A driving risk quantification method, system and device based on driver perception

Abstract

The application discloses a driving risk quantification method, system and device based on driver perception, wherein the driving risk quantification method comprises the following steps: obtaining a risk perception factor of a vehicle; and obtaining a driving risk value representing driving risk through a driving risk quantification expression based on the risk perception factor, wherein the risk perception factor of the vehicle is represented by instantaneous acceleration of the vehicle. Compared with the prior art, the driving risk quantification method of the application takes the acceleration adopted by the driver as the risk factor perceived by the driver from the perspective of driver perception, and the driving risk quantification method based on the risk factor can express the influence of other traffic vehicles in the environment on the driving safety of the vehicle from the perspective of driver perception, which is beneficial to the vehicle to find early safety hazards, can provide a basis for a driving early warning mechanism, and can quantify the driving risk that an automatic driving vehicle may face when executing a certain strategy from the perspective of driver perception, thereby providing a reference basis for decision-making.

Description

Technical Field

[0001] This invention belongs to the field of safety risk assessment technology, specifically relating to a method, system, and device for quantifying driving risks based on driver perception. Background Technology

[0002] In the field of traffic safety, quantifying driving risks can provide vehicles with information about their driving situation in the traffic environment, which helps vehicles make timely judgments, avoid risks in advance, and ensure driving safety.

[0003] As the operator and passenger of a vehicle, human drivers are the first to perceive the level of danger in driving based on their driving experience, and continuously adjust the vehicle to reduce driving risks. Assessing driving risks from the perspective of driver perception is beneficial for the vehicle to detect early safety hazards and avoid them in time. However, most current driving risk assessment technologies focus on vehicle collisions and use common kinematic indicators to quantify driving risks, with few methods for quantifying risks from the perspective of driver perception. Summary of the Invention

[0004] To address the problems in the prior art, this invention discloses a method for quantifying driving risks based on driver perception, the specific technical solution of which is as follows:

[0005] A method for quantifying driving risks based on driver perception includes the following steps:

[0006] Obtain risk perception factors for the vehicle;

[0007] Based on the risk perception factor, the driving risk value, which represents driving risk, is obtained through the driving risk quantification expression.

[0008] Among them, the risk perception factor of the vehicle is characterized by the instantaneous acceleration of the vehicle.

[0009] Furthermore, the self-driving vehicle is a manually driven vehicle, the risk perception factor of the self-driving vehicle is the actual instantaneous acceleration of the self-driving vehicle, and the driving risk quantification expression is:

[0010]

[0011] Among them, R human denoted as α, where α is the driving risk value of the manually driven vehicle, and α is the actual instantaneous acceleration of the manually driven vehicle.

[0012] Furthermore, the vehicle in question is an autonomous vehicle, and the method for obtaining the risk perception factor of the vehicle is as follows:

[0013] The input and output sets of the risk perception factor prediction model are constructed based on the NGSIM database; the input and output sets of the risk perception factor prediction model are used to train the regression prediction model based on Gaussian process to construct the risk perception factor prediction model; and the risk perception factor is predicted based on the risk perception factor prediction model.

[0014] The risk perception factor prediction model includes a horizontal risk perception factor prediction model and a vertical risk perception factor prediction model. The risk perception factors include horizontal risk perception factors predicted by the horizontal risk perception factor prediction model and vertical risk perception factors predicted by the vertical risk perception factor prediction model.

[0015] Furthermore, the method for constructing the input and output sets of the risk perception factor prediction model based on the NGSIM database is as follows:

[0016] An initial impact feature set of vehicle driving risk is constructed based on the NGSIM database; the initial impact feature set is filtered according to the feature impact degree and feature redundancy to obtain the impact feature set; the mean and standard deviation of the impact feature set in the time period [t-ΔT,t] are obtained as inputs to the risk perception factor prediction model to construct the input set; the instantaneous acceleration at time t is obtained as the output of the model to construct the output set; ΔT is the driver's reaction lag time.

[0017] The initial impact feature set includes an initial horizontal impact feature set and an initial vertical impact feature set. The impact feature set includes a horizontal impact feature set used to train the horizontal risk perception factor prediction model and a vertical impact feature set used to train the vertical risk perception factor prediction model.

[0018] Furthermore, the method for obtaining the influence feature set by filtering the initial influence feature set based on feature influence degree and feature redundancy is as follows:

[0019] Calculate the Pearson correlation coefficient between the initial features in the initial influence feature set and their corresponding instantaneous acceleration at time t. Select initial features with a Pearson correlation coefficient ≥ 0.6 as initial features with high influence. Calculate the Pearson correlation coefficient between the selected initial features. Remove the initial features with low influence from the two initial features with strong mutual correlation to obtain the influence feature set. Strong mutual correlation means that the Pearson correlation coefficient between the two initial features is ≥ 0.8.

[0020] Furthermore, the method for predicting instantaneous acceleration based on the risk perception factor prediction model is as follows:

[0021] Based on the future driving trajectories of the vehicle and surrounding vehicles, obtain the vehicle's future lateral and longitudinal impact feature sets; based on the vehicle's future lateral impact feature sets and lateral risk perception factor prediction model, predict the lateral risk perception factor; based on the vehicle's future longitudinal impact feature sets and longitudinal risk perception factor prediction model, predict the longitudinal risk perception factor.

[0022] Furthermore, the expression for quantifying driving risk is as follows:

[0023]

[0024]

[0025]

[0026] When the lane-changing vehicle cuts in from the left side of the vehicle, ΔX = x leftboder -x LCV ;

[0027] When the lane-changing vehicle cuts in from the right side of the vehicle, ΔX = x LCV -x rightboder ;

[0028] Among them, a EV,LCV For horizontal risk perception factors, a EV,LV For longitudinal risk perception factor, R humanlike,lat For lateral driving risk value, R humanlike,lon For longitudinal driving risk value, R humanlike For autonomous vehicles, the driving risk value, x LCV For the lateral position of the lane-changing vehicle, x leftboder The lateral position of the left lane line of the lane where the vehicle is located, x rightboder ρ represents the lateral position of the right lane line of the lane where the vehicle is located, and ρ is the probability of the vehicle detecting a lane-changing vehicle cutting in.

[0029] This invention also discloses a driving risk quantification system based on driver perception, comprising:

[0030] The risk perception factor acquisition module is configured to acquire the risk perception factors of the vehicle; the driving risk quantification module is configured to obtain the driving risk value representing the driving risk based on the risk perception factors and through the driving risk quantification expression.

[0031] Furthermore, when the vehicle is an autonomous vehicle, the risk perception factor acquisition module includes:

[0032] The first submodule is configured to provide the input and output sets for building a risk perception factor prediction model based on the NGSI M database; the second submodule is configured to use the input and output sets of the risk perception factor prediction model to train a prediction model based on Gaussian process regression and build a risk perception factor prediction model; the third submodule is configured to predict risk perception factors based on the risk perception factor prediction model.

[0033] The present invention also discloses a driving risk quantification device based on driver perception, including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the steps of the driving risk quantification method described above.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] The driving risk quantification method of the present invention starts from the driver's perception perspective and uses the acceleration taken by the driver as the risk factor perceived by the driver. The driving risk quantification method based on this risk factor can express the impact of other vehicles in the environment on the driving safety of the vehicle from the driver's perception perspective. It is conducive to the vehicle to discover early safety hazards, can provide a basis for driving early warning mechanism, and can also provide a theoretical reference for the intervention mechanism of driving assistance system.

[0036] When the vehicle is an autonomous vehicle, this invention uses a human driving database and machine learning data mining capabilities to summarize the risk perception patterns of human drivers in the driving environment, and establishes a risk perception factor prediction model. This helps autonomous vehicles quantify the risk level in the environment from a human-like perception perspective, so as to serve the vehicle's decision-making process. Attached Figure Description

[0037] Figure 1 A flowchart for quantifying driving risks of manually driven vehicles;

[0038] Figure 2 Flowchart for quantifying driving risks of autonomous vehicles;

[0039] Figure 3 This is a line graph showing the feature influence degree of the lateral initial influence feature in Example 2;

[0040] Figure 4 This is a line graph showing the feature influence degree of the longitudinal initial influence feature in Example 2;

[0041] Figure 5 The test results are shown in the figure for a test example of the horizontal risk perception factor prediction model in Example 2.

[0042] Figure 6This is a test result diagram of another test example of the horizontal risk perception factor prediction model in Example 2;

[0043] Figure 7 The test results are shown in the figure for a test example of the longitudinal risk perception factor prediction model in Example 2.

[0044] Figure 8 This is a test result diagram of another test example of the longitudinal risk perception factor prediction model in Example 2. Detailed Implementation

[0045] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1

[0047] This embodiment discloses a method for quantifying driving risks based on the driver's perception perspective. This embodiment is aimed at manually driven vehicles, such as... Figure 1 As shown, it includes the following steps:

[0048] S1. Obtain risk perception factors for the vehicle;

[0049] Human drivers constantly adjust their acceleration during driving to regulate the perceived real-time risk level. This is a risk perception ability unique to human drivers. Therefore, in this embodiment, the instantaneous acceleration of the vehicle is used as the risk perception factor of the vehicle.

[0050] The instantaneous acceleration of a manually driven vehicle can be obtained in real time during driving (i.e., the actual instantaneous acceleration during driving), for example, through the vehicle's acceleration sensor.

[0051] S2. Based on the risk perception factor, the driving risk value representing driving risk is obtained through the driving risk quantification expression;

[0052] The expression for quantifying driving risk is:

[0053]

[0054] Among them, R human denoted as the driving risk value of manually driven vehicles, ranging from [0,1], and 'a' as the actual instantaneous acceleration of the manually driven vehicle.

[0055] The quantified driving risk value can be used for risk warning. For example, a risk threshold can be preset (the setting of the risk threshold depends on the risk tolerance in actual application, such as 0.6 or 0.8). When the quantified driving risk value exceeds the risk threshold, a warning will be issued to prompt the driver to adjust the driving strategy in time and avoid the risk.

[0056] Example 2

[0057] This embodiment discloses a method for quantifying driving risks based on the driver's perception perspective. This embodiment is aimed at autonomous vehicles, such as... Figure 2 As shown, it includes the following steps:

[0058] S1. Obtain risk perception factors for the vehicle;

[0059] S2. Based on the risk perception factor, the driving risk value representing driving risk is obtained through the driving risk quantification expression;

[0060] The risk perception factor of autonomous vehicles is derived from an anthropomorphic perspective and is based on the instantaneous acceleration predicted by the risk perception factor prediction model. Since the driver's perceived risk mainly comes from two aspects: vehicles in front of the vehicle in the lane and vehicles in adjacent lanes, but considering that the interaction between these two and the vehicle is different, the resulting risks are also different. Therefore, this invention establishes risk perception factor prediction models from both the lateral and longitudinal perspectives to predict the lateral risk perception factor and the longitudinal risk perception factor of autonomous vehicles, respectively, to characterize the risk perception factor of the vehicle.

[0061] Among them, the lateral risk perception factor refers to the instantaneous acceleration of the vehicle when a lane-changing vehicle cuts in, at which time the lane-changing vehicle is cutting between the vehicle and the vehicle in front, and the instantaneous acceleration of the vehicle is mainly affected by the lane-changing vehicle; the longitudinal risk perception factor refers to the instantaneous acceleration of the vehicle when no lane-changing vehicle cuts in, at which time the instantaneous acceleration of the vehicle is mainly affected by the vehicle in front.

[0062] Specifically, step S1 includes the following sub-steps:

[0063] S11. Construct the input and output sets of a risk perception factor prediction model based on the NGSIM database;

[0064] S111. Construct an initial impact feature set of vehicle driving risks based on the NGSIM database;

[0065] The method for constructing the initial lateral influence feature set is as follows:

[0066] (1) Obtain a vehicle lane-changing sample set based on the NGSIM database. The specific method can be found in the invention patent application document with publication number CN114633750A entitled "A method for extracting lane-changing process and analyzing lane-changing characteristics based on unsupervised technology".

[0067] (2) From the vehicle lane-changing sample set, a sample set of vehicles that have entered the lane is selected. The specific criteria for defining the entry behavior are as follows:

[0068] When the front of the vehicle touches the lane line, the time interval between the vehicle and the vehicle changing lanes is THW≤2s;

[0069] Throughout the entire lane-changing process (within 10 seconds before and after crossing the line as defined in patent application CN114633750A), the vehicle engaged in braking behavior, meaning the minimum acceleration of the vehicle during the lane-changing process was < -0.92 m / s². 2 ;

[0070] (3) Based on the sample set of vehicles entering the lane, a lateral initial impact feature set is constructed to characterize the influence of the lane-changing vehicle's LCV on the vehicle's EV. The lateral initial impact feature set constructed in this embodiment is shown in Table 1:

[0071] Table 1. Initial Lateral Influence Feature Set

[0072]

[0073] The method for constructing the initial vertical influence feature set is as follows:

[0074] (1) Vehicle following samples were obtained based on the NGSIM dataset. For specific methods, please refer to the published literature "Zhu Ting, Yang Hongtai, Zhong Xinzhi, Zou Yajie. Analysis of dangerous following behavior characteristics based on intelligent driver model [J]. Transportation and Transport, 2021, 37(3):87-91". It will not be repeated here.

[0075] (2) Based on the car-following samples, a longitudinal initial impact feature set is constructed to characterize the influence of the preceding vehicle's LV on the vehicle's EV. The longitudinal initial impact feature set constructed in this embodiment is shown in Table 2:

[0076] Table 2. Initial Vertical Influence Feature Set

[0077]

[0078] S112. Based on the feature influence degree and feature redundancy, the initial influence feature set is filtered to obtain the influence feature set;

[0079] To reduce computational complexity, it is necessary to find features that have a significant impact on the vehicle. Therefore, this invention performs feature filtering on the initial impact feature set based on feature impact and feature redundancy to obtain features that have a significant impact on the vehicle, i.e., the impact feature set.

[0080] The characteristic influence degree of this invention refers to the correlation between initial influence features and risk perception factors. This invention obtains the characteristic influence degree by calculating the Pearson correlation coefficient between the initial features in the initial influence feature set and their corresponding instantaneous acceleration at time t. The characteristic influence degree of the horizontal initial influence feature in this embodiment is as follows: Figure 3 As shown, the feature influence degree of the initial vertical influence feature is as follows: Figure 4 As shown.

[0081] This invention uses feature influence calculation to select features with a Pearson coefficient value ≥ 0.6 as features strongly correlated with the risk perception factor for prediction. Based on this, in this embodiment, the initial horizontal feature selected based on feature influence is v. LCV v y,LV a x,LCV , dx-1, dy-1, THW-1, d-1; initial vertical feature is v LV v y,LV , THW-2, dx-2, dy-2, d_θ-2, dv-2, dvy-2, d-2;

[0082] To reduce the cross-correlation between influencing features and thus decrease the complexity of the risk factor prediction model, this invention performs feature redundancy analysis on the initially selected features. By calculating the Pearson correlation coefficient between the initial features, this invention removes the initial feature with lower influence from two highly correlated initial features, thus obtaining the influencing feature set. The cross-correlation results of the horizontal initial features in this embodiment are shown in Table 3, and the cross-correlation results of the vertical initial features are shown in Table 4.

[0083] Table 3 Results of initial cross-correlation of horizontal features

[0084] f1 1 0.99 0.61 -0.41 -0.24 0.21 -0.24 f4 1 0.60 -0.39 -0.22 0.22 -0.23 f5 1 -0.52 -0.73 -0.51 -0.73 f8 1 0.37 0.75 0.37 f9 1 0.48 1.00 f13 1 0.57 f16 1

[0085] Table 4. Results of initial longitudinal feature cross-correlation.

[0086] f1’ 1 0.99 -0.33 -0.55 -0.48 -0.24 0.37 -0.42 -0.54 f4’ 1 -0.35 -0.53 -0.46 -0.10 0.36 -0.44 -0.52 f8’ 1 -0.31 -0.39 0.61 0.24 0.75 -0.32 f9’ 1 0.94 -0.63 -0.73 0.23 1 f11’ 1 -0.80 -0.51 0.25 0.94 f13’ 1 0.19 0.44 -0.64 f14’ 1 -0.10 -0.73 f15’ 1 0.22 f16’ 1

[0087] In this invention, two initial features with a Pearson correlation coefficient greater than or equal to 0.8 are selected as features with strong cross-correlation. Based on the cross-correlation results of the initial features, features f4 and f9 in the initial lateral features are excluded in this embodiment. The constructed lateral influence feature set includes: the absolute distance d-1 between the lane-changing vehicle and the vehicle, the relative lateral distance dx-1 between the lane-changing vehicle and the vehicle, the headway THW-1 between the lane-changing vehicle and the vehicle, and the lateral acceleration a of the lane-changing vehicle. x,LCV Lane change vehicle speed v LCV , that is, Fi-1={d-1, dx-1, THW-1, a x,LCV v LCVAfter excluding features f1', f11', and f16' from the initial longitudinal features, the constructed longitudinal influence feature set includes: the longitudinal speed v of the preceding vehicle. y,LV The relative headway between the preceding and following vehicles is THW-2, the relative heading angle between the preceding and following vehicles is d_θ-2, the relative lateral distance between the preceding and following vehicles is dx-2, the relative longitudinal distance between the preceding and following vehicles is dy-2, and the relative resultant velocity between the preceding and following vehicles is dv-2, i.e., Fi-2={v y,LV , THW-2, d_θ-2, dx-2, dy-2, dv-2};.

[0088] S113. Obtain the mean and standard deviation of the influencing feature set in the time interval [t-ΔT,t] as input to the risk perception factor prediction model, construct the input set, obtain the instantaneous acceleration at time t as the output of the model, and construct the output set.

[0089] The inputs and outputs of the horizontal risk perception factor prediction model are obtained from the horizontal impact feature set, while the inputs and outputs of the vertical risk perception factor prediction model are obtained from the vertical impact feature set.

[0090] Assuming the moment when the vehicle reacts to acceleration or deceleration when faced with a vehicle cutting in is t, considering the driver's reaction time and the braking system's reaction time, we can conclude that there is a lag of △T between the impact of the lane-changing vehicle or the vehicle in front on the vehicle and the driver's actual reaction. In this invention, △T is set to 1s, and this value is determined with reference to the conclusions on driver reaction time in existing research.

[0091] S12. Use the input and output sets of the risk perception factor prediction model to train the regression prediction model based on Gaussian process and construct the risk perception factor prediction model.

[0092] Gaussian Process Regression (GPR) is a regression prediction model based on Gaussian processes. Because Gaussian processes support the modeling of univariate or multivariate normally distributed random variables and have a high degree of flexibility in their model structure, compared with traditional machine learning methods such as ANN and SVM, the GPR model reduces the number of parameters that need to be identified, simplifies the parameter optimization process, and enhances the convergence of the model. It has gradually developed into a mainstream machine learning model. Therefore, this invention constructs a risk perception factor prediction model based on the GPR algorithm to achieve risk perception factor prediction.

[0093] The input and output sets of the risk perception factor prediction model constructed in step S11 are divided into a training set and a validation set in an 8:2 ratio.

[0094] S121. Using the GPR toolkit in MATLAB software, construct a risk perception factor prediction model for autonomous vehicles based on the training set.

[0095] S122. The trained risk perception factor prediction model is validated using the validation set. Table 5 shows the indicators used to evaluate the model of this invention.

[0096] Table 5

[0097]

[0098] The results show that the ACU of the horizontal risk perception factor prediction model in this embodiment is 0.9815 on the validation set, and the ACU of the vertical risk perception factor prediction model is 0.9690 on the validation set, indicating that both the horizontal and vertical risk perception factor prediction models of this invention have excellent prediction performance.

[0099] Furthermore, to further evaluate the advantages of the prediction model of this invention, this embodiment compares its advantages with other prediction models. Since neural network prediction algorithms require a large sample size, we use Support Vector Machines (SVMs), which do not have strict requirements on sample size, as a comparison model. Using Python's built-in SVM toolbox, we built a Support Vector Machine Regression (SVR) prediction model. Based on the same validation set, and using RMSE and MAE as evaluation metrics, the model performance comparison is shown in Table 6. It can be concluded that the prediction error of the GPR-based model is much smaller than that of the SVR model.

[0100] Table 6 Model Performance Comparison

[0101] SVR 0.6868 0.5433 GPR 0.1475 0.2986 Error reduction 78.52％ 45.04％

[0102] S123. Test the constructed risk perception factor prediction model.

[0103] The predictive performance of the horizontal risk perception factor prediction model is tested using samples outside the training set. Two test samples are randomly selected, and the prediction results are as follows. Figure 5 , Figure 6 As shown, where, Figure 5 The test sample had a MAE value of 0.0115 and an RMSE value of 0.0139. Figure 6 The test sample in the sample has a MAE value of 0.0237 and an RMSE value of 0.0322.

[0104] The predictive performance of the longitudinal risk perception factor prediction model was tested using samples outside the training set. Two test samples were randomly selected, and the prediction results are as follows. Figure 7 , Figure 8 As shown, where, Figure 7The test sample had a MAE value of 0.0922 and an RMSE value of 0.1113. Figure 8 The test sample in the model has an MAE value of 0.0711 and an RMSE value of 0.0969. These test results indicate that the risk perception factor prediction model of this invention has superior predictive performance.

[0105] S13. Predict risk perception factors based on the risk perception factor prediction model;

[0106] S131. Based on the future driving trajectories of the vehicle and surrounding vehicles, obtain the future lateral influence feature set and longitudinal influence feature set of the vehicle;

[0107] The future driving trajectories of the vehicle and surrounding vehicles can be obtained according to the method described in the published paper "Yuan-YuanRen,LanZhao,Xue-Lian Zheng,et al.A Method for Predicting Diverse Lane-Changing Trajectories of Surrounding Vehicles Based on Early Detection of Lane Change[J].IEEE Access,2022,Vol.10:17451-17472.", which will not be elaborated here.

[0108] S132. Based on the future horizontal impact feature set of the self-driving vehicle and the horizontal risk perception factor prediction model, predict the horizontal risk perception factor; based on the future vertical impact feature set of the self-driving vehicle and the vertical risk perception factor prediction model, predict the vertical risk perception factor.

[0109] The expression for quantifying driving risk in step S2 of this embodiment is:

[0110]

[0111]

[0112] When the lane-changing vehicle cuts in from the left side of the vehicle, ΔX = x leftboder -x LCV ;

[0113] When the lane-changing vehicle cuts in from the right side of the vehicle, ΔX = x LCV -x rightboder ;

[0114] Among them, a EV,LCV For horizontal risk perception factors, a EV,LV For longitudinal risk perception factor, R humanlike,lat For lateral driving risk value, Rhumanlike,lon For longitudinal driving risk value, R humanlike For autonomous vehicles, the driving risk value, x LCV For the lateral position of the lane-changing vehicle, x leftboder The lateral position of the left lane line of the lane where the vehicle is located, x rightboder ρ represents the lateral position of the right lane line of the lane where the vehicle is located, and ρ is the probability of the vehicle detecting a lane-changing vehicle cutting in.

[0115] Among them, ρ can be obtained from the lane-changing intent recognition model of other vehicles in the published literature "Yuan-Yuan Ren, Lan Zhao, Xue-LianZheng, et al. AMethod for Predicting Diverse Lane-Changing Trajectoriesof SurroundingVehicles Based on Early Detection of Lane Change[J].IEEE Access,2022,Vol.10:17451-17472."

[0116] The driving risk quantification method of the present invention adopts anthropomorphism and quantifies the driving risk of autonomous vehicles from the perspective of driver perception. The risk quantification method of the present invention can quantify the driving risk that may be faced when a certain strategy is executed from the perspective of driver perception, and can be used as the basis for judgment when choosing a strategy.

[0117] Example 3

[0118] This embodiment discloses a driving risk quantification system based on driver perception, including: a risk perception factor acquisition module configured to acquire risk perception factors of the vehicle; and a driving risk quantification module configured to obtain a driving risk value representing driving risk based on the risk perception factors and a driving risk quantification expression.

[0119] Wherein, when the vehicle is an autonomous vehicle, the risk perception factor acquisition module includes: a first submodule configured to construct an input set and an output set for a risk perception factor prediction model based on the NGSIM database; a second submodule configured to train a prediction model based on Gaussian process regression using the input set and output set of the risk perception factor prediction model to construct a risk perception factor prediction model; and a third submodule configured to predict risk perception factors based on the risk perception factor prediction model.

[0120] Example 4

[0121] This embodiment discloses a driving risk quantification device based on driver perception, including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the steps of any of the driving risk quantification methods described above.

[0122] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for quantifying driving risks based on driver perception, characterized in that, Includes the following steps: Obtain risk perception factors for the vehicle; Based on the risk perception factor, the driving risk value, which represents driving risk, is obtained through the driving risk quantification expression. Among them, the risk perception factor of the vehicle is characterized by the instantaneous acceleration of the vehicle; The vehicle in question is an autonomous vehicle, and the method for obtaining the risk perception factor of the vehicle is as follows: The input and output sets of a risk perception factor prediction model are constructed based on the NGSIM database. The risk perception factor prediction model is constructed by training a regression prediction model based on Gaussian process using the input and output sets of the risk perception factor prediction model. Predict risk perception factors based on the risk perception factor prediction model; The risk perception factor prediction model includes a horizontal risk perception factor prediction model and a vertical risk perception factor prediction model. The risk perception factors include horizontal risk perception factors predicted by the horizontal risk perception factor prediction model and vertical risk perception factors predicted by the vertical risk perception factor prediction model. The method for constructing the input and output sets of a risk perception factor prediction model based on the NGSIM database is as follows: An initial impact feature set of vehicle driving risks was constructed based on the NGSIM database; The initial impact feature set is filtered based on feature impact degree and feature redundancy to obtain the impact feature set; Obtaining the feature set of influence The mean and standard deviation over the time period are used as inputs to the risk perception factor prediction model to construct the input set and obtain... t The instantaneous acceleration at time t is used as the model's output to construct the output set. This refers to the driver's reaction time delay. The initial impact feature set includes an initial horizontal impact feature set and an initial vertical impact feature set. The impact feature set includes a horizontal impact feature set used to train the horizontal risk perception factor prediction model and a vertical impact feature set used to train the vertical risk perception factor prediction model. The method for predicting instantaneous acceleration based on the risk perception factor prediction model is as follows: Based on the future driving trajectories of the vehicle and surrounding vehicles, obtain the vehicle's future lateral and longitudinal influence feature sets. Based on the future horizontal impact feature set and the horizontal risk perception factor prediction model of autonomous vehicles, the horizontal risk perception factor is predicted. Based on the future vertical impact feature set and the vertical risk perception factor prediction model of autonomous vehicles, the vertical risk perception factor is predicted.

2. The method for quantifying driving risks according to claim 1, characterized in that, The method for obtaining the influence feature set by feature filtering of the initial influence feature set based on feature influence degree and feature redundancy is as follows: Calculate the Pearson correlation coefficient between the initial features in the initial influence feature set and their corresponding instantaneous accelerations at time t, and select features with Pearson coefficient values ​​≥ An initial feature with a value of 0.6 is selected as an initial feature with high influence. Calculate the Pearson correlation coefficient between the selected initial features, and remove the initial features with lower influence from the two initial features with strong correlation to obtain the set of influential features. Strong correlation means that the Pearson correlation coefficient between the two initial features is ≥0.

8.

3. The method for quantifying driving risks according to claim 1, characterized in that, The expression for quantifying driving risk is: When a lane-changing vehicle cuts in from the left side of the vehicle... ; When a lane-changing vehicle cuts in from the right side of the vehicle... ; in, For horizontal risk perception factors, For longitudinal risk perception factors, For lateral driving risk value, For longitudinal driving risk value, For autonomous vehicles, driving risk value, For the lateral position of the lane-changing vehicle, The lateral position of the left lane line of the lane where the vehicle is located. This refers to the lateral position of the right lane line of the lane where the vehicle is located. ρ This represents the probability of a lane-changing vehicle cutting in, detected by the autonomous vehicle.

4. A driving risk quantification system based on driver perception, used to execute the driving risk quantification method according to any one of claims 1-3, characterized in that, include: The risk perception factor acquisition module is configured to acquire risk perception factors for the vehicle. The driving risk quantification module is configured to obtain a driving risk value that represents driving risk based on risk perception factors and through a driving risk quantification expression.

5. The driving risk quantification system according to claim 4, characterized in that, When the vehicle is an autonomous vehicle, the risk perception factor acquisition module includes: The first submodule configures the input and output sets for building a risk perception factor prediction model based on the NGSIM database; The second submodule is configured to train the prediction model based on Gaussian process regression using the input and output sets of the risk perception factor prediction model, and to construct the risk perception factor prediction model. The third submodule is configured to predict risk perception factors based on the risk perception factor prediction model.

6. A device for quantifying driving risks based on driver perception, characterized in that, It includes a memory, a processor, and a computer program stored in the memory, which, when executed by the processor, implements the steps of the driving risk quantification method as described in any one of claims 1-3.

Images