A method for evaluating driver's potential danger perception ability

Abstract

The invention relates to the field of traffic potential hazard identification, and discloses a method for evaluating a driver's potential hazard perception ability, comprising the following steps: S1, collecting eye movement data of a driver in different potential risk scenarios through a driving simulation experiment, and converting the eye movement data into specific evaluation parameter values; S2, screening out evaluation parameters with significant differences as key evaluation parameters according to the characteristics of each risk scenario and the differences in evaluation parameters among different groups of people in the risk scenario; S3, using an entropy weight method to perform weight calculation on the key evaluation parameters in each risk scenario; S4, substituting the values ​​of the driver's key evaluation parameters into a TOPSIS model, and calculating the degree of proximity between the evaluation results of the driver in each risk scenario and positive and negative ideal solutions, so as to obtain the potential hazard perception ability score of the driver in each risk scenario, so as to solve the technical problem of complex quantization models when evaluating the potential hazard perception ability in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic safety, and in particular to a method for evaluating a driver's potential danger perception ability. Background Art

[0002] Potential hazard perception ability is a manifestation of the driver's driving skills. It reflects the driver's ability to identify and distinguish potential hazard information in risky traffic scenes during driving. It is closely related to the driver's driving experience and the ability to capture potential hazard factors in risky traffic scenes. The potential hazard perception ability of the driver can be specifically quantified through the constructed driver's potential hazard perception ability evaluation method. Drivers with weak potential hazard perception ability can be screened out through scoring to conduct corresponding interventions to reduce the probability of accidents. And by analyzing the evaluation factors that affect the driver's potential hazard perception ability, the driver's behavior pattern and decision-making basis when facing different risks can be obtained, which helps the autonomous driving system to more accurately simulate the decision-making process of human drivers. By learning and drawing on these evaluation methods, the autonomous driving system under the artificial intelligence system can better predict and respond to potential hazards in complex traffic environments, thereby improving the safety and stability of the system, optimizing the system design, and improving the autonomous driving system's ability to identify and respond to potential hazards.

[0003] Existing potential hazard perception assessment methods, such as the Chinese patent "A driver's potential hazard perception assessment method based on a real vehicle" with publication number CN117068181A, discloses that potential hazard research is divided into visible potential hazards and hidden potential hazards, and a real vehicle and an eye tracker are used to drive the real vehicle and collect eye movement data. Then, multiple typical potential hazard traffic scenarios are selected according to different potential hazard characteristics, and gaze characteristic indicators and scanning characteristic indicators are set to establish a gray near-optimal comprehensive evaluation model, quantify the driver's potential hazard perception ability, and obtain a comprehensive evaluation value of potential hazard perception.

[0004] However, this scheme uses actual vehicle tests and videos of real road environments, but these scenarios cannot fully cover the more representative potential risk scenarios in reality. It also uses a grey near-optimal comprehensive evaluation model for quantitative assessment. This model has certain advantages in dealing with incomplete information and uncertainty, but may be too complex and not easy to apply. Summary of the invention

[0005] The present invention aims to provide a method for evaluating a driver's ability to perceive potential danger, so as to solve the technical problem of complex quantification models when evaluating the ability to perceive potential danger in the prior art.

[0006] To achieve the above object, the present invention adopts the following technical solution: a method for evaluating a driver's potential danger perception ability, comprising the following steps:

[0007] S1. Collecting the eye movement data of the driver in different potential risk scenarios, and extracting evaluation parameters for evaluating the driver's potential risk perception ability from the eye movement data; the evaluation parameters include gaze feature parameters and glance feature parameters; the gaze feature parameters include gaze reaction time, average gaze time and attention ratio; the glance feature parameters include horizontal search breadth, vertical search depth, glance amplitude and glance number;

[0008] S2. Calculate the probability values ​​of the evaluation parameters between the driving groups in the potential risk scenario by using a statistical difference test method, and select the key evaluation parameters for quantifying the potential danger perception ability of the driver in the potential risk scenario from the evaluation parameters for each potential risk scenario, wherein the key evaluation parameters are evaluation parameters whose probability values ​​are less than a preset significance level value;

[0009] S3. Use the entropy weight method to calculate the weights of each key evaluation parameter in each potential risk scenario;

[0010] S4. Substitute the values ​​of each key evaluation parameter into the TOPSIS model, and calculate the degree of proximity between the driver's evaluation results in the potential risk scenario and the positive and negative ideal solutions to obtain the driver's potential danger perception ability score in each risk scenario;

[0011] Wherein, in S4, in the TOPSIS model, the positive ideal solution is set as the 90th percentile of each key evaluation parameter, and the negative ideal solution is set as the 10th percentile of each key evaluation parameter.

[0012] The principles and advantages of this scheme are: through driving simulation experiments, various potential risk scenarios can be simulated in a well-controlled environment to ensure the consistency and repeatability of the experiment; the relationship between the evaluation parameters of each scenario and the final score is known, and the entropy weight-TOPSIS method is used to lower the threshold for model construction and use; based on the characteristics of each risk scenario and the differences in evaluation parameters among people, the most critical evaluation parameters for evaluating the ability to perceive potential danger are screened out; this process ensures the pertinence and effectiveness of the evaluation;

[0013] Traditionally, the determination of positive and negative ideal solutions in the entropy weight-TOPSIS method relies on the actual maximum and minimum values ​​of each indicator. However, this method is easily affected by outliers in the data, which may lead to deviations in the decision-making process. Based on this, in order to improve the robustness of decision analysis, this scheme makes corresponding improvements to the determination of positive and negative ideal solutions in this method, and determines the positive and negative ideal solutions of indicators based on the decision of the quantile range to reduce the potential impact of outliers;

[0014] According to the observation of the indicator distribution, this scheme defines the positive ideal solution as the 90th percentile of the indicator, and the negative ideal solution as the 10th percentile of the indicator. The 90th percentile, that is, the 90th percentile, means that after the data is sorted, 90% of the data items are less than or equal to this value, and only 10% of the data items are greater than this value; the 10th percentile, that is, the 10th percentile, means that after the data is sorted, 10% of the data items are less than or equal to this value, and 90% of the data items are greater than this value. In this way, the performance of most decision-making schemes on various indicators is more accurately reflected, while the distortion effect caused by extreme values ​​is reduced, the impact of outliers on the evaluation results is effectively reduced, and the robustness and accuracy of the evaluation are improved;

[0015] The evaluation parameters are set to refine the evaluation dimensions, making the evaluation process more detailed and comprehensive. This can better capture the visual behavior characteristics of drivers when facing potential risk scenarios, and improve the accuracy and reliability of the evaluation.

[0016] This evaluation method is simple and easy to implement, and can be easily applied and promoted in practice, providing an important reference for the optimization of intelligent driving systems and the development of autonomous driving technology.

[0017] Preferably, as an improvement, it also includes:

[0018] S5. Determine the weight of each potential risk scenario through a questionnaire survey. The subjective questionnaire used for the questionnaire survey includes multiple evaluation dimensions. The evaluation dimensions of the subjective questionnaire include the frequency of occurrence of potential risk scenarios, the possibility of accidents caused by the potential risk scenarios, and the degree of casualties caused;

[0019] S6. Calculate the driver's comprehensive score of potential risk perception ability by combining the weight of each potential risk scenario and the driver's potential risk perception ability score in each potential risk scenario.

[0020] The beneficial effect of this improvement is: determining the weight of each risk scenario through a questionnaire survey, highlighting the importance of the risk scenario itself in actual driving, and making the evaluation results more comprehensive.

[0021] Preferably, as an improvement, S2 also includes dividing the driver's sight area into multiple interest areas by combining the traditional mechanical division method and the gaze point statistics method, and selecting key interest areas for each potential risk scene according to the difference in gaze point distribution of the driving group, and the gaze feature parameters in the key interest areas are used as calculation parameters in S3 and S4;

[0022] The areas of interest include the road ahead, the left side of the road, the right side of the road, the left rearview mirror, the right rearview mirror, and the instrument panel.

[0023] The beneficial effect of this improvement is that by dividing the interest areas according to the driver's visual characteristics and attention distribution, the driver's visual attention focus when facing different risk scenarios can be captured more accurately. Selecting the key interest areas corresponding to each risk scenario as the source of key evaluation parameters for evaluation can ensure that the evaluation process is more focused on the driver's performance at key risk points, thereby improving the accuracy of the evaluation; during the evaluation process, using the key evaluation parameters in the key interest areas as calculation parameters can simplify the evaluation process and reduce unnecessary data processing and analysis.

[0024] Preferably, as an improvement, the potential risk scenarios include illegal pedestrian crossing scenarios, pedestrian crossing scenarios on unsignaled road sections, unsignaled T-junction scenarios, marked transition section scenarios, highway confluence scenarios and parallel overtaking scenarios.

[0025] The beneficial effect of this improvement is that it makes the assessment process more comprehensive and covers more risk situations that may be encountered in actual driving. This helps to more comprehensively assess the driver's ability to perceive potential dangers and ensure the accuracy and reliability of the assessment results. By incorporating these specific scenarios into the assessment scope, the driver's ability to perceive potential dangers in different situations can be more targeted, providing more specific guidance for subsequent training and intervention.

[0026] Preferably, as an improvement, in S5, the method for determining the weight ratio of each potential risk scenario includes:

[0027] S501, taking the product of the scores of each evaluation dimension of the potential risk scenario as the risk score of the potential risk scenario;

[0028] S502. Calculate the ratio of the risk score of each potential risk scenario to the sum of the risk scores of all potential risk scenarios in a single subjective questionnaire;

[0029] S503. Calculate the average of the proportions of the risk scores of each potential risk scenario in all subjective questionnaires as the weight proportion of the potential risk scenario.

[0030] The beneficial effects of this improvement are: the risk score is obtained by calculating the product of the scores of each risk scenario in multiple evaluation dimensions. This method comprehensively considers multiple aspects of the risk scenario, making the weight determination more comprehensive and objective; avoiding the influence of a single dimension or subjective judgment on the weight, and improving the accuracy and reliability of the weight;

[0031] The average of the risk scores in all questionnaires is used to determine the weight ratio of the risk scenarios, making the weights more representative and universal; the weights determined in this way can better reflect the importance and impact of different risk scenarios in actual driving.

[0032] Preferably, as an improvement, the key area of ​​interest in the illegal pedestrian crossing scene is the left side of the road; the key area of ​​interest in the pedestrian crossing scene on the unsignaled road section is the right side of the road; the key area of ​​interest in the unsignaled T-junction scene is the right side of the road; the key area of ​​interest in the marking transition section scene is the right rearview mirror; the key area of ​​interest in the highway confluence scene is the left rearview mirror, and the key area of ​​interest in the parallel overtaking scene is the right rearview mirror.

[0033] Preferably, as an improvement, the key evaluation parameters in the illegal pedestrian crossing scene are the average fixation time of the interest area, the horizontal search breadth and the vertical search depth;

[0034] The key evaluation parameters in the pedestrian crossing scene on the non-signal-controlled road section are the fixation reaction time of the interest area, the proportion of attention to the interest area, the horizontal search breadth and the number of glances;

[0035] The key evaluation parameters in the non-signal controlled T-junction scenario are the interest area fixation reaction time, the interest area average fixation time, the interest area attention ratio, the horizontal search breadth and the number of glances;

[0036] The key evaluation parameters in the line transition scene are the interest area fixation reaction time, the interest area average fixation time, the interest area attention ratio and the horizontal search breadth;

[0037] The key evaluation parameters in the highway entrance scene are the fixation reaction time of the interest area, the average fixation time of the interest area, the attention ratio of the interest area, the horizontal search breadth and the scanning amplitude;

[0038] The key evaluation parameters in the parallel overtaking scenario are the proportion of interest area attention, horizontal search breadth, scanning amplitude and vertical search depth.

[0039] Preferably, as an improvement, in S2, the specific method for determining the key evaluation parameters is:

[0040] S201, identifying a potential risk scenario, and selecting associated evaluation parameters according to the driver's visual attention allocation and environmental safety risk in the potential risk scenario;

[0041] S202, selecting a key region of interest from the region of interest based on the distribution map of the driver's gaze points in the potential risk scenario according to the difference in gaze point distribution of different groups of people, and then determining the region of interest to which the gaze feature parameter in the associated evaluation parameter belongs;

[0042] S203, using statistical methods to verify the differences, and then selecting the associated evaluation parameters whose probability values ​​are less than the significance level values ​​from the associated evaluation parameters as key evaluation parameters.

[0043] The beneficial effect of this improvement is that by combining the gaze point distribution map to observe the differences in gaze point distribution of different groups of people in the same risk scenario, it is possible to more accurately locate those areas that are critical to the perception of potential dangers, namely the key areas of interest. This method makes the determined key evaluation parameters more targeted and can more accurately reflect the driver's ability to perceive potential dangers in specific risk scenarios; a statistical method is used to verify the significance of the differences in gaze point distribution. This method reduces the impact of subjective judgment on the evaluation results and improves the scientificity and objectivity of the evaluation. Through statistical verification, it can be more confident that the selected key evaluation parameters are effective indicators that can truly reflect the driver's ability to perceive potential dangers. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of an embodiment of the present invention.

[0045] Figure 2 Partition the map for the region of interest.

[0046] Figure 3 This is the distribution map of the driver's gaze points in scene one.

[0047] Figure 4 This is the distribution map of the driver's gaze points in scene 2.

[0048] Figure 5 This is the distribution map of the driver's gaze points in scene three.

[0049] Figure 6 This is the distribution map of the driver's gaze points in scene four.

[0050] Figure 7 This is the distribution map of the driver's gaze points in scene five.

[0051] Figure 8 This is the distribution map of the driver's gaze points in scene six.

[0052] Fig. 9 This is a schematic diagram of the risk scenario for scenario one.

[0053] Fig.10 This is a schematic diagram of the risk scenario for scenario two.

[0054] Fig.11 This is a schematic diagram of the risk scenario for scenario three.

[0055] Fig.12 This is a schematic diagram of the risk scenario for scenario four.

[0056] Fig.13 This is a schematic diagram of the risk scenario for scenario five.

[0057] Fig.14 This is a schematic diagram of the risk scenario for scenario six. DETAILED DESCRIPTION

[0058] The following is further described in detail through specific implementation methods:

[0059] Example

[0060] Basically as attached Figure 1 As shown, a method for evaluating a driver's potential danger perception ability comprises the following steps:

[0061] S1. Collect the driver's eye movement data in different potential risk scenarios, and extract evaluation parameters for evaluating the driver's potential risk perception ability from the eye movement data; the evaluation parameters include gaze feature parameters and glance feature parameters; the gaze feature parameters include gaze reaction time, average gaze time and attention ratio; the glance feature parameters include horizontal search breadth, vertical search depth, glance amplitude and glance number

[0062] S2. Calculate the probability values ​​of the evaluation parameters between the driving groups in the potential risk scenario by using a statistical difference test method, and select the key evaluation parameters for quantifying the potential danger perception ability of the driver in the potential risk scenario from the evaluation parameters for each potential risk scenario, wherein the key evaluation parameters are evaluation parameters whose probability values ​​are less than a preset significance level value;

[0063] Among them, the probability value is the p value used in statistics to express the difference between two data, and the significance level is 0.05.

[0064] S3. Use the entropy weight method to calculate the weights of each key evaluation parameter in each potential risk scenario;

[0065] S4. Substitute the values ​​of each key evaluation parameter into the TOPSIS model, and calculate the degree of proximity between the driver's evaluation results in the potential risk scenario and the positive and negative ideal solutions to obtain the driver's potential danger perception ability score in each risk scenario;

[0066] In S4, in the TOPSIS model, the positive ideal solution is set as the 90th percentile of each key evaluation parameter, and the negative ideal solution is set as the 10th percentile of each key evaluation parameter.

[0067] S5. Determine the weight of each potential risk scenario through a questionnaire survey. The subjective questionnaire used for the questionnaire survey includes multiple evaluation dimensions. The evaluation dimensions of the subjective questionnaire include the frequency of occurrence of potential risk scenarios, the possibility of accidents caused by the potential risk scenarios, and the degree of casualties caused;

[0068] S6. Calculate the driver's comprehensive score of potential risk perception ability by combining the weight of each potential risk scenario and the driver's potential risk perception ability score in each potential risk scenario.

[0069] Among them, in S1, this embodiment adopts a three-degree-of-freedom driving simulator developed by Ergoneers of Germany, which includes a simulated cockpit Mock-up, scene projection Sight Simulation and motion platform Motion Cueing System, namely, the SIMLAB automobile driving simulation system to simulate potential risk scenarios.

[0070] The wearable eye tracker DG3, namely Dikablis Glasses 3, developed by Ergoneers of Germany, is used as a visual information collection device to collect the driver's eye movement data. The sampling rate of the eye tracker is 60Hz, the tracking accuracy is 0.3°, and the eye movement collection field of view is 116° horizontally and 60° vertically. In addition, the eye tracker is equipped with a full HD scene camera with a resolution of 1920×1080P, which can achieve high-quality shooting and tracking of the scene.

[0071] Potential risk scenarios include:

[0072] (1) Scenario 1: Pedestrians crossing the street illegally; Fig. 9 As shown in the figure, the test vehicle is driving on a lane close to the center of a two-way four-lane road. There is a barrier in the middle of the road. A pedestrian crosses the road illegally in front of the test vehicle and climbs over the barrier to the lane of the test vehicle. When passing this scene, the driver should pay attention to whether the "pedestrian" suddenly crosses the road sideways when the vehicle passes by. Pedestrians who cross the road illegally have unpredictable behaviors and movement trajectories. At the same time, they are also a vulnerable group. In this scene, the driver needs to observe the dynamics of pedestrians to avoid accidents.

[0073] Scenario start time: the time frame when the driving vehicle is 100 meters away from the illegal pedestrian crossing the street; scenario end time: the time frame when the driving vehicle passes the illegal pedestrian.

[0074] (2) Scenario 2: Pedestrian crossing on a road section without signal control; Fig.10 As shown, the test vehicle is traveling on a two-way four-lane lane close to the center of the road. There is a crosswalk (without signal control) in front of the test vehicle in the direction of travel, and there is a large truck in front of the right of the test vehicle, which passes the crosswalk in the same time period as the test vehicle. The presence of the large truck will affect the driver's observation of the environment in front of the right. When the test vehicle passes the crosswalk, it will collide with pedestrians who may be crossing the street due to blocked vision.

[0075] The start time of the scene: the time frame when the vehicle is 50 meters away from the pedestrian crossing; the end time of the scene: the time frame when the vehicle passes the pedestrian crossing.

[0076] (3) Scenario 3: T-shaped intersection without signal control; Fig.11 As shown in the figure, the test vehicle is driving normally on a two-way two-lane urban road. There is a T-shaped intersection without signal control in front of the vehicle's driving direction, and the driver needs to complete a left turn at the intersection. When the test vehicle passes through a T-shaped intersection without signal control, the driver needs to judge the order of passage according to traffic rules, such as turning to give way to vehicles going straight. According to the analysis of urban road traffic accidents, it can be seen that when drivers pass through similar road scenes, they often pay more attention to the road environment on the side they are about to turn to, and ignore the situation of vehicles coming from the opposite direction, which may cause the driver to collide with the opposite straight-moving vehicle when turning, increasing the risk of accidents.

[0077] Scenario start time: the time frame when the driving vehicle is 50 meters away from the T-junction; Scenario end time: the time frame when the driving vehicle completes the turn.

[0078] (4) Scenario 4: Marking transition section; as shown in the attached Fig.12 As shown in the figure, the test vehicle is driving on a six-lane bidirectional lane close to the center of the road. There is a cross intersection in front of the vehicle's driving direction, and the test vehicle intends to go straight through. When the road width remains unchanged, the three lanes become four guide lanes, and there is an unmarked section in between for vehicles to choose a guide lane. If vehicles in two adjacent lanes choose the same guide lane at the same time, if they are not carefully observed, a lateral collision may occur between the two parties. When the driver passes through the unmarked section and enters the guide lane, he should pay attention to the dynamics of the surrounding vehicles.

[0079] The start time of the scene: the time frame when the driving vehicle is 30 meters away from the unmarked line segment; the end time of the scene: the time frame when the driving vehicle enters the guide lane line of the intersection.

[0080] (5) Scenario 5: Highway confluence entrance; as shown in the attached Fig.13 As shown, the test vehicle merges into the main road of the expressway from the branch road. During the merging process, attention should be paid to the vehicle on the main road that the vehicle is about to merge into, so as to avoid lateral collision or rear-end collision with vehicles on the main road.

[0081] Scenario start time: the time frame when the driving vehicle enters the acceleration section; Scenario end time: the time frame when the driving vehicle completes merging into the main road.

[0082] (6) Scenario 6: Parallel transcendence; as shown in the attached Fig.14As shown, the test vehicle is driving normally on the main road of a two-way four-lane highway. There is a slower car in front of the adjacent lane. The test vehicle intends to overtake the car in parallel. During the parallel overtaking process, when there is a large gap between the test vehicle and the overtaken vehicle, there may be other vehicles behind the test vehicle that accelerate and change lanes to overtake through the gap, which may cause the test vehicle to collide with the other vehicles. During this process, the driver of the test vehicle needs to pay attention to whether there are vehicles behind the test vehicle and their dynamics.

[0083] Scenario start time: the time frame when the driving vehicle is 150 meters away from the vehicle being overtaken; scenario end time: the time frame when the driving vehicle completes parallel overtaking.

[0084] Use the simulation software SILAB to create potential risk scenarios. The steps are as follows:

[0085] Step 1: Create roads and environmental factors. According to relevant specifications, the road traffic signs and markings, number of lanes, lane width and lane length are designed in accordance with different scene contents. In order to improve the realism of the scene, the vegetation, buildings and weather in the scene are reasonably designed.

[0086] Step 2: Setting the behavior of vehicles and pedestrians in the road environment. SILAB has a rich material library. The motor vehicle model covers cars, trucks, buses, etc.; the non-motor vehicle model consists of two-wheeled bicycles and electric vehicles; and the pedestrian model can choose to set different genders and ages.

[0087] Step 3: Set the moving target parameters and add its driving trajectory. Select the target and right-click to select Properties window. In this window, set the speed and acceleration of the corresponding moving target, and use the Insert Flowpoint and Insert setup point for pedestrians functions to set the moving target path of vehicles and pedestrians, such as going straight, changing lanes or turning.

[0088] Step 4: Set floating points. Based on the test scenario requirements, set activation floating points for environmental vehicles and pedestrians at specific sections or locations. When the driving vehicle passes the floating points, the environmental vehicles and pedestrians will perform the pre-set behaviors. This ensures that the dynamic scenes encountered by each test subject at the same location are basically the same, and improves the correlation between the data of different test subjects.

[0089] Step 5: Generate 3D visualization video. The created plane scene is converted into a 3D visualization traffic scene video through the SILAB software main program. In the formal test, the 3D visualization video of the potential risk scene is displayed to the driver through the ring screen projection technology.

[0090] The different groups selected in this embodiment are skilled drivers and unskilled drivers. Among them, 17 unskilled drivers with a driving experience of less than 3 years or a driving mileage of less than 10,000 kilometers are selected, and 15 skilled drivers with a driving experience of more than 3 years and a driving mileage of more than 50,000 kilometers are selected.

[0091] Seven gaze and scanning characteristic parameters closely related to the driver's visual behavior were screened out for quantitative analysis of risk scenarios, namely evaluation parameters.

[0092] The evaluation parameters include three gaze feature parameters. Through gaze, drivers can obtain necessary information more effectively and make appropriate driving decisions.

[0093] The three gaze feature parameters are:

[0094] (1) The AOI gaze reaction time is the difference between the time when the driver first looks at the potential danger source and the time when the scene starts. It represents the driver's sensitivity to potential danger sources in unfamiliar traffic environments. Since the total time for different drivers to complete dangerous scenes varies, the gaze reaction time is normalized accordingly.

[0095]

[0096]

[0097] Where, T RT is the AOI gaze reaction time (s), α is the start time of the potential risk scenario, β is the end time of the potential risk scenario, and ξ is the first time to gaze at the potential hazard source. norm-RT is the normalized AOI gaze reaction time (s), T RT-min T before normalization RT The minimum value in T RT-max T before normalization RT If the driver fails to successfully focus on the potential danger source in a potential danger scene, the driver's T norm-RT is 1.

[0098] (2) Average fixation time on the AOI, hereinafter referred to as the average fixation time on the AOI, is the ratio of the total fixation time on a certain AOI to the number of fixations within a certain period of time. The larger the value, the greater the difficulty of information processing, and the observer needs more time to interpret the information in the AOI.

[0099]

[0100] Where M represents the average fixation time of AOI (ms), P represents the total time that the subject fixates on a certain area of ​​interest within a certain period of time, and Z represents the number of fixations.

[0101] (3) The percentage of attention paid to the area of ​​interest, hereinafter referred to as the AOI percentage, is the percentage of time that the subject looks at a certain area of ​​interest within a certain period of time. It is used to measure the importance of the area of ​​interest. The higher the AOI percentage, the more attention the driver pays to the area.

[0102]

[0103] In the formula, N represents the proportion of AOI attention, r represents the duration of a fixation (i=1, 2, 3, ..., n), and R represents the duration of the selected time interval.

[0104] The evaluation parameters include 4 scanning feature parameters. Through scanning, the driver can fully understand the situation in the entire field of vision in a short time.

[0105] The four gaze feature parameters are:

[0106] (1) Horizontal search breadth refers to the standard deviation of the observer’s gaze point in the horizontal pixels in the video image captured by the eye tracker scene camera. The larger the value, the wider the horizontal observation range and the more effective information can be obtained.

[0107] (2) Vertical search depth refers to the standard deviation of the observer’s gaze point in the vertical pixels in the video captured by the eye tracker scene camera. The larger the value, the wider the vertical observation range and the more effective information can be obtained.

[0108] (3) Saccade amplitude: The saccade amplitude refers to the range covered from the beginning to the end of a saccade. It also refers to the range of eye jumps from the end of one fixation to the beginning of the next fixation. It is usually expressed in degrees of visual angle.

[0109] The scanning range is an indicator of the depth of attention. If a fixation can include a lot of information, after processing, it will jump a large distance when it is transferred to the next fixation. If a fixation can only obtain limited and little information, the subsequent scanning distance will be very small.

[0110] (4) Number of glances refers to the number of glances in the viewing direction within a certain period of time, reflecting the frequency of the driver's information search during driving.

[0111] Among them, the exported raw data has problems of gaze point loss and data anomaly due to the interference of driver activities and occlusion of eye image acquisition during the test. Linear difference is used to compensate for signal loss and replace abnormal values, and sliding mean filtering is used for noise reduction.

[0112] In S2, the driver's interest area is divided by combining the traditional mechanical division method with the gaze point one by one statistical method, as shown in the attached figure. Figure 2 The specific division is into 6 areas of interest, as shown in Table 1 below:

[0113] Table 1 Division of interest areas

[0114]

[0115] In S2, the specific method for determining the key evaluation parameters is:

[0116] S201, identifying a potential risk scenario, and selecting associated evaluation parameters according to the driver's visual attention allocation and environmental safety risk in the potential risk scenario;

[0117] S202, selecting a key region of interest from the region of interest based on the distribution map of the driver's gaze points in the potential risk scenario according to the difference in gaze point distribution of different groups of people, and then determining the region of interest to which the gaze feature parameter in the associated evaluation parameter belongs;

[0118] S203, using statistical methods to verify the difference, and then selecting the associated evaluation parameters whose probability values ​​are less than the significance level value from the associated evaluation parameters as key evaluation parameters.

[0119] (1) Scenario 1: Pedestrians crossing the street illegally. Pedestrians are a vulnerable group in road traffic and are easily injured in traffic accidents. In addition, the behavior of pedestrians is relatively unpredictable and they may suddenly appear on the road at any place and at any time. When interacting with pedestrians, motor vehicle drivers should always remain alert, pay attention to the movements of pedestrians, and take corresponding preventive measures to reduce the occurrence of traffic accidents. Based on this, AOI gaze reaction time, AOI average gaze time and AOI attention ratio represent the driver's sensitivity, information processing ability and attention level to pedestrians crossing the street illegally; horizontal search breadth and vertical search depth reflect the comprehensiveness of the driver's observation of the surrounding environment information; AOI gaze reaction time, AOI average gaze time, AOI attention ratio, horizontal search breadth and vertical search depth are used as the associated evaluation parameters of scenario 1.

[0120] As attached Figure 3As shown in the figure, the gaze distribution diagram of unskilled drivers and skilled drivers when driving past illegal pedestrians crossing the street. Both types of drivers paid attention to illegal pedestrians crossing the street, but the unskilled drivers' gaze points were mainly concentrated on the front of the road and the illegal pedestrians crossing the street, and lacked attention to other targets. In addition to paying attention to illegal pedestrians crossing the street and vehicles on the road ahead, skilled drivers also pay attention to traffic participants in the right area of ​​the road. Compared with unskilled drivers, the gaze points in the horizontal position cover a wider range. If potential danger occurs (illegal pedestrians crossing the street start to cross the road), skilled drivers have a better grasp of the information about the surrounding environment of the vehicle than unskilled drivers, and may take better countermeasures to avoid the occurrence of danger. In this scenario, AOI2 (the area on the left side of the current lane) is an area with potential danger sources, which needs to be perceived and paid attention to by the driver during driving. Based on this, AOI2 is selected as the key area of ​​interest in the scene below.

[0121] To further clarify whether the difference is statistically significant, the AOI2 gaze reaction time, AOI2 average gaze time and AOI2 attention ratio of skilled and unskilled drivers all obey normal distribution, so the independent sample t-test is used for difference analysis. The difference analysis of the gaze feature parameters of scene 1 is shown in Table 2 below:

[0122] Table 2 Analysis of differences in gaze feature parameters in scene 1

[0123]

[0124] There is a significant difference in the average AOI2 gaze time (p=0.011<0.05) between the two types of drivers in scenario 1, so the average AOI2 gaze time is selected as the key evaluation parameter for scenario 1.

[0125] The difference analysis of horizontal search breadth and vertical search depth of skilled and unskilled drivers is performed. The difference analysis of scanning feature parameters in scenario 1 is shown in Table 3 below:

[0126] Table 3 Analysis of differences in scanning feature parameters in scenario 1

[0127]

[0128] In scenario 1, there are significant differences between the two types of drivers in vertical search depth (p=0.033<0.05) and horizontal search breadth (p=0.028>0.05), so vertical search depth and horizontal search breadth are selected as the key evaluation indicators of scenario 1.

[0129] (2) Scenario 2: Pedestrians on uncontrolled roads usually have priority when crossing a pedestrian crossing on an uncontrolled road. Therefore, when a driver passes through a similar scenario, he should pay attention to whether there are pedestrians preparing to cross the road or crossing the road. In particular, when the driver's vision is blocked by other vehicles in the surrounding environment or trees on the roadside before reaching the crosswalk, he should pay more attention to the blocked area to prevent pedestrians from suddenly crossing the road. Based on this, AOI fixation reaction time and AOI attention ratio reflect the driver's perception and attention to potential danger areas; horizontal search breadth and number of glances represent the driver's information search range and frequency; AOI fixation reaction time, AOI attention ratio, horizontal search breadth and number of glances are used as related evaluation parameters for scenario 2.

[0130] As attached Figure 4 As shown in the figure, the distribution of gaze points of two types of drivers passing through the pedestrian crossing on the road without signal control, the unskilled driver focuses on the vehicles in front of the road and on the opposite lane in this potential risk scenario, while ignoring the blind spots blocked by adjacent vehicles. Compared with unskilled drivers, the skilled driver has a wider horizontal visual search range in this scenario. In addition to the front of the road, the skilled driver focuses more on the adjacent vehicles and the front of the adjacent vehicles. Based on this, AOI3 is selected as the key area of ​​interest in scenario 2.

[0131] To further clarify whether the difference is statistically significant, the AOI3 gaze reaction time of both skilled and unskilled drivers does not obey the normal distribution. The independent sample Mann Whitney test is used to test their statistical differences, while the AOI3 attention ratio meets the Shapiro-Wilk test. The t-test is used for difference analysis. The difference analysis of the gaze feature parameters of scene 2 is shown in Table 4 below:

[0132] Table 4 Analysis of differences in gaze feature parameters in scene 2

[0133]

[0134] There are significant differences in AOI3 gaze reaction time (p=0.019<0.05) and AOI3 attention ratio (p=0.002<0.05) between the two types of drivers in scenario 2. Therefore, AOI3 gaze reaction time and AOI3 attention ratio are selected as the key evaluation parameters for scenario 2.

[0135] The difference analysis of horizontal search breadth and scanning times of skilled and unskilled drivers is conducted. The difference analysis of scanning feature parameters in scenario 2 is shown in Table 5 below:

[0136] Table 5 Analysis of differences in scanning feature parameters in scene 2

[0137]

[0138] There are significant differences in the number of glances (p=0.042<0.05) and horizontal search breadth (p=0.049>0.05) between the two types of drivers in scenario 2, so the number of glances and horizontal search breadth are selected as the key evaluation indicators of scenario 2.

[0139] (3) Scenario 3: T-shaped intersection without signal lights. When passing through a T-shaped intersection without signal lights, the driver needs to determine the timing of passing based on his own judgment. Drivers generally need to abide by a series of basic traffic rules, such as turning left to go straight or turning right to turn left, to ensure safe travel at the intersection. Drivers need to adopt a more comprehensive and flexible visual search mode to perceive and process information about potential dangerous targets. Based on this, in this scenario, AOI fixation reaction time, AOI average fixation time and AOI attention ratio are used to characterize the driver's sensitivity to the potential dangerous area, key information extraction ability and attention level; horizontal search breadth, scanning amplitude and scanning number are used to quantify the driver's visual search range, attention allocation strategy and information search frequency at the intersection; AOI fixation reaction time, AOI average fixation time, AOI attention ratio, horizontal search breadth, scanning amplitude and scanning number are used as the correlation evaluation parameters of scenario 3.

[0140] As attached Figure 5 As shown in the figure, the gaze distribution of unskilled drivers and skilled drivers when preparing to turn left at a T-shaped intersection without signal lights. From the distribution of gaze points, it can be seen that the gaze points of unskilled drivers are mainly concentrated in the front and left front of the road to ensure that they will not collide with the straight or right-turning vehicles in the left front when turning left, while ignoring the vehicles that may be going straight or turning left in the opposite direction; in contrast, the horizontal visual search of skilled drivers is more extensive, focusing on the vehicles on both sides, and can more comprehensively analyze the situation of the entire intersection. Based on this, AOI3 is selected as the key area of ​​interest in scenario three.

[0141] To further clarify whether the difference is statistically significant, the AOI3 gaze reaction time and AOI3 attention ratio of unskilled drivers do not conform to the normal distribution, and the AOI3 average gaze time of skilled drivers does not conform to the normal distribution. Therefore, the three indicators are all tested using the independent sample Mann Whitney test to study their statistical differences. The difference analysis of the gaze feature parameters of scene three is shown in Table 6 below:

[0142] Table 6 Analysis of differences in gaze feature parameters in scene 3

[0143]

[0144] There are significant differences in AOI3 gaze reaction time (p=0.002<0.05), AOI3 average gaze time (p=0.021<0.05) and AOI3 attention ratio (p=0.018<0.05) between the two types of drivers in scenario three. Therefore, AOI3 gaze reaction time, AOI3 average gaze time and AOI3 attention ratio are selected as the key evaluation parameters of scenario three.

[0145] The difference analysis of horizontal search breadth, scanning amplitude and scanning frequency between skilled and unskilled drivers is performed. The difference analysis of scanning characteristic parameters in scenario three is shown in Table 7 below:

[0146] Table 7 Analysis of differences in scanning feature parameters in scenario 3

[0147]

[0148] In scenario three, there are significant differences between the two types of drivers in horizontal search breadth (p=0.025<0.05) and number of glances (p=0.022<0.05); there is no significant difference in glance amplitude (p=0.349>0.05), so horizontal search breadth and number of glances are selected as the key evaluation indicators of scenario three.

[0149] (4) Scenario 4: Marking transition section. The marking transition section exists between the front lane and the guide lane of the intersection entrance. The purpose is to allow drivers to have enough space and time to choose the appropriate guide lane when approaching the intersection so that they can enter the intersection according to their own driving direction. The potential risk of this scenario is that if vehicles on two adjacent lanes choose the same guide lane in the same time period, there may be a risk of rear-end collision or lateral collision. Based on this, in this risk scenario, AOI gaze reaction time, AOI average gaze time, AOI attention ratio, horizontal search breadth and number of glances are selected to reflect the different visual search patterns of the two types of drivers and serve as the associated evaluation parameters of scenario 4.

[0150] As attached Figure 6 The figure shows the distribution of gaze points of unskilled drivers and skilled drivers when they enter the guide lane from the roadway before the intersection entrance. From the distribution of gaze points of the two types of drivers in the figure, it can be seen that before entering the guide lane, unskilled drivers focus their gaze on the traffic participants in front and lack attention to the surrounding vehicles; in contrast, skilled drivers have a wider horizontal search and observe the traffic conditions of the adjacent lanes through the right rearview mirror. Based on this, AOI5 is selected as the key area of ​​interest in scene 4.

[0151] To further clarify whether the difference is statistically significant, the AOI5 gaze reaction time of unskilled drivers does not conform to the normal distribution, so the independent sample Mann Whitney test is used to test the statistical difference of the AOI5 gaze reaction time of the two types of drivers; the AOI5 average gaze time and AOI5 attention ratio of the two types of drivers both meet the normality, and the t-test is used for difference analysis. The calculation method is the same as that in scenes one, two, and three, and no further records are given.

[0152] In the traffic scene of the transition section of the marking section, there were significant differences in the AOI5 gaze reaction time (p=0.002<0.05), the AOI5 average gaze time (p=0.037<0.05) and the AOI5 attention ratio (p=0.009<0.05) between the two types of drivers, and all three were selected as key evaluation parameters.

[0153] The horizontal search spans of the two types of drivers met the normality, and the independent sample t-test was used for difference analysis; the number of glances of skilled drivers did not conform to the normal distribution, and the independent sample Mann Whitney test was used to test the statistical difference in the number of glances between the two types of drivers; there was a significant difference in the horizontal search span (p=0.025<0.05) between the two types of drivers; while there was no significant difference in the number of glances (p=0.155>0.05); only the horizontal search span was selected as the key evaluation parameter.

[0154] (5) Scenario 5: Highway confluence. The highway confluence refers to the area connecting the ramp and the main road of the highway. Before the vehicle merges from the ramp to the main road, the driver should pay attention to the traffic on the main road of the highway to ensure that there is enough space and time to merge onto the main road, and at the same time increase the vehicle speed to a reasonable range. Compared with low-speed urban roads, drivers need to adopt a more efficient visual search strategy. Based on this, in this scenario, AOI gaze response time, AOI average gaze time, AOI attention ratio, horizontal search breadth, scan amplitude and scan number are selected to compare the advantages and disadvantages of the visual search strategies of the two types of drivers, and used as the associated evaluation parameters of scenario 5.

[0155] As attached Figure 7 The figure shows the distribution of the gaze points of two types of drivers before merging into the main road of the expressway. From the distribution of the gaze points of the drivers in the figure, it can be seen that the gaze points of unskilled drivers are mainly concentrated on the front of the road and the dashboard before merging into the main road, and they lack observation of the vehicles coming from behind on the main road; while the gaze points of skilled drivers are more evenly distributed, and they pay attention to the front of the road, the left rearview mirror and the dashboard area. Based on this, AOI4 is selected as the key area of ​​interest in scene 5.

[0156] To further clarify whether the difference is statistically significant, the three gaze characteristic indicators of the two types of drivers all conform to the normal distribution, and the independent sample t test is used to test their statistical differences; in this potential risk scenario, the two types of drivers have significant differences in AOI4 gaze reaction time (p=0.015<0.05), AOI4 average gaze time (p=0.023<0.05) and AOI4 attention ratio (p=0.005<0.05); all are selected as the key evaluation parameters of scenario five.

[0157] The horizontal search breadth, scanning amplitude and scanning number satisfy the normal distribution. In this potential risk scenario, the horizontal search breadth statistics of the two types of drivers are t=-2.677, p=0.015<0.05, the scanning amplitude statistics are t=-2.649, p=0.014<0.05, and the scanning number statistics are t=-1.463, p=0.158>0.05. Therefore, there are significant differences in horizontal search breadth and scanning amplitude between skilled drivers and unskilled drivers, and both are selected as key evaluation parameters for scenario five.

[0158] (6) Scenario 6: Parallel overtaking. When parallel overtaking a vehicle in an adjacent lane on a highway, pay attention to the movement status of other vehicles around you to ensure that no other vehicles in the adjacent lane are overtaking in the same time period to avoid cross-flow conflicts. The speed on the highway is high, and drivers need more agile reaction capabilities and more efficient visual search strategies. Therefore, in this scenario, AOI gaze reaction time, AOI average gaze time, AOI attention ratio, horizontal search breadth, vertical search depth and scanning amplitude are selected to quantify the difference in potential danger perception between the two types of drivers and are selected as the associated evaluation parameters of scenario 6.

[0159] As attached Figure 8 The following is a distribution diagram of the gaze points of two types of drivers when they are parallel overtaking. From the distribution of the gaze points of the drivers in the figure, it can be seen that during the parallel overtaking process, the gaze points of unskilled drivers are mainly distributed in front of the road and the target vehicle; while skilled drivers not only pay attention to the traffic environment in front, but also observe the right rearview mirror to obtain the traffic flow conditions behind the adjacent lane. Based on this, AOI5 is selected as the key area of ​​interest in scenario 6.

[0160] To further clarify whether the difference is statistically significant, the independent sample Mann Whitney test was used to study the statistical difference of the AOI5 gaze reaction time of the two types of drivers; the AOI5 average gaze time and AOI5 attention ratio of the two types of drivers obeyed the normal distribution, and the independent sample t-test was used for difference analysis; in this risk traffic scenario, there was no significant difference in the AOI5 gaze reaction time (p=0.282>0.05) and AOI5 average gaze time (p=0.588>0.05) between the two types of drivers; there was a significant difference in AOI5 attention ratio (p=0.006<0.05); AOI5 attention ratio was selected as the key evaluation parameter of scenario six.

[0161] The horizontal search breadth, scanning amplitude and vertical search depth all conform to the normal distribution, and the independent sample t-test is used to analyze the difference of the data; in this potential risk scenario, there are significant differences between the two types of drivers in horizontal search breadth (p=0.026<0.05), scanning amplitude (p=0.039<0.05) and vertical search depth (p=0.041<0.05); all of them are selected as the key evaluation parameters of scenario six.

[0162] In summary, the key evaluation parameters of each risk scenario finally obtained in S2 are shown in Table 8 below:

[0163] Table 8 Selection of key evaluation parameters for risk scenarios

[0164]

[0165]

[0166] S3 and S4 use the entropy weight-TOPSIS method. The index weights are first determined using the entropy weight method, and then the TOPSIS method is used for ranking. The specific steps are as follows:

[0167] (1) Determine the original data matrix

[0168] Use the original data of the evaluation object to establish the initial matrix, n is the evaluation object, m is the evaluation parameter, then x in the matrix X ij It represents the value of the jth indicator under the i-th evaluation object, where i = 1, 2, 3, ..., n; j = 1, 2, 3, ..., m.

[0169]

[0170] (2) Matrix Normalization

[0171] In order to avoid the influence of different index properties on the results, the matrix must be normalized before calculation, and the range method is used to perform both positive and normalization on the data.

[0172] When the indicator is a positive indicator:

[0173]

[0174] When the indicator is negative:

[0175]

[0176] The normalized matrix Y is obtained. To ensure the validity of the numerical value, if there is a 0 value, the matrix needs to be shifted.

[0177] (3) Entropy calculation

[0178] According to the difference degree of each evaluation parameter and its importance to the comprehensive evaluation, the entropy value of each indicator is calculated using information entropy. The larger the entropy value of the indicator, the smaller the utility value of the information provided. j is the entropy value of index j, and the formula is as follows:

[0179]

[0180] Where P ij is the characteristic weight of the j index in the i-th sample

[0181] (4) Weight calculation

[0182] Calculate the weight W of the jth indicator j :

[0183]

[0184] (5) Indicator empowerment

[0185] The indicator weight W calculated by the entropy weight method j , calculate the weighted index value Z ij .

[0186]

[0187] (6) Determine the positive ideal solution and the negative ideal solution

[0188] When the evaluation parameter is benefit-based, the larger the value, the better; when the evaluation parameter is cost-based, the smaller the value, the better.

[0189] Positive ideal solution:

[0190]

[0191] Negative ideal solution:

[0192]

[0193] In order to reduce the impact of outliers in the data on the evaluation results, this paper defines the positive ideal solution as the 90th percentile of the indicator, and the negative ideal solution as the 10th percentile of the indicator.

[0194] (7) Calculate the distance between the evaluation object and the positive ideal solution and the negative ideal solution

[0195] Positive ideal solution distance:

[0196]

[0197] Negative ideal solution distance:

[0198]

[0199] (8) Calculate the relative closeness between the evaluation object and the optimal solution

[0200]

[0201] (9) Comprehensive ranking of evaluation values

[0202] S i The driver's potential danger perception ability score in this risk scenario is calculated based on the relative proximity S i The size of the evaluation object is sorted, S i The larger the value, the better the driver's ability to perceive potential danger in the scene. i The smaller the value, the worse the driver's ability to perceive potential danger in this scenario.

[0203] The entropy weight assignment results of each key evaluation parameter in each scenario are shown in Table 9:

[0204] Table 9 Entropy weight assignment results of key evaluation parameters

[0205]

[0206]

[0207] From this calculation, it can be seen that in scenario one, the best score for the unskilled driver's ability to perceive potential dangers was 75.2 points, the worst was 29.1 points, and the average score was 52.5 points; the best score for the skilled driver's ability to perceive potential dangers was 93.8 points, the worst was 39.9 points, and the average score was 60.3 points.

[0208] In scenario two, the best score for unskilled drivers in terms of their ability to perceive potential dangers was 67.6 points, the worst was 26.9 points, and the average score was 45.2 points; the best score for skilled drivers in terms of their ability to perceive potential dangers was 93.8 points, the worst was 48.1 points, and the average score was 66.2 points.

[0209] In scenario three, the best score for unskilled drivers in terms of their ability to perceive potential dangers was 78.6 points, the worst was 35.6 points, and the average score was 51.7 points; the best score for skilled drivers in terms of their ability to perceive potential dangers was 91.5 points, the worst was 41.8 points, and the average score was 67.8 points.

[0210] In scenario four, the best score for unskilled drivers in terms of their ability to perceive potential dangers was 67.2 points, the worst was 38.1 points, and the average score was 52.8 points; the best score for skilled drivers in terms of their ability to perceive potential dangers was 90.2 points, the worst was 48.1 points, and the average score was 72.2 points.

[0211] In scenario five, the best score for unskilled drivers in terms of their ability to perceive potential dangers was 69.3 points, the worst was 16.6 points, and the average score was 44.3 points; the best score for skilled drivers in terms of their ability to perceive potential dangers was 82.3 points, the worst was 49.9 points, and the average score was 69.3 points.

[0212] In scenario six, the best score for unskilled drivers in terms of their ability to perceive potential dangers was 71.7 points, the worst was 26.5 points, and the average score was 40.3 points; the best score for skilled drivers in terms of their ability to perceive potential dangers was 85.2 points, the worst was 49.6 points, and the average score was 65.6 points.

[0213] Different scenarios have different degrees of risk, and the magnitude of the risk is related to the driver's driving behavior and is also affected by factors such as the road environment and traffic environment. The higher the risk of a scenario, the greater the proportion of the driver's potential danger perception score in that scenario.

[0214] In S5, the method for determining the weight ratio of each risk scenario includes:

[0215] S501, taking the product of the scores of each evaluation dimension of each risk scenario as the risk score of the risk scenario;

[0216] S502, calculating the ratio of the risk score of each risk scenario to the total score in a single questionnaire;

[0217] S503. Calculate the average of the risk score proportions of each risk scenario in all questionnaires as the weight proportion of the risk scenario.

[0218] The specific methods of S5 and S6 are:

[0219] By assigning risk weights to six typical risk scenarios, the driver's comprehensive score of potential danger perception when facing typical risk scenarios is calculated.

[0220] The scenario risk assessment method using subjective questionnaires is used. Based on the operating condition hazard assessment method, subjective scenario risk assessment is conducted from three dimensions. They are the frequency of each potential risk scenario in daily life, the possibility of an accident, and the degree of casualties. For example, "How often do you think drivers encounter similar scenarios in daily life?", "When drivers encounter similar scenarios, do you think there is a possibility of an accident or danger?", and "If a traffic accident occurs in this scenario, how serious do you think the casualties may be?"

[0221] Respondents scored the three questions in each risk scenario on a scale of 1 to 5, and the product of the scores of the three questions was the risk score of the scenario. The ratio of the score of each scenario in each questionnaire to the total score was calculated, and then the average of the score ratio of each scenario in the 20 questionnaires was calculated as the weight ratio of the scenario.

[0222] Among them, 20 questionnaires were sent to 10 experts in related fields and 10 professional drivers to evaluate the risk level of each risk scenario based on their experience and intentions. The risk level weight ratio of each risk scenario is shown in Table 10:

[0223] Table 10 Risk weight ratio of risk scenarios

[0224] Scene 1 Scene 2 Scene 3 Scene 4 Scene 5 Scene 6 Weight % 8.300 13.082 6.906 16.356 31.524 23.829

[0225] Finally, it was found that there was one skilled driver with a comprehensive score of over 80 points in potential hazard perception, and his comprehensive potential hazard perception in typical risk scenarios performed well;

[0226] The comprehensive scores of potential hazard perception ability were between 70 and 80 points, including 6 skilled drivers and 4 unskilled drivers, whose comprehensive potential hazard perception performance in typical risk scenarios was medium;

[0227] There are 4 skilled and 4 unskilled drivers with comprehensive potential hazard perception scores between 60 and 70 points, and their comprehensive potential hazard perception performance in typical risk scenarios is qualified;

[0228] There were 13 drivers with scores below 60 points (failing), including 9 unskilled drivers and 4 skilled drivers.

[0229] This program screens out people whose comprehensive scores for potential hazard perception ability are lower than 60 points (failing) and conducts intervention education to improve the driver's potential hazard perception ability.

[0230] In the past, most of the scenarios used in the research on drivers' potential danger perception were shot through real road environment videos or real vehicle tests, which cannot fully cover the more representative potential risk scenarios in reality. This embodiment collects the driver's eye movement behavior data when facing typical risk scenarios through driving simulation tests, explores the driver's potential danger perception level in different typical risk scenarios, and provides more comprehensive data support for improving the driver's safe driving level.

[0231] This embodiment can provide an important reference for the intelligent driving system by studying the relationship between the driver's eye movement parameters and the ability to perceive potential dangers, optimize the system design, and improve the automatic driving system's ability to identify and respond to potential dangers. The comprehensive evaluation method for the driver's ability to perceive potential dangers constructed in this embodiment can specifically quantify the driver's ability to perceive potential dangers, and through the score, drivers with weak ability to perceive potential dangers can be screened out for corresponding intervention. It provides decision-making references for traffic management agencies, and relevant managers can formulate corresponding preventive measures for a typical risk scenario to reduce the incidence of such traffic accidents. It has certain support and reference value for the theoretical knowledge test of driving schools, and sets corresponding intervention measures for the driver's perception of potential dangers to improve their ability to perceive potential dangers in typical risk scenarios.

[0232] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A method for evaluating a driver's potential danger perception ability, characterized by: The following steps are involved: S1. Collecting eye movement data of the driver in different potential risk scenarios, and extracting evaluation parameters for evaluating the driver's potential risk perception ability from the eye movement data; the evaluation parameters include gaze feature parameters and scan feature parameters; The gaze characteristic parameters include gaze reaction time, average gaze time and attention ratio; The scanning characteristic parameters include horizontal search breadth, vertical search depth, scanning amplitude and scanning number; S2. Calculate the probability values ​​of the evaluation parameters between the driving groups in the potential risk scenario by using a statistical difference test method, and select the key evaluation parameters for quantifying the potential danger perception ability of the driver in the potential risk scenario from the evaluation parameters for each potential risk scenario, wherein the key evaluation parameters are evaluation parameters whose probability values ​​are less than a preset significance level value; S3. Use the entropy weight method to calculate the weights of each key evaluation parameter in each potential risk scenario; S4. Substitute the values ​​of each key evaluation parameter into the TOPSIS model, and calculate the degree of proximity between the driver's evaluation results in the potential risk scenario and the positive and negative ideal solutions to obtain the driver's potential danger perception ability score in each risk scenario; Wherein, in said S4, in the TOPSIS model, the positive ideal solution is set as the 90th percentile of each key evaluation parameter, and the negative ideal solution is set as the 10th percentile of each key evaluation parameter; S5. Determine the weight of each potential risk scenario through a questionnaire survey. The subjective questionnaire used for the questionnaire survey includes multiple evaluation dimensions. The evaluation dimensions of the subjective questionnaire include the frequency of occurrence of potential risk scenarios, the possibility of accidents caused by the potential risk scenarios, and the degree of casualties caused; S6. Calculate a comprehensive score of the driver's potential risk perception ability by combining the weight of each potential risk scenario and the driver's potential risk perception ability score under each potential risk scenario; The potential risk scenarios include illegal pedestrian crossing scenarios, pedestrian crossing scenarios on unsignaled roads, unsignaled T-junction scenarios, marked transition section scenarios, highway confluence scenarios and parallel overtaking scenarios.

2. A method for evaluating a driver's potential danger perception ability according to claim 1, characterized in that: The S2 also includes dividing the driver's sight area into multiple interest areas by combining the traditional mechanical division method and the gaze point statistics method, and selecting key interest areas for each potential risk scene according to the difference in gaze point distribution of the driving group, and the gaze feature parameters in the key interest areas are used as calculation parameters in S3 and S4; The areas of interest include the road ahead, the left side of the road, the right side of the road, the left rearview mirror, the right rearview mirror, and the instrument panel.

3. A method for evaluating a driver's potential danger perception ability according to claim 2, characterized in that: In S5, the method for determining the weight ratio of each potential risk scenario includes: S501, taking the product of the scores of each evaluation dimension of the potential risk scenario as the risk score of the potential risk scenario; S502. Calculate the ratio of the risk score of each potential risk scenario to the sum of the risk scores of all potential risk scenarios in a single subjective questionnaire; S503. Calculate the average of the proportions of the risk scores of each potential risk scenario in all subjective questionnaires as the weight proportion of the potential risk scenario.

4. A method for evaluating a driver's potential danger perception ability according to claim 3, characterized in that: The key area of ​​interest in the illegal pedestrian crossing scenario is the left side of the road; the key area of ​​interest in the pedestrian crossing scenario on the unsignaled road section is the right side of the road; the key area of ​​interest in the unsignaled T-junction scenario is the right side of the road; the key area of ​​interest in the marking transition section scenario is the right rearview mirror; the key area of ​​interest in the highway confluence scenario is the left rearview mirror, and the key area of ​​interest in the parallel overtaking scenario is the right rearview mirror.

5. A method for evaluating a driver's potential danger perception ability according to claim 4, characterized in that: The key evaluation parameters in the illegal pedestrian crossing scene are the average fixation time of the interest area, the horizontal search breadth and the vertical search depth; The key evaluation parameters in the pedestrian crossing scene on the non-signal-controlled road section are the fixation reaction time of the interest area, the proportion of attention to the interest area, the horizontal search breadth and the number of glances; The key evaluation parameters in the non-signal controlled T-junction scenario are the interest area fixation reaction time, the interest area average fixation time, the interest area attention ratio, the horizontal search breadth and the number of glances; The key evaluation parameters in the line transition scene are the interest area fixation reaction time, the interest area average fixation time, the interest area attention ratio and the horizontal search breadth; The key evaluation parameters in the highway entrance scene are the fixation reaction time of the interest area, the average fixation time of the interest area, the attention ratio of the interest area, the horizontal search breadth and the scanning amplitude; The key evaluation parameters in the parallel overtaking scenario are the proportion of interest area attention, horizontal search breadth, scanning amplitude and vertical search depth.

6. A method for evaluating a driver's potential danger perception ability according to claim 5, characterized in that: In S2, the specific method for determining the key evaluation parameters is: S201, identifying a potential risk scenario, and selecting associated evaluation parameters according to the driver's visual attention allocation and environmental safety risk in the potential risk scenario; S202, selecting a key region of interest from the region of interest based on the distribution map of the driver's gaze points in the potential risk scenario according to the difference in gaze point distribution of different groups of people, and then determining the region of interest to which the gaze feature parameter in the associated evaluation parameter belongs; S203, using statistical methods to verify the difference, and then selecting the associated evaluation parameters whose probability values ​​are less than the significance level value from the associated evaluation parameters as key evaluation parameters.

Images