Intelligent driving school mode virtual driving scene generation method based on AI model

Abstract

The invention discloses an intelligent driving school mode virtual driving scene generation method based on an AI model, relates to the technical field of virtual driving scene generation, and is used for solving the problems of feedback lag and poor individualized adaptability of a training scheme caused by difficulty in dynamically adapting to the current stress level and operation habit of a driver. The method comprises the steps of monitoring the instantaneous speed and direction control amplitude of a simulated vehicle in real time, calculating a longitudinal G value and a steering change rate, collecting physiological data of a tested driver to analyze heart rate variability and cognitive load degree, generating driving behavior characteristics and stress characteristics by using an LSTM model, and judging whether to trigger a specific scene or not through a support vector machine based on the characteristics. The method comprises the following steps of: acquiring response time and misoperation frequency, generating a risk response score through fusion function joint quantification, and judging a driving level grade according to a score threshold value, thereby realizing screening and marking of tested drivers, dynamically adapting to actual performance, and improving pertinence and effectiveness of virtual training.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual driving scene generation, and more specifically, to a method for generating a virtual driving scene in an intelligent driving school mode based on an AI model. Background Art

[0002] With the development of artificial intelligence and intelligent driving simulation technology, virtual scenario-based driving training and assessment systems have gradually become an important part of the driver training industry. Traditional driving school teaching relies primarily on manual instruction and physical vehicle operation. This fails to fully quantify students' behavioral characteristics and psychological workload during driving. Furthermore, it is difficult to simulate complex, dangerous, or unexpected driving scenarios under safety constraints, thus hindering the development of students' comprehensive driving skills.

[0003] The existing technology has the following deficiencies:

[0004] At present, some existing virtual driving systems train basic driving skills by setting static training tasks, such as starting, changing lanes, and reversing into a parking space. However, they lack the ability to deeply model the driver's real-time behavioral status and personalized scenario scheduling, making it difficult to dynamically adapt to the driver's current stress level and operating habits, resulting in delayed feedback and poor individual adaptability of the training program. Therefore, a virtual driving scenario generation method for an intelligent driving school mode based on an AI model is proposed.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for generating virtual driving scenes in an intelligent driving school mode based on an AI model, which solves the problems raised in the above-mentioned background technology by using a driving behavior and stress feature recognition algorithm jointly modeled by LSTM and SVM, a dynamic virtual scene triggering mechanism, and a driving level judgment model based on fusion scoring.

[0007] To achieve the above objectives, the present invention provides the following technical solutions: a method for generating a virtual driving scene in an intelligent driving school mode based on an AI model, comprising the following methods:

[0008] Step S1: In the primary simulation scenario, the instantaneous speed and steering amplitude of the simulated vehicle are monitored and the longitudinal G-force and steering rate of change are calculated respectively. The physiological status data of the test driver is collected and analyzed for heart rate variability and cognitive load.

[0009] Step S2: Using the LSTM model, a comprehensive analysis of longitudinal G-force and steering rate of change is performed to generate the test driver's driving behavior characteristics. The stress characteristics of the test driver are assessed by combining heart rate variability and cognitive load.

[0010] Step S3: determining whether a specific simulation scenario is triggered based on the driving behavior characteristics and stress characteristics, and obtaining the test driver's reaction time and frequency of incorrect operations in the specific simulation scenario;

[0011] Step S4: Generate a risk response score based on the reaction time and the frequency of incorrect operations and determine the driving proficiency level of the candidate driver. Screen and mark the candidate driver according to the driving proficiency level.

[0012] In a preferred embodiment, in step S1, the test driver controls the simulated vehicle to move in the forward direction in the primary simulation scenario, and a monitoring period is preset and divided into multiple collection moments;

[0013] Monitor the instantaneous speed of the simulated vehicle at each acquisition moment through the data acquisition interface in the simulated driving system;

[0014] The speed change is obtained by subtracting the instantaneous vehicle speeds at adjacent collection moments, and the ratio of the speed change to the time interval between the collection moments is taken as the instantaneous acceleration;

[0015] The ratio of instantaneous acceleration to gravitational acceleration is taken as the longitudinal G value;

[0016] The steering wheel angle sensor is used to obtain the steering control amplitude of the simulated vehicle at each acquisition moment.

[0017] In a preferred embodiment, in step S1, the angle change is obtained by subtracting the direction control amplitudes at adjacent acquisition moments;

[0018] The ratio of the angle change to the time interval between adjacent acquisition moments is taken as the steering change rate;

[0019] During the monitoring period, the ECG signal of the test driver is monitored by the ECG sensor, and the R-wave detection algorithm is used to identify the R-wave peak in the ECG signal.

[0020] In a preferred embodiment, in step S1, the timestamps of adjacent R-wave peaks are subtracted to obtain R-wave intervals, and the standard deviation of each R-wave interval is calculated to obtain heart rate variability;

[0021] The pulse sensor is used to monitor the pulse signal of the test driver, and the pulse peak value in the pulse signal is identified by a pulse peak detection algorithm;

[0022] The ratio of the total number of pulse peaks to the monitoring period was used as the cognitive load.

[0023] In a preferred embodiment, in step S2, the longitudinal G value and the steering change rate at the same acquisition time within the monitoring period are normalized and then combined into a feature vector;

[0024] Combine each feature vector within the monitoring period into an input sample;

[0025] The LSTM model is used to perform time series modeling on the input samples to extract driving behavior features.

[0026] In a preferred embodiment, in step S2, the heart rate variability and cognitive load are normalized, and the logistic regression parameters are calculated using a weighting function;

[0027] Substitute the logistic regression parameters into the logistic regression formula to calculate the stress characteristics of the test driver.

[0028] In a preferred embodiment, in step S3, the driving behavior characteristics and stress characteristics are constructed as a trigger discrimination feature vector;

[0029] Input the trigger discriminant feature vector into the trained support vector machine model, calculate its classification result on the high-dimensional hyperplane, execute the discriminant function, and output the discrimination result 1 or 0;

[0030] When the judgment result output is 1, a specific simulation scenario is triggered;

[0031] When the judgment result output is 0, the specific simulation scenario is not triggered.

[0032] In a preferred embodiment, in step S3, after a specific simulation scenario is triggered, the reaction time and frequency of incorrect operations of the test driver are collected in real time;

[0033] Reaction time is the time interval between the triggering moment of a specific traffic event and the moment when the test driver first makes an effective action;

[0034] The frequency of incorrect operations refers to the number of times that the candidate's actual control behavior deviates from the standard control action sequence set for the specific traffic event during a specific simulation scenario.

[0035] In a preferred embodiment, in step S4, based on the reaction time and the frequency of wrong operation of the test driver, a logarithmic reciprocal compression function is used as a fusion function to construct a risk response score;

[0036] Based on the risk response score, the driving level is determined by an interval judgment method, and the driving level levels include A, B and C;

[0037] Screen and mark candidate drivers according to their driving proficiency level:

[0038] Filter out test drivers with driving level A and mark them as passed;

[0039] Mark the test driver with driving level grade B as pending assessment;

[0040] The test driver with a driving level grade of C is marked as failed.

[0041] Technical effects and advantages of the present invention:

[0042] The present invention monitors the instantaneous speed and steering amplitude of the simulated vehicle in real time in a primary simulation scenario, calculates the longitudinal G-value and steering change rate respectively, collects the physiological status data of the test driver at the same time, analyzes the heart rate variability and cognitive load, and uses the LSTM model for comprehensive processing to generate the driving behavior characteristics and stress characteristics of the test driver. Based on the driving behavior characteristics and stress characteristics, a support vector machine model is used to determine whether a specific simulation scenario is triggered, and the reaction time and error operation frequency of the test driver are collected. The reaction time and error operation frequency are jointly quantified through a fusion function to generate a risk response score. The driving level of the test driver is determined based on the score level determination threshold, thereby realizing the screening and marking of the test driver, dynamically adapting to the actual performance of the test driver, and improving the pertinence and effectiveness of virtual driving training. It provides technical support and theoretical basis for intelligent driving school teaching and driver ability assessment, and effectively improves the accuracy of driving ability assessment and feedback efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flowchart of the implementation method of the virtual driving scene generation method of the intelligent driving school mode based on the AI ​​model of the present invention.

[0044] Figure 2 This is a schematic diagram of the steps of the method for generating a virtual driving scene in an intelligent driving school mode based on an AI model of the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] The present invention monitors the instantaneous speed and steering amplitude of the simulated vehicle in real time in a primary simulation scenario, calculates the longitudinal G-value and steering change rate respectively, collects the physiological status data of the test driver at the same time, analyzes the heart rate variability and cognitive load, and uses the LSTM model for comprehensive processing to generate the driving behavior characteristics and stress characteristics of the test driver. Based on the driving behavior characteristics and stress characteristics, a support vector machine model is used to determine whether a specific simulation scenario is triggered, and the reaction time and error operation frequency of the test driver are collected. The reaction time and error operation frequency are jointly quantified through a fusion function to generate a risk response score, and the driving level of the test driver is determined based on the score level determination threshold, thereby realizing the screening and marking of the test driver, dynamically adapting to the actual performance of the test driver, and improving the pertinence and effectiveness of virtual driving training, providing technical support and theoretical basis for intelligent driving school teaching and driver ability assessment.

[0047] Example 1, a method for generating a virtual driving scene in an intelligent driving school mode based on an AI model, such as Figures 1 to 2 As shown, the following steps are included:

[0048] Step S1: In the primary simulation scenario, the instantaneous speed and steering amplitude of the simulated vehicle are monitored and the longitudinal G-force and steering rate of change are calculated respectively. The physiological status data of the test driver is collected and analyzed for heart rate variability and cognitive load.

[0049] Step S2: Using the LSTM model, a comprehensive analysis of longitudinal G-force and steering rate of change is performed to generate the test driver's driving behavior characteristics. The stress characteristics of the test driver are assessed by combining heart rate variability and cognitive load.

[0050] Step S3: determining whether a specific simulation scenario is triggered based on the driving behavior characteristics and stress characteristics, and obtaining the test driver's reaction time and frequency of incorrect operations in the specific simulation scenario;

[0051] Step S4: Generate a risk response score based on the reaction time and the frequency of incorrect operations and determine the driving proficiency level of the candidate driver. Screen and mark the candidate driver according to the driving proficiency level.

[0052] The specific implementation is as follows:

[0053] In step S1, the primary simulation scenario refers to a basic training scenario provided for the test driver in a virtual driving scenario;

[0054] The test driver controls the simulated vehicle in the forward direction in the primary simulation scenario. A monitoring cycle is preset and divided into multiple collection moments. The test driver monitors the instantaneous speed of the simulated vehicle at each collection moment through the data collection interface in the simulated driving system. The instantaneous speed is the speed of the simulated vehicle in the forward direction at the collection moment.

[0055] The speed change is obtained by subtracting the instantaneous vehicle speeds at adjacent collection moments, and the ratio of the speed change to the time interval between the collection moments is taken as the instantaneous acceleration;

[0056] The ratio of instantaneous acceleration to gravitational acceleration is used as the longitudinal G value. The longitudinal G value reflects the intensity of instantaneous acceleration during the motion of the simulated vehicle. The unit of instantaneous acceleration is converted into dimensionless form through gravitational acceleration.

[0057] The larger the longitudinal G value, the more significant the test driver's acceleration operation of the simulated vehicle is, and the smaller the longitudinal G value, the smoother the test driver's operation of the simulated vehicle is.

[0058] The steering wheel angle sensor is used to obtain the steering amplitude of the simulated vehicle at each sampling moment. The steering amplitude refers to the steering angle of the test driver during the simulated driving process.

[0059] The angle change is obtained by subtracting the steering amplitudes at adjacent acquisition moments. The ratio of the angle change to the time interval between adjacent acquisition moments is taken as the steering change rate. The steering change rate reflects the speed of change of the test-taker's steering control behavior.

[0060] It should be noted that the preset monitoring period refers to the monitoring time window set by professionals. For example, the preset monitoring period is 5-10 minutes, which is determined according to the primary scenario training objectives: basic straight driving and turning training is set to 5 minutes, and multi-link training including starting, acceleration and deceleration is set to 10 minutes, ensuring at least 300 effective collection moments (sampling frequency ≥ 1Hz) to meet the time series data volume requirements of model analysis, which is used to limit the time range for collecting the instantaneous speed, steering control amplitude and physiological data of the simulated vehicle during the simulated driving process; the simulated driving system is a device that simulates the real driving environment based on the AI ​​model, including a virtual driving scene, a data acquisition interface, and a user interface. Among them, the data acquisition interface refers to the connection point between the simulated driving system and the external device, which is used to transmit data; the steering wheel angle sensor is used to detect the steering wheel rotation angle. It is installed on the steering wheel of the simulated vehicle and is used to monitor the steering wheel rotation angle in real time.

[0061] Synchronously collect the driver's physiological status data based on the monitoring cycle, including electrocardiogram and pulse signals;

[0062] During the monitoring period, the ECG sensor monitors the driver's ECG signal. The ECG signal is stored in the form of a time series. The R-wave detection algorithm is used to identify the R-wave peak in the ECG signal. The timestamps of adjacent R-wave peaks are subtracted to obtain the R-wave interval. The standard deviation of each R-wave interval is calculated to obtain the heart rate variability: in, is the mean of the R wave intervals, RRi is the i-th R-wave interval, n is the total number of R-wave intervals, and SDNN is the heart rate variability;

[0063] The peak of the R wave represents the starting point of the heartbeat. The heart rate variability is calculated by the R wave interval, which reflects the change of the time interval between consecutive heartbeats over time.

[0064] Cognitive load is used to reflect the information processing stress on the driver's brain during training in primary simulation scenarios. A pulse sensor is used to monitor the driver's pulse signal. A pulse peak detection algorithm is used to identify pulse peaks in the pulse signal. The ratio of the total number of pulse peaks to the monitoring period is used as the cognitive load.

[0065] Pulse peaks reflect changes in heart rate, which is regulated by the autonomic nervous system and is related to the brain's cognitive activity. Under cognitive load, the human nervous system becomes active, causing the heart rate to increase and the frequency of pulse peaks to increase. The frequency of pulse peaks reflects the level of information processing stress in the driver's brain.

[0066] It should be noted that the ECG sensor is a device used to monitor and record cardiac electrical activity and is used to collect ECG signals from the test driver's body; the R-wave detection algorithm is used to identify the R-wave peak position from the collected ECG signal. The R-wave detection algorithm runs in a computer operating system environment to automatically detect the R-wave peak in the ECG signal; the pulse sensor is a physiological sensing device used to detect pulse signals and is used to collect the test driver's pulse signal; the pulse peak detection algorithm is used to identify each pulse peak from the pulse signal. The algorithm runs in a computer operating system environment to automatically detect the pulse peak in the pulse signal. For example, the sampling frequency of the pulse sensor is set to 50Hz to ensure that the peak characteristics of the pulse signal can be accurately captured; experimental verification shows that when cognitive load increases, the pulse peak frequency is significantly positively correlated with the intensity of cognitive activity (correlation coefficient r=0.78, p<0.01). Based on this quantitative relationship, it is clear and reasonable to use the ratio of the total number of pulse peaks to the monitoring period to calculate cognitive load.

[0067] In step S2, the longitudinal G value and the steering change rate at the same acquisition time within the monitoring period are normalized and combined into a feature vector, and the feature vectors within the monitoring period are arranged in chronological order to form an input sample;

[0068] The LSTM model is used to perform time series modeling on the input samples and extract driving behavior features. The specific steps are as follows:

[0069] Input layer: receives the normalized feature vector and transmits the input sample to the hidden layer;

[0070] Hidden layer: performs time series modeling and state memory on the input samples. Based on the feature vector at the current moment and the hidden state at the previous moment, the hidden state at the current moment is calculated. The hidden layer includes a forget gate, an input gate, and an output gate, which perform the following functions respectively:

[0071] The forget gate calculates the feature vector in the current input sample and the previous hidden state to generate a weight coefficient, which is used to control the degree of retention of the memory information at the previous moment. The input gate also receives the feature vector of the current moment and the hidden state of the previous moment, generates a write weight coefficient, and updates the current memory content in combination with the candidate state. The candidate state is temporary memory information calculated by nonlinear processing based on the current feature vector and the hidden state of the previous moment through the tanh activation function. The output gate weights the memory content by the weight coefficient to obtain the hidden state at the current moment, and then takes the weighted average of the hidden states of all time steps to output the driving behavior feature vector.

[0072] Output layer: Perform linear transformation on the driving behavior feature vector through matrix multiplication and map it into a scalar value as the driving behavior feature.

[0073] It should be explained that the LSTM model is a recurrent neural network with long-term dependency learning capabilities, used to model the temporal characteristics and changing patterns of input samples. After receiving input samples, the LSTM model is trained to output driving behavior characteristics. Standardization methods include but are not limited to standard linear transformation based on interval scaling, statistical Z-Score standardization methods, or normalization methods based on nonlinear mapping functions. The application methods of standardization are not detailed here.

[0074] Driving behavior characteristics reflect the comprehensive stability of the test driver's operation. The larger the longitudinal G value and the steering change rate, the greater the fluctuation of the simulated vehicle's motion state and motion steering. The higher the driving behavior characteristics, the more frequent the test driver's operation.

[0075] Heart rate variability and cognitive load are normalized, and logistic regression parameters are calculated using a weighting function: in, is the normalized heart rate variability, is the standardized cognitive load, β0 is the preset bias term, β1 and β2 are the preset regression coefficients; Y is the logistic regression parameter;

[0076] Substituting the logistic regression parameters into the logistic regression formula, the stress characteristics of the test driver are calculated: Among them, L is the stress characteristic of the test driver;

[0077] Higher heart rate variability indicates a stronger autonomic nervous system regulation ability, a more stable physiological state, and lower stress characteristics. Higher cognitive load indicates greater information processing pressure on the driver's brain and higher stress characteristics.

[0078] The stress characteristics of the test driver are calculated based on heart rate variability and cognitive load, reflecting the stress state of the test driver at the physiological and psychological levels during the simulated driving process;

[0079] It should be noted that the preset regression coefficient ranges from 0 to 1, which is used to reflect the degree of influence of heart rate variability and cognitive load on the stress characteristics of the test driver. It is specifically set by professionals according to actual application requirements. The preset bias term is a constant term in the logistic regression parameter, which shifts or adjusts the results of the logistic regression parameter. It is specifically set by professionals and will not be elaborated here.

[0080] For example, the specific structural parameters of the LSTM model are set as follows: the hidden layer contains three layers, with 64 neurons in each layer; the mean square error (MSE) is used as the loss function during training, the Adam optimizer is selected, the initial learning rate is set to 0.001, the number of iterations is 500, and training is terminated when the validation set loss does not decrease after 10 consecutive iterations. With these parameter settings, the time series features of driving behavior can be accurately extracted, ensuring that the model has good convergence and generalization capabilities when processing time series data of longitudinal G-force and steering rate of change.

[0081] In step S3, a binary classification judgment model is jointly constructed based on driving behavior characteristics and stress characteristics to determine whether a specific simulation scenario needs to be triggered, ensuring that simulation system resources are concentrated on test-takers with advanced evaluation value.

[0082] The binary classification judgment model is a support vector machine model. During the training phase of the support vector machine model, a training sample set is constructed based on historical data samples. Among them, historical data samples refer to structured data samples obtained based on the behavioral data of test-takers collected during historical simulated driving. i is the feature vector of the i-th data sample, y i ∈{0,1} represents the corresponding discrimination result, 0 and 1 are discrimination labels, 0 indicates that the triggering condition of a specific simulation scenario is not met, and 1 indicates that the triggering condition of a specific simulation scenario is met.

[0083] The training sample set is marked according to the discriminant label, and the optimal classification hyperplane is calculated in the feature space so that data samples of different categories are separated by the maximum interval on both sides of the classification hyperplane. The minimum distance from the support vector to the classification hyperplane is taken as the optimization goal. An optimization problem with constraints is constructed and the Lagrange multiplier is introduced to solve the optimization problem through the Lagrange dual method, so as to calculate the Lagrange multiplier corresponding to each data sample, screen out data samples with Lagrange multipliers greater than 0 to obtain support vectors, and construct a discriminant function based on the support vectors to perform classification judgment on the newly input feature vectors.

[0084] It should be noted that the support vector machine model is a binary classification discriminant model built based on the principle of maximizing the margin. It is used to construct the optimal classification hyperplane in a given feature space, thereby effectively distinguishing samples of different categories. The Lagrange multiplier is an introduced variable used to solve optimization problems with equality or inequality constraints. It is used to embed the constraints into the objective function, thereby constructing a Lagrangian function so that the optimization process obtains the optimal solution under the premise of satisfying the constraints. The Lagrangian dual method is an optimization method based on the Lagrangian function to construct a dual problem. It is used to convert the original constrained optimization problem into its dual form, thereby simplifying the computational complexity and facilitating the introduction of kernel functions to handle nonlinear classification tasks.

[0085] The driving behavior characteristics and stress characteristics are constructed into a trigger discriminant feature vector X = [x1, x2], where x1 represents the driving behavior characteristics and x2 represents the stress characteristics. The trigger discriminant feature vector X = [x1, x2] is input into the trained support vector machine model, the classification results on the classification hyperplane are calculated, and the following discriminant function is executed:

[0086]

[0087] Among them, f(X) is the discrimination result, α i is the Lagrange multiplier, K(·,·) is the kernel function, and b is the bias term.

[0088] It should be explained that the kernel function refers to a function that maps data samples in the original input space to a high-dimensional feature space; the bias term is a constant value in the discriminant function calculated by the support vector that satisfies the interval constraint.

[0089] When the judgment result output is 1, it means that the driving behavior characteristics and stress characteristics meet the trigger conditions, and the candidate driver enters a specific simulation scene for further testing.

[0090] When the discrimination result output is 0, it means that the driving behavior characteristics and stress characteristics do not meet the trigger conditions and the specific simulation scenario will not be entered.

[0091] Specific simulation scenarios include preset specific traffic events, such as sudden crossing pedestrians, suddenly stopped vehicles ahead, temporary lane changes due to road construction, and other specific traffic events with uncertainty. Each specific traffic event corresponds to a standard control action sequence. The standard control action sequence is a preset set of response actions and their execution order that meet safe driving requirements. The operation data includes action type, operation timing, amplitude range and response time threshold, which are used to characterize the standard response behavior of the test driver in this situation.

[0092] Among them, the values ​​of the amplitude range and response time threshold are determined comprehensively based on driving behavior standards, traffic safety research and historical driving simulation test results, and are not elaborated here.

[0093] During the operation of a specific simulation scenario, the reaction time and frequency of incorrect operations of the test driver are collected in real time. The reaction time is the time interval between the triggering moment of a specific traffic event and the moment when the test driver makes the first effective operation. Among them, effective operations include behaviors such as the steering wheel angle change exceeding the change threshold and the brake pedal depression depth exceeding the depth threshold. It is used to reflect the test driver's perception and reaction speed to emergencies.

[0094] Among them, the change threshold is set based on the sensitivity of the vehicle control system, the sensor resolution and the statistical analysis of the driver's regular operation amplitude; the depth threshold is set comprehensively by the force feedback characteristics of the vehicle braking system, the driver's regular braking intensity distribution and driving safety standards, which will not be elaborated here.

[0095] The frequency of incorrect operations refers to the number of operational errors made by the candidate driver in a specific simulation scenario. An operational error is defined as the candidate driver's actual control behavior deviating from the response action in the standard control action sequence set for the specific traffic event, including operation time exceeding the specified response time threshold, operation amplitude not meeting the amplitude range, incorrect execution sequence, execution of redundant actions that conflict with the standard control action sequence, and simultaneous execution of conflicting operations.

[0096] The identification of operational errors is based on the standard operating action sequence. By comparing the actual operating data of the candidate driver with the preset operating data in the standard operating action sequence, it is determined whether there is any deviation behavior. The number of errors made by the candidate driver in a specific simulation scenario is then counted and used as the frequency of incorrect operations.

[0097] For example, the support vector machine training sample set contains 5,000 sets of historical data samples, covering the driving behavior characteristics and stress characteristics of test-takers with different driving levels in various primary simulation scenarios; the kernel function uses the radial basis function (RBF kernel), where the kernel parameter γ is set to 0.1, which is used to map the original feature space to a high-dimensional space to handle nonlinear classification problems; during the model training process, the cross-validation method (5-fold cross-validation) is used to determine the optimal penalty parameter C, with a value of 10, which is used to balance the classification accuracy and generalization ability of the model.

[0098] In step S4, based on the reaction time and frequency of incorrect operations of the candidate driver, a risk response score is constructed through a fusion function to quantify the candidate driver's dynamic response ability in a specific simulation scenario and determine his or her driving level accordingly.

[0099] The fusion function is a logarithmic inverse compression function, which is defined as follows:

[0100]

[0101] Among them, R s Score the risk response, T r is the reaction time, f e is the frequency of misoperation, T0 and f0 are the benchmark standard values ​​of reaction time and misoperation frequency, respectively. They are set according to the statistical mean of reaction time and misoperation frequency of test-takers marked as passed in the historical simulated driving process, and are used to perform dimensionless normalization on the input variables.

[0102] After the reaction time and the frequency of erroneous operations are normalized respectively by the logarithmic inverse compression function, the linkage enhancement effect is realized through the product method, and the natural logarithmic transformation is applied to realize the monotonically decreasing response of the reaction time and the frequency of erroneous operations to the risk reaction score, thereby avoiding the over-discreteness of the risk reaction score, ensuring the continuity and controllability of the fusion function in the entire definition domain, and ensuring that the output risk reaction score decreases with the increase of reaction time and the frequency of erroneous operations.

[0103] The specific method for setting the benchmark values ​​for reaction time and error frequency is as follows: 1,000 test-takers marked as "passed" in historical simulated driving were selected as a sample group, and their reaction time and error frequency data in the same specific simulated scenarios were collected; the mean of the reaction time was calculated using the arithmetic mean method;

[0104] Risk response score R s ∈(0,1], the closer the value is to 1, the better the test driver's response to risk.

[0105] Based on the risk response score, the driving level is determined by using an interval judgment method. The classification rules are as follows:

[0106] If R s >θ1, then L d =A;

[0107] If θ2 <R s ≤θ1, then L d =B;

[0108] If ≤θ2, then L d =C;

[0109] Among them, θ1 and θ2 are the scoring level determination thresholds, satisfying 1>θ1>θ2>0. Their values ​​are determined based on the industry safety standards and reference to the requirements for "Emergency Response Capability Assessment" in the "Motor Vehicle Driver Training Teaching and Examination Outline" (Jiaoyunfa

[2022] No. 36), and are not elaborated here.

[0110] Applicants are screened and marked according to their driving proficiency level, including:

[0111] Filter out test drivers with driving level A and mark them as passed;

[0112] Mark the test driver with driving level grade B as pending assessment;

[0113] The test driver with a driving level grade of C is marked as failed.

[0114] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0115] Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0116] As used herein, the singular forms "a," "an," and "the" may also include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "include," "comprising," "having," and the like specify the presence of stated features, integers, steps, operations, components, parts, or combinations thereof, but do not preclude the presence or addition of one or more other features, integers, steps, operations, components, parts, or combinations thereof. Furthermore, the term "and / or" as used in this specification includes any and all combinations of the relevant listed items.

[0117] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0118] The above description of the disclosed embodiments will enable those skilled in the art to implement or use various modifications of these embodiments, and it will be apparent to those skilled in the art that the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating virtual driving scenarios in an intelligent driving school mode based on an AI model, characterized by: The following methods are included: Step S1: In the primary simulation scenario, the instantaneous speed and steering amplitude of the simulated vehicle are monitored and the longitudinal G-force and steering rate of change are calculated respectively. The physiological status data of the test driver is collected and analyzed for heart rate variability and cognitive load. Step S2: Using the LSTM model, a comprehensive analysis of longitudinal G-force and steering rate of change is performed to generate the test driver's driving behavior characteristics. The stress characteristics of the test driver are assessed by combining heart rate variability and cognitive load. Step S3: determining whether a specific simulation scenario is triggered based on the driving behavior characteristics and stress characteristics, and obtaining the test driver's reaction time and frequency of incorrect operations in the specific simulation scenario; Step S4: Generate a risk response score based on the reaction time and the frequency of incorrect operations and determine the driving proficiency level of the candidate driver. Screen and mark the candidate driver according to the driving proficiency level.

2. The method for generating a virtual driving scene in an intelligent driving school mode based on an AI model according to claim 1, characterized in that: In step S1, the test driver controls the simulated vehicle in the forward direction in the primary simulation scenario, and the monitoring period is preset and divided into multiple collection moments; Monitor the instantaneous speed of the simulated vehicle at each acquisition moment through the data acquisition interface in the simulated driving system; The speed change is obtained by subtracting the instantaneous vehicle speeds at adjacent collection moments, and the ratio of the speed change to the time interval between the collection moments is taken as the instantaneous acceleration; The ratio of instantaneous acceleration to gravitational acceleration is taken as the longitudinal G value; The steering wheel angle sensor is used to obtain the steering control amplitude of the simulated vehicle at each acquisition moment.

3. The method for generating a virtual driving scene in an intelligent driving school mode based on an AI model according to claim 2, characterized in that: In step S1, the angle change is obtained by subtracting the direction control amplitudes at adjacent acquisition moments; The ratio of the angle change to the time interval between adjacent acquisition moments is taken as the steering change rate; During the monitoring period, the ECG signal of the test driver is monitored by the ECG sensor, and the R-wave detection algorithm is used to identify the R-wave peak in the ECG signal.

4. The method for generating a virtual driving scene in an intelligent driving school mode based on an AI model according to claim 3, characterized in that: In step S1, the timestamps of adjacent R-wave peaks are subtracted to obtain the R-wave intervals, and the standard deviation of each R-wave interval is calculated to obtain the heart rate variability; The pulse sensor is used to monitor the pulse signal of the test driver, and the pulse peak value in the pulse signal is identified by a pulse peak detection algorithm; The ratio of the total number of pulse peaks to the monitoring period was used as the cognitive load.

5. The method for generating a virtual driving scene in an intelligent driving school mode based on an AI model according to claim 3, characterized in that: In step S2, the longitudinal G value and the steering change rate at the same acquisition time within the monitoring period are normalized and combined into a feature vector; Combine each feature vector within the monitoring period into an input sample; The LSTM model is used to perform time series modeling on the input samples to extract driving behavior features.

6. The method for generating a virtual driving scene in an intelligent driving school mode based on an AI model according to claim 4, characterized in that: In step S2, the heart rate variability and cognitive load are normalized, and the logistic regression parameters are calculated using a weighting function; Substitute the logistic regression parameters into the logistic regression formula to calculate the stress characteristics of the test driver.

7. The method for generating a virtual driving scene in an intelligent driving school mode based on an AI model according to claim 6, characterized in that: In step S3, the driving behavior characteristics and stress characteristics are constructed into a trigger discrimination feature vector; Input the trigger discriminant feature vector into the trained support vector machine model, calculate its classification result on the high-dimensional hyperplane, execute the discriminant function, and output the discrimination result 1 or 0; When the judgment result output is 1, a specific simulation scenario is triggered; When the judgment result output is 0, the specific simulation scenario is not triggered.

8. The method for generating a virtual driving scene in an intelligent driving school mode based on an AI model according to claim 1, characterized in that: In step S3, after a specific simulation scenario is triggered, the reaction time and frequency of incorrect operations of the test driver are collected in real time; Reaction time is the time interval between the triggering moment of a specific traffic event and the moment when the test driver first makes an effective action; The frequency of incorrect operations refers to the number of times during a specific simulation scenario that the candidate's actual control behavior deviates from the standard control action sequence set for the corresponding specific traffic event.

9. The method for generating a virtual driving scene in an intelligent driving school mode based on an AI model according to claim 8, characterized in that: In step S4, based on the reaction time and the frequency of wrong operation of the test driver, a logarithmic reciprocal compression function is used as a fusion function to construct a risk response score; Based on the risk response score, the driving level is determined by an interval judgment method, and the driving level levels include A, B and C; Screen and mark candidate drivers according to their driving proficiency level: Filter out test drivers with driving level A and mark them as passed; Mark the test driver with driving level grade B as pending assessment; The test driver with a driving level grade of C is marked as failed.

Images