Method for predicting takeover performance based on driver characteristics and application thereof

Abstract

The invention belongs to the technical field of traffic safety, and particularly discloses a method for predicting takeover performance based on driver characteristics and application thereof.The method comprises the steps that physiological data and driving behavior data of a driver are collected, takeover performance indexes are graded through a cumulative frequency method, and prediction models based on random forests are constructed respectively; through the steps of normalizing input features, dividing a data set, optimizing model parameters and the like, the performance of the model is evaluated through indexes such as the accuracy rate and the precision rate, finally, according to a prediction result, the takeover request time is prolonged or an active safety strategy is activated for poor performance, and the control right is handed over according to preset for good performance and excellent performance. According to the method, multi-dimensional data fusion, dynamic optimization of takeover decision and quantitative evaluation of the driver state are realized, the problem that a traditional method is single in feature and lacks evaluation standards is solved, the method can be applied to L3-level automatic driving man-machine control right switching scenes, decision basis is provided for automatic driving system control right switching, and the system reliability is improved. And technical reserve is provided for higher-level automatic driving strategies.

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic safety, and in particular relates to a method for predicting takeover performance based on driver characteristics and its application. Background Art

[0002] With the rapid development of intelligent driving technology, traditional driving models are undergoing profound changes. However, due to technical bottlenecks, regulatory gaps, and safety risks, the current development of the intelligent driving industry is mainly concentrated in the L0-L3 stage, and the realization of fully autonomous driving still faces many technical barriers.

[0003] Level 3 autonomous driving is known as conditional autonomous driving. At this level, most driving tasks can be completed independently by the autonomous driving system, eliminating the need for driver intervention. However, due to limited technological maturity, Level 3 conditional autonomous driving systems have limitations. Factors such as complex road conditions, sudden weather changes, and fluctuations in system performance can cause operating conditions to exceed design limits or trigger system failures. This necessitates a transfer of control from the driver to the vehicle to mitigate risks and ensure driving safety. It is worth noting that the driver in a Level 3 conditionally autonomous vehicle is under passive monitoring for extended periods of time. This necessitates a recovery period for the human driver to retake control of the vehicle, which can be risky and reduce vehicle performance. Therefore, leveraging driver characteristics to predict takeover performance in Level 3 autonomous driving scenarios is of great practical significance.

[0004] In view of this, this invention is proposed. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a method and application for predicting takeover performance based on driver characteristics. It mainly collects the driver's physiological data and the system request takeover time during the automatic driving process, predicts the takeover performance when the vehicle control is switched, and provides a decision-making basis for the switching timing and smooth transition of the automatic driving system control.

[0006] The purpose of the present invention is to solve the problem through the following technical solutions:

[0007] In a first aspect, the present invention provides a method for predicting takeover performance based on driver characteristics, comprising the following steps:

[0008] Step 1: collecting physiological data of the driver during the autonomous driving process, wherein the physiological data includes heart rate growth rate and heart rate variability indicators;

[0009] Step 2: Acquire driving behavior data during the vehicle control handover process and calculate takeover performance indicators, which include reaction time, operation time, minimum collision time, maximum longitudinal deceleration, and maximum lateral acceleration;

[0010] Step 3: Use the cumulative frequency method to grade the takeover performance indicators into three levels: poor, good, and excellent;

[0011] Step 4: Using the physiological data and system takeover request time from step 1 as input features and the takeover performance level from step 3 as the target variable, a random forest-based takeover performance prediction model is constructed.

[0012] Step 5: Dynamically adjust the takeover request time or activate the proactive security policy based on the output of the prediction model.

[0013] Furthermore, in step 1, the calculation formula of the heart rate growth rate is shown in the following formula (1):

[0014]

[0015] In formula (1), △HR represents the heart rate growth rate; HR y Indicates the heart rate value when the event occurs; HR0 indicates the driver's resting heart rate value.

[0016] Furthermore, in step 1, the heart rate variability index includes time domain analysis indexes: RMSSD, SDSD and PNN50, and their corresponding calculation formulas are shown in the following equations (2) to (4):

[0017]

[0018] In formulas (2) to (4), RR i is the interval between adjacent heartbeats, NN 50 is the number of adjacent RR intervals with a difference greater than 50ms, and NN is the total number of RR intervals.

[0019] Furthermore, in step 2, the reaction time refers to the time from the system issuing a takeover request to the driver turning the steering wheel angle greater than 0° or pressing the brake pedal to a degree greater than 0%; the operation time refers to the time from the driver starting the takeover operation to completing the lane change and obstacle avoidance operation; the minimum collision time refers to the minimum value of the ratio of the relative distance between the vehicle and the obstacle vehicle to the speed difference between the vehicle and the obstacle vehicle after the driver takes over the vehicle when the system issues a takeover request.

[0020] Furthermore, the calculation formula of the minimum collision time is shown in the following formula (5):

[0021]

[0022] In formula (5), minTTC is the minimum collision time; d is the distance between the current vehicle and the preceding vehicle (unit: meter); v rel is the speed difference between the current vehicle and the preceding vehicle (unit: m / s).

[0023] Furthermore, in step 3, the specific steps of grading the takeover performance indicators using the cumulative frequency method are as follows:

[0024] Step 3.1: Arrange the original data of each takeover performance indicator in ascending order, construct a frequency distribution table, and calculate the interval frequency and frequency;

[0025] Step 3.2, calculate the upward cumulative frequency and generate a cumulative frequency distribution curve;

[0026] Step 3.3: Using the 25th and 75th percentiles as thresholds, divide the takeover performance into:

[0027] Poor grade: The values ​​of reaction time, operation time, maximum longitudinal deceleration, and maximum lateral acceleration are greater than the 75th percentile, or the minimum collision time is less than the 25th percentile;

[0028] Good level: Reaction time, operation time, maximum longitudinal deceleration, and maximum lateral acceleration values ​​are between the 25th and 75th percentiles;

[0029] Excellent level: The values ​​of reaction time, operation time, maximum longitudinal deceleration, and maximum lateral acceleration are less than the 25th percentile, or the minimum collision time is greater than the 75th percentile.

[0030] Furthermore, in step 4, when constructing the takeover performance prediction model based on the random forest, the specific steps are as follows:

[0031] Step 4.1: Normalize the input features;

[0032] Step 4.2: Divide the dataset into training and test sets according to the set ratio, and use ten-fold cross validation on the training set;

[0033] Step 4.3: Optimize the random forest parameters using grid search, including the number of trees, depth, and number of split features.

[0034] Step 4.4: Use the Bagging ensemble method to fuse the results of ten rounds of cross-validation to generate the final prediction model.

[0035] Furthermore, the performance of the prediction model is evaluated by the following indicators (6) to (9):

[0036] Accuracy:

[0037]

[0038] Accuracy:

[0039]

[0040] Recall:

[0041]

[0042] F1 score:

[0043]

[0044] In formulas (6) to (9), TP represents the number of samples that are actually positive and predicted to be positive; FP represents the number of samples that are actually negative and predicted to be positive; TN represents the number of samples that are actually positive and predicted to be negative; and FN represents the number of samples that are actually negative and predicted to be negative.

[0045] Furthermore, in step 5, dynamically adjusting the takeover request time includes: if the predicted performance is poor, extending the takeover request time or directly activating the active safety policy; if the predicted performance is good or excellent, handing over control rights at a preset time;

[0046] The active safety strategy includes emergency braking or lane keeping assistance.

[0047] In a second aspect, the present invention provides an application of a method for predicting takeover performance based on driver characteristics, which is applicable to a human-machine control switching scenario in an L3 autonomous driving system.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention collects physiological data such as the driver's heart rate growth rate and heart rate variability, combines it with driving behavior data such as reaction time and minimum collision time during the vehicle control switching process, and uses the cumulative frequency method to grade the takeover performance indicators. It also uses physiological data and system takeover request time as input to construct prediction models based on random forests. According to the prediction results, the takeover request time is dynamically adjusted or the active safety strategy is activated. This method can be directly applied to the human-machine control switching scenario of the L3 autonomous driving system, can provide a safety optimization strategy, improve the effectiveness of the takeover, and provide a decision-making basis for the control switching of the autonomous driving system. At the same time, it provides theoretical and technical reserves for the human-machine hybrid takeover strategy of higher-level autonomous driving (L4 / L5). In addition, the present invention has the following advantages:

[0050] (1) Multidimensional data fusion improves prediction accuracy. Compared with traditional methods that rely solely on takeover reaction time or a single physiological indicator (such as heart rate growth rate), the present invention integrates the time domain characteristics of heart rate variability (RMSSD, SDSD, PNN50) with the system takeover request time, and uses the random forest algorithm to mine the nonlinear correlation of multidimensional data. Through the joint modeling of multidimensional features, the accuracy of the prediction model is greatly improved compared to the single-factor model, effectively solving the misjudgment problem caused by the single feature dimension in the existing technology.

[0051] (2) Dynamic takeover decision-making optimizes system robustness. Traditional fixed-threshold takeover strategies (such as a uniformly set 3-second takeover window) are difficult to adapt to fluctuations in driver status. However, the present invention constructs a graded response mechanism by predicting the takeover performance level in real time: when the prediction is "poor", the system automatically extends the takeover request time and activates lane keeping assist, increasing the driver's cognitive recovery time; if the prediction is "excellent", a progressive prompt strategy is adopted to transfer control in stages and reduce the frequency of invalid alarms. Actual road tests have shown that this dynamic strategy significantly improves the takeover success rate and significantly alleviates human-machine conflicts. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are incorporated in and constitute a part of this specification and, together with the description, serve to explain the principles of the invention.

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0054] Figure 1 A flowchart of a method for predicting takeover performance based on driver characteristics according to the present invention;

[0055] Figure 2 It is a cumulative frequency distribution curve of reaction time in the takeover performance index of the present invention;

[0056] Figure 3 A cumulative frequency distribution curve of operation time in the takeover performance indicator of the present invention;

[0057] Figure 4 This is a cumulative frequency distribution curve of the minimum collision time in the takeover performance indicator of the present invention;

[0058] Figure 5 This is a cumulative frequency distribution curve of the maximum lateral acceleration in the takeover performance index of the present invention;

[0059] Figure 6 This is a cumulative frequency distribution curve of the maximum longitudinal deceleration in the takeover performance index of the present invention;

[0060] Figure 7 A flow chart for constructing a takeover performance prediction model of the present application is shown in Fig. 1. DETAILED DESCRIPTION

[0061] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The following description is made with reference to the accompanying drawings in which like reference numerals refer to like elements, unless the context of use indicates otherwise. The following exemplary embodiments described are not representative of all embodiments consistent with the present application. Rather, they are merely examples consistent with some aspects of the present application as detailed in the appended claims.

[0062] In order to better understand the technical solutions of the present application, the present application will be described in further detail below with reference to the accompanying drawings and examples.

[0063] Referring to Fig. 1, a method for predicting takeover performance based on driver characteristics is provided, which specifically includes the following steps: Figures 1 to 7

[0064] 1) Collecting physiological data of the driver during the automatic driving process, the physiological data including a heart rate increase rate and a heart rate variability index.

[0065] Specifically, the calculation formula of the heart rate increase rate is shown in the following formula (1):

[0066]

[0067] In formula (1), △HR represents the heart rate increase rate; HR represents the heart rate value at the time of the event; and HR0 represents the resting heart rate value of the driver. y

[0068] The heart rate variability index includes time domain analysis indexes: RMSSD, SDSD and PNN50, the corresponding calculation formulas of which are shown in the following formulas (2)-(4):

[0069]

[0070] In formulas (2)-(4), the RMSSD index means the root mean square of adjacent RR interval differences, reflecting the speed of heart rate change; the SDSD index means the standard deviation of adjacent RR interval differences; and the PNN50 index means the percentage of the number of adjacent RR interval differences greater than 50 ms in the total number of RR intervals, the greater the value, the higher the heart rate variability; RRi is the adjacent heartbeat interval (the basis for measuring each index of heart rate variability); NN50 is the number of adjacent RR interval differences greater than 50 ms; and NN is the total number of RR intervals.

[0071] ​​It should be noted that the relevant data in the heart rate growth rate and heart rate variability index can be collected by existing wearable smart devices, which will not be described in detail here.

[0072] 2) Obtain driving behavior data in the vehicle control right switching process, calculate takeover performance indicators, including reaction time, operation time, minimum collision time, maximum longitudinal deceleration and maximum lateral acceleration.

[0073] Specifically, the reaction time refers to the time taken from the system issuing a takeover request to the driver turning the steering wheel angle greater than 0° or the opening and closing degree of the brake pedal exceeding 0%; the operation time refers to the time taken from the driver starting to perform the takeover operation to the driver completing the lane changing and obstacle avoidance operation; the minimum collision time refers to the minimum value of the ratio of the relative distance between the ego vehicle and the obstacle vehicle to the speed difference between the ego vehicle and the obstacle vehicle during the process of avoiding collision risk by intervention measures (braking or steering, etc.) after the driver takes over the vehicle after the system issues a takeover request, which is calculated as shown in the following formula (5):

[0074]

[0075] In formula (5), minTTC is the minimum collision time; d is the distance between the current vehicle and the front vehicle (unit: meters); v rel is the speed difference between the current vehicle and the front vehicle (unit: meters / second).

[0076] 3) Build an automatic driving takeover scenario, select subjects to conduct driving simulation experiments, and collect physiological state data of the driver before the takeover request is triggered and driving behavior data when the vehicle control right is switched.

[0077] 4) Use the cumulative frequency method to classify the takeover performance indicators into three levels: poor, good and excellent.

[0078] Specifically, the steps for classifying the takeover performance using the cumulative frequency method are as follows: first, arrange the original data in ascending order and construct a frequency distribution table, and count the frequency (absolute frequency) and frequency (relative frequency) of each data interval; second, calculate the upward cumulative frequency (cumulative frequency from the minimum value to the maximum value), form a cumulative frequency distribution curve, as shown in Figures 2 to 6 ; then, determine the key points according to the number of classification groups, and use the upper and lower limits of the commonly used quartiles as the points, which correspond to the key nodes of the cumulative frequencies of 25% and 75%; finally, divide the data into three categories: excellent, good and poor, according to the points.

[0079] The grading criteria for each indicator are based on the actual meaning of the performance indicator. A longer minimum collision time indicates higher takeover safety, and a higher value for this indicator indicates better performance. Therefore, a value below the 25th percentile is classified as poor performance, a value above the 75th percentile is classified as excellent, and a value between the 25th and 75th percentiles is classified as good performance. For the other takeover performance indicators, lower values ​​indicate faster driver reaction, quicker control, and smoother lateral and longitudinal maneuvers, thus improving performance. Therefore, for the remaining four indicators, values ​​below the 25th percentile are classified as excellent performance, values ​​above the 75th percentile are classified as poor performance, and values ​​between the 25th and 75th percentiles are classified as good performance. The grading criteria for each indicator are shown in Table 1.

[0080] Table 1 Classification criteria for takeover performance

[0081]

[0082] 5) If Figure 7 As shown in FIG, the physiological data and the system takeover request time in step 1) are used as input features, and the takeover performance level in step 4) is used as the target variable to construct a takeover performance prediction model based on random forest.

[0083] When building a takeover performance prediction model based on random forest, the specific steps are as follows:

[0084] 5.1) Normalize the input features to eliminate the interference of dimensional differences on model training and ensure that feature variables of different scales have the same weight contribution;

[0085] 5.2) For validation set construction, a 10-fold cross-validation mechanism was used to construct the training and validation sets, splitting the original dataset into a training set (70%) and a test set (30%) in a 7:3 ratio. The training set was stratified with K = 10, with one subset selected as the validation set in each iteration, and the remaining nine subsets forming the iterative training set, resulting in 10 mutually exclusive training-validation data pairs for model optimization.

[0086] 5.3) For parameter optimization and cross-validation, a search space containing core parameters such as the number of trees, tree depth, and number of split features was constructed. A grid search was performed using ten-fold cross-validation, and the accuracy was used as the evaluation metric to select the optimal parameter combination.

[0087] 5.4) Train the optimal parameter model for each of the 10 training-validation data pairs. Use the bagging ensemble method to fuse the results of the ten training rounds. Generate the final prediction output through mean voting. Save the optimal model parameters and the ensemble results for subsequent use.

[0088] 6) Evaluate the predictive effectiveness of the model and use the widely used accuracy, precision, recall and F1 value of machine learning models to evaluate model performance.

[0089] The performance of the prediction model is evaluated by the following indicators (6) to (9):

[0090] Accuracy:

[0091]

[0092] Accuracy:

[0093]

[0094] Recall:

[0095]

[0096] F1 score:

[0097]

[0098] In formulas (6) to (9), accuracy is the most basic indicator for measuring the overall performance of the model, which indicates the proportion of samples correctly predicted by the model to the total number of samples; precision indicates the proportion of samples that are actually positive among the samples predicted by the model to be positive; recall is also called true positive rate, which indicates the proportion of samples that are correctly predicted to be positive among all samples that are actually positive; F1 score is the harmonic mean of precision and recall, which is used to comprehensively evaluate the performance of the model (the F1 score takes into account both precision and recall and is suitable for application scenarios with high requirements for both. When both precision and recall are high, the F1 score will also be high; conversely, if one of the indicators is low, the F1 score will also be affected). The above indicators are calculated through the mixing matrix, which includes four cases: TP when the actual positive class is predicted to be positive; FP when the actual negative class is predicted to be positive; TN when the actual positive class is predicted to be negative; FN when the actual negative class is predicted to be negative.

[0099] 7) Dynamically adjust the takeover request time or activate the proactive security strategy based on the output of the prediction model.

[0100] Dynamically adjust the takeover request time, including: if the predicted performance is poor, extend the takeover request time or directly activate the proactive security policy; if the predicted performance is good or excellent, transfer control at the preset time;

[0101] Among them, active safety strategies include emergency braking or lane keeping assistance.

[0102] In order to further verify the efficacy of the present invention, the inventors conducted the following experiments:

[0103] Embodiment

[0104] This embodiment uses a physiological polygraph to collect the physiological data of the driver during the automatic driving process, and extracts and calculates the heart rate growth rate and the time domain indicators RMSSD, SDSD and PNN50 related values of heart rate variability according to existing research.

[0105] The driving simulator is used to collect vehicle and driver behavior related data, and the driver's reaction time, operation time, minimum collision time, maximum longitudinal deceleration and maximum lateral acceleration during the takeover process are calculated according to existing research.

[0106] The automatic driving takeover scenario is constructed, and the driving simulation experiment is carried out to collect the physiological state data of the driver before the takeover request is triggered and the driving behavior data when the vehicle control right is switched.

[0107] The cumulative frequency method is used to classify the calculated takeover performance related data into three levels of poor, good and excellent. The values corresponding to each index quantile are shown in Table 2.

[0108] Table 2 Cumulative frequency corresponding table of takeover performance

[0109]

[0110] The physiological data of the driver and the system takeover request time are used as input features, and the takeover performance level is used as the target variable. The random forest-based takeover performance prediction model is constructed, and the accuracy, precision, recall and F1 value are used to evaluate the model performance. The prediction results of each takeover performance indicator are shown in Table 3.

[0111] Table 3 Prediction results of random forest model

[0112]

[0113]

[0114] As shown in Table 3, the prediction accuracy of the random forest prediction model for the five takeover performance indicators is 80% or more, indicating that the random forest model has good prediction effect.

[0115] Further, the takeover request time is dynamically adjusted or the active safety strategy is activated according to the prediction results. The dynamic adjustment of the takeover request time includes: if the predicted performance is poor, the takeover request time is extended or the active safety strategy is directly activated; if the predicted performance is good or excellent, the control right is handed over according to the preset time.

[0116] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention.

[0117] It should be understood that the present invention is not limited to the above description and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for predicting takeover performance based on driver characteristics, characterized in that: The following steps are involved: Step 1: collecting physiological data of the driver during the autonomous driving process, wherein the physiological data includes heart rate growth rate and heart rate variability indicators; Step 2: Acquire driving behavior data during the vehicle control handover process and calculate takeover performance indicators, which include reaction time, operation time, minimum collision time, maximum longitudinal deceleration, and maximum lateral acceleration; Step 3: Use the cumulative frequency method to grade the takeover performance indicators into three levels: poor, good, and excellent; Step 4: Using the physiological data and system takeover request time from step 1 as input features and the takeover performance level from step 3 as the target variable, a random forest-based takeover performance prediction model is constructed. Step 5: Dynamically adjust the takeover request time or activate the proactive security policy based on the output of the prediction model.

2. The method for predicting takeover performance based on driver characteristics according to claim 1, characterized in that: In step 1, the calculation formula of the heart rate growth rate is shown in the following formula (1): In formula (1), △HR represents the heart rate growth rate; HR y Indicates the heart rate value when the event occurs; HR0 represents the driver's resting heart rate.

3. The method for predicting takeover performance based on driver characteristics according to claim 1, characterized in that: In step 1, the heart rate variability index includes time domain analysis indexes: RMSSD, SDSD and PNN50, and their corresponding calculation formulas are shown in the following equations (2) to (4): In formulas (2) to (4), RR i is the interval between adjacent heartbeats, NN 50 is the number of adjacent RR intervals with a difference greater than 50ms, and NN is the total number of RR intervals.

4. The method for predicting takeover performance based on driver characteristics according to claim 1, characterized in that: In step 2, the reaction time refers to the time from the system issuing a takeover request to the driver turning the steering wheel angle greater than 0° or pressing the brake pedal to an opening and closing degree greater than 0%; the operation time refers to the time from the driver starting the takeover operation to completing the lane change and obstacle avoidance operation; the minimum collision time refers to the minimum value of the ratio of the relative distance between the vehicle and the obstacle vehicle to the speed difference between the vehicle and the obstacle vehicle during the process of the driver taking over the vehicle after the system issues a takeover request and avoiding the collision risk through intervention measures.

5. The method for predicting takeover performance based on driver characteristics according to claim 4, characterized in that: The calculation formula of the minimum collision time is shown in the following formula (5): In formula (5), minTTC is the minimum collision time; d is the distance between the current vehicle and the preceding vehicle (unit: meter); v rel is the speed difference between the current vehicle and the preceding vehicle (unit: m / s).

6. The method for predicting takeover performance based on driver characteristics according to claim 1, characterized in that: In step 3, the specific steps of grading the takeover performance indicators using the cumulative frequency method are as follows: Step 3.1: Arrange the original data of each takeover performance indicator in ascending order, construct a frequency distribution table, and calculate the interval frequency and frequency; Step 3.2, calculate the upward cumulative frequency and generate a cumulative frequency distribution curve; Step 3.3: Using the 25th and 75th percentiles as thresholds, divide the takeover performance into: Poor grade: The values ​​of reaction time, operation time, maximum longitudinal deceleration, and maximum lateral acceleration are greater than the 75th percentile, or the minimum collision time is less than the 25th percentile; Good level: Reaction time, operation time, maximum longitudinal deceleration, and maximum lateral acceleration values ​​are between the 25th and 75th percentiles; Excellent level: The values ​​of reaction time, operation time, maximum longitudinal deceleration, and maximum lateral acceleration are less than the 25th percentile, or the minimum collision time is greater than the 75th percentile.

7. The method for predicting takeover performance based on driver characteristics according to claim 1, characterized in that: In step 4, when constructing the takeover performance prediction model based on the random forest, the specific steps are as follows: Step 4.1: Normalize the input features; Step 4.2: Divide the dataset into training and test sets according to the set ratio, and use ten-fold cross validation on the training set; Step 4.3: Optimize the random forest parameters using grid search, including the number of trees, depth, and number of split features. Step 4.4: Use the Bagging ensemble method to fuse the results of ten rounds of cross-validation to generate the final prediction model.

8. The method for predicting takeover performance based on driver characteristics according to claim 7, characterized in that: The performance of the prediction model is evaluated by the following indicators (6) to (9): Accuracy: Accuracy: Recall: F1 score: In formulas (6) to (9), TP represents the number of samples that are actually positive and predicted to be positive; FP represents the number of samples that are actually negative and predicted to be positive; TN represents the number of samples that are actually positive and predicted to be negative; and FN represents the number of samples that are actually negative and predicted to be negative.

9. The method for predicting takeover performance based on driver characteristics according to claim 1, characterized in that: In step 5, dynamically adjusting the takeover request time includes: if the predicted performance is poor, extending the takeover request time or directly activating the active security policy; if the predicted performance is good or excellent, issuing the takeover request at the preset time; The active safety strategy includes emergency braking or lane keeping assistance.

10. An application of the method for predicting takeover performance based on driver characteristics according to any one of claims 1 to 9, characterized in that: The method is applicable to the human-machine control switching scenario of L3 autonomous driving system.

Images