Intelligent vehicle emergency obstacle avoidance braking control method and system

Abstract

The application discloses a kind of intelligent vehicle emergency obstacle avoidance brake control method and system, it is related to automatic driving technical field, including: obtaining historical vehicle driving data and historical vehicle user monitoring data, constructs driving behavior analysis model, assesses the reaction ability of vehicle driving user under collision warning;Obtain historical emergency obstacle avoidance brake control trigger scene big data and the collision warning reaction ability of vehicle driving user, construct personalized dynamic braking redundancy space evaluation model, generate the collision warning reaction personalized redundancy space of vehicle driving user;Real-time vehicle surrounding environment data is acquired, with the collision warning reaction personalized redundancy space of vehicle driving user as radius, establish car-road environment collaborative risk field function, output vehicle travel dynamic risk field, establish vehicle emergency obstacle avoidance brake control mechanism, generate the decision of dynamic virtual tow rope.The beneficial effects of the application are: strong robustness is considered to deal with uncertainty, significantly improve the success rate of risk avoidance.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of automatic driving, in particular to an intelligent vehicle emergency obstacle avoidance braking control method and system. BACKGROUND

[0002] Current intelligent vehicle emergency obstacle avoidance braking control schemes generally rely on predefined fixed thresholds or decision mechanisms based on traditional perception models. Such methods are not adaptive enough in dynamic complex scenarios: they have poor robustness to sensor noise, extreme weather or sudden obstacle shape changes, which can easily lead to misjudgment or response delay. On the other hand, traditional control strategies are mostly based on linear assumptions and cannot accurately handle the nonlinear characteristics of vehicle-environment coupling. Moreover, they do not fully integrate real-time road conditions and driver intentions in collaborative decision-making, resulting in a mismatch between braking timing or force and actual risk levels, which reduces obstacle avoidance efficiency and even causes secondary accidents. SUMMARY

[0003] To solve the above technical problems, an intelligent vehicle emergency obstacle avoidance braking control method and system are provided, which solve the above problems.

[0004] To achieve the above purposes, the technical scheme adopted by the application is as follows:

[0005] An intelligent vehicle emergency obstacle avoidance braking control method comprises the following steps:

[0006] S1, based on the ECU database and the ADAS database, historical vehicle driving data and historical vehicle user monitoring data are obtained, the user state under the collision warning of the vehicle driving is analyzed, a driving behavior analysis model is constructed, and the reaction ability of the vehicle driving user under the collision warning is evaluated;

[0007] S2, based on the Internet and the vehicle enterprise background database, historical emergency obstacle avoidance braking control trigger scene big data and collision warning reaction ability of the vehicle driving user are obtained, a personalized dynamic braking redundancy space evaluation model is constructed, and a personalized redundancy space of the collision warning reaction of the vehicle driving user is generated;

[0008] S3, real-time vehicle surrounding environment data is obtained, the personalized redundancy space of the collision warning reaction of the vehicle driving user is taken as a radius, a vehicle-road environment collaborative risk field function is established, a vehicle travel dynamic risk field is output, a vehicle emergency obstacle avoidance braking control mechanism is established, and a dynamic virtual traction rope decision is generated.

[0009] Preferably, step S1 specifically comprises:

[0010] Based on the ECU database and the DSM database, historical vehicle driving data and historical vehicle user monitoring data are obtained;

[0011] Data cleaning and time alignment preprocessing are performed on historical vehicle driving data and historical vehicle user monitoring data;

[0012] Based on a sliding window, the collision warning trigger time in the historical vehicle driving data is taken as an anchor point, the historical vehicle driving data and the historical vehicle user monitoring data in the unit time window under each collision anchor point are intercepted, and historical vehicle driving collision warning-user behavior feature data is established;

[0013] PCA principal component analysis is used to reduce the dimension of the historical vehicle driving collision warning-user behavior feature data;

[0014] According to the historical vehicle driving collision warning-user behavior feature data, CUSUM point change monitoring is used to identify and extract the mutation point on the driving user brake pedal stroke time sequence from the historical vehicle driving collision warning trigger unit timestamp, and the user reaction time feature under the historical vehicle driving collision warning event is extracted;

[0015] According to the historical vehicle driving collision warning-user behavior feature data, the driver state feature in the historical vehicle user monitoring data under the historical vehicle driving collision warning event unit time node is extracted according to the historical vehicle driving collision warning trigger unit timestamp;

[0016] According to the correlation coefficient, the correlation coefficient between the driver state feature vector in the historical vehicle user monitoring data under the historical vehicle driving collision warning event unit time node and the user reaction time feature vector under the historical vehicle driving collision warning event is verified, which is recorded as the driving fatigue reaction time increment factor;

[0017] According to the Gaussian process regression, a driving behavior analysis model is established, the historical vehicle driving collision warning-user behavior feature data is taken as the input, the driving fatigue reaction time increment factor is taken as the intervention condition, the user reaction time probability distribution under the influence of the driver state feature in the historical vehicle user monitoring data is verified according to the maximum likelihood estimation, and the vehicle driving user reaction ability under the collision warning is obtained.

[0018] Preferably, step S2 specifically comprises:

[0019] Based on historical emergency obstacle avoidance braking control trigger scene big data, each emergency obstacle avoidance braking control trigger scene type and vehicle speed interval is marked, the braking redundancy space mean and variance under each emergency obstacle avoidance braking control trigger scene type and vehicle speed interval are verified according to the Bayesian prior, and the braking redundancy space prior knowledge of several emergency obstacle avoidance braking control trigger scenes is generated by substituting the probability density function of normal distribution; The braking redundancy space includes: time redundancy and distance redundancy;

[0020] According to the braking redundancy space prior knowledge of several emergency obstacle avoidance braking control trigger scenes, the physical parameters in the corresponding emergency obstacle avoidance braking control trigger scenes are labeled, the Bayesian likelihood function is substituted, the braking redundancy space of several emergency obstacle avoidance braking control trigger scenes is taken as the observation condition, and the observation probability distribution of the physical parameters in the emergency obstacle avoidance braking control trigger scene under the given observation condition is verified.

[0021] According to the basic braking redundancy space prior knowledge of several emergency obstacle avoidance braking control trigger scenes and the observation probability distribution of the physical parameters in the emergency obstacle avoidance braking control trigger scene under the given observation condition, the Bayesian posterior is substituted, and the benchmark braking redundancy space posterior knowledge of several emergency obstacle avoidance braking control trigger scenes is generated.

[0022] According to the benchmark braking redundancy space posterior knowledge of several emergency obstacle avoidance braking control trigger scenes, the collision warning reaction behavior actions of the vehicle driving user in the several emergency obstacle avoidance braking control trigger scenes are labeled, and the benchmark braking redundancy space associated collision warning reaction behavior action feature data of the several emergency obstacle avoidance braking control trigger scenes is assembled.

[0023] Preferably, step S2 further comprises:

[0024] Based on SVR support vector regression, an individualized dynamic braking redundancy space evaluation model is established;

[0025] The benchmark braking redundancy space reaction behavior hyperplane boundary of each emergency obstacle avoidance braking control trigger scene is generated by training the individualized dynamic braking redundancy space evaluation model using the benchmark braking redundancy space associated collision warning reaction behavior action feature data of several emergency obstacle avoidance braking control trigger scenes.

[0026] According to the Euclidean distance formula, the minimum value of the spatial distance between the collision warning reaction ability of the vehicle driving user and the benchmark braking redundancy space reaction behavior hyperplane boundary of each emergency obstacle avoidance braking control trigger scene is calculated, and the collision warning reaction individualized redundancy space of the vehicle driving user is determined.

[0027] Preferably, step S3 specifically comprises:

[0028] Based on the vehicle multi-source heterogeneous sensor, real-time vehicle surrounding environment data is obtained, the collision warning reaction individualized redundancy space of the vehicle driving user is taken as the radius, and the initial measurement vector of the tracking target within the collision warning reaction individualized redundancy space of the vehicle driving user is labeled; the initial measurement vector of the tracking target includes position, velocity, and acceleration.

[0029] Based on constant rotation rate and acceleration motion model, the state transition scheme of the tracking target is constructed, the initial measurement vector of the tracking target in the personalized redundancy space of the collision warning response of the vehicle driving user is taken as the input, the state trend of the tracking target at the future unit timestamp is predicted, the Jacobian matrix of the state trend of the tracking target is calculated, and the state trend covariance matrix of the tracking target in the personalized redundancy space of the collision warning response of the vehicle driving user is obtained.

[0030] By using the Ralman gain, the optimal estimation of the uncertainty influence of the real-time vehicle surrounding environment data on the state trend covariance matrix of the tracking target in the personalized redundancy space of the collision warning response of the vehicle driving user is calculated, the real-time state trend covariance matrix of the tracking target in the personalized redundancy space of the collision warning response of the vehicle driving user is obtained, and the future step trajectory of the tracking target in the personalized redundancy space of the collision warning response of the vehicle driving user is determined.

[0031] Based on the Cartesian coordinate system, according to the future step trajectory of the tracking target in the personalized redundancy space of the collision warning response of the vehicle driving user, the longitudinal direction of the personalized redundancy space coordinate of the collision warning response of the vehicle driving user is taken as the X axis, and the transverse direction is taken as the Y axis, the tracking target risk field coordinate system in the personalized redundancy space of the collision warning response of the vehicle driving user is constructed.

[0032] According to the tracking target risk field coordinate system in the personalized redundancy space of the collision warning response of the vehicle driving user, a two-dimensional elliptical Gaussian function is used to establish the tracking target risk diffusion field in the personalized redundancy space of the collision warning response of the vehicle driving user, the long axis is along the obstacle motion direction, and the short axis is perpendicular to the motion direction.

[0033] According to the tracking target risk diffusion field in the personalized redundancy space of the collision warning response of the vehicle driving user, the angle between the obstacle velocity direction and the X axis is calculated, the coordinates of the field midpoint are rotated relative to the obstacle position, the major axis of the ellipse is consistent with the velocity direction, the obstacle motion direction and the risk diffusion degree in the obstacle motion direction are determined, and the elliptical coefficient of the two-dimensional elliptical Gaussian distribution of the tracking target risk diffusion field in the personalized redundancy space of the collision warning response of the vehicle driving user is determined.

[0034] The obstacle motion direction and the risk diffusion degree in the obstacle motion direction are linearly superimposed to obtain the total obstacle risk field, the elliptical coefficient of the two-dimensional elliptical Gaussian distribution of the tracking target risk diffusion field in the personalized redundancy space of the collision warning response of the vehicle driving user is applied to the two-dimensional elliptical Gaussian function, the car-road environment cooperative risk field function is constructed, the risk value of each obstacle in the total obstacle risk field is obtained, and the vehicle driving dynamic risk field is established.

[0035] Preferably, step S3 further comprises:

[0036] Based on the Khatib artificial potential field method, the real-time vehicle travel target path attractive field and the obstacle repulsive field in the vehicle travel dynamic risk field are marked;

[0037] Based on the collision warning response personalized redundancy space of the vehicle driving user, the driving behavior characteristic parameters of the vehicle driving user are marked, the conservative type, the ordinary type and the aggressive type driving clustering clusters are preset according to the K-Means clustering algorithm, the driving behavior characteristics are clustered and divided, and the driving category attribution cluster of the vehicle driving user is determined;

[0038] The centroid of the driving category attribution cluster of the vehicle driving user is normalized to determine the driving category index of the vehicle driving user;

[0039] According to the driving category attribution cluster of the vehicle driving user, the personalized instant risk coefficient of the vehicle driving user is calculated, the real-time vehicle travel target path attractive field and the obstacle repulsive field in the vehicle travel dynamic risk field are dynamically compensated to obtain the real-time vehicle travel target path personalized attractive field and the obstacle personalized repulsive field in the vehicle travel dynamic risk field, and the method is as follows:

[0040] ;

[0041] Wherein, is the real-time vehicle travel target obstacle personalized repulsive field in the vehicle travel dynamic risk field, is the real-time vehicle travel target path personalized attractive field in the vehicle travel dynamic risk field, is the personalized instant risk coefficient of the vehicle driving user, is the driving category index of the vehicle driving user, is the collision warning response personalized redundancy space of the vehicle driving user, is the real-time distance between the vehicle and the obstacle, is the amplification coefficient, is the individual inhibition item weight, is the target abandonment rate, and ρ is the target persistence coefficient;

[0042] Based on the real-time vehicle travel target path individualized attractive force field and the obstacle individualized repulsive force field in the vehicle travel dynamic risk field, a tracking target virtual coupling force in the collision warning reaction individualized redundant space of the vehicle driving user is constructed, the vehicle emergency obstacle avoidance braking control direction is determined, the vehicle emergency obstacle avoidance braking control mechanism is established according to the MPPI model prediction path integral, the current position of the vehicle is taken as the starting point, the vehicle emergency obstacle avoidance braking control direction is taken, a plurality of future trajectories are generated, for each trajectory, the minimum real-time vehicle travel target path individualized attractive force field and the real-time vehicle travel target obstacle individualized repulsive force field are taken as the safety cost and the target error cost, the optimal future trajectory is determined by weighted average solving according to the total cost of each path, and the dynamic virtual traction rope decision is generated by substituting the MPC trajectory tracking controller.

[0043] Further, an intelligent vehicle emergency obstacle avoidance braking control system comprises:

[0044] The capability evaluation module obtains historical vehicle driving data and historical vehicle user monitoring data based on the ECU database and the ADAS database, analyzes the user state under the collision warning of the historical vehicle driving, constructs a driving behavior analysis model, and evaluates the reaction capability of the vehicle driving user under the collision warning.

[0045] The redundant space division module obtains historical emergency obstacle avoidance braking control triggering scene big data and the collision warning reaction capability of the vehicle driving user based on the Internet and the vehicle enterprise background database, constructs a personalized dynamic braking redundant space evaluation model, and generates the collision warning reaction individualized redundant space of the vehicle driving user.

[0046] The traction traction module obtains real-time vehicle surrounding environment data, takes the collision warning reaction individualized redundant space of the vehicle driving user as the radius, establishes a vehicle-road environment cooperative risk field function, outputs a vehicle travel dynamic risk field, establishes a vehicle emergency obstacle avoidance braking control mechanism, and generates a dynamic virtual traction rope decision.

[0047] Compared with the prior art, the beneficial effects of the present application are that:

[0048] The present application proposes an intelligent vehicle emergency obstacle avoidance braking control scheme. By dynamically evaluating the state of the driver and constructing a personalized safety redundant space, the safety protection from one-size-fits-all to person-specific is realized; by using the prospective risk field prediction and virtual traction rope decision mechanism, the traditional passive reaction is changed into active prevention, a smooth optimal path is generated by cooperative lateral obstacle avoidance and longitudinal generation, the strong robustness of coping with uncertainty is considered, and the risk avoidance success rate is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 It is an intelligent vehicle emergency obstacle avoidance braking control method flow chart;

[0050] Figure 2 A framework diagram of an intelligent vehicle emergency obstacle avoidance braking control system. DETAILED DESCRIPTION

[0051] The following description is provided to enable those skilled in the art to practice the invention. The preferred embodiments described below are only examples of the invention and other obvious variants can be thought of by those skilled in the art.

[0052] Referring to Figure 1 As shown in the figure, an intelligent vehicle emergency obstacle avoidance braking control method comprises:

[0053] S1, based on the ECU database and the ADAS database, obtaining historical vehicle driving data and historical vehicle user monitoring data, analyzing the user state under the historical vehicle driving collision warning, constructing a driving behavior analysis model, and evaluating the vehicle driving user reaction ability under the collision warning;

[0054] Step S1 specifically comprises:

[0055] Based on the ECU database and the DSM database, historical vehicle driving data and historical vehicle user monitoring data are obtained;

[0056] As further content, the historical vehicle driving data includes CAN bus data and CAN bus data, and the historical vehicle user monitoring data includes face orientation, eyelid opening, gaze direction, blink frequency, and head posture.

[0057] The historical vehicle driving data and the historical vehicle user monitoring data are preprocessed by data cleaning and time alignment;

[0058] Based on the sliding window, the collision warning trigger time in the historical vehicle driving data is taken as the anchor point, the historical vehicle driving data and the historical vehicle user monitoring data in the unit time window under each collision anchor point are intercepted, and the historical vehicle driving collision warning-user behavior feature data is assembled;

[0059] Using PCA principal component analysis method, the historical vehicle driving collision warning-user behavior feature data is processed by dimension reduction;

[0060] According to the historical vehicle driving collision warning-user behavior feature data, the CUSUM point change monitoring is used to identify and extract the mutation point on the driving user brake pedal stroke time sequence from the historical vehicle driving collision warning trigger unit timestamp, and the user reaction time feature under the historical vehicle driving collision warning event is extracted;

[0061] According to the historical vehicle driving collision warning-user behavior feature data, the driver state features in the historical vehicle user monitoring data at the historical vehicle driving collision warning event unit time node are extracted according to the historical vehicle driving collision warning trigger unit timestamp;

[0062] According to the correlation coefficient, the correlation coefficient between the driver state feature vector in the historical vehicle user monitoring data at the historical vehicle driving collision warning event unit time node and the user reaction time feature vector under the historical vehicle driving collision warning event is verified, which is recorded as the driving fatigue reaction time increment factor;

[0063] According to the Gaussian process regression, a driving behavior analysis model is established, the historical vehicle driving collision warning-user behavior feature data is taken as the input, the driving fatigue reaction time increment factor is taken as the intervention condition, the user reaction time probability distribution under the influence of the driver state feature in the historical vehicle user monitoring data is verified according to the maximum likelihood estimation, and the vehicle driving user reaction ability under the collision warning is obtained.

[0064] In use, the contents in the above steps are combined:

[0065] As further content, through a data-driven method, vehicle CAN bus data and driver state monitoring data are deeply fused, and the driver reaction time under the historical collision warning is accurately extracted by using the CUSUM algorithm, and then a probability model is constructed by Gaussian process regression to dynamically quantify the mapping relationship between the real-time physiological state (such as fatigue) of the driver and the reaction ability of the driver in an emergency. Realize the forward-looking and personalized safety protection, can predict the possible reaction delay of the driver according to the current state of the driver, and generate a dynamic, probabilistic and personalized redundant space accordingly, so that the subsequent emergency obstacle avoidance can be intervened earlier and more smoothly, significantly improving the success rate of risk avoidance, and breaking the traditional fixed threshold.

[0066] Further, as an embodiment of step S1:

[0067] A driver named Mr. Wang who is driving long distance at night is being served.

[0068] Step 1: Real-time data acquisition and preprocessing;

[0069] CAN data: vehicle speed 100km / h, distance to the front vehicle 60 meters.

[0070] DSM data: eyelid opening degree (PERCLOS) reaches 80% (indicating drowsiness), line of sight deviates from the road center for more than 2 seconds, and blink frequency abnormally increases.

[0071] Processing: Real-time cleaning and alignment of these data to form a current Mr. Wang state feature vector X_wang_now.

[0072] Step 2: State assessment and reaction capability prediction;

[0073] Input X_wang_now into the trained Gaussian Process Regression model.

[0074] The model queries the historical database and finds that in historical cases with similar state features as Wang's current state, the reaction time of the driver is generally longer.

[0075] Model output: The brake reaction time of Wang in the current state is not the standard 1.0 seconds, but a normal distribution with a mean of 1.8 seconds and a standard deviation of 0.2 seconds. That is, T_reaction ~ N(1.8, 0.2^2).

[0076] Step 3: Generate personalized redundancy space;

[0077] The traditional safety distance will be calculated based on the 1.0 second reaction time: S_std = 100km / h ×1.0s ≈ 27.8 meters.

[0078] In order to cover higher probability of safety, the 95% quantile reaction time is used: T_95 = 1.8 + 1.645×0.2 ≈ 2.13 seconds.

[0079] Accordingly, the radius of the personalized redundancy space is calculated as: S_personal = 100km / h ×2.13s ≈ 59.2 meters.

[0080] S2, based on the Internet and the vehicle enterprise background database, obtain historical emergency obstacle avoidance braking control triggering scene big data and vehicle driving user collision warning reaction capability, build a personalized dynamic braking redundancy space evaluation model, and generate a personalized dynamic braking redundancy space for vehicle driving users;

[0081] Step S2 specifically includes:

[0082] Based on the historical emergency obstacle avoidance braking control triggering scene big data, mark each emergency obstacle avoidance braking control triggering scene type and vehicle speed interval, verify the braking redundancy space mean and variance under each emergency obstacle avoidance braking control triggering scene type and vehicle speed interval according to the Bayesian prior, and substitute into the probability density function of normal distribution to generate braking redundancy space prior knowledge of several emergency obstacle avoidance braking control triggering scenes; the braking redundancy space includes: time redundancy and distance redundancy;

[0083] According to the braking redundancy space prior knowledge of several emergency obstacle avoidance braking control trigger scenes, the physical parameters in the corresponding emergency obstacle avoidance braking control trigger scenes are labeled, the Bayesian likelihood function is substituted, the braking redundancy space of several emergency obstacle avoidance braking control trigger scenes is taken as the observation condition, and the observation probability distribution of the physical parameters in the emergency obstacle avoidance braking control trigger scene under the given observation condition is verified.

[0084] According to the basic braking redundancy space prior knowledge of several emergency obstacle avoidance braking control trigger scenes and the observation probability distribution of the physical parameters in the emergency obstacle avoidance braking control trigger scene under the given observation condition, the Bayesian posterior is substituted, and the benchmark braking redundancy space posterior knowledge of several emergency obstacle avoidance braking control trigger scenes is generated.

[0085] According to the benchmark braking redundancy space posterior knowledge of several emergency obstacle avoidance braking control trigger scenes, the collision warning reaction behavior actions of the vehicle driving user in several emergency obstacle avoidance braking control trigger scenes are marked, and the benchmark braking redundancy space associated collision warning reaction behavior action feature data of several emergency obstacle avoidance braking control trigger scenes is established.

[0086] Step S2 further comprises:

[0087] Based on SVR support vector regression, an individualized dynamic braking redundancy space evaluation model is established.

[0088] The benchmark braking redundancy space reaction behavior hyperplane boundary of each emergency obstacle avoidance braking control trigger scene is generated by training the individualized dynamic braking redundancy space evaluation model using the benchmark braking redundancy space associated collision warning reaction behavior action feature data of several emergency obstacle avoidance braking control trigger scenes.

[0089] According to the Euclidean distance formula, the minimum value of the spatial distance between the collision warning reaction ability of the vehicle driving user and the benchmark braking redundancy space reaction behavior hyperplane boundary of each emergency obstacle avoidance braking control trigger scene is calculated, and the collision warning reaction individualized redundancy space of the vehicle driving user is determined.

[0090] In use, the contents in the above steps are combined:

[0091] As a further content, the benchmark braking redundancy space posteriori knowledge fused with specific physical parameters is refined from historical scene big data through Bayesian learning, and a complex hyperplane boundary between the benchmark space and the driver's reaction behavior in different scenes is constructed using support vector regression (SVR). Finally, by calculating the minimum Euclidean distance between the real-time driver reaction ability and these boundaries, the personalized redundancy space is dynamically matched and generated. The double personalization safety protection of scene adaptation and driver state adaptation is realized, which not only can be based on the real-time reaction ability of the driver, but also can intelligently and accurately adjust the safety boundary according to different driving risk scenes (such as high-speed rear-end and intersection crossing), so as to provide the optimal avoidance decision in complex and variable environment.

[0092] Further, as an embodiment of step S2:

[0093] Driver Li is being served, who is driving at a speed of 80 km / h on the highway.

[0094] Step 1: Determine the scene and the benchmark:

[0095] Through camera and radar perception, it is found that Li is following the front car stably, and it is identified as a high-speed rear-end risk scene.

[0096] By calling the benchmark braking redundancy space posteriori knowledge of this scene, which integrates factors such as average reaction time and current dry road (high adhesion coefficient), it is calculated that the benchmark is: at least 2.0 seconds of braking time redundancy.

[0097] Step 2: Evaluate the driver's individualized ability:

[0098] The DSM data from S1 shows that Li is tired, and the mean of the probability distribution of his reaction time is 1.8 seconds (slightly longer than that of an average person).

[0099] Running the SVR model, it is found that the current reaction ability characteristics of Li have the minimum Euclidean distance with the reaction behavior hyperplane boundary of the high-speed rear-end scene. It is determined that the current greatest risk is high-speed rear-end, and Li's state makes this risk more severe.

[0100] Step 3: Generate personalized redundancy space and make decisions:

[0101] Finally, the personalized redundancy space generated for Li is the benchmark based on the high-speed rear-end scene, i.e. 2.0 seconds of time redundancy.

[0102] The actual time redundancy corresponding to the current following distance is calculated in real time. Assuming that the current vehicle distance is 40 meters, the corresponding TTC is 40m / (80 / 3.6 m / s) ≈ 1.8 seconds.

[0103] Decision: Since actual TTC (1.8 seconds) < personalized redundancy space requirement (2.0 seconds), it is determined that there is high risk.

[0104] Trigger early and aggressive primary visual and haptic alerts.

[0105] Slightly pre-tighten the seatbelt and pre-fill the brakes, in preparation for possible emergency braking.

[0106] S3, acquire real-time vehicle surrounding environment data, take the personalized redundancy space of the collision warning reaction of the vehicle driving user as the radius, establish a vehicle-road environment cooperative risk field function, output a vehicle travel dynamic risk field, establish a vehicle emergency obstacle avoidance braking control mechanism, and generate a dynamic virtual traction rope decision;

[0107] Step S3 specifically includes:

[0108] Based on the vehicle multi-source heterogeneous sensor, real-time vehicle surrounding environment data is acquired, the personalized redundancy space of the collision warning reaction of the vehicle driving user is taken as the radius, and the initial measurement vector of the tracking target within the personalized redundancy space of the collision warning reaction of the vehicle driving user is marked; the initial measurement vector of the tracking target includes position, velocity, and acceleration;

[0109] Based on the constant rotation rate and acceleration motion model, a tracking target state transition scheme is constructed, the initial measurement vector of the tracking target within the personalized redundancy space of the collision warning reaction of the vehicle driving user is taken as the input, the tracking target state trend at the future unit timestamp is predicted, the tracking target state trend Jacobian matrix is calculated, and the tracking target state trend covariance matrix within the personalized redundancy space of the collision warning reaction of the vehicle driving user is obtained;

[0110] Using the Kalman gain, the optimal estimation of the uncertainty influence of the real-time vehicle surrounding environment data relative to the tracking target state trend covariance matrix within the personalized redundancy space of the collision warning reaction of the vehicle driving user is calculated, the real-time tracking target state trend covariance matrix within the personalized redundancy space of the collision warning reaction of the vehicle driving user is obtained, and the future step trajectory of the tracking target within the personalized redundancy space of the collision warning reaction of the vehicle driving user is determined;

[0111] Based on the Cartesian coordinate system, according to the future step trajectory of the tracking target within the personalized redundancy space of the collision warning reaction of the vehicle driving user, taking the longitudinal direction of the personalized redundancy space of the collision warning reaction of the vehicle driving user as the X axis and the transverse direction as the Y axis, a tracking target risk field coordinate system within the personalized redundancy space of the collision warning reaction of the vehicle driving user is constructed;

[0112] According to the tracking target risk field coordinate system in the personalized redundancy space of the collision warning reaction of the vehicle driving user, a two-dimensional elliptical Gaussian function is used to establish the tracking target risk diffusion field in the personalized redundancy space of the collision warning reaction of the vehicle driving user, wherein the long axis is along the obstacle movement direction, and the short axis is perpendicular to the movement direction.

[0113] According to the tracking target risk diffusion field in the personalized redundancy space of the collision warning reaction of the vehicle driving user, the angle between the obstacle speed direction and the X axis is calculated to rotate the coordinates of the field center point relative to the obstacle position, so that the major axis of the ellipse is consistent with the speed direction, the obstacle movement direction and the risk diffusion degree in the obstacle movement direction are determined, and the elliptical coefficient of the two-dimensional elliptical Gaussian distribution of the tracking target risk diffusion field in the personalized redundancy space of the collision warning reaction of the vehicle driving user is determined.

[0114] The obstacle movement direction and the risk diffusion degree in the obstacle movement direction are linearly superimposed to obtain a total obstacle risk field, so as to apply the elliptical coefficient of the two-dimensional elliptical Gaussian distribution of the tracking target risk diffusion field in the personalized redundancy space of the collision warning reaction of the vehicle driving user to the two-dimensional elliptical Gaussian function, construct a vehicle-road environment cooperative risk field function, obtain the risk value of each obstacle in the total obstacle risk field, and form a vehicle driving dynamic risk field.

[0115] Step S3 further includes:

[0116] Based on the Khatib artificial potential field method, the real-time vehicle driving target path attractive field and the obstacle repulsive field in the vehicle driving dynamic risk field are marked.

[0117] Based on the personalized redundancy space of the collision warning reaction of the vehicle driving user, the driving behavior characteristic parameters of the vehicle driving user are marked, the conservative, ordinary and aggressive driving clustering clusters are preset according to the K-Means clustering algorithm, the driving behavior characteristics are clustered and divided, and the driving category attribution cluster of the vehicle driving user is determined.

[0118] The centroid of the driving category attribution cluster of the vehicle driving user is normalized to determine the driving category index of the vehicle driving user.

[0119] According to the driving category attribution cluster of the vehicle driving user, the personalized instantaneous risk coefficient of the vehicle driving user is calculated, the real-time vehicle driving target path attractive field and the obstacle repulsive field in the vehicle driving dynamic risk field are dynamically compensated to obtain the personalized attractive field of the real-time vehicle driving target path and the personalized repulsive field of the obstacle in the vehicle driving dynamic risk field, and the method is as follows:

[0120] ;

[0121] Wherein, Personalized repulsive force field of real-time vehicle travel target obstacle in vehicle travel dynamic risk field, Personalized attractive force field of real-time vehicle travel target path in vehicle travel dynamic risk field, Personalized instantaneous risk coefficient of vehicle driving user, Driving category index of vehicle driving user, Personalized redundancy space of collision warning reaction of vehicle driving user, Real-time distance between vehicle and obstacle, Amplification coefficient, Personalized inhibition term weight, Target abandonment rate, and p is a target persistence coefficient.

[0122] Based on the personalized attractive force field of the real-time vehicle travel target path and the personalized repulsive force field of the obstacle in the vehicle travel dynamic risk field, a tracking target virtual attractive force in the personalized redundancy space of the collision warning reaction of the vehicle driving user is constructed, the direction of the vehicle emergency obstacle avoidance braking control is determined, the path integral is predicted according to the MPPI model, the vehicle emergency obstacle avoidance braking control mechanism is established, a plurality of future trajectories are generated from the current position of the vehicle as the starting point and in the direction of the vehicle emergency obstacle avoidance braking control, for each trajectory, the safety cost and the target error cost are minimized by taking the personalized attractive force field of the real-time vehicle travel target path and the personalized repulsive force field of the real-time vehicle travel target obstacle as the safety cost and the target error cost, the optimal future trajectory is determined by weighted average solving according to the total cost of each path, and the dynamic virtual traction rope decision is generated by substituting the MPC trajectory tracking controller.

[0123] In use, the contents in the above steps are combined:

[0124] As further content, the future trajectory and uncertainty of the obstacle in the personalized redundancy space are predicted and quantified by Kalman filtering, and an elliptical Gaussian risk field with the long axis aligned with the motion direction of the obstacle is constructed accordingly, realizing accurate and anisotropic modeling of dynamic threats; then, the improved artificial potential field method and the MPPI model prediction path integral algorithm are fused, and after the risk field is converted into a virtual attractive force, instead of simple tracking, a probability-optimal smooth path, i.e. a dynamic virtual traction rope, is searched out under the premise of considering vehicle dynamics and random disturbances. An obstacle avoidance decision is generated that has forward-looking, smoothness and strong robustness, can predict risks in advance, accurately quantify threats, and output an optimal path that is feasible in vehicle dynamics and can cope with uncertainties, thereby realizing comfortable, reliable and humanized advanced intelligent obstacle avoidance in complex dynamic scenarios.

[0125] Further, as an embodiment of step S3:

[0126] Assume the vehicle is driving on an urban road, and the driver is Wang Shifu, whose reaction capability is at average level (personalized redundancy space is 50 meters).

[0127] Step 1: Perception and Prediction:

[0128] Sensors: Detect that within 50 meters ahead, a bicycle (Target A) is moving at a constant speed in the right lane, and a oncoming car (Target B) is overtaking by cutting into the lane.

[0129] Kalman Filter: Predict that Target A will maintain its current state with small covariance (trajectory is certain). Predict that Target B will reach the current position of the vehicle in 2 seconds, but its trajectory covariance is large (because the overtaking behavior may be aborted).

[0130] Step 2: Build Dynamic Risk Field:

[0131] Build a risk field centered on the ego vehicle with a radius of 50 meters.

[0132] Target A (Bicycle): Generate a narrow elliptical risk field with the long axis pointing in the direction of motion (forward), but with a small σ_long due to slow speed.

[0133] Target B (Oncoming Car): Generate a large and diffuse elliptical risk field with the long axis pointing towards the vehicle (high risk), and with large σ_long and σ_lat due to large prediction uncertainty, the risk field range is wider.

[0134] Fusion: The two elliptical risk fields overlap in the area directly in front of the ego vehicle, forming a region of extremely high risk.

[0135] Step 3: Decision and Generate Tow Rope:

[0136] Artificial Potential Field Method: Calculate virtual resultant force. The repulsive force from Target B is extremely large, pointing to the right side of the vehicle; the attractive force and the repulsive force from Target A together produce a force pointing to the left side of the vehicle. The direction of the resultant force may indicate a slight left avoidance.

[0137] MPPI Optimization:

[0138] The MPPI controller generates thousands of random paths starting from the current state: some sharply turn left, some slightly turn left, and some brake first and then turn.

[0139] The cost function evaluates each path, penalizing: ① passing through high-risk areas, ② controlling instructions are drastic, ③ deviating from the lane.

[0140] After calculation, a path that brakes moderately and slows down at first, and then turns left slightly, bypassing the target A from behind, is selected as the optimal one. Because a direct left turn may collide with the target A, and a sudden brake may be rear-ended by the following vehicle.

[0141] Output: This optimal path is the dynamic virtual towrope. The expected steering angle and expected deceleration required to track this path are output to the underlying controller to execute and generate the decision of the dynamic virtual towrope.

[0142] Referring to Figure 2 As shown in the figure, an intelligent vehicle emergency obstacle avoidance braking control system comprises:

[0143] The capability evaluation module obtains historical vehicle driving data and historical vehicle user monitoring data based on the ECU database and the ADAS database, analyzes the user state under the collision warning of historical vehicle driving, constructs a driving behavior analysis model, and evaluates the reaction capability of the vehicle driving user under the collision warning.

[0144] The redundancy space division module obtains historical emergency obstacle avoidance braking control triggering scene big data and collision warning reaction capability of the vehicle driving user based on the Internet and the vehicle enterprise background database, constructs a personalized dynamic braking redundancy space evaluation model, and generates the collision warning reaction personalized redundancy space of the vehicle driving user.

[0145] The tow traction module takes the collision warning reaction personalized redundancy space of the vehicle driving user as the radius, establishes a vehicle-road environment collaborative risk field function, outputs a vehicle travel dynamic risk field, establishes a vehicle emergency obstacle avoidance braking control mechanism, and generates the decision of the dynamic virtual towrope.

[0146] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent vehicle emergency obstacle avoidance braking control, characterized in that, Comprise: S1, based on ECU database and ADAS database, obtain historical vehicle driving data and historical vehicle user monitoring data, analyze user state under historical vehicle driving collision warning, build driving behavior analysis model, evaluate vehicle driving user reaction ability under collision warning; S2, based on Internet and vehicle enterprise background database, obtain historical emergency obstacle avoidance braking control trigger scene big data and vehicle driving user collision warning reaction ability, build personalized dynamic braking redundancy space evaluation model, generate vehicle driving user collision warning reaction personalized redundancy space; S3, obtain real-time vehicle surrounding environment data, take the collision warning reaction personalized redundancy space of vehicle driving user as the radius, establish vehicle-road environment collaborative risk field function, output vehicle driving dynamic risk field, establish vehicle emergency obstacle avoidance braking control mechanism, generate dynamic virtual traction rope decision. 2.The intelligent vehicle emergency obstacle avoidance braking control method according to claim 1, characterized in that, Step S1 specifically includes: Based on ECU database and DSM database, obtain historical vehicle driving data and historical vehicle user monitoring data; Data cleaning and time alignment preprocessing are carried out for historical vehicle driving data and historical vehicle user monitoring data; Based on sliding window, taking the collision warning trigger time in historical vehicle driving data as anchor point, the historical vehicle driving data and historical vehicle user monitoring data in each collision anchor point unit time window are intercepted, and the historical vehicle driving collision warning-user behavior feature data are established; PCA principal component analysis method is used to reduce the dimension of historical vehicle driving collision warning-user behavior feature data; According to the historical vehicle driving collision warning-user behavior feature data, CUSUM point change monitoring is used to identify and extract the mutation point from the driving user brake pedal stroke time sequence from the historical vehicle driving collision warning trigger unit timestamp, and the user reaction time feature under the historical vehicle driving collision warning event is extracted; According to the historical vehicle driving collision warning-user behavior feature data, the driver state feature in the historical vehicle user monitoring data at the historical vehicle driving collision warning event unit time node is extracted according to the historical vehicle driving collision warning trigger unit timestamp; According to the correlation coefficient, the correlation coefficient between the driver state feature vector in the historical vehicle user monitoring data at the historical vehicle driving collision warning event unit time node and the user reaction time feature vector under the historical vehicle driving collision warning event is verified, which is recorded as the driving fatigue reaction time increment factor; According to Gaussian process regression, a driving behavior analysis model is established, the historical vehicle driving collision warning-user behavior feature data are taken as input, the driving fatigue reaction time increment factor is taken as intervention condition, the user reaction time probability distribution under the influence of the driver state feature in the historical vehicle user monitoring data is verified according to maximum likelihood estimation, and the vehicle driving user reaction ability under collision warning is obtained. 3.The intelligent vehicle emergency obstacle avoidance braking control method of claim 2, wherein, Step S2 specifically includes: Based on the big data of a plurality of emergency obstacle avoidance braking control triggering scenes, mark each emergency obstacle avoidance braking control triggering scene type and vehicle speed interval, verify the braking redundancy space mean and variance of each emergency obstacle avoidance braking control triggering scene type and vehicle speed interval according to the Bayesian prior, and substitute into the probability density function of the normal distribution to generate braking redundancy space prior knowledge of a plurality of emergency obstacle avoidance braking control triggering scenes; the braking redundancy space includes: time redundancy and distance redundancy; According to the braking redundancy space prior knowledge of a plurality of emergency obstacle avoidance braking control triggering scenes, mark the physical parameters in the corresponding emergency obstacle avoidance braking control triggering scene, substitute into the Bayesian likelihood function, and take the braking redundancy space of a plurality of emergency obstacle avoidance braking control triggering scenes as the observation condition to verify the physical parameter observation probability distribution in the emergency obstacle avoidance braking control triggering scene under the given observation condition; According to the basic braking redundancy space prior knowledge of a plurality of emergency obstacle avoidance braking control triggering scenes and the physical parameter observation probability distribution in the emergency obstacle avoidance braking control triggering scene under the given observation condition, substitute into the Bayesian posterior to generate the benchmark braking redundancy space posterior knowledge of a plurality of emergency obstacle avoidance braking control triggering scenes; According to the benchmark braking redundancy space posterior knowledge of a plurality of emergency obstacle avoidance braking control triggering scenes, mark the collision warning reaction behavior action of the vehicle driving user in a plurality of emergency obstacle avoidance braking control triggering scenes, and assemble the benchmark braking redundancy space associated collision warning reaction behavior action feature data of a plurality of emergency obstacle avoidance braking control triggering scenes.

4. The intelligent vehicle emergency obstacle avoidance braking control method according to claim 3, characterized in that, Step S2 further includes: Based on SVR support vector regression, an individualized dynamic braking redundancy space evaluation model is established; The benchmark braking redundancy space reaction behavior hyperplane boundary of each emergency obstacle avoidance braking control triggering scene is generated by training the individualized dynamic braking redundancy space evaluation model using the benchmark braking redundancy space associated collision warning reaction behavior action feature data of a plurality of emergency obstacle avoidance braking control triggering scenes; According to the Euclidean distance formula, the minimum value of the spatial distance between the collision warning reaction ability of the vehicle driving user and the benchmark braking redundancy space reaction behavior hyperplane boundary of each emergency obstacle avoidance braking control triggering scene is calculated to determine the collision warning reaction individualized redundancy space of the vehicle driving user.

5. The intelligent vehicle emergency obstacle avoidance braking control method according to claim 4, characterized in that, Step S3 specifically includes: Based on the vehicle multi-source heterogeneous sensor, real-time vehicle surrounding environment data is obtained, and the collision warning reaction individualized redundancy space of the vehicle driving user is taken as the radius to mark the initial measurement vector of the tracking target within the collision warning reaction individualized redundancy space of the vehicle driving user; the initial measurement vector of the tracking target includes position, velocity, and acceleration; Based on the constant rotation rate and acceleration motion model, a tracking target state transition scheme is constructed, the initial measurement vector of the tracking target within the collision warning reaction individualized redundancy space of the vehicle driving user is taken as the input, the tracking target state trend at the future unit timestamp is predicted, the tracking target state trend Jacobian matrix is calculated, and the tracking target state trend covariance matrix within the collision warning reaction individualized redundancy space of the vehicle driving user is obtained; The tracking target state trend covariance matrix in the personalized redundancy space of the collision warning reaction of the vehicle driving user is obtained by utilizing the Larmor gain to calculate the optimal estimation of the influence of the uncertainty of the tracking target state trend covariance matrix in the personalized redundancy space of the collision warning reaction of the vehicle driving user on real-time vehicle surrounding environment data, and the future step trajectory of the tracking target in the personalized redundancy space of the collision warning reaction of the vehicle driving user is determined. Based on the Cartesian coordinate system, the future step trajectory of the tracking target in the personalized redundancy space of the collision warning reaction of the vehicle driving user is constructed, the longitudinal direction of the personalized redundancy space of the collision warning reaction of the vehicle driving user is taken as the X axis, and the transverse direction is taken as the Y axis, to construct the tracking target risk field coordinate system in the personalized redundancy space of the collision warning reaction of the vehicle driving user. According to the tracking target risk field coordinate system in the personalized redundancy space of the collision warning reaction of the vehicle driving user, a two-dimensional elliptical Gaussian function is used to establish the tracking target risk diffusion field in the personalized redundancy space of the collision warning reaction of the vehicle driving user, the long axis is along the obstacle movement direction, and the short axis is perpendicular to the movement direction. According to the tracking target risk diffusion field in the personalized redundancy space of the collision warning reaction of the vehicle driving user, the angle between the obstacle velocity direction and the X axis is calculated, the coordinates of the field center point are rotated relative to the obstacle position, the major axis of the ellipse is consistent with the velocity direction, the obstacle movement direction and the risk diffusion degree in the obstacle movement direction are determined, and the elliptical coefficient of the two-dimensional elliptical Gaussian distribution of the tracking target risk diffusion field in the personalized redundancy space of the collision warning reaction of the vehicle driving user is determined. The obstacle movement direction and the risk diffusion degree in the obstacle movement direction are linearly superimposed to obtain the total obstacle risk field, the elliptical coefficient of the two-dimensional elliptical Gaussian distribution of the tracking target risk diffusion field in the personalized redundancy space of the collision warning reaction of the vehicle driving user is applied to the two-dimensional elliptical Gaussian function, the car-road environment cooperative risk field function is constructed, the risk value of each obstacle in the total obstacle risk field is obtained, and the vehicle travel dynamic risk field is established.

6. The intelligent vehicle emergency obstacle avoidance braking control method according to claim 5, characterized in that, Step S3 further comprises: Based on the Khatib artificial potential field method, the real-time vehicle travel target path attractive field and the obstacle repulsive field in the vehicle travel dynamic risk field are marked; Based on the personalized redundancy space of the collision warning reaction of the vehicle driving user, the driving behavior characteristic parameters of the vehicle driving user are marked, the conservative, ordinary and aggressive driving clustering clusters are preset according to the K-Means clustering algorithm, the driving behavior characteristics are clustered and divided, and the driving category attribution cluster of the vehicle driving user is determined; The centroid of the driving category attribution cluster of the vehicle driving user is normalized to determine the driving category index of the vehicle driving user; According to the driving category attribution cluster of the vehicle driving user, the personalized instantaneous risk coefficient of the vehicle driving user is calculated, the real-time vehicle travel target path attractive field and the obstacle personalized repulsive field in the vehicle travel dynamic risk field are dynamically compensated, and the personalized attractive field and the personalized repulsive field of the real-time vehicle travel target path in the vehicle travel dynamic risk field are obtained, in the following manner: ； wherein, is a real-time vehicle travel target obstacle individualized repulsive force field in a vehicle travel dynamic risk field, is a real-time vehicle travel target path individualized attractive force field in a vehicle travel dynamic risk field, is a personalized instant risk coefficient of a vehicle driving user, is a driving category index of a vehicle driving user, is a collision warning response personalized redundancy space of a vehicle driving user, is a real-time distance between a vehicle and an obstacle, is a magnification coefficient, is a personalized inhibition term weight, is a target abandonment rate, and ρ is a target persistence coefficient. Based on the real-time vehicle travel target path individualized attractive force field and the obstacle individualized repulsive force field in the vehicle travel dynamic risk field, a tracking target virtual cohesive force in the collision warning reaction individualized redundant space of the vehicle driving user is constructed, the vehicle emergency obstacle avoidance braking control direction is determined, the vehicle emergency obstacle avoidance braking control mechanism is established according to the MPPI model prediction path integral, the current position of the vehicle is taken as the starting point, the vehicle emergency obstacle avoidance braking control direction is taken, a plurality of future trajectories are generated, for each trajectory, the minimum real-time vehicle travel target path individualized attractive force field and the real-time vehicle travel target obstacle individualized repulsive force field are taken as the safety cost and the target error cost, the optimal future trajectory is determined by weighted average solution according to the total cost of each path, and the dynamic virtual traction rope decision is generated by substituting the MPC trajectory tracking controller.

7. An intelligent vehicle emergency obstacle avoidance braking control system, characterized by, The intelligent vehicle emergency obstacle avoidance braking control method comprises the following steps: The capability evaluation module obtains historical vehicle driving data and historical vehicle user monitoring data based on the ECU database and the ADAS database, analyzes the user state under the collision warning of historical vehicle driving, constructs a driving behavior analysis model, and evaluates the reaction capability of the vehicle driving user under the collision warning. The redundant space division module obtains historical emergency obstacle avoidance braking control triggering scene big data and the collision warning reaction capability of the vehicle driving user based on the Internet and the vehicle enterprise background database, constructs an individualized dynamic braking redundant space evaluation model, and generates the collision warning reaction individualized redundant space of the vehicle driving user. The traction traction module obtains real-time vehicle surrounding environment data, takes the collision warning reaction individualized redundant space of the vehicle driving user as the radius, establishes a vehicle-road environment cooperative risk field function, outputs a vehicle travel dynamic risk field, establishes a vehicle emergency obstacle avoidance braking control mechanism, and generates a dynamic virtual traction rope decision.

Images