otherwise, the default value is assigned to the basic unit value of one. By introducing the directional derivative algorithm and freight attribute weights, this step not only organically couples the driver's physiological sluggish reaction with the external physical collision distance mathematically, but also incorporates the high-risk attributes of hazardous chemicals unique to the road transport industry into the early warning gain system. It keenly captures the deterioration derivative of minor danger trends, completely solving the core pain points of traditional mechanical threshold alarm mechanisms in high-speed, heavy-load dangerous goods transport scenarios, such as delayed early warning and insufficient alarm intervention, fully conforming to and deepening the design intent of the safety alarm system.

[0043] Specifically, the process of converting the time-varying risk warning gradient into an alarm trigger signal envelope according to a nonlinear mapping rule and inputting it into the vehicle multi-channel alarm controller, and triggering the multimodal audio-visual alarm device and the vehicle seat tactile vibration device by analyzing the amplitude of the high and low frequency components of the alarm trigger signal envelope, is as follows: The time-varying risk warning gradient is mapped to a preset oscillation waveform generator to generate an alarm trigger signal envelope whose amplitude changes with the gradient. The alarm trigger signal envelope is then spectrum-stripped using a bandpass filter with specific frequency response characteristics to separate the high-frequency and low-frequency characteristic components. The high-frequency characteristic components are encapsulated into audio-visual driving commands and sent to the multimodal audio-visual alarm device, while the low-frequency characteristic components are encapsulated into tactile sequence pulse commands and sent to the vehicle seat tactile vibration device.

[0044] In this implementation scheme, the time-varying risk warning gradient is mapped to a preset oscillation waveform generator, aiming to transform purely algorithmic mathematical variables into a basic electrical signal waveform that can be directly recognized and driven by physical hardware. To ensure that the physical intensity and stimulus level of the alarm signal perfectly match the rate of risk deterioration, the system generates an alarm trigger signal envelope whose amplitude varies with the gradient. The specific calculation formula for this nonlinear envelope mapping is as follows: In the formula, Indicates the amplitude of the alarm trigger signal envelope; This represents the driving reference static voltage of the oscillation waveform generator; This represents the time-varying risk warning gradient obtained from the prior calculation; This represents the power of the risk gradient exponent; The lower cutoff angular frequency of a bandpass filter that represents its specific frequency response characteristics; This indicates the upper cutoff angular frequency of the bandpass filter; This represents the gain function of the bandpass filter's amplitude-frequency response in the frequency domain. Represents the integral frequency variable; This represents the nonlinear mapping weight coefficient. The method for determining the nonlinear mapping weight coefficient is as follows: The acoustic background noise level in decibels inside the cabin is collected in real time by an onboard array microphone deployed on the top of the vehicle cabin. The difference between this noise level and the preset standard silent state decibel value is calculated, and this difference is input into a pre-calibrated exponential compensation growth function. Through environmental noise compensation mapping calculation, it can be ensured that the generated composite alarm signal still has strong penetration in extremely noisy high-speed heavy-load transportation environments. Subsequently, a bandpass filter is used to precisely strip the spectrum of the signal envelope. The stripping operation completely separates the high-frequency and low-frequency characteristic components in the waveform. Due to its strong auditory penetration and visual flashing stimulation properties, the high-frequency component is directly encapsulated as an audio-visual driving command and sent to the multimodal audio-visual alarm device for the highest level of mandatory warning intervention. The low-frequency component, with its excellent solid medium directional penetration and musculoskeletal resonance characteristics, is encapsulated as a tactile sequence pulse command and sent to the vehicle seat tactile vibration device. It outputs directional low-frequency physical vibrations to the driver's torso through the seat back and cushion. This multimodal diversion triggering mechanism based on frequency stripping realizes the use of low-frequency vibrations to awaken the subconscious mind without fright in the early stage of danger, and the use of high-frequency audio-visual stimulation to forcibly block behavior at the edge of danger and loss of control, thus constructing a hierarchical and non-abrupt active safety closed-loop response.

[0045] Specifically, the process of synchronously encoding the cognitive electromechanical decoupling index and the time-varying risk warning gradient into an alarm telemetry data packet and sending it to the remote monitoring center is as follows: The cognitive electromechanical decoupling index, the time-varying risk warning gradient, and the current global positioning coordinates are asynchronously aligned using multi-source data. The aligned data is serialized into a bitstream using a preset industrial communication protocol, and a vehicle identification code and alarm level identifier are inserted. Error detection is performed on the serialized data using a cyclic redundancy check algorithm. The data packet is then encapsulated to generate an alarm telemetry data packet. The priority scheduling algorithm of the vehicle wireless communication link is called to load the alarm telemetry data packet into a high-priority transmission queue and upload it to the remote monitoring center.

[0046] In this implementation scheme, the cognitive electromechanical decoupling index, time-varying risk warning gradient, and current global positioning coordinates are asynchronously aligned using multi-source data. This is to overcome the slight misalignment of arrival timestamps caused by the asynchronous refresh rates of various heterogeneous sensors and underlying algorithms in the vehicle controller area network bus. This ensures that the hazard assessment parameters and physical geographical location have absolutely strict spatiotemporal attributes in multidimensional data representation. Subsequently, the aligned data is serialized into a bitstream using a preset industrial communication protocol, and vehicle identification codes and alarm level identifiers are inserted. This operation reconstructs non-standard local floating-point arithmetic values ​​into standard structured binary communication frames that can be stably transmitted over long distances across physical network segments. Before data packets are encapsulated, the serialized data is error-detected using a cyclic redundancy check algorithm. This detection mechanism automatically identifies and removes bit-flipping scrambled bits that may occur under strong electromagnetic radiation interference in the vehicle compartment by generating a high-order generator polynomial remainder, thereby encapsulating and generating alarm telemetry data packets with extremely high data fidelity and anti-tampering characteristics. During the remote upload phase, given that mobile cellular network channels are easily obstructed by terrain during travel, leading to bandwidth attenuation, the system invokes the priority scheduling algorithm of the vehicle-mounted wireless communication link to dynamically calculate the transmission priority of alarm telemetry data packets in the local buffer queue, and accordingly loads them into the high-priority transmission queue. The specific formula for calculating the scheduling priority is as follows: In the formula, This represents the priority weight coefficient for sending queue scheduling; Factors indicating the urgency level of the warning; Indicates the time-varying risk warning gradient; Indicates the cognitive-mechanical decoupling index; This indicates the available uplink network bandwidth of the vehicle-mounted wireless link at the current moment; Indicates the total length of the serialized bitstream; Indicates the real-time signal-to-noise ratio of the wireless channel; The above-mentioned warning urgency factor is determined by matching the actual road topology interval where the vehicle's global positioning coordinates are located through the vehicle navigation and positioning module, retrieving the risk level index of historical traffic accident hotspots in that road interval stored in the high-precision dynamic map server, and inputting the risk level index into the built-in ladder transition addressing mapping table for comparison. By introducing real road segment risk attributes and real-time channel quality for comprehensive queuing and scheduling calculation, the system can ensure that in bandwidth congestion or poor network conditions, alarm telemetry data packets located in high-risk accident-prone road segments and showing an extremely high risk of loss of control are prioritized, ignoring the underlying conventional communication queue queuing mechanism, and forcibly seizing scarce channel resources to achieve second-level low-latency uploading. This allows the back-end administrator of the remote monitoring center to intervene in the scheduling and emergency command of extreme traffic risk situations with zero time difference.

[0047] Please see Figure 5 The integrated monitoring platform for dangerous driving behavior of road transport vehicles includes: a synchronous acquisition module, used to capture the spatial coordinate sequence of the driver's gaze point, the timing sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis through an on-board sensor network, and perform time-domain alignment processing through a hardware clock hard synchronization mechanism; a state calculation module, used to estimate the cross-power spectral density and identify parameters of the time-domain aligned sequence, extract the phase hysteresis of the intended action, the low-frequency coupling coefficient, and the vehicle's mechanical response hysteresis, and extract the cognitive-mechanical decoupling index through state-space feature decomposition; a warning generation module, used to fuse on-board radar and visual data to fit the longitudinal collision avoidance safety boundary to obtain the continuous collision time margin, and input the cognitive-mechanical decoupling index and the continuous collision time margin into the alarm modulation transfer function to solve and generate the time-varying risk warning gradient; and an alarm telemetry module, used to separate the high and low frequency components of the signal envelope corresponding to the time-varying risk warning gradient through a filter, independently trigger the multimodal audio-visual alarm device and the on-board seat tactile vibration device, and simultaneously encapsulate the warning data into an alarm telemetry data packet and upload it to the remote monitoring center.

[0048] In this implementation plan, the synchronous acquisition module is mainly responsible for the real-time acquisition and benchmark alignment of multi-source heterogeneous data at the system's underlying layer. By coordinating the driver-facing visual sensors inside the vehicle and the kinematic sensors on the chassis outside the vehicle, this module physically converges and integrates the originally independent physiological behavior signals and physical control signals. Utilizing a hardware-level hard synchronization mechanism, it completely eliminates the timing misalignment caused by the differences in sampling clocks between different sensors. This strict time-domain alignment process provides an absolutely reliable data input base for the system to accurately trace the causal time delay relationship between the driver's cognitive intention and the vehicle's actual movement trajectory, fundamentally eliminating the risk of false danger misjudgments caused by the asynchrony of underlying data. The state calculation module, as the core algorithm analysis engine for human-vehicle interaction risks, is responsible for in-depth system dynamics deconstruction of synchronously collected basic time-series data. This module not only independently quantifies the driver's physiological reaction sluggishness and the mechanical hysteresis characteristics under full vehicle load, but also further decomposes these two types of hysteresis parameters into a unified multi-dimensional state observation space for feature matrix decomposition. Its core function is to penetrate the interference of complex road conditions and cabin environment, directly address the intrinsic nature of dangerous driving, and accurately calculate the dynamic instability of the current human-machine collaborative control system. By outputting a quantified cognitive-mechanical decoupling index, this module makes the driver's hidden micro-fatigue and distraction hazards explicit, providing a solid endogenous state assessment basis for anti-interference alarms. The warning generation module plays the role of a central hub for the fusion of internal and external risk factors and dynamic decision-making. This module breaks the limitation of traditional alarm systems that only focus on the driver's state inside the vehicle. By introducing external spatial perception data fused with forward radar and vision, it calculates in real time the allowable limit of collision avoidance tolerance boundary under the current physical environment. By inputting the cognitive-mechanical decoupling index (representing internal hazards) and the continuous collision time margin (representing external environmental constraints) into the transfer function for modulation, this module can adaptively and dynamically adjust the sensitivity of alarm triggering based on the urgency of external road conditions. This allows the system to flexibly output time-varying risk warning gradients, ensuring a rapid increase in warning intensity when a rear-end collision is imminent, while effectively suppressing frequent false alarms when the vehicle is traveling on safe, open roads. The alarm telemetry module acts as the physical execution terminal for hazard prevention and the communication bridge for cloud monitoring. This module uses filtering and separation technology to decompose the warning gradient signal into high and low frequency bands, transforming abstract algorithmic decisions into physical stimuli targeting different sensory channels of the driver. Independently triggering directional tactile vibrations and multimodal audio-visual devices for different frequency components achieves a scientifically graded alarm system, ranging from gentle subconscious arousal to forced auditory and visual intervention, effectively avoiding the stress, fright, and misoperation risks caused to heavy truck drivers by a single, abrupt high-decibel alarm.At the same time, the module encapsulates and packages the core early warning parameters and uploads them to the remote monitoring center, completely breaking down the information silos between the front-end alarm of a single vehicle and the back-end supervision of the fleet, and realizing three-dimensional closed-loop supervision and remote rapid response to dangerous driving behavior of road transport vehicles.

[0049] In summary, this application has at least the following effects:

[0050] An integrated identification method and monitoring platform for dangerous driving behavior in road transport vehicles simultaneously collects data on driver gaze, steering wheel fine-tuning torque, and vehicle chassis yaw rate. It performs underlying dynamic fusion calculations to integrate the driver's physiological cognitive delays and the vehicle's mechanical response lags, accurately extracting a decoupling index characterizing the degree of human-machine collaborative control failure. Combined with forward collision avoidance time margin data obtained from onboard radar and vision equipment, the system dynamically modulates warning sensitivity, generating a risk warning gradient that changes with real road conditions and vehicle load status. Furthermore, based on the risk deterioration rate, the system decomposes the warning signal into a low-frequency component driving seat tactile vibrations and a high-frequency component triggering audio-visual devices. This enables physical subconscious awakening in the early stages of potential danger and forced audio-visual intervention in emergency situations, while prioritizing the remote telemetry uploading of core warning parameters to a remote backend. This solution effectively overcomes the technical shortcomings of traditional single-phenomenon monitoring, which frequently results in false alarms in complex transportation environments. It completely solves the industry pain point that fixed threshold alarms are prone to startling drivers and causing secondary accidents. It builds a complete active safety protection closed loop for the road transportation industry, from the underlying decoupling judgment of people and vehicles and the dynamic fault-tolerant calculation of the environment to multimodal physical hierarchical blocking and cloud-based collaborative supervision.

[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0056] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An integrated method for identifying dangerous driving behaviors of road transport vehicles, characterized in that, Includes the following steps: S1. Obtain the spatial coordinate sequence of the driver's gaze point, the timing sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis; perform cross-power spectral density estimation on the spatial coordinate sequence of the gaze point and the timing sequence of the steering wheel fine-tuning torque, and extract the phase hysteresis of the intended action and the low-frequency coupling coefficient. S2. Construct an autoregressive moving average model between the steering wheel fine-tuning torque time series and the vehicle chassis yaw rate series, analyze the principal poles of the autoregressive moving average model to extract the vehicle mechanical response hysteresis; construct the intention action phase hysteresis, low-frequency coupling coefficient and vehicle mechanical response hysteresis into a dynamic state space matrix and perform eigenvalue decomposition, extract the maximum Lyapunov exponent as the cognitive electromechanical decoupling exponent. S3. Collect the relative distance sequence and relative speed sequence between the vehicle and the target obstacle in front. Based on the time decay characteristics of the relative distance and relative speed sequences, and combined with the maximum longitudinal braking deceleration of the vehicle, fit the longitudinal collision avoidance safety boundary to obtain the continuous collision time margin. Construct the alarm modulation transfer function with the cognitive electromechanical decoupling index as the state variable and the continuous collision time margin as the boundary constraint, and solve the time-varying risk warning gradient. S4. Convert the time-varying risk warning gradient into an alarm trigger signal envelope according to a nonlinear mapping rule and input it into the vehicle multi-channel alarm controller. By analyzing the amplitude of the high and low frequency components of the alarm trigger signal envelope, trigger the multimodal sound and light alarm device and the vehicle seat tactile vibration device respectively. Simultaneously encode the cognitive electromechanical decoupling index and the time-varying risk warning gradient into an alarm telemetry data packet and send it to the remote monitoring center.

2. The integrated identification method for dangerous driving behavior of road transport vehicles according to claim 1, characterized in that: The specific process for obtaining the spatial coordinate sequence of the driver's gaze point, the timing sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis is as follows: The driver's facial feature image is captured by the cockpit infrared binocular vision sensor. The facial feature image is transformed into a spatial coordinate system to extract the three-dimensional coordinates of the gaze point. The gaze point spatial coordinate sequence is generated by resampling at a fixed sampling frequency. The initial torque electrical signal is acquired by the steering column torque sensor, and the initial torque electrical signal is denoised and filtered to extract the effective steering intervention signal, generating a steering wheel fine-tuning torque timing sequence. The vehicle inertial measurement unit collects six-degree-of-freedom motion data of the chassis, separates the longitudinal and roll components, extracts the yaw rate data to generate a vehicle chassis yaw rate sequence, and uses a hardware clock hard synchronization mechanism to perform time-domain alignment of the gaze point spatial coordinate sequence, the steering wheel fine-tuning torque timing sequence, and the vehicle chassis yaw rate sequence.

3. The integrated identification method for dangerous driving behavior of road transport vehicles according to claim 1, characterized in that: The specific process of estimating the cross-power spectral density of the gaze point spatial coordinate sequence and the steering wheel fine-tuning torque temporal sequence, and extracting the intentional action phase hysteresis and low-frequency coupling coefficient is as follows: A sliding time window is applied to the time-domain aligned gaze point spatial coordinate sequence and the steering wheel fine-tuning torque time sequence to truncate the data, eliminating spectral leakage at the sequence edges. The truncated gaze point spatial coordinate sequence and steering wheel fine-tuning torque time sequence are then converted to the frequency domain, and the cross-power spectral density function is extracted by analyzing their cross-period graph. The real and imaginary parts of the cross-power spectral density function are decomposed. Within the preset low-frequency band of human-machine interaction, the principal values ​​of the arctangent of the real and imaginary components are extracted as the phase hysteresis of the intended action. Based on the independent power spectral density functions of the spatial coordinate sequence of the gaze point and the timing sequence of the steering wheel fine-tuning torque, the cross-power spectral density function is analyzed by spectral coherence normalization to extract the low-frequency coupling coefficient.

4. The integrated identification method for dangerous driving behavior of road transport vehicles according to claim 1, characterized in that: The specific process of constructing an autoregressive moving average model between the time series of steering wheel fine-tuning torque and the time series of vehicle chassis yaw rate, and extracting the vehicle mechanical response hysteresis from the principal poles of the autoregressive moving average model is as follows: The timing sequence of steering wheel fine-tuning torque is used as the forward excitation input and the yaw rate sequence of vehicle chassis is used as the state feedback output to establish a discrete difference equation. The augmented least squares algorithm is used to identify the parameters of the discrete difference equation, determine the autoregression order and the moving average order to generate an autoregressive moving average model. The zero-pole distribution of the autoregressive moving average model is mapped, and the eigenvalue closest to the unit circle in the complex plane is located as the principal pole of the model. The decay time constant corresponding to the principal pole of the model is extracted, and the decay time constant is transformed into the time dimension according to the sampling period of the chassis yaw rate sequence to obtain the vehicle mechanical response hysteresis.

5. The integrated identification method for dangerous driving behavior of road transport vehicles according to claim 1, characterized in that: The specific process of constructing a dynamic state-space matrix from the intentional action phase hysteresis, low-frequency coupling coefficient, and vehicle mechanical response hysteresis, and then performing eigenvalue decomposition to extract the maximum Lyapunov exponent as the cognitive-electromechanical decoupling exponent is as follows: The intentional action phase hysteresis, low-frequency coupling coefficient and vehicle mechanical response hysteresis are embedded as independent state variables into the multi-dimensional state observation space to reconstruct the human-machine interaction state evolution trajectory and construct a dynamic state space matrix. The Jacobian matrix sequence of the dynamic state space matrix is ​​deduced along the evolution direction of the human-computer interaction state evolution trajectory. The Jacobian matrix sequence is orthogonally decomposed and updated. The exponential divergence rate of the neighboring trajectories in the observation space is tracked to obtain the Lyapunov exponential spectrum of the dynamic state space matrix. The scalar extrema characterizing the maximum divergence trend of human-machine collaboration instability are extracted from the Lyapunov exponent spectrum of the dynamic state space matrix and used as the cognitive-electromechanical decoupling index.

6. The integrated identification method for dangerous driving behavior of road transport vehicles according to claim 1, characterized in that: The specific process of collecting the relative distance and relative velocity sequences between the vehicle and the target obstacle ahead, and then fitting the longitudinal collision avoidance safety boundary based on the temporal decay characteristics of the relative distance and relative velocity sequences, combined with the vehicle's maximum longitudinal braking deceleration, to obtain the continuous collision time margin is as follows: By integrating target point cloud data collected by vehicle-mounted millimeter-wave radar with visual feature points collected by vehicle-mounted forward-looking camera, the motion vector of the target obstacle in front is extracted and the relative distance sequence and relative speed sequence between the vehicle and the target obstacle in front are calculated. The relative distance sequence is mapped by first-order difference using the relative velocity sequence. The temporal decay features of the relative distance sequence and the relative velocity sequence are extracted. The maximum longitudinal braking deceleration of the vehicle and the vehicle mechanical response hysteresis are input into the kinematic prediction model and projected to generate the longitudinal collision avoidance safety boundary. By solving the intersection of the relative distance sequence and the longitudinal collision avoidance safety boundary in the spatiotemporal dimension, the remaining time length before the vehicle reaches the collision critical state is extracted as the continuous collision time margin.

7. The integrated identification method for dangerous driving behavior of road transport vehicles according to claim 1, characterized in that: The specific process of constructing the alarm modulation transfer function by using the cognitive electromechanical decoupling exponent as the state variable and the continuous collision time margin as the boundary constraint, and solving the output time-varying risk warning gradient is as follows: The cognitive electromechanical decoupling index, which is a state variable, is converted into a warning dynamic response coefficient through a nonlinear gain function. The continuous collision time margin is mapped to a time-varying risk probability distribution field, and the boundary threshold constraint for the evolution of dangerous states is defined. An alarm modulation transfer function is constructed by using the early warning dynamic response coefficient as the numerator and the continuous collision time margin as the denominator. The amplitude-frequency response characteristics of the alarm modulation transfer function in the complex frequency domain are analyzed. The directional derivative of the output envelope of the alarm modulation transfer function is calculated to extract the time-varying risk early warning gradient.

8. The integrated identification method for dangerous driving behavior of road transport vehicles according to claim 1, characterized in that: The process of converting the time-varying risk warning gradient into an alarm trigger signal envelope according to a nonlinear mapping rule and inputting it into the vehicle multi-channel alarm controller, and then triggering the multimodal audible and visual alarm device and the vehicle seat tactile vibration device by analyzing the amplitude of the high and low frequency components of the alarm trigger signal envelope, is as follows: The time-varying risk warning gradient is mapped to a preset oscillation waveform generator to generate an alarm trigger signal envelope whose amplitude changes with the gradient. A bandpass filter with specific frequency response characteristics is used to perform spectrum stripping on the alarm trigger signal envelope to separate high-frequency and low-frequency characteristic components. The high-frequency characteristic components are encapsulated into audio-visual driving commands and sent to the multimodal audio-visual alarm device, while the low-frequency characteristic components are encapsulated into tactile sequence pulse commands and sent to the vehicle seat tactile vibration device.

9. The integrated identification method for dangerous driving behavior of road transport vehicles according to claim 1, characterized in that: The specific process of simultaneously encoding the cognitive electromechanical decoupling index and the time-varying risk warning gradient into an alarm telemetry data packet and sending it to the remote monitoring center is as follows: The cognitive electromechanical decoupling index, time-varying risk warning gradient and current global positioning coordinates are asynchronously aligned from multiple sources. The aligned data is then serialized into a bit stream using a preset industrial communication protocol, and vehicle identification code and alarm level flag are inserted. Error detection is performed on the serialized data using a cyclic redundancy check algorithm. The data is then encapsulated to generate alarm telemetry data packets. The priority scheduling algorithm of the vehicle wireless communication link is then invoked to load the alarm telemetry data packets into a high-priority transmission queue and upload them to the remote monitoring center.

10. An integrated monitoring platform for dangerous driving behavior of road transport vehicles, used to execute the integrated identification method for dangerous driving behavior of road transport vehicles as described in any one of claims 1-9, characterized in that, include: The synchronous acquisition module is used to capture the spatial coordinate sequence of the driver's gaze point, the timing sequence of the steering wheel fine-tuning torque, and the yaw rate sequence of the vehicle chassis through the vehicle sensor network, and perform time-domain alignment processing through a hardware clock hard synchronization mechanism. The state solution module is used to estimate the cross power spectral density and identify parameters of the time-domain aligned sequence, extract the intentional action phase hysteresis, low-frequency coupling coefficient and vehicle mechanical response hysteresis, and extract the cognitive electromechanical decoupling index through state space feature decomposition. The warning generation module is used to fuse vehicle radar and visual data to fit the longitudinal collision avoidance safety boundary to obtain the continuous collision time margin, and input the cognitive electromechanical decoupling index and the continuous collision time margin into the alarm modulation transfer function to solve and generate the time-varying risk warning gradient. The alarm telemetry module is used to separate the high and low frequency components of the signal envelope corresponding to the time-varying risk warning gradient through a filter, and independently trigger the multimodal sound and light alarm device and the vehicle seat tactile vibration device, and simultaneously encapsulate the warning data into an alarm telemetry data packet and upload it to the remote monitoring center.

Images