Curve driving collision early warning method and system based on multi-risk field fusion

Abstract

The invention discloses a curve driving collision early warning method and system based on multi-risk field fusion, and the method comprises the steps: 1, collecting vehicle dynamics data in real time, and carrying out the short-time prediction; step 2, geometric modeling and rasterization of a road; 3, constructing a geometric risk field; 4, constructing a kinetic energy risk field; 5, constructing a behavior risk field; step 6, constructing a comprehensive risk field; step 7, collision risk judgment; and step 8, generating an active intervention strategy. Collaborative perception and risk prediction of vehicles, roads and driving behaviors are realized through multi-risk field fusion, and potential collision hidden dangers of curve driving are identified in advance; a vehicle dynamics model and risk correction based on road adhesion are introduced to realize accurate risk assessment of the vehicle; multi-time window trajectory prediction and behavior modeling are adopted, so that the recognition capability of complex driving behaviors is improved; a closed-loop system from risk prediction and decision making to active control is realized, and the safety and stability of the vehicle driving in the curve scene are effectively improved.

Description

Technical Field

[0001] This invention relates to a method and system for early warning of vehicle collisions on curves based on multi-risk field fusion, belonging to the field of intelligent transportation and vehicle safety control technology. Background Technology

[0002] With the development of intelligent connected vehicle technology, vehicle safety control is gradually evolving from traditional passive protection to proactive early warning and intelligent decision-making. In complex road environments, especially curved sections, there are characteristics such as large changes in road curvature, limited visibility, and significant fluctuations in road surface adhesion coefficient, making vehicle driving risks significantly higher than in straight road scenarios. In existing technologies, AEB systems mainly rely on the relative speed and distance between the vehicle and obstacles for early warning. However, in curved scenarios, due to the large changes in road curvature and limited visibility, a single method of determining relative distance and speed is prone to high false alarm rates and long warning delays, making it difficult to achieve accurate warnings. Furthermore, the dynamic stability of a vehicle is usually affected by multiple factors such as longitudinal acceleration, yaw rate, lateral force distribution, and road surface adhesion conditions, making it difficult to comprehensively characterize actual driving risks by relying on a single model. The driver's maneuvering behavior in curves (such as deceleration, steering, or lane changing) also has significant uncertainties, making it difficult for traditional rule-based or threshold-based early warning algorithms to balance accuracy and real-time performance.

[0003] Therefore, there is an urgent need for a multi-dimensional risk modeling method that can integrate vehicle dynamics, road geometry, and driving behavior prediction information to achieve accurate assessment, early warning, and proactive intervention of potential vehicle collision risks in curve scenarios. Summary of the Invention

[0004] Purpose of the invention: To address the shortcomings of existing technologies, this invention provides a method and system for cornering collision warning based on multi-risk field fusion. This invention achieves multi-dimensional modeling and dynamic assessment of driving risks by real-time modeling of vehicle dynamics, road geometry, and driving behavior characteristics, thereby identifying potential collision risks in advance and taking timely warning and proactive intervention strategies to improve vehicle safety and stability.

[0005] Technical solution: A curve collision warning method based on multi-risk field fusion, comprising the following steps:

[0006] Step 1: Collect vehicle dynamics data in real time and make short-term predictions of vehicle position, attitude and trajectory based on the vehicle dynamics model;

[0007] Step 2: Road geometry modeling and rasterization;

[0008] Step 3: Construct a geometric risk field on the rasterized road geometry model in Step 2: Calculate the road curvature, lane width and lateral deviation, and generate a geometric risk distribution using the lateral deviation, boundary distance and road curvature as inputs to obtain the geometric risk field;

[0009] Step 4: Construct a kinetic risk field based on the geometric risk field in Step 3: Based on the vehicle's dynamic data, road surface adhesion conditions, and vehicle center of gravity height, adopt a spatial distribution with the vehicle as a reference and superimpose the risk amplification factor of the working condition to reflect the dynamic limit risk under different working conditions, thus forming a kinetic risk field.

[0010] Step 5: Based on the geometric risk field and kinetic risk field of Step 3 and Step 4, construct the behavioral risk field: extrapolate the trajectory within the prediction time window, model the uncertainty around the trajectory using probability distribution, and superimpose the weights of each time window to obtain the behavioral risk field that reflects the uncertainty of driving behavior.

[0011] Step 6: Construct a comprehensive risk field: Normalize and weight the geometric risk field, kinetic risk field, and behavioral risk field from Steps 3, 4, and 5 to obtain a comprehensive risk field;

[0012] Step 7: Collision Risk Assessment: Extract the comprehensive risk value from the area in front of the vehicle's path, calculate the maximum risk, average risk, and cumulative risk along the path, determine the risk level based on preset multi-level risk thresholds, and introduce a collision time margin to correct the risk level.

[0013] Step 8: Active intervention strategy generation: Based on the risk level obtained in Step 7, generate intervention commands for braking, steering, or a combination of both, and return to Step 1 to continuously update the vehicle status and risk field according to the sampling cycle, so as to achieve real-time early warning and closed-loop control.

[0014] The preferred option, step one, specifically includes:

[0015] Obtain vehicle dynamics data:

[0016] Acquire vehicle bus and sensor data: vehicle speed Longitudinal acceleration lateral acceleration yaw rate Steering wheel angle ;

[0017] Obtain road information: Road centerline curvature Lane width Inner and outer boundary curves, road surface adhesion coefficient ;

[0018] Obtain the relative quantity with the vehicle in front: relative distance Relative velocity ;

[0019] longitudinal acceleration lateral acceleration yaw rate Perform Kalman filtering to correct the sensor's output error in the zero-input state and unify the timestamps;

[0020] Vehicle dynamics model was selected to unify the time. speed Longitudinal acceleration lateral acceleration yaw rate Steering wheel angle and total vehicle mass Yaw moment of inertia Distance from center of gravity to front axle and the distance from the center of gravity to the rear axle Front axle equivalent lateral stiffness and rear axle equivalent lateral stiffness For input,

[0021] Establish system state vector ,in For longitudinal velocity, For the lateral velocity, the slip angle and lateral force are calculated using a linear tire model, derived from the dynamic equations with step sizes. Numerical integration yields In the world coordinate system, the pose update equation is used. Down propagation pose ,in The vehicle's heading angle is given by a rotation matrix. Complete vehicle coordinates and Alignment:

[0022] Before calculating the lateral forces of the front and rear wheels, the tire slip angle is first calculated based on the velocity decomposition in the vehicle's center-of-gravity coordinate system; the front wheel contact point is in the vehicle body coordinate system. The velocity below is expressed as:

[0023]

[0024]

[0025] The rear wheel contact point speed is:

[0026]

[0027]

[0028] in, For longitudinal velocity, For lateral velocity, The yaw rate is angular velocity. and These are the distances from the center of mass to the front and rear axles, respectively.

[0029] Based on the small-angle linearization assumption The slip angles of the front and rear wheels are defined as follows:

[0030]

[0031]

[0032] in, This represents the front wheel steering angle; the above expression reflects the angle between the local velocity direction of the tire and the tire's pointing direction.

[0033] After the slip angle is calculated, the tire lateral force is calculated using a linear tire model:

[0034]

[0035]

[0036] in, and These represent the lateral stiffness of the front and rear wheels, respectively; a negative sign indicates that the lateral force of the tire always acts in the direction that reduces the lateral slip angle.

[0037] Define the longitudinal dynamic equation:

[0038]

[0039] In the formula, For the overall vehicle quality, The rate of change of longitudinal velocity, Projection of longitudinal force on the front wheel This is the projection of the lateral force on the front wheel. For the longitudinal force of the rear wheel, For air resistance, For rolling resistance;

[0040] Define the transverse dynamic equation:

[0041]

[0042] In the formula, For the vehicle body The actual acceleration in the axial direction, This is the projection of the lateral force on the front wheel. The lateral force is from the rear wheel. Projection of longitudinal force on the front wheel;

[0043] Define the yaw moment balance equation:

[0044]

[0045] In the formula, Let be the yaw moment of inertia of the vehicle about its center of mass. This is the yaw acceleration. This is the distance from the center of gravity to the front axle. It is the resultant lateral force component of the front wheel. This is the distance from the center of mass to the rear axle. This refers to the lateral force on the rear wheel;

[0046] Define derived quantity :

[0047]

[0048] Define speed limit:

[0049]

[0050] After obtaining the vehicle's velocity state and performing amplitude limiting, the vehicle's dynamic state is transformed into a motion trajectory in the world coordinate system through the pose update equation; the vehicle's kinematic update equation in the global coordinate system is:

[0051]

[0052]

[0053]

[0054] in, Let this be the position of the vehicle's center of mass in the world coordinate system. For the vehicle's heading angle, For longitudinal velocity, For lateral velocity, This refers to the yaw rate;

[0055] Discretizing the above kinematic equations using the Euler method yields:

[0056]

[0057]

[0058]

[0059] in, The discrete time step is used; through the above pose update process, the predicted position and heading of the vehicle at the next moment are obtained, providing input for subsequent short-term trajectory prediction and behavioral risk field construction;

[0060] Will Projected onto the center line of the road quantity ,in Let the coordinates be the arc length. For lateral error, For heading error, and the current curvature of the road centerline. Using the same model, extrapolate the trajectory within a 2-3 second prediction time window, applying physical constraints and velocity limits during the prediction process;

[0061] In the discrete system, the first Time points Output and derivatives lateral acceleration , combined With 2-3 second predicted trajectory It includes a quality flag indicating whether physical constraints / speed limits have been triggered, which can be used for subsequent risk field construction and early warning determination.

[0062] in, Geometric state quantities of the vehicle in the Frenet curve coordinate system of the road; variables This indicates the arc length position of the vehicle on the center line of the road; Indicates the lateral deviation of the vehicle relative to the centerline; This indicates the deviation between the vehicle's heading angle and the road tangential angle; This indicates the curvature of the road at that location.

[0063] The preferred option, step two, specifically includes:

[0064] At the time of unity The system reconstructs the road centerline based on online lane detection results and uses spline interpolation to obtain a continuous curve. ;

[0065] To ensure numerical stability, the centerline is spaced according to arc length intervals. The system resamples the centerline points. After resampling, the system smooths the resampled centerline point set using Gaussian smoothing filtering. The centerline point set is a discrete sequence of road centerline points.

[0066] Finally, the road tangential angle was calculated. , Road tangent direction vector and curvature This is used for the subsequent construction of the geometric risk field.

[0067] Road tangent angle:

[0068]

[0069] In the formula, For the road at the arc length position The tangential angle at that point, i.e., the heading of the centerline. The first derivative of the curve indicates the direction of the tangent.

[0070] Road curvature:

[0071]

[0072] Tangential vector:

[0073]

[0074] In the formula, It is the tangent vector, pointing in the direction of the road's movement;

[0075] Normal vector:

[0076]

[0077] In the formula, It is the normal vector, perpendicular to the road direction, and points to the left side of the road;

[0078] According to lane width , thus obtaining the inner and outer boundaries:

[0079]

[0080]

[0081] In the formula, The coordinates of the inner boundary of the road. The coordinates of the outer boundary of the road. Lane width;

[0082] Covering the centerline and extending the forward sight distance outwards Establish a regular grid within the rectangular area .

[0083] The preferred option, step three, specifically includes:

[0084] For each grid point The system uses the nearest point projection method to calculate its arc length position on the center line. This is used for subsequent curvature and risk value mapping calculations.

[0085]

[0086] In the formula, For point The projection position on the center line, i.e., the arc length. Choose the one that minimizes the distance. , The distance from the point to the center line;

[0087] Therefore, the lateral deviation is calculated:

[0088]

[0089] In the formula, For lateral deviation, vehicle or point lateral distance relative to the centerline The normal direction is used to calculate the offset.

[0090] Lateral distance from the vehicle to the lane edge:

[0091]

[0092] In the formula, For point Distance to the nearest lane boundary It is half the width of the lane. The absolute lateral deviation of a point relative to the centerline;

[0093] Road mask ,when ,otherwise ,in This is for redundant width;

[0094] The output data is:

[0095]

[0096] The output data is used as input for constructing the geometric risk field and subsequent risk fusion;

[0097] For each grid point Based on the lateral deviation obtained in step three Define the horizontal risk quantity:

[0098]

[0099] In the formula, This is an adjustment factor used to control the sensitivity of risk to changes in lateral deviation;

[0100] Based on the distance within the band Define the boundary risk components:

[0101]

[0102] In the formula, This is the attenuation coefficient, used to control the rate at which boundary risk decays with distance;

[0103] Based on the road curvature of the corresponding projection point Define curvature risk:

[0104]

[0105] In the formula, For reference curvature, Controlling sensitivity;

[0106] The three components are normalized and then weighted and summed:

[0107]

[0108] In the formula, the weights ,

[0109] For points not within the road, i.e. The risk field value is directly set to zero or marked as invalid; only the risk values ​​of grid points within the road are retained.

[0110] The output data is:

[0111]

[0112] In the formula, This represents the geometric risk value of each grid point, and the output data serves as the geometric factor input for the overall risk field.

[0113] The preferred option, step four, specifically includes:

[0114] With the vehicle's center of mass as the origin, the vehicle coordinate system For reference, define the elliptic kernel function:

[0115]

[0116] In the formula, These are the coordinates of a point in the vehicle's coordinate system. Indicates the longitudinal scale. Indicates the longitudinal scale;

[0117] Based on the vehicle's motion conditions, the longitudinal direction window is superimposed with an elliptical kernel, emphasizing the front during braking and the rear during driving. Simultaneously superimposed curve direction window ,

[0118] Based on the vehicle's lateral acceleration Define the lateral acceleration gain:

[0119]

[0120] In the formula, meters per square second This is the lateral acceleration gain coefficient. Sensitivity index;

[0121] Based on yaw rate Define the yaw rate gain:

[0122]

[0123] In the formula, For reference yaw rate, This is the yaw gain coefficient. Sensitivity index;

[0124] Based on longitudinal acceleration Define the longitudinal acceleration gain:

[0125]

[0126] In the formula, The longitudinal acceleration gain coefficient, Sensitivity index;

[0127] Introducing the road surface adhesion coefficient Define the attachment utilization rate:

[0128]

[0129]

[0130] In the formula, As a risk amplification factor under adhesion conditions, For the attachment correction gain coefficient, Sensitivity index;

[0131] Considering the height of the center of mass With wheelbase Define the rollover index:

[0132]

[0133]

[0134] In the formula, As a factor amplifying the risk of rollover, This is the side-flip gain coefficient. Sensitivity index;

[0135] Multiplying each gain factor by the elliptic kernel yields the kinetic risk field:

[0136]

[0137] In the formula, This indicates a trimming operation, ensuring the risk value is within a certain range. ;

[0138] Output kinetic energy risk field and the corresponding longitudinal acceleration lateral acceleration yaw rate Adhesion coefficient Center of mass height Wheelbase The output data serves as the input for the fusion of comprehensive risk fields.

[0139] The preferred option, step five, specifically includes:

[0140] At any moment The system is based on the planar position of the vehicle's center of mass in the world coordinate system. The angle between the vehicle's longitudinal axis and the x-axis of the world coordinate system Longitudinal velocity lateral velocity yaw rate And step two, kinematic model to generate future trajectory Set multiple forecast time windows:

[0141]

[0142] Within each time window, use step size Second integration yields a series of trajectory points Mapped to raster ;

[0143] Considering the uncertainties caused by driving operations and environmental disturbances, the trajectory is modeled as a Gaussian distribution:

[0144]

[0145] In the formula, For grid points; In the time window Interior point prediction trajectory; For point To trajectory The minimum distance; The radius of the trajectory spread.

[0146] For each time window, the risk function is defined as:

[0147]

[0148] In the formula, For point In the time window The following behavioral risks; As time window weight, satisfying ;

[0149] Taking into account both short-term and long-term forecasts, the risks of different time windows are superimposed:

[0150]

[0151] If the system detects a driving intention, it will weight the direction accordingly:

[0152]

[0153] In the formula, This is a correction factor; For the indicator function, if point If it is in the expected direction of behavior, it is 1; otherwise, it is 0.

[0154] The revised risk is:

[0155]

[0156] right Perform two-dimensional Gaussian smoothing to eliminate jagged edges and abrupt changes; then normalize the entire field to... This ensures direct integration with the geometric risk field and kinetic risk field; the output is:

[0157]

[0158] In the formula, This represents the behavioral risk value of a grid point, and the output data serves as the input for behavioral factors in the fusion of the comprehensive risk field.

[0159] The preferred option, step six, specifically includes:

[0160] Since the geometric risk field, kinetic risk field, and behavioral risk field have different numerical ranges and dimensions, each risk field is first normalized:

[0161]

[0162]

[0163]

[0164] By normalizing, the value ranges of the three risk fields are ensured to be uniform. ;

[0165] Therefore, a weighted summation method is used to obtain the comprehensive risk:

[0166]

[0167] In the formula, geometric risk weight kinetic risk weight and behavioral risk weights satisfy ;

[0168] Considering the variance of the probability distribution in behavioral prediction, the risk field is corrected as follows:

[0169]

[0170] In the formula, Indicates behavioral risk at a point Uncertain variance, These are the weighting coefficients;

[0171] The resulting comprehensive risk field is:

[0172]

[0173] In the formula, This represents the overall risk value for each grid point.

[0174] The preferred option, step seven, specifically includes:

[0175] Get comprehensive risk field Afterwards, the system enters the collision risk assessment and threshold triggering phase; high-risk areas in front of the vehicle are extracted from the comprehensive risk distribution, risk indicators are calculated, and compared with set thresholds to trigger the corresponding collision warning level; within the vehicle's forward-looking range... Within, capture the comprehensive risk field; select grid points located within the lane range. Forming candidate risk areas For the region The risk value within is statistically analyzed, and various indicators are defined:

[0176] Define the maximum risk value:

[0177]

[0178] Define the average risk value:

[0179]

[0180] Define the risk of the path integral ahead:

[0181]

[0182] In the formula, Predict vehicle trajectories;

[0183] Three risk thresholds are set: Mild risk threshold: Moderate risk threshold: Severe risk threshold: ;when or When the corresponding threshold is exceeded, different warning levels are triggered:

[0184] when The status is safe and there are no warnings.

[0185] when This is a Level 1 warning, alerting drivers to take precautions.

[0186] when This is a Level II warning, a strong warning, accompanied by audible and visual alerts;

[0187] when This is a Level 3 warning, indicating a serious danger, triggering braking or intervention.

[0188] Further adjustments based on collision time margin (TTC):

[0189]

[0190] In the formula, The relative distance between the vehicle and the obstacle. For relative velocity; when = Every second, the risk level is raised by one level; once the risk level is determined, the system triggers the corresponding warning action.

[0191] The output result is:

[0192]

[0193] In the formula, These correspond to no risk, mild risk, moderate risk, and severe risk, respectively.

[0194] In the preferred embodiment, step eight specifically includes:

[0195] Based on the risk level and predicted trajectory, appropriate longitudinal, lateral, or combined intervention measures are dynamically selected to minimize collision risk and maintain vehicle stability.

[0196] Based on the output of step six Define control objectives at different levels:

[0197] Level 1 warning It only alerts the driver and does not take control of the vehicle;

[0198] Level II Warning Maintain a safe distance and slow down;

[0199] Level III Warning Take immediate mandatory measures, such as emergency braking or steering;

[0200] If the predicted trajectory carries a risk of forward collision, i.e. If the value is less than the set threshold, a vertical control strategy is implemented:

[0201] Define braking intervention:

[0202]

[0203] In the formula, Slowing down demand, This is the adhesion limit;

[0204] Define the safe distance model:

[0205]

[0206] If the actual vehicle distance If so, braking will be triggered;

[0207] If the risk field is concentrated in the direction of the curve or laterally, then a lateral control strategy should be formulated:

[0208] By correcting the steering to bring the vehicle back closer to the centerline, directional intervention is defined.

[0209]

[0210] In the formula, This is the lateral deviation. For heading error, , To control the gain;

[0211] When outputting intervention commands, the system must impose stability constraints:

[0212]

[0213] In the formula, The rollover index;

[0214] The output is a set of intervention instructions:

[0215]

[0216] In the formula, This is a longitudinal acceleration command; a negative value indicates braking. For steering angle command, {No intervention, braking only, steering only, braking + steering}.

[0217] A system for implementing a curve collision warning method based on multi-risk field fusion, including

[0218] The data acquisition module obtains vehicle dynamics data and road environment information, providing an input basis for risk assessment.

[0219] The state prediction module performs state estimation and short-term prediction on the collected data based on the vehicle dynamics model, and outputs the vehicle's future position, attitude and trajectory.

[0220] The road geometry modeling and rasterization module reconstructs the road centerline based on the perceived lane lines, boundary lines and road curvature information, calculates the road curvature, lane width and lateral deviation, and generates a rasterized representation of the road geometry model.

[0221] The geometric risk field construction module calculates lateral deviation, boundary distance, and road curvature risk factors based on the road geometric model, and establishes a geometric risk field that reflects the road's geometric constraints.

[0222] The kinetic energy risk field construction module constructs a kinetic energy risk field that reflects the vehicle's dynamic limits based on factors such as the vehicle's longitudinal and lateral acceleration, yaw rate, road adhesion coefficient, and vehicle center of gravity height.

[0223] The behavioral risk field construction module extrapolates the trajectory within multiple prediction time windows, models the uncertainty around the trajectory in probabilistic form, and generates a behavioral risk field that reflects differences in driving behavior and operational stability.

[0224] The comprehensive risk field fusion module normalizes and weights the geometric risk field, kinetic risk field and behavioral risk field to form a comprehensive risk distribution for subsequent risk level assessment.

[0225] The risk level determination module calculates risk indicators based on the comprehensive risk distribution and compares them with multi-level thresholds to determine the current risk level.

[0226] The active intervention module generates corresponding longitudinal braking or lateral steering intervention commands based on the risk level, enabling active safety control of the vehicle in curve scenarios.

[0227] The system's loop and real-time update module periodically samples vehicle status data and road environment information, and updates risk fields and intervention commands in real time to achieve closed-loop operation and dynamic control of the system.

[0228] The modules are connected via data communication to enable collision risk prediction and active safety control of vehicles in curve scenarios.

[0229] Beneficial effects: This invention achieves collaborative perception and risk prediction of vehicles, roads, and driving behavior through multi-risk field fusion, and identifies potential collision hazards in advance when driving on curves; it introduces vehicle dynamics models and risk correction based on road surface adhesion to achieve accurate risk assessment of vehicles; it adopts multi-time window trajectory prediction and behavior modeling to improve the ability to identify complex driving behaviors; and it realizes a closed-loop system from risk prediction and decision-making to active control, effectively improving the safety and stability of vehicles driving in curve scenarios. Attached Figure Description

[0230] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0231] Figure 1 This is a schematic diagram of a cornering collision warning system.

[0232] Figure 2 A schematic diagram is constructed for the geometric risk field;

[0233] Figure 3 A schematic diagram for constructing a kinetic energy risk field;

[0234] Figure 4 A schematic diagram of constructing a behavioral risk field;

[0235] Figure 5 A logic diagram of vertical and horizontal intervention strategies;

[0236] Figure 6 This is a block diagram of a cornering collision warning system. Detailed Implementation

[0237] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0238] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0239] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0240] like Figure 1 As shown, the curve collision warning method based on multi-risk field fusion includes the following steps:

[0241] Step 1: Collect vehicle dynamics data in real time and make short-term predictions of vehicle position, attitude and trajectory based on the vehicle dynamics model;

[0242] Obtain vehicle dynamics data:

[0243] Acquire vehicle bus and sensor data: vehicle speed Longitudinal acceleration lateral acceleration yaw rate Steering wheel angle ;

[0244] Obtain road information: Road centerline curvature Lane width Inner and outer boundary curves, road surface adhesion coefficient ;

[0245] Obtain the relative quantity with the vehicle in front: relative distance Relative velocity ;

[0246] longitudinal acceleration lateral acceleration yaw rate Perform Kalman filtering to correct the sensor's output error in the zero-input state and unify the timestamps;

[0247] Vehicle dynamics model was selected to unify the time. speed Longitudinal acceleration lateral acceleration yaw rate Steering wheel angle and total vehicle mass Yaw moment of inertia Distance from center of gravity to front axle and the distance from the center of gravity to the rear axle Front axle equivalent lateral stiffness and rear axle equivalent lateral stiffness For input,

[0248] Establish system state vector ,in For longitudinal velocity, For the lateral velocity, the slip angle and lateral force are calculated using a linear tire model, derived from the dynamic equations with step sizes. Numerical integration yields In the world coordinate system, the pose update equation is used. Down propagation pose ,in The vehicle's heading angle is given by a rotation matrix. Complete vehicle coordinates and Alignment:

[0249] Before calculating the lateral forces of the front and rear wheels, the tire slip angle is first calculated based on the velocity decomposition in the vehicle's center-of-gravity coordinate system; the front wheel contact point is in the vehicle body coordinate system. The velocity below is expressed as:

[0250]

[0251]

[0252] The rear wheel contact point speed is:

[0253]

[0254]

[0255] in, For longitudinal velocity, For lateral velocity, The yaw rate is angular velocity. and These are the distances from the center of mass to the front and rear axles, respectively.

[0256] Based on the small-angle linearization assumption The slip angles of the front and rear wheels are defined as follows:

[0257]

[0258]

[0259] in, This represents the front wheel steering angle; the above expression reflects the angle between the local velocity direction of the tire and the tire's pointing direction.

[0260] After the slip angle is calculated, the tire lateral force is calculated using a linear tire model:

[0261]

[0262]

[0263] in, and These represent the lateral stiffness of the front and rear wheels, respectively; a negative sign indicates that the lateral force of the tire always acts in the direction that reduces the lateral slip angle.

[0264] Define the longitudinal dynamic equation:

[0265]

[0266] In the formula, For the overall vehicle quality, The rate of change of longitudinal velocity, Projection of longitudinal force on the front wheel This is the projection of the lateral force on the front wheel. For the longitudinal force of the rear wheel, For air resistance, For rolling resistance;

[0267] Define the transverse dynamic equation:

[0268]

[0269] In the formula, For the vehicle body The actual acceleration in the axial direction, This is the projection of the lateral force on the front wheel. The lateral force is from the rear wheel. Projection of longitudinal force on the front wheel;

[0270] Define the yaw moment balance equation:

[0271]

[0272] In the formula, Let be the yaw moment of inertia of the vehicle about its center of mass. This is the yaw acceleration. This is the distance from the center of gravity to the front axle. It is the resultant lateral force component of the front wheel. This is the distance from the center of mass to the rear axle. This refers to the lateral force on the rear wheel;

[0273] Define derived quantity :

[0274]

[0275] Define speed limit:

[0276]

[0277] After obtaining the vehicle's velocity state and performing amplitude limiting, the vehicle's dynamic state is transformed into a motion trajectory in the world coordinate system through the pose update equation; the vehicle's kinematic update equation in the global coordinate system is:

[0278]

[0279]

[0280]

[0281] in, Let this be the position of the vehicle's center of mass in the world coordinate system. For the vehicle's heading angle, For longitudinal velocity, For lateral velocity, This refers to the yaw rate;

[0282] Discretizing the above kinematic equations using the Euler method yields:

[0283]

[0284]

[0285]

[0286] in, The discrete time step is used; through the above pose update process, the predicted position and heading of the vehicle at the next moment are obtained, providing input for subsequent short-term trajectory prediction and behavioral risk field construction;

[0287] Will Projected onto the center line of the road quantity ,in Let the coordinates be the arc length. For lateral error, For heading error, and the current curvature of the road centerline. Using the same model, extrapolate the trajectory within a 2-3 second prediction time window, applying physical constraints and velocity limits during the prediction process;

[0288] In the discrete system, the first Time points Output and derivatives lateral acceleration , combined With 2-3 second predicted trajectory It includes a quality flag indicating whether physical constraints / speed limits have been triggered, which can be used for subsequent risk field construction and early warning determination.

[0289] in, Geometric state quantities of the vehicle in the Frenet curve coordinate system of the road; variables This indicates the arc length position of the vehicle on the center line of the road; Indicates the lateral deviation of the vehicle relative to the centerline; This indicates the deviation between the vehicle's heading angle and the road tangential angle; This indicates the curvature of the road at that location.

[0290] Step 2: Road geometry modeling and rasterization;

[0291] At the time of unity The system reconstructs the road centerline based on online lane detection results and uses spline interpolation to obtain a continuous curve. ;

[0292] To ensure numerical stability, the centerline is spaced according to arc length intervals. Resampling is performed at intervals that ensure the representation of road geometric details while avoiding noise amplification caused by overly dense sampling. After resampling, the system smooths the resampled centerline point set using Gaussian smoothing filtering, thereby reducing the impact of visual detection errors on curvature calculation and improving the continuity and robustness of road geometric feature calculation. The centerline point set is a discrete sequence of road centerline points.

[0293] Finally, the road tangential angle was calculated. , Road tangent direction vector and curvature This is used for the subsequent construction of the geometric risk field.

[0294] Road tangent angle:

[0295]

[0296] In the formula, For the road at the arc length position The tangential angle at that point, i.e., the heading of the centerline. The first derivative of the curve indicates the direction of the tangent.

[0297] Road curvature:

[0298]

[0299] Tangential vector:

[0300]

[0301] In the formula, It is the tangent vector, pointing in the direction of the road's movement;

[0302] Normal vector:

[0303]

[0304] In the formula, This is the normal vector, perpendicular to the road direction, pointing to the left side of the road. In this embodiment, it is defined as positive on the left and negative on the right.

[0305] According to lane width , thus obtaining the inner and outer boundaries:

[0306]

[0307]

[0308] In the formula, The coordinates of the inner boundary of the road. The coordinates of the outer boundary of the road. Lane width;

[0309] Covering the centerline and extending the forward sight distance outwards Establish a regular grid within the rectangular area .in Preferred The distance is set at 1 meter to strike a balance between timely warnings and computational complexity. This distance ensures that the system has a warning time of approximately 2-3 seconds when the vehicle speed is less than 80 km / h, while avoiding computational delays caused by an excessively large forward-looking field of view.

[0310] The preferred raster resolution This resolution, measured in meters, can reduce the number of redundant points and improve real-time performance while ensuring the spatial continuity of the risk field.

[0311] like Figure 2 As shown, in step three, a geometric risk field is constructed on the rasterized road geometry model in step two: the road curvature, lane width and lateral deviation are calculated, and the geometric risk distribution is generated with the lateral deviation, boundary distance and road curvature as inputs to obtain the geometric risk field;

[0312] For each grid point The system uses the nearest point projection method to calculate its arc length position on the center line. This is used for subsequent curvature and risk value mapping calculations.

[0313]

[0314] In the formula, For point The projection position on the center line, i.e., the arc length. Choose the one that minimizes the distance. , The distance from the point to the center line;

[0315] Therefore, the lateral deviation is calculated:

[0316]

[0317] In the formula, For lateral deviation, vehicle or point The lateral distance relative to the center line is positive on the left and negative on the right. The normal direction is used to calculate the offset.

[0318] Lateral distance from the vehicle to the lane edge:

[0319]

[0320] In the formula, For point Distance to the nearest lane boundary It is half the width of the lane. The absolute lateral deviation of a point relative to the centerline;

[0321] Road mask ,when ,otherwise ,in To avoid redundant widths and numerical gaps, the inner and outer boundaries can be closed into a road polygon, and a consistency check can be performed on the rasterization results.

[0322] The output data is:

[0323]

[0324] The output data is used as input for constructing the geometric risk field and subsequent risk fusion;

[0325] After completing the road geometry modeling and rasterization, the system... A geometric risk field is constructed on the grid. The basic idea is that geometric features such as the position of the road centerline, lane boundaries, and curvature directly affect the driving safety of vehicles. Therefore, in the grid space, the risk value of each point is closely related to its lateral deviation, distance from the boundary, and the road curvature at the corresponding location.

[0326] For each grid point Based on the lateral deviation obtained in step three Define the horizontal risk quantity:

[0327]

[0328] In the formula, This is an adjustment coefficient used to control the sensitivity of risk to changes in lateral deviation; preferably... This ensures a balance between risk sensitivity and numerical smoothness. At the same time, the risk function can produce a significant response in the early stage of vehicle deviation from the center line, thereby enhancing the system's ability to identify lateral deviation trends in the early stage, while avoiding excessive amplification of risk under small deviations, and ensuring the stability and computational convergence of the model.

[0329] Based on the distance within the band Define the boundary risk components:

[0330]

[0331] In the formula, The attenuation coefficient is used to control the rate of change of risk with boundary distance. rice;

[0332] Based on the road curvature of the corresponding projection point Define curvature risk:

[0333]

[0334] In the formula, Reference curvature (corresponding radius) rice), Control sensitivity; the smaller the curve radius, the higher the risk.

[0335] The three components are normalized (range [0,1]) and then weighted and superimposed:

[0336]

[0337] In the formula, the weights It can be set based on scenario experience, for example .

[0338] For points not within the road, i.e. The risk field value is directly set to zero or marked as invalid; only the risk values ​​of grid points within the road are retained; at the same time, to avoid discrete jagged edges, the risk field can be... Perform a two-dimensional Gaussian smoothing, standard deviation In order to obtain a continuous and stable risk distribution.

[0339] The output data is:

[0340]

[0341] In the formula, This represents the geometric risk value of each grid point, and the output data serves as the geometric factor input for the overall risk field.

[0342] like Figure 3 As shown, in step four, a kinetic risk field is constructed based on the geometric risk field in step three: based on the vehicle's dynamic data, road surface adhesion conditions, and vehicle center of gravity height, a spatial distribution with the vehicle as a reference is adopted and a risk amplification factor for the working condition is superimposed to reflect the dynamic limit risk under different working conditions, thus forming a kinetic risk field.

[0343] With the vehicle's center of mass as the origin, the vehicle coordinate system For reference, define the elliptic kernel function:

[0344]

[0345] In the formula, These are the coordinates of a point in the vehicle's coordinate system. Representing longitudinal scale rice, Representing longitudinal scale Meters; the ellipse describes the basic attenuation distribution of risk in the front-to-back and left-to-right directions of the vehicle.

[0346] Based on the vehicle's motion conditions, the longitudinal direction window is superimposed with an elliptical kernel, emphasizing the front during braking and the rear during driving. Simultaneously superimposed curve direction window This directional window is used to enhance the risk weight of the road curvature center side (i.e., the inside of the curve) under turning conditions, reflecting the characteristic that the vehicle has a higher centrifugal risk in the direction of the curve center (curvature center direction).

[0347] Based on the vehicle's lateral acceleration Define the lateral acceleration gain:

[0348]

[0349] In the formula, meters per square second This is the lateral acceleration gain coefficient. It is a sensitivity index; the greater the lateral acceleration, the greater the risk in the curve direction.

[0350] Based on yaw rate Define the yaw rate gain:

[0351]

[0352] In the formula, For reference yaw rate radians per second This is the yaw gain coefficient. This is a sensitivity index; when the vehicle's yaw rate is too high, the risk of cornering increases significantly.

[0353] Based on longitudinal acceleration Define the longitudinal acceleration gain:

[0354]

[0355] In the formula, The longitudinal acceleration gain coefficient, It is a sensitivity index; when a vehicle accelerates or brakes suddenly, the risk in the corresponding direction is amplified.

[0356] Introducing the road surface adhesion coefficient Define the attachment utilization rate:

[0357]

[0358]

[0359] In the formula, As a risk amplification factor under adhesion conditions, For the attachment correction gain coefficient, It is a sensitivity index; when the lateral force approaches the adhesion limit, the risk increases rapidly.

[0360] Considering the height of the center of mass With wheelbase Define the rollover index:

[0361]

[0362]

[0363] In the formula, As a factor amplifying the risk of rollover, This is the side-flip gain coefficient. This is a sensitivity index; when lateral acceleration is too high, the risk of potential rollover increases significantly.

[0364] Multiplying each gain factor by the elliptic kernel yields the kinetic risk field:

[0365]

[0366] In the formula, This indicates a trimming operation, ensuring the risk value is within a certain range. ;

[0367] Output kinetic energy risk field and the corresponding longitudinal acceleration lateral acceleration yaw rate Adhesion coefficient Center of mass height Wheelbase The output data serves as the input for the fusion of comprehensive risk fields.

[0368] Step 5: Based on the geometric risk field and kinetic risk field of Step 3 and Step 4, construct the behavioral risk field: extrapolate the trajectory within the prediction time window, model the uncertainty around the trajectory using probability distribution, and superimpose the weights of each time window to obtain the behavioral risk field that reflects the uncertainty of driving behavior.

[0369] After the geometric risk field and kinetic risk field are constructed, the system further generates a behavioral risk field, as shown in the diagram below. Figure 4 As shown, this system reflects the evolution of the vehicle's future trajectory over time and the uncertainty of driving operations. First, based on the vehicle's current dynamic state and kinematic model, the system predicts possible driving trajectories within different time windows of 0.5 seconds, 1.0 seconds, and 2.0 seconds. Then, within each time window, the uncertainty surrounding the predicted trajectory is modeled as a Gaussian distribution, thus obtaining the risk probability of each grid point near the trajectory. The risk distributions of each time window are superimposed according to set weights to form a comprehensive risk field across multiple time scales, enabling short-term risks to reflect immediate dangers and long-term risks to provide early warnings. Finally, based on the driving intention recognition results (such as lane changing, braking, or acceleration behaviors), a correction coefficient is applied to the risk value in the corresponding direction to more closely approximate real driving behavior patterns. Through these steps, the behavioral risk field obtained by the system comprehensively reflects the short-term and long-term motion uncertainties of the vehicle and provides input for comprehensive risk assessment.

[0370] At any moment The system is based on the planar position of the vehicle's center of mass in the world coordinate system. The angle between the vehicle's longitudinal axis and the x-axis of the world coordinate system Longitudinal velocity lateral velocity yaw rate And step two, kinematic model to generate future trajectory Set multiple forecast time windows:

[0371]

[0372] Within each time window, use step size Second integration yields a series of trajectory points Mapped to raster ;

[0373] Considering the uncertainties caused by driving operations and environmental disturbances, the trajectory is modeled as a Gaussian distribution:

[0374]

[0375] In the formula, For grid points; In the time window Interior point prediction trajectory; For point To trajectory The minimum distance; The trajectory spread radius increases with the prediction time. The farther the trajectory is, the greater the uncertainty and the more "fuzzy" the steps become.

[0376] For each time window, the risk function is defined as:

[0377]

[0378] In the formula, For point In the time window The following behavioral risks; As time window weight, satisfying ;

[0379] Taking into account both short-term and long-term forecasts, the risks of different time windows are superimposed:

[0380]

[0381] The result of risk aggregation is a risk distribution that unfolds over time. Short-term risks determine the accuracy, while long-term risks ensure early warning.

[0382] If the system detects a driving intention (such as changing lanes, braking, or overtaking), it weights the corresponding direction:

[0383]

[0384] In the formula, For correction factor, =0.2-0.5; For the indicator function, if point If it is in the expected direction of behavior, it is 1; otherwise, it is 0.

[0385] The revised risk is:

[0386]

[0387] right Perform 2D Gaussian smoothing (standard deviation) (meters), eliminating jagged edges and abrupt changes; then normalizing the entire field to This ensures direct integration with the geometric risk field and kinetic risk field; the output is:

[0388]

[0389] In the formula, This represents the behavioral risk value of a grid point, and the output data serves as the input for behavioral factors in the fusion of the comprehensive risk field.

[0390] Step 6: Construct a comprehensive risk field: Normalize and weight the geometric risk field, kinetic risk field, and behavioral risk field from Steps 3, 4, and 5 to obtain a comprehensive risk field;

[0391] After the geometric risk field, kinetic risk field, and behavioral risk field are calculated separately, the system enters the risk field fusion stage, which normalizes and weights risk factors from different sources in the same grid coordinate system to obtain a comprehensive risk field that can simultaneously reflect road geometric constraints, vehicle dynamic limits, and driving behavior uncertainties.

[0392] Since the geometric risk field, kinetic risk field, and behavioral risk field have different numerical ranges and dimensions, each risk field is first normalized:

[0393]

[0394]

[0395]

[0396] By normalizing, the value ranges of the three risk fields are ensured to be uniform. ;

[0397] Therefore, a weighted summation method is used to obtain the comprehensive risk:

[0398]

[0399] In the formula, geometric risk weight kinetic risk weight and behavioral risk weights satisfy ;

[0400] Considering the variance of the probability distribution in behavioral prediction, the risk field is corrected as follows:

[0401]

[0402] In the formula, Indicates behavioral risk at a point Uncertain variance, These are the weighting coefficients. = ;

[0403] The resulting comprehensive risk field is:

[0404]

[0405] In the formula, This represents the overall risk value for each grid point. This risk field will be used in the subsequent warning and judgment module to determine whether the vehicle is in a dangerous state.

[0406] Step 7: Collision Risk Assessment: Extract the comprehensive risk value from the area in front of the vehicle's path, calculate the maximum risk, average risk, and cumulative risk along the path, determine the risk level based on preset multi-level risk thresholds, and introduce a collision time margin to correct the risk level.

[0407] Get comprehensive risk field Afterwards, the system enters the collision risk assessment and threshold triggering phase; high-risk areas in front of the vehicle are extracted from the comprehensive risk distribution, risk indicators are calculated, and compared with set thresholds to trigger the corresponding collision warning level; within the vehicle's forward-looking range... ( Within a range of meters, capture the comprehensive risk field; select grid points located within the lane area. Forming candidate risk areas For the region The risk value within is statistically analyzed, and various indicators are defined:

[0408] Define the maximum risk value:

[0409]

[0410] Define the average risk value:

[0411]

[0412] Define the risk of the path integral ahead:

[0413]

[0414] In the formula, Predict vehicle trajectories;

[0415] Three risk thresholds are set: Mild risk threshold: Moderate risk threshold: Severe risk threshold: ;when or When the corresponding threshold is exceeded, different warning levels are triggered:

[0416] when The status is safe and there are no warnings.

[0417] when This is a Level 1 warning, alerting drivers to take precautions.

[0418] when This is a Level II warning, a strong warning, accompanied by audible and visual alerts;

[0419] when This is a Level 3 warning, indicating a serious danger, triggering braking or intervention.

[0420] Further adjustments based on collision time margin (TTC):

[0421]

[0422] In the formula, The relative distance between the vehicle and the obstacle. For relative velocity; when , = Every second, the risk level is raised by one level; once the risk level is determined, the system triggers the corresponding warning action:

[0423] Level 1 warning: Human-machine interface prompts (visual + voice prompts);

[0424] Level 2 Warning: Increase audible and visual intensity, activate vibration alarm, and prepare for braking if necessary (the system pre-charges the braking system to prepare the braking actuators, thereby shortening the response time of potential Automatic Emergency Braking (AEB). This action does not directly cause vehicle deceleration, but serves as a preparatory measure before Level 3 risk triggers AEB, improving system reaction speed and safety).

[0425] Level 3 warning: Directly triggers the active safety system (AEB / ESC), active braking / steering;

[0426] The output result is:

[0427]

[0428] In the formula, These correspond to no risk, mild risk, moderate risk, and severe risk, respectively.

[0429] like Figure 5 As shown, step eight, active intervention strategy generation: Based on the risk level obtained in step seven, generate intervention commands for braking, steering, or a combination of both, and return to step one to continuously update the vehicle status and risk field according to the sampling cycle to achieve real-time early warning and closed-loop control.

[0430] Based on the risk level and predicted trajectory, appropriate longitudinal, lateral, or combined intervention measures are dynamically selected to minimize collision risk and maintain vehicle stability.

[0431] Based on the output of step six Define control objectives at different levels:

[0432] Level 1 warning Only visual and voice prompts are given to the driver; no vehicle control is performed.

[0433] Level II Warning Increase the intensity of sound and light, activate the vibration alarm, and maintain a safe distance and reduce speed if necessary;

[0434] Level III Warning Take immediate and forceful measures, including emergency braking / steering;

[0435] If the predicted trajectory carries a risk of forward collision, i.e. If the value is less than the set threshold, a vertical control strategy is implemented:

[0436] Define braking intervention:

[0437]

[0438] In the formula, Slowing down demand, This is the adhesion limit;

[0439] Define the safe distance model:

[0440]

[0441] If the actual vehicle distance If so, braking will be triggered;

[0442] If the risk field is concentrated in the direction of the curve or laterally, then a lateral control strategy should be formulated:

[0443] By correcting the steering to bring the vehicle back closer to the centerline, directional intervention is defined.

[0444]

[0445] In the formula, This is the lateral deviation. For heading error, , To control the gain;

[0446] In practical applications, vertical and horizontal interventions need to be coordinated:

[0447] like If the obstacle originates from a forward obstacle, braking should be prioritized.

[0448] like If the obstacle originates from a lateral / boundary obstacle, then steer first.

[0449] If both exceed the threshold simultaneously, comprehensive intervention will be implemented, that is, braking and steering will be performed simultaneously, and stability will be ensured through ESC / ABS.

[0450] When outputting intervention commands, the system must impose stability constraints:

[0451]

[0452] In the formula, The rollover index is used; when it is about to exceed the limit, longitudinal braking intensity should be reduced or steering angle should be limited first.

[0453] The output is a set of intervention instructions:

[0454]

[0455] In the formula, This is a longitudinal acceleration command; a negative value indicates braking. For steering angle command, {No intervention, braking only, steering only, braking + steering}. This output is sent through the vehicle's actuators (braking system, steering actuator, ESC / ABS module) to achieve active safety intervention.

[0456] The system uses a sampling period milliseconds (corresponding frequency) The system continuously acquires vehicle sensor data (speed, acceleration, yaw rate, steering wheel angle), environmental data (lane lines, road curvature, obstacle information), and actuator status (braking, steering feedback). All data is synchronized through a unified timestamp to form a real-time input stream.

[0457] Within each cycle, the vehicle state, pose, and road geometry parameters are updated; the geometric risk field, kinetic risk field, and behavioral risk field are recalculated; and the risk fields are fused to obtain the comprehensive risk field. It also extracts risk areas in real time, calculates risk indicators and compares them with thresholds, and updates the current risk level. New longitudinal intervention commands are generated based on risk levels and road conditions. and lateral intervention commands This is achieved through the braking and steering systems.

[0458] The system checks the performance in each cycle, verifying whether the longitudinal braking effect achieves the expected deceleration, whether the lateral steering response meets the set angle, and monitoring the lateral acceleration. Rollover index This ensures stability does not exceed limits. If the actuator does not perform as expected or malfunctions, the system triggers a degradation strategy, which reduces the control amplitude to maintain minimum safety. This means keeping the vehicle driving stably within the current lane, avoiding further acceleration or aggressive maneuvers, until the risk is eliminated or the driver takes over.

[0459] The above process is repeated in each sampling period, forming a complete closed loop. The system will adaptively adjust parameters (such as risk threshold and weight) based on historical data. Gain coefficient This improves early warning sensitivity and reduces false alarms. It outputs real-time risk levels. Intervention model The system also predicts trajectories and stores risk field data and intervention records in logs for subsequent verification, optimization, and algorithm retraining.

[0460] like Figure 6 As shown, a system for implementing a curve collision warning method based on multi-risk field fusion includes: a data acquisition module to acquire vehicle dynamics data and road environment information, providing an input basis for risk assessment; a state prediction module to perform state estimation and short-term prediction on the acquired data based on the vehicle dynamics model, outputting the vehicle's future position, attitude, and trajectory; a road geometry modeling and rasterization module to reconstruct the road centerline based on perceived lane lines, boundary lines, and road curvature information, calculate road curvature, lane width, and lateral deviation, and generate a rasterized representation of the road geometry model; a geometric risk field construction module to calculate lateral deviation, boundary distance, and road curvature risk factors based on the road geometry model, establishing a geometric risk field reflecting road geometric constraints; a kinetic energy risk field construction module to construct a kinetic energy risk field reflecting the vehicle's dynamic limits based on factors such as vehicle longitudinal and lateral acceleration, yaw rate, road adhesion coefficient, and vehicle center of gravity height; and a behavioral risk... The system comprises several modules: a risk field construction module, a risk field construction module, and a risk field fusion module. The former performs trajectory extrapolation within multiple prediction time windows, modeling the uncertainties around the trajectory in probabilistic form to generate a behavioral risk field reflecting differences in driving behavior and operational stability. The latter normalizes and weights the geometric risk field, kinetic risk field, and behavioral risk field to form a comprehensive risk distribution for subsequent risk level assessment. The former calculates risk indicators based on the comprehensive risk distribution and compares them with multi-level thresholds to determine the current risk level. The latter generates corresponding longitudinal braking or lateral steering intervention commands based on the risk level, enabling active safety control of the vehicle in curve scenarios. The former performs system looping and real-time updating, periodically sampling vehicle status data and road environment information, and updating the risk field and intervention commands in real time to achieve closed-loop operation and dynamic control of the system. All modules are connected via data communication to achieve collision risk prediction and active safety control of the vehicle in curve scenarios.

Claims

1. A method for cornering collision warning based on multi-risk field fusion, characterized in that: Includes the following steps: Step 1: Collect vehicle dynamics data in real time and make short-term predictions of vehicle position, attitude and trajectory based on the vehicle dynamics model; Step 2: Road geometry modeling and rasterization; Step 3: Construct a geometric risk field on the rasterized road geometry model in Step 2: Calculate the road curvature, lane width and lateral deviation, and generate a geometric risk distribution using the lateral deviation, boundary distance and road curvature as inputs to obtain the geometric risk field; Step 4: Construct a kinetic risk field based on the geometric risk field in Step 3: Based on the vehicle's dynamic data, road surface adhesion conditions, and vehicle center of gravity height, adopt a spatial distribution with the vehicle as a reference and superimpose the risk amplification factor of the working condition to reflect the dynamic limit risk under different working conditions, thus forming a kinetic risk field. Step 5: Based on the geometric risk field and kinetic risk field of Step 3 and Step 4, construct the behavioral risk field: extrapolate the trajectory within the prediction time window, model the uncertainty around the trajectory using probability distribution, and superimpose the weights of each time window to obtain the behavioral risk field that reflects the uncertainty of driving behavior. Step 6: Construct a comprehensive risk field: Normalize and weight the geometric risk field, kinetic risk field, and behavioral risk field from Steps 3, 4, and 5 to obtain a comprehensive risk field; Step 7: Collision Risk Assessment: Extract the comprehensive risk value from the area in front of the vehicle's path, calculate the maximum risk, average risk, and cumulative risk along the path, determine the risk level based on preset multi-level risk thresholds, and introduce a collision time margin to correct the risk level. Step 8: Active intervention strategy generation: Based on the risk level obtained in Step 7, generate intervention commands for braking, steering, or a combination of both, and return to Step 1 to continuously update the vehicle status and risk field according to the sampling cycle, so as to achieve real-time early warning and closed-loop control.

2. The method for cornering collision warning based on multi-risk field fusion according to claim 1, characterized in that: Step one specifically involves: Obtain vehicle dynamics data: Acquire vehicle bus and sensor data: vehicle speed Longitudinal acceleration lateral acceleration yaw rate Steering wheel angle ; Obtain road information: Road centerline curvature Lane width Inner and outer boundary curves, road surface adhesion coefficient ; Obtain the relative quantity with the vehicle in front: relative distance Relative velocity ; longitudinal acceleration lateral acceleration yaw rate Perform Kalman filtering to correct the sensor's output error in the zero-input state and unify the timestamps; Vehicle dynamics model was selected to unify the time. speed Longitudinal acceleration lateral acceleration yaw rate Steering wheel angle and total vehicle mass Yaw moment of inertia Distance from center of gravity to front axle and the distance from the center of gravity to the rear axle Front axle equivalent lateral stiffness and rear axle equivalent lateral stiffness For input, Establish system state vector ,in For longitudinal velocity, For the lateral velocity, the slip angle and lateral force are calculated using a linear tire model, derived from the dynamic equations with step sizes. Numerical integration yields In the world coordinate system, the pose update equation is used. Down propagation pose ,in The vehicle's heading angle is given by a rotation matrix. Complete vehicle coordinates and Alignment: Before calculating the lateral forces of the front and rear wheels, the tire slip angle is first calculated based on the velocity decomposition in the vehicle's center of gravity coordinate system. The front wheel contact point in the vehicle coordinate system The velocity below is expressed as: ； ； The rear wheel contact point speed is: ； ； in, For longitudinal velocity, For lateral velocity, The yaw rate is angular velocity. and These are the distances from the center of mass to the front and rear axles, respectively. Based on the small-angle linearization assumption The slip angles of the front and rear wheels are defined as follows: ； ； in, This represents the front wheel steering angle; the above expression reflects the angle between the local velocity direction of the tire and the tire's pointing direction. After the slip angle is calculated, the tire lateral force is calculated using a linear tire model: ； ； in, and These represent the lateral stiffness of the front and rear wheels, respectively; a negative sign indicates that the lateral force of the tire always acts in the direction that reduces the lateral slip angle. Define the longitudinal dynamic equation: ； In the formula, For the overall vehicle quality, The rate of change of longitudinal velocity, Projection of longitudinal force on the front wheel This is the projection of the lateral force on the front wheel. For the longitudinal force of the rear wheel, For air resistance, For rolling resistance; Define the transverse dynamic equation: ； In the formula, For the vehicle body The actual acceleration in the axial direction, This is the projection of the lateral force on the front wheel. The lateral force is from the rear wheel. Projection of longitudinal force on the front wheel; Define the yaw moment balance equation: ； In the formula, Let be the yaw moment of inertia of the vehicle about its center of mass. This is the yaw acceleration. This is the distance from the center of gravity to the front axle. It is the resultant lateral force component of the front wheel. This is the distance from the center of mass to the rear axle. This refers to the lateral force on the rear wheel; Define derived quantity : ； Define speed limit: ； After obtaining the vehicle's velocity state and performing amplitude limiting, the vehicle's dynamic state is transformed into a motion trajectory in the world coordinate system through the pose update equation; the vehicle's kinematic update equation in the global coordinate system is: ； ； ； in, Let this be the position of the vehicle's center of mass in the world coordinate system. For the vehicle's heading angle, For longitudinal velocity, For lateral velocity, This refers to the yaw rate; Discretizing the above kinematic equations using the Euler method yields: ； ； ； in, The discrete time step is used; through the above pose update process, the predicted position and heading of the vehicle at the next moment are obtained, providing input for subsequent short-term trajectory prediction and behavioral risk field construction; Will Projected onto the center line of the road quantity ,in Let the coordinates be the arc length. For lateral error, For heading error, and the current curvature of the road centerline. Using the same model, extrapolate the trajectory within a 2-3s prediction time window, applying physical constraints and velocity limits during the prediction process; In the discrete system, the first Time points Output and derivatives lateral acceleration , combined With 2-3 second predicted trajectory It includes a quality flag indicating whether physical constraints / speed limits have been triggered, which can be used for subsequent risk field construction and early warning determination. in, Geometric state quantities of the vehicle in the Frenet curve coordinate system of the road; variables This indicates the arc length position of the vehicle on the center line of the road; Indicates the lateral deviation of the vehicle relative to the centerline; This indicates the deviation between the vehicle's heading angle and the road tangential angle; This indicates the curvature of the road at that location.

3. The method for cornering collision warning based on multi-risk field fusion according to claim 1, characterized in that: Step two specifically involves: At the time of unity The system reconstructs the road centerline based on online lane detection results and uses spline interpolation to obtain a continuous curve. ; To ensure numerical stability, the centerline is spaced according to arc length intervals. The system resamples the centerline points. After resampling, the system smooths the resampled centerline point set using Gaussian smoothing filtering. The centerline point set is a discrete sequence of road centerline points. Finally, the road tangential angle was calculated. , Road tangent direction vector and curvature This is used for the subsequent construction of the geometric risk field. Road tangent angle: ； In the formula, For the road at the arc length position The tangential angle at that point, i.e., the heading of the centerline. The first derivative of the curve indicates the direction of the tangent. Road curvature: ； Tangential vector: ； In the formula, It is the tangent vector, pointing in the direction of the road's movement; Normal vector: ； In the formula, It is the normal vector, perpendicular to the road direction, and points to the left side of the road; According to lane width , thus obtaining the inner and outer boundaries: ； ； In the formula, The coordinates of the inner boundary of the road. The coordinates of the outer boundary of the road. Lane width; Covering the centerline and extending the forward sight distance outwards Establish a regular grid within the rectangular area .

4. The method for cornering collision warning based on multi-risk field fusion according to claim 1, characterized in that: Step three specifically involves: For each grid point The system uses the nearest point projection method to calculate its arc length position on the center line. This is used for subsequent curvature and risk value mapping calculations. ； In the formula, For point The projection position on the center line, i.e., the arc length. Choose the one that minimizes the distance. , The distance from the point to the center line; Therefore, the lateral deviation is calculated: ； In the formula, For lateral deviation, vehicle or point lateral distance relative to the centerline The normal direction is used to calculate the offset. Lateral distance from the vehicle to the lane edge: ； In the formula, For point Distance to the nearest lane boundary It is half the width of the lane. The absolute lateral deviation of a point relative to the centerline; Road mask ,when ,otherwise ,in This is for redundant width; The output data is: ； The output data is used as input for constructing the geometric risk field and subsequent risk fusion; For each grid point Based on the lateral deviation obtained in step three Define the horizontal risk quantity: ； In the formula, This is an adjustment factor used to control the sensitivity of risk to changes in lateral deviation; Based on the distance within the band Define the boundary risk components: ； In the formula, The attenuation coefficient; Based on the road curvature of the corresponding projection point Define curvature risk: ； In the formula, For reference curvature, Controlling sensitivity; The three components are normalized and then weighted and summed: ； In the formula, the weights , For points not within the road, i.e. The risk field value is directly set to zero or marked as invalid; only the risk values ​​of grid points within the road are retained. The output data is: ； In the formula, This represents the geometric risk value of each grid point, and the output data serves as the geometric factor input for the overall risk field.

5. The method for cornering collision warning based on multi-risk field fusion according to claim 1, characterized in that: Step four specifically involves: With the vehicle's center of mass as the origin, the vehicle coordinate system For reference, define the elliptic kernel function: ； In the formula, These are the coordinates of a point in the vehicle's coordinate system. Indicates the longitudinal scale. Indicates the longitudinal scale; Based on the vehicle's motion conditions, the longitudinal direction window is superimposed with an elliptical kernel, emphasizing the front during braking and the rear during driving. Simultaneously superimposed curve direction window , Based on the vehicle's lateral acceleration Define the lateral acceleration gain: ； In the formula, meters per square second This is the lateral acceleration gain coefficient. Sensitivity index; Based on yaw rate Define the yaw rate gain: ； In the formula, For reference yaw rate, This is the yaw gain coefficient. Sensitivity index; Based on longitudinal acceleration Define the longitudinal acceleration gain: ； In the formula, The longitudinal acceleration gain coefficient, Sensitivity index; Introducing the road surface adhesion coefficient Define the attachment utilization rate: ； ； In the formula, As a risk amplification factor under adhesion conditions, For the attachment correction gain coefficient, Sensitivity index; Considering the height of the center of mass With wheelbase Define the rollover index: ； ； In the formula, As a factor amplifying the risk of rollover, This is the side-flip gain coefficient. Sensitivity index; Multiplying each gain factor by the elliptic kernel yields the kinetic risk field: ； In the formula, This indicates a trimming operation, ensuring the risk value is within a certain range. ; Output kinetic energy risk field and the corresponding longitudinal acceleration lateral acceleration yaw rate Adhesion coefficient Center of mass height Wheelbase The output data serves as the input for the fusion of comprehensive risk fields.

6. The method for cornering collision warning based on multi-risk field fusion according to claim 1, characterized in that: Step five specifically involves: At any moment The system is based on the planar position of the vehicle's center of mass in the world coordinate system. The angle between the vehicle's longitudinal axis and the x-axis of the world coordinate system Longitudinal velocity lateral velocity yaw rate And step two, kinematic model to generate future trajectory Set multiple forecast time windows: ； Within each time window, use step size Second integration yields a series of trajectory points Mapped to raster ; Considering the uncertainties caused by driving operations and environmental disturbances, the trajectory is modeled as a Gaussian distribution: ； In the formula, For grid points; In the time window Interior point prediction trajectory; For point To trajectory The minimum distance; The radius of the trajectory spread. For each time window, the risk function is defined as: ； In the formula, For point In the time window The following behavioral risks; As time window weight, satisfying ; Taking into account both short-term and long-term forecasts, the risks of different time windows are superimposed: ； If the system detects a driving intention, it will weight the direction accordingly: ； In the formula, This is a correction factor; For the indicator function, if point If it is in the expected direction of behavior, it is 1; otherwise, it is 0. The revised risk is: ； right Perform two-dimensional Gaussian smoothing to eliminate jagged edges and abrupt changes; then normalize the entire field to... This ensures direct integration with the geometric risk field and the kinetic risk field; The output is: ； In the formula, This represents the behavioral risk value of a grid point, and the output data serves as the input for behavioral factors in the fusion of the comprehensive risk field.

7. The method for cornering collision warning based on multi-risk field fusion according to claim 1, characterized in that: Step six specifically involves: Since the geometric risk field, kinetic risk field, and behavioral risk field have different numerical ranges and dimensions, each risk field is first normalized: ； ； ； By normalizing, the value ranges of the three risk fields are ensured to be uniform. ; Therefore, a weighted summation method is used to obtain the comprehensive risk: ； In the formula, geometric risk weight kinetic risk weight and behavioral risk weights satisfy ; Considering the variance of the probability distribution in behavioral prediction, the risk field is corrected as follows: ； In the formula, Indicates behavioral risk at a point Uncertain variance, These are the weighting coefficients; The resulting comprehensive risk field is: ； In the formula, This represents the overall risk value for each grid point.

8. The method for cornering collision warning based on multi-risk field fusion according to claim 1, characterized in that: Step seven specifically involves: Get comprehensive risk field Then, the system enters the collision risk assessment and threshold triggering stage; high-risk areas in front of the vehicle are extracted from the comprehensive risk distribution, risk indicators are calculated and compared with the set threshold, thereby triggering the corresponding collision warning level. Within the vehicle's forward visibility range Within, capture the comprehensive risk field; Select grid points located within the lane area Forming candidate risk areas For the region The risk value within is statistically analyzed, and various indicators are defined: Define the maximum risk value: ； Define the average risk value: ； Define the risk of the path integral ahead: ； In the formula, Predict vehicle trajectories; Three risk thresholds are set: Mild risk threshold: Moderate risk threshold: Severe risk threshold: ;when or When the corresponding threshold is exceeded, different warning levels are triggered: when The status is safe and there are no warnings. when This is a Level 1 warning, alerting drivers to take precautions. when This is a Level II warning, a strong warning, accompanied by audible and visual alerts; when This is a Level 3 warning, indicating a serious danger, triggering braking or intervention. Further adjustments based on collision time margin (TTC): ； In the formula, The relative distance between the vehicle and the obstacle. For relative velocity; when , = Every second, the risk level is raised by one level; once the risk level is determined, the system triggers the corresponding warning action. The output result is: ； In the formula, These correspond to no risk, mild risk, moderate risk, and severe risk, respectively.

9. The method for cornering collision warning based on multi-risk field fusion according to claim 1, characterized in that: Step eight specifically involves: Based on the risk level and predicted trajectory, appropriate longitudinal, lateral, or combined intervention measures are dynamically selected to minimize collision risk and maintain vehicle stability. Based on the output of step six Define control objectives at different levels: Level 1 warning It only alerts the driver and does not take control of the vehicle; Level II Warning Maintain a safe distance and slow down; Level III Warning Take immediate mandatory measures, such as emergency braking or steering; If the predicted trajectory carries a risk of forward collision, i.e. If the value is less than the set threshold, a vertical control strategy is implemented: Define braking intervention: ； In the formula, Slowing down demand, This is the adhesion limit; Define the safe distance model: ； If the actual vehicle distance If so, braking will be triggered; If the risk field is concentrated in the direction of the curve or laterally, then a lateral control strategy should be formulated: By correcting the steering to bring the vehicle back closer to the centerline, directional intervention is defined. ； In the formula, This is the lateral deviation. For heading error, , To control the gain; When outputting intervention commands, the system must impose stability constraints: ； In the formula, The rollover index; The output is a set of intervention instructions: ； In the formula, This is a longitudinal acceleration command; a negative value indicates braking. For steering angle command, {No intervention, braking only, steering only, braking + steering}.

10. A system for implementing the curve collision warning method based on multi-risk field fusion as described in any one of claims 1-9, characterized in that, include The data acquisition module obtains vehicle dynamics data and road environment information, providing an input basis for risk assessment. The state prediction module performs state estimation and short-term prediction on the collected data based on the vehicle dynamics model, and outputs the vehicle's future position, attitude and trajectory. The road geometry modeling and rasterization module reconstructs the road centerline based on the perceived lane lines, boundary lines and road curvature information, calculates the road curvature, lane width and lateral deviation, and generates a rasterized representation of the road geometry model. The geometric risk field construction module calculates lateral deviation, boundary distance, and road curvature risk factors based on the road geometric model, and establishes a geometric risk field that reflects the road's geometric constraints. The kinetic energy risk field construction module constructs a kinetic energy risk field that reflects the vehicle's dynamic limits based on factors such as the vehicle's longitudinal and lateral acceleration, yaw rate, road adhesion coefficient, and vehicle center of gravity height. The behavioral risk field construction module extrapolates the trajectory within multiple prediction time windows, models the uncertainty around the trajectory in probabilistic form, and generates a behavioral risk field that reflects differences in driving behavior and operational stability. The comprehensive risk field fusion module normalizes and weights the geometric risk field, kinetic risk field and behavioral risk field to form a comprehensive risk distribution for subsequent risk level assessment. The risk level determination module calculates risk indicators based on the comprehensive risk distribution and compares them with multi-level thresholds to determine the current risk level. The active intervention module generates corresponding longitudinal braking or lateral steering intervention commands based on the risk level, enabling active safety control of the vehicle in curve scenarios. The system's loop and real-time update module periodically samples vehicle status data and road environment information, and updates risk fields and intervention commands in real time to achieve closed-loop operation and dynamic control of the system. The modules are connected via data communication to enable collision risk prediction and active safety control of vehicles in curve scenarios.

Images