Driving safety comprehensive evaluation method under cognitive distraction and potential danger scenes

Abstract

The invention relates to the technical field of driving safety evaluation, in particular to a driving safety comprehensive evaluation method under a cognitive distraction and potential danger scene, which comprises the following steps of: acquiring sub-dimension data of actual driver operation response characteristics corresponding to a driver and actual driver potential danger perception and psychological characteristics in real time; real-time recording is carried out through an automobile data recorder on a vehicle driven by the driver, and whether the scene where the driver is located currently is a potential dangerous scene or not is judged; determining a potential dangerous scene type corresponding to a scene where the driver is located currently, calling a dangerous scene evaluation index table corresponding to the potential dangerous scene type from a database, and determining dimension importance corresponding to each sub-dimension in the dangerous scene evaluation index table, determining sub-dimension data corresponding to each sub-dimension in the dangerous scene evaluation index table at the current moment; and determining a corresponding driving safety comprehensive evaluation result when the driver is in cognitive distraction in the potential dangerous scene.

Description

Technical Field

[0001] This invention relates to the field of driving safety assessment technology, specifically to a comprehensive driving safety assessment method under cognitive distraction and potentially dangerous situations. Background Technology

[0002] With the continuous growth of motor vehicle ownership and the number of drivers, road traffic safety has become a major global concern. Distracted driving, as one of the significant causes of traffic accidents, has received widespread attention. Distracted driving severely interferes with a driver's normal thinking and operation, making it impossible for them to react quickly in emergency situations, greatly increasing the risk of traffic accidents. For example, when driving normally, a driver can react to a sudden braking of the vehicle in front within a few seconds, but when driving distractedly, the reaction time is significantly prolonged, leading to frequent rear-end collisions and other accidents.

[0003] Current research on driver cognitive distraction and driving safety often suffers from a lack of comprehensive analysis. Most studies focus only on a specific type of driver distraction, such as detecting and analyzing only visual distraction, lacking a holistic consideration of the "human-vehicle-road" information under cognitive distraction conditions. This makes it difficult to construct accurate and effective driving safety risk assessment models. Furthermore, existing driving safety assessments often fail to adequately consider the types of potential hazardous scenarios. In actual driving, different potential hazardous scenarios, such as a stopped vehicle in the middle of the road, passing a school zone, or entering a construction zone, have varying degrees of impact on driver cognitive load and driving safety, but current technologies have failed to provide targeted driving safety assessments based on these scenario types.

[0004] Therefore, there is an urgent need for a comprehensive assessment method for driving safety under cognitive distraction and potential dangerous situations, which can accurately and reliably assess the driver's safe driving across both cognitive distraction and potential dangerous situations, thereby greatly improving the accuracy of driving safety assessment. Summary of the Invention

[0005] One of the objectives of this invention is to provide a comprehensive assessment method for driving safety under cognitive distraction and potentially dangerous situations, which can accurately and reliably assess the driver's safe driving in both cognitive distraction and potentially dangerous situations, thereby greatly improving the accuracy of driving safety assessment.

[0006] To achieve the above objectives, a comprehensive assessment method for driving safety under cognitive distraction and potentially dangerous scenarios is provided, including the following steps: S1. During the process of the driver driving the vehicle, the data of each sub-dimensional of the actual driver operation response characteristics and the data of each sub-dimensional of the actual driver's potential danger perception and psychological characteristics are collected in real time. The actual driver operation response characteristics and the actual driver's potential danger perception and psychological characteristics are all related to cognitive distraction. S2. During the driver's driving, the dashcam on the vehicle records the scene in real time. Based on the recorded information, it is determined whether the driver's current scene is a potentially dangerous scene. If so, S3 is executed; otherwise, the judgment continues. S3. Determine the type of potential dangerous scenario corresponding to the driver's current location, and based on the type of potential dangerous scenario, retrieve the dangerous scenario assessment index table corresponding to the type of potential dangerous scenario from the database, and determine the dimension importance corresponding to each sub-dimension in the dangerous scenario assessment index table; the higher the ranking of the sub-dimension in the dangerous scenario assessment table, the greater the dimension importance, and the greater the impact on cognitive distraction. S4. Based on the sub-dimensions corresponding to the sub-dimensions in the hazard scenario assessment index table corresponding to the retrieved potential hazard scenario type, determine the sub-dimension data corresponding to each sub-dimension in the hazard scenario assessment index table at the current moment. S5. Based on the sub-dimensional data and corresponding dimension importance of each sub-dimensional in the dangerous scenario assessment index table at the current moment, determine the comprehensive driving safety assessment result corresponding to the driver's cognitive distraction in the potential dangerous scenario.

[0007] The technical principles and effects of this solution are as follows: In this solution, step S1 focuses on two core types of data directly related to cognitive distraction: driver operational response characteristic data, which directly reflects the driver's operational delays or deviations caused by cognitive distraction; and driver potential hazard perception and psychological characteristic data, which accurately captures changes in the driver's sensitivity to potential hazards under cognitive distraction. Real-time acquisition of these two types of data provides fundamental data support for subsequent analysis of the correlation between cognitive distraction and hazard perception ability.

[0008] Step S2 uses the real-time recording information from the dashcam to trigger the assessment process when the scenario meets the criteria for a potential dangerous scenario, thereby achieving dynamic switching recognition from "normal driving" to "dangerous scenario".

[0009] Step S3 establishes a mapping relationship between "potential hazardous scenario type - assessment index table - dimension importance" based on the differentiated assessment requirements of scenario types: different potential hazardous scenario types correspond to exclusive hazardous scenario assessment index tables to ensure the matching of indicators with scenario risk characteristics. By matching the exclusive hazardous scenario assessment index tables, it is possible to quickly know the sub-dimensions that have the greatest impact on cognitive distraction for a certain potential hazardous scenario type, as well as the dimension importance of each sub-dimension. This provides reliable data support for the subsequent comprehensive assessment results of driving safety when cognitive distraction occurs in the potential hazardous scenario.

[0010] Step S4 determines the dimensional importance and sub-dimensional data of each sub-dimensional under this potential hazardous scenario type.

[0011] Step 5 transforms real-time data of each indicator into quantifiable security assessment results based on "sub-dimension data + dimension importance".

[0012] The most significant innovation of this approach lies in its first systematic study of the key scientific question of "the impact mechanism of cognitive distraction on a driver's ability to perceive potential hazards." This breaks through the limitations of traditional driving safety assessments that "only focus on operational behavior or single-scenario risks," establishing a causal relationship model between cognitive distraction and hazard perception ability. It reveals the attenuation pattern of a driver's ability throughout the entire chain of "hazard information reception - information processing - perception judgment - decision output" under cognitive distraction, providing a scientific basis for scenario-based assessments. This enables accurate and reliable assessments of driver safety across the dual dimensions of cognitive distraction and potential hazard scenarios, significantly improving the accuracy of driving safety assessments.

[0013] Furthermore, the steps for creating the indicator table corresponding to the hazard assessment indicator table for the potential hazard scenario types in S3 are as follows: S300: Based on a certain type of potential dangerous scenario, a driving simulation scenario corresponding to that type of potential dangerous scenario is built; S301. Based on the constructed driving simulation scenario, several drivers simulate normal driving and perform corresponding cognitive sub-tasks in the driving simulation scenario; S302. When the driver is simulating normal driving and performing the corresponding cognitive sub-task, the data of each sub-dimensional of the virtual driver's operation response characteristics and the data of each sub-dimensional of the virtual driver's potential danger perception and psychological characteristics are collected in real time to form the corresponding virtual sub-dimensional dataset. S303. Divide the formed virtual sub-dimension dataset into a training set and a test set according to a preset ratio; S304. Based on the training set and the test set, and using the preset hyperparameter optimization strategy and the random forest hyperparameter optimization model, determine the optimal hyperparameter set of the random forest hyperparameter optimization model corresponding to the potential hazardous scenario type. S305. Based on the random forest hyperparameter optimization model corresponding to the determined optimal hyperparameter group, as well as the training set and test set, determine the dimensional importance of each sub-dimension under the potential dangerous scenario type based on the preset dimensional importance determination strategy, sort the sub-dimensions according to their dimensional importance, and select the sub-dimensions ranked in the preset position to form the dangerous scenario evaluation index table corresponding to the potential dangerous scenario type.

[0014] Beneficial effects: By building dedicated driving simulation scenarios for specific potential hazards (such as pedestrians violating traffic rules or merging at high speeds), and combining the virtual simulated driving data (operational response, perception, and psychological characteristics) of drivers in these scenarios to select indicators, we can ensure that the evaluation indicator table is highly matched with the characteristics of the scenario (such as including "horizontal search breadth" in blind spot scenarios and "longitudinal acceleration standard deviation" in high-speed merging scenarios). This avoids the general indicators from masking the specificity of the scenario and improves the accuracy of the evaluation.

[0015] The process, including the division of training and testing sets, hyperparameter optimization, and importance ranking, is based on standardized data processing logic. It can be quickly reused for new potential hazardous scenarios (such as tunnel entrances and exits, and construction sections): simply build a new scenario and collect data, and the same algorithm can be used to generate a unique evaluation index table without redesigning the core screening logic, thus improving the versatility and scalability of the method.

[0016] Furthermore, the sub-dimensional data of the actual driver's operational response characteristics include AOPH fixation frequency, AOPH single fixation average time, AOPH first fixation time, horizontal search breadth, vertical search breadth, saccade speed, heart rate growth rate, and heart rate variability; the sub-dimensional data of the actual driver's potential danger perception and psychological characteristics include steering wheel rotation rate, braking response time, speed, longitudinal acceleration, and lateral offset standard deviation.

[0017] Beneficial effects: It simultaneously covers two major sub-dimensions: "potential danger perception and psychological characteristics" (eye movement and electrocardiogram data) and "operational response characteristics" (vehicle control and motion data), enabling multi-perspective assessment of driving safety.

[0018] Furthermore, the preset hyperparameter optimization strategy is as follows: The set of hyperparameters to be optimized and their corresponding value ranges in the random forest hyperparameter optimization model are determined, and the value ranges are dynamically adjusted for different types of potential hazardous scenarios. The training and test sets are standardized to unify the metric units of each sub-dimension data to between 0 and 1. The training set is divided into several subsets, and the proportion of normal driving and cognitive distraction samples in each subset is consistent with that in the original training set. Randomly generate several hyperparameter groups and determine the parameter values ​​corresponding to each hyperparameter group; The data in the subset is used as input data and fed into the random forest hyperparameter optimization model corresponding to each hyperparameter group for iterative training. After each iteration of training, the trained random forest hyperparameter optimization model is output. The training effect of the trained random hyperparameter optimization model is verified by inputting data from the test set to determine whether the training effect meets the standard. If it does, the corresponding hyperparameter set is the optimal hyperparameter set of the random forest hyperparameter optimization model. Otherwise, the hyperparameter set is randomly generated again.

[0019] Beneficial Effects: By defining the hyperparameter set to be optimized and its value range, and dynamically adjusting the value range for different types of potential hazardous scenarios, hyperparameter optimization becomes more closely aligned with the characteristics of specific scenarios. Different potential hazardous scenarios place varying demands on the model; dynamically adjusting the value range allows hyperparameters to be optimized within a more reasonable range, avoiding limitations on model performance due to fixed value ranges, thereby improving the model's adaptability and accuracy across various scenarios. Standardization of the training and test sets unifies the metrics of each sub-dimension data to between 0 and 1, eliminating interference caused by differences in metrics during model training. This ensures the model treats each sub-dimension data more fairly during learning, reducing training bias caused by data volume issues, guaranteeing data consistency and comparability, and laying a solid foundation for effective model training. By randomly generating hyperparameter sets and iteratively training them, combined with testing the test set to verify the training effect, the optimal hyperparameter set can be quickly selected based on its training performance. This approach eliminates the need for manual adjustment of hyperparameters, reducing the influence of human experience on hyperparameter selection. Furthermore, through continuous iteration and verification, it ensures that the optimal set of hyperparameters found enables the model to achieve ideal training results, thereby improving the efficiency and scientific rigor of hyperparameter optimization.

[0020] Furthermore, the cognitive subtask includes setting up a continuous subtraction operation task 1-back delayed digit recall task to simulate spontaneous distraction and external interference distraction, respectively. Attached Figure Description

[0021] Figure 1 This is a flowchart of the comprehensive driving safety assessment method under cognitive distraction and potential dangerous situations in Embodiment 1 of the present invention.

[0022] Figure 2These are a three-dimensional simulation scene diagram and a two-dimensional schematic diagram of the illegal pedestrian crossing scenario in Embodiment 1 of the present invention; Figure 3 This is a three-dimensional simulation scene diagram and a planar schematic diagram of the blind spot scenario of a pedestrian crossing without traffic control in Embodiment 1 of the present invention; Figure 4 This is a three-dimensional simulation scene diagram and a planar schematic diagram of the lane marking transition section scene in Embodiment 1 of the present invention; Figure 5 This is a three-dimensional simulation scene diagram and a plan view of the highway entrance ramp in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram showing the division of potential hazard sources corresponding to various potential hazardous scenarios in Embodiment 1 of the present invention. Detailed Implementation

[0023] The following detailed description illustrates the specific implementation method: Example 1 A comprehensive assessment method for driving safety under cognitive distraction and potentially dangerous situations, basically as follows: Figure 1 As shown, it includes the following steps: S1. During the driver's operation of the vehicle, data on each sub-dimensional of the actual driver's operational response characteristics and each sub-dimensional of the actual driver's potential hazard perception and psychological characteristics are collected in real time. The actual driver's operational response characteristics and the actual driver's potential hazard perception and psychological characteristics are all related to cognitive distraction. The sub-dimensional data of the actual driver's operational response characteristics include AOPH fixation frequency, AOPH average fixation time per fixation, AOPH first fixation time, horizontal search breadth, vertical search breadth, saccade speed, heart rate growth rate, and heart rate variability. The sub-dimensional data of the actual driver's potential hazard perception and psychological characteristics include steering wheel rotation rate, braking response time, speed, longitudinal acceleration, and lateral offset standard deviation.

[0024] S2. During the driver's operation, the dashcam on the vehicle records real-time. Based on the recorded information, it is determined whether the driver's current location is a potentially hazardous scene. If so, proceed to S3; otherwise, continue the assessment. In this embodiment, potentially hazardous scene types include scenarios involving illegally crossing pedestrians, blind spots at uncontrolled pedestrian crossings, lane marking transition sections, and highway entrance ramps. Each scenario has corresponding start and end times. For example, in the scenario of illegally crossing pedestrians: Start time: 150 meters from the illegally crossing pedestrian; End time: 150 meters after the illegally crossing pedestrian. In the scenario of blind spots at uncontrolled pedestrian crossings: Start time: 150 meters from the pedestrian crossing; End time: 150 meters after the pedestrian crossing. In this embodiment, the AOPH area represents the area containing the potential hazard source corresponding to each type of potentially hazardous scene, specifically as follows: Figure 6 As shown in the figure, the AOPH area refers to the area where potential hazards belong in each scenario.

[0025] S3. Determine the type of potential hazardous scenario corresponding to the driver's current location, and based on the type of potential hazardous scenario, retrieve the corresponding hazardous scenario assessment index table from the database, and determine the importance of each sub-dimension in the hazardous scenario assessment index table; the higher the ranking of the sub-dimension in the hazardous scenario assessment table, the greater the dimension importance, and the greater the impact on cognitive distraction; the specific hazardous scenario assessment index tables corresponding to each type of potential hazardous scenario are shown in the following tables:

[0026] Table 1 Evaluation Indicators for Illegal Pedestrian Crossings

[0027] Table 2 Evaluation Indicators for Blind Spot Scenarios at Pedestrian Crossings Without Traffic Control

[0028] Table 3 Evaluation Indicators for Lane Marking Transition Section Scenarios

[0029] Table 4 Evaluation Indicators for Highway Entrance Ramps The steps for creating the indicator table corresponding to the potential hazardous scenario types in S3 are as follows: S300. Based on a certain type of potential hazardous scenario, a driving simulation scenario corresponding to that scenario is constructed. In this embodiment, the corresponding driving simulation scenario is constructed using the SIMLAB driving simulator. The SIMLAB system consists of three parts: a driving simulation cabin, a 6-DOF platform, and a scenario display system. The driving simulation cabin provides a highly realistic driving environment, equipped with the same control devices and sensors as a real vehicle. The 6-DOF platform is equipped with motion simulations such as pitch, roll, and yaw to increase the immersion of driving and can acquire vehicle operation data in real time. The scenario display system consists of a projector and a screen, presenting the driver with a 360-degree variable virtual driving environment. To collect psychological data of the test subjects during the driving process and analyze their neural activity, this paper selects to use the PhysioLAB physiological instrument to collect the psychological data of the drivers during the experiment. This physiological instrument can integrate data acquisition and data analysis, and can simultaneously collect multiple physiological data such as electrocardiogram and skin conductance. This paper only collects the electrocardiogram signal of the test subjects, with a sampling frequency of 1000Hz. In this embodiment, as... Figure 2 , 3 Figures 4 and 5 show the three-dimensional simulation scene diagrams and risk diagrams corresponding to each potential hazardous scene type.

[0030] S301. Based on the constructed driving simulation scenario, several drivers simulate normal driving and perform corresponding cognitive sub-tasks in the driving simulation scenario; in this embodiment, the driver needs to complete one driving task and two tasks, namely normal driving and distracted driving.

[0031] S302. During simulated normal driving and the execution of corresponding cognitive sub-tasks, real-time data of each sub-dimensional of the virtual driver's operational response characteristics and each sub-dimensional of the virtual driver's potential danger perception and psychological characteristics are collected to form corresponding virtual sub-dimensional datasets. In this implementation, the cognitive sub-tasks include setting up a continuous subtraction operation task and a 1-back delayed digit recall task to simulate spontaneous distraction and external interference distraction, respectively. The combinations of continuous subtraction operations are diverse; this paper selects a combination of subtracting a single-digit number from a three-digit number. To ensure that each driver experiences the same driving load, the same instruction is given to each participant, i.e., the same combination of numbers is calculated. During the formal experiment, the staff provides the participants with an audio set consisting of random numbers (0 to 9). The number interval is set to 2.25 seconds. When a new number is played, the participants need to loudly pronounce the number before the most recent number (1-back). To ensure the same applied load, each participant receives the same auditory stimulus.

[0032] S303. Divide the formed virtual sub-dimension dataset into a training set and a test set according to a preset ratio; S304. Based on the training set and the test set, and using the preset hyperparameter optimization strategy and the random forest hyperparameter optimization model, determine the optimal hyperparameter set of the random forest hyperparameter optimization model corresponding to the potential hazardous scenario type. The preset hyperparameter optimization strategy is as follows: The set of hyperparameters to be optimized and their corresponding value ranges in the random forest hyperparameter optimization model are determined, and the value ranges are dynamically adjusted for different types of potential hazardous scenarios. The training and test sets are standardized to unify the metrics of each sub-dimension data to between 0 and 1. In this embodiment, to avoid model bias towards categories with a large number of samples, sample balancing is required for the original data. Common methods include oversampling, undersampling, and algorithms based on synthetic samples. This paper uses the SMOTE (Synthetic Minority Over-sampling Technique) algorithm for sample balancing.

[0033] The training set is divided into several subsets, and the proportion of normal driving and cognitive distraction samples in each subset is consistent with that in the original training set. Several hyperparameter sets are randomly generated, and the parameter values ​​corresponding to each hyperparameter set are determined. In this embodiment, hyperparameters refer to parameters that need to be manually set before training the machine learning model, such as the number of trees in the random forest (n_estimators), the maximum depth (max_depth), and the learning rate. These parameters are not learned from data but are set by the user. Hyperparameter optimization is to find an optimal combination of hyperparameters to make the model perform best on a given task (e.g., highest accuracy and lowest error). Common methods include manual search, random search, and grid search. This study uses grid search combined with five-fold cross-validation to tune the key hyperparameters of the model to ensure that the model achieves optimal generalization ability and performance. This paper uses grid search and five-fold cross-validation for hyperparameter optimization.

[0034] The data in the subset is used as input data and fed into the random forest hyperparameter optimization model corresponding to each hyperparameter group for iterative training. After each iteration of training, the trained random forest hyperparameter optimization model is output. The training effect of the trained random hyperparameter optimization model is verified by inputting data from the test set to determine whether the training effect meets the standard. If it does, the corresponding hyperparameter set is the optimal hyperparameter set of the random forest hyperparameter optimization model. Otherwise, the hyperparameter set is randomly generated again.

[0035] S305. Based on the random forest hyperparameter optimization model corresponding to the determined optimal hyperparameter group, as well as the training set and test set, determine the dimensional importance of each sub-dimension under the potential dangerous scenario type based on the preset dimensional importance determination strategy, sort the sub-dimensions according to their dimensional importance, and select the sub-dimensions ranked in the preset position to form the dangerous scenario evaluation index table corresponding to the potential dangerous scenario type.

[0036] S4. Based on the sub-dimensions corresponding to the sub-dimensions in the hazard scenario assessment index table corresponding to the retrieved potential hazard scenario type, determine the sub-dimension data corresponding to each sub-dimension in the hazard scenario assessment index table at the current moment. S5. Based on the sub-dimensional data and corresponding dimension importance of each sub-dimensional in the dangerous scenario assessment index table at the current moment, determine the comprehensive driving safety assessment result corresponding to the driver's cognitive distraction in the potential dangerous scenario.

[0037] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, based on the guidance provided in this application, improve and implement this solution in combination with their own capabilities. Some typical well-known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A comprehensive assessment method for driving safety under cognitive distraction and potentially dangerous scenarios, characterized in that: Includes the following steps: S1. During the process of the driver driving the vehicle, the data of each sub-dimensional of the actual driver operation response characteristics and the data of each sub-dimensional of the actual driver's potential danger perception and psychological characteristics are collected in real time. The actual driver operation response characteristics and the actual driver's potential danger perception and psychological characteristics are all related to cognitive distraction. S2. During the driver's driving, the dashcam on the vehicle records the scene in real time. Based on the recorded information, it is determined whether the driver's current scene is a potentially dangerous scene. If so, S3 is executed; otherwise, the judgment continues. S3. Determine the type of potential dangerous scenario corresponding to the driver's current location, and based on the type of potential dangerous scenario, retrieve the dangerous scenario assessment index table corresponding to the type of potential dangerous scenario from the database, and determine the dimension importance corresponding to each sub-dimension in the dangerous scenario assessment index table; the higher the ranking of the sub-dimension in the dangerous scenario assessment table, the greater the dimension importance, and the greater the impact on cognitive distraction. S4. Based on the sub-dimensions corresponding to the sub-dimensions in the hazard scenario assessment index table corresponding to the retrieved potential hazard scenario type, determine the sub-dimension data corresponding to each sub-dimension in the hazard scenario assessment index table at the current moment. S5. Based on the sub-dimensional data and corresponding dimension importance of each sub-dimensional in the dangerous scenario assessment index table at the current moment, determine the comprehensive driving safety assessment result corresponding to the driver's cognitive distraction in the potential dangerous scenario.

2. The comprehensive driving safety assessment method under cognitive distraction and potentially dangerous scenarios according to claim 1, characterized in that: The steps for creating the indicator table corresponding to the potential hazardous scenario types in S3 are as follows: S300: Based on a certain type of potential dangerous scenario, a driving simulation scenario corresponding to that type of potential dangerous scenario is built; S301. Based on the constructed driving simulation scenario, several drivers simulate normal driving and perform corresponding cognitive sub-tasks in the driving simulation scenario; S302. When the driver is simulating normal driving and performing the corresponding cognitive sub-task, the data of each sub-dimensional of the virtual driver's operation response characteristics and the data of each sub-dimensional of the virtual driver's potential danger perception and psychological characteristics are collected in real time to form the corresponding virtual sub-dimensional dataset. S303. Divide the formed virtual sub-dimension dataset into a training set and a test set according to a preset ratio; S304. Based on the training set and the test set, and using the preset hyperparameter optimization strategy and the random forest hyperparameter optimization model, determine the optimal hyperparameter set of the random forest hyperparameter optimization model corresponding to the potential hazardous scenario type. S305. Based on the random forest hyperparameter optimization model corresponding to the determined optimal hyperparameter group, as well as the training set and test set, determine the dimensional importance of each sub-dimension under the potential dangerous scenario type based on the preset dimensional importance determination strategy, sort the sub-dimensions according to their dimensional importance, and select the sub-dimensions ranked in the preset position to form the dangerous scenario evaluation index table corresponding to the potential dangerous scenario type.

3. The comprehensive driving safety assessment method under cognitive distraction and potential dangerous situations according to claim 2, characterized in that: The sub-dimensional data of the actual driver's operational response characteristics include AOPH fixation frequency, AOPH single fixation average time, AOPH first fixation time, horizontal search breadth, vertical search breadth, saccade speed, heart rate growth rate, and heart rate variability; the sub-dimensional data of the actual driver's potential danger perception and psychological characteristics include steering wheel rotation rate, braking response time, speed, longitudinal acceleration, and lateral offset standard deviation.

4. The comprehensive driving safety assessment method under cognitive distraction and potentially dangerous scenarios according to claim 3, characterized in that: The preset hyperparameter optimization strategy is as follows: The set of hyperparameters to be optimized and their corresponding value ranges in the random forest hyperparameter optimization model are determined, and the value ranges are dynamically adjusted for different types of potential hazardous scenarios. The training and test sets are standardized to unify the metric units of each sub-dimension data to between 0 and 1. The training set is divided into several subsets, and the proportion of normal driving and cognitive distraction samples in each subset is consistent with that in the original training set. Randomly generate several hyperparameter groups and determine the parameter values ​​corresponding to each hyperparameter group; The data in the subset is used as input data and fed into the random forest hyperparameter optimization model corresponding to each hyperparameter group for iterative training. After each iteration of training, the trained random forest hyperparameter optimization model is output. The training effect of the trained random hyperparameter optimization model is verified by inputting data from the test set to determine whether the training effect meets the standard. If it does, the corresponding hyperparameter set is the optimal hyperparameter set of the random forest hyperparameter optimization model. Otherwise, the hyperparameter set is randomly generated again.

5. The comprehensive driving safety assessment method under cognitive distraction and potentially dangerous scenarios according to claim 4, characterized in that: The cognitive subtasks include setting up a continuous subtraction operation task and a delayed digit recall task, which respectively simulate spontaneous distraction and external interference distraction.

Images