A method for detecting cognitive distraction in lane-changing behavior based on real-time driving

Abstract

The invention relates to the technical field of traffic safety, and discloses a method for detecting cognitive distraction in lane-changing behavior based on real-time driving, comprising S1, sequentially collecting lane-changing driving data in a base state and a cognitive distraction state; S2, dividing the lane-changing driving into an intention stage and an execution stage according to node moments, and extracting characterization parameters from the lane-changing driving data; S3, selecting characterization parameters with significant differences as key parameters according to the differences in characterization parameters of a driving group in a base state and a cognitive distraction state; S4, constructing a cognitive distraction discrimination model for identifying cognitive distraction behaviors of drivers; and S5, extracting key parameters of real-time driving in a current intention stage or an execution stage and inputting them into a trained cognitive distraction discrimination model to obtain a binary decision result of a distraction state or a base state, so as to solve the technical problem that the prior art lacks accurate detection of cognitive distraction in real-time lane-changing driving.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic safety, and in particular to a method for detecting cognitive distraction in lane-changing behavior based on real-time driving. Background Art

[0002] Driving cognitive distraction refers to the phenomenon that drivers lose their focus on driving due to internal psychological factors such as distraction, road rage, and fatigue driving. This distraction is different from visual distraction (such as looking at a mobile phone) or auditory distraction (such as talking to passengers). It mainly involves the decline of the driver's cognitive ability and information processing ability, thereby increasing the risk of traffic accidents. Especially when performing complex driving operations such as changing lanes, cognitive distraction may cause the driver to misjudge the surrounding environment, thereby causing traffic accidents.

[0003] Current research on driver distraction detection mainly focuses on visual distraction. Cognitive distraction is more difficult to identify than other types of distraction, mainly because it is an intrinsic psychological process that is not easy to directly observe. Although traditional monitoring methods such as video monitoring and eye tracking can reflect the driver's behavioral changes to a certain extent, it is difficult to directly capture the changes in the driver's cognitive state. Especially in scenes that require high concentration, such as lane changing, the driver's cognitive distraction may be more hidden and difficult to detect.

[0004] At present, driving distraction identification is mainly carried out by using driving performance indicators, physiological signal monitoring and video image analysis; however, driving performance indicators have poor real-time performance, physiological signal monitoring is too difficult, and video image analysis, especially eye tracking technology, is easily affected by environmental factors such as weather and lighting; and there is a lack of cognitive distraction detection for lane changing behavior; the lane changing process involves not only the lateral movement of the target vehicle, but also dynamic interaction with multiple environmental vehicles in multiple lanes, which places high demands on the driver's attention allocation, and there is an urgent need for a cognitive distraction detection for lane changing behavior dedicated to real-time driving.

[0005] Accurate identification of cognitive driver distraction is of great significance to improving road traffic safety and is also one of the key links in the development of artificial intelligence driverless technology. The driverless system needs to be able to perceive the driver's driving status in real time and accurately so that it can take over control when necessary to avoid accidents. However, due to the limitations of existing technologies in identifying cognitive distraction during lane change driving, the safety and reliability of driverless systems are affected. Summary of the invention

[0006] The present invention aims to provide a method for detecting cognitive distraction in lane-changing behavior based on real-time driving, so as to solve the technical problem that the prior art lacks accurate detection of cognitive distraction in real-time lane-changing driving.

[0007] To achieve the above object, the present invention adopts the following technical solution: a lane-changing behavior cognitive distraction detection method based on real-time driving, comprising the following steps:

[0008] S1. sequentially collecting lane-changing driving data in a base state and a cognitive distraction state, wherein the base state is lane-changing driving without cognitive distraction behavior, and the cognitive distraction state is lane-changing driving with cognitive distraction behavior, and the lane-changing driving data includes eye movement data and vehicle operation status data;

[0009] S2. Dividing the lane-changing driving into an intention phase and an execution phase according to the node time, extracting characterization parameters for quantifying the driver's cognitive distraction behavior from the lane-changing driving data, extracting multiple characterization parameters of the intention phase as an intention characterization parameter set, and extracting multiple characterization parameters of the execution phase as an execution characterization parameter set;

[0010] The node moments include the intention phase start time, the execution phase start time and the execution phase end time. The intention phase start time is the moment when the driver first pays attention to the rearview mirror or turns on the turn signal for the first time. The execution phase start time is also the intention phase end time, which is the moment when the vehicle first starts to make a large lateral displacement or the first starts to turn the steering wheel significantly. The execution phase end time is the moment when the vehicle ends the large lateral displacement or the steering wheel returns to the center position.

[0011] S3. Calculate the probability values ​​of the characterization parameters of the driving group in the intention stage or the execution stage under the base state and the cognitive distraction state by using a statistical difference test method, and select the key parameters used to quantify the cognitive distraction behavior of the driver in the intention stage or the execution stage from the characterization parameters. The key parameters are the characterization parameters whose probability values ​​are less than the preset significance level values; multiple key parameters selected from the intention characterization parameter set are used as the intention key parameter set, and multiple key parameters selected from the execution characterization parameter set are used as the execution key parameter set;

[0012] S4. constructing a cognitive distraction discrimination model for identifying cognitive distraction behaviors of drivers, and training the cognitive distraction discrimination model using a key parameter set;

[0013] S5. Identify whether the current real-time driving data meets the node time, extract the key parameters of the current intention stage or execution stage and input them into the trained cognitive distraction discrimination model to obtain the binary decision result of the distraction state or the base state.

[0014] The principles and advantages of this scheme are as follows: this method is specifically used to detect cognitive distraction in complex driving operations such as lane changing. Since the driver's focus is different at different stages in lane changing driving, the intention stage and the execution stage of lane changing behavior are accurately divided through node moments, so that the key parameters screened out by differences can more accurately characterize the state of the current driving operation, thereby improving the accuracy and reliability of detection; at the same time, statistical methods are used to perform difference tests to ensure that the screened key parameters are significant; by collecting eye movement data and vehicle operation status data in real time, the method can instantly analyze and identify the driver's cognitive distraction state during driving; the screened key parameter set is used to train the cognitive distraction discrimination model, ensuring the accuracy and reliability of the model in identifying cognitive distraction behaviors;

[0015] This method not only provides a powerful tool for traditional driving safety monitoring, but also provides important support for the development of intelligent driving systems. Intelligent driving systems need to perceive the driver's driving status in real time and accurately so that they can take over control when necessary. This method provides key driving status information for intelligent driving systems by detecting the driver's cognitive distraction status in real time, which helps to improve the safety and reliability of unmanned driving systems.

[0016] Preferably, as an improvement, the cognitive distraction detection process of S5 includes:

[0017] Step 1: Trigger the intention phase start time detection, detect whether the real-time driving triggers the intention phase start time, if triggered, determine that the current driving enters the intention phase, and enter the intention phase detection process;

[0018] Step 2: Detection process of the intention stage: continuously obtain lane change driving data of the intention stage to form a time series until the width of the time series reaches the preset sliding window width, extract the key parameters of the intention stage from the sliding window, input the cognitive distraction discrimination model, and judge whether there is cognitive distraction behavior in the current driving. If there is cognitive distraction behavior, exit the cognitive distraction detection process; if there is no cognitive distraction behavior, move the sliding window to the next time window, and re-extract the key parameters for detection by combining the current sliding time window and the historical sliding time window;

[0019] Step 3: Continuous monitoring during the intention phase, and continuous looping of step 2 during the intention phase until cognitive distraction behavior or lane change driving data is identified to trigger the start of the execution phase;

[0020] Step 4: Trigger the execution phase start time detection, detect whether the real-time driving triggers the execution phase start time, if triggered, determine that the current driving enters the execution phase, and enter the execution phase detection process;

[0021] Step 5: Detection process of the execution phase: continuously obtain the lane change driving data of the execution phase to form a time series until the width of the time series reaches the width of the sliding window, extract the key parameters of the execution phase from the sliding window, input the cognitive distraction discrimination model, and determine whether there is cognitive distraction behavior in the current driving. If there is cognitive distraction behavior, exit the cognitive distraction detection process. If there is no cognitive distraction behavior, move the sliding window to the next time window, and re-extract the key parameters for detection by combining the current sliding time window and the historical sliding time window;

[0022] Step 6: Continuous monitoring during the execution phase. Step 5 is continuously repeated during the execution phase until cognitive distraction behavior or lane change driving data is identified, triggering the execution phase deadline.

[0023] The beneficial effects of this improvement are: by continuously acquiring and processing lane change driving data through a sliding window, the detection process can track the driver's driving state changes in real time. This continuity and real-time nature helps to promptly detect and respond to the driver's cognitive distraction behavior; by combining the data of the current sliding window and the historical sliding window to extract and detect key parameters, it can reduce misjudgments caused by data fluctuations at a single time point;

[0024] The detection process covers the intention and execution stages of lane-changing behavior, and realizes the full-process monitoring of the driver's cognitive distraction behavior during the lane-changing process. This full-process monitoring helps to fully understand the driver's driving state changes and provide comprehensive data support for the decision-making of the intelligent driving system;

[0025] Compared with the commonly used dynamic time window, the sliding time window used in this solution allows the window to slide in time, that is, to continuously process and update the data. This flexibility enables the detection process to track the driver's driving state changes in real time, capture more subtle and short-term change characteristics, reduce repeated calculations, and ensure data continuity and real-time detection.

[0026] Preferably, as an improvement, the width of the sliding window is selected in the range of [0.5, 2.5] seconds.

[0027] The beneficial effect of this improvement is determined by analyzing the distribution range of the intention time window under different states of the driver in S1 and comprehensively considering the effectiveness of recognition.

[0028] Preferably, as an improvement, from the sliding window width selection range, the optimal sliding window width for the intention stage is selected as 0.5 seconds, and the optimal sliding window width for the execution stage is selected as 0.6 seconds.

[0029] The beneficial effect of this improvement is that the selection of sliding window width value has an important impact on distraction detection. If the sliding window width value is too small, the data in a single time window is not enough to determine the distraction state, and the recognition accuracy is low. If the sliding window width value is too large, it may cause a delay in the response time of distraction behavior, reducing the real-time performance of detection;

[0030] According to the accuracy analysis of the detection results under different sliding window widths in the intention stage and the execution stage, when the sliding window width is 1.2 seconds in the intention stage, the model has the highest accuracy, reaching 96.6%, and when the sliding window width is 0.5 seconds in the intention stage, the model accuracy is 96.1%; when the sliding window width is 1.3 seconds in the intention stage, the model accuracy is the highest, reaching 99.2%, and when the sliding window width is 0.6 seconds in the intention stage, the model accuracy is 97.2%; however, only by detecting cognitive distraction before the driver crosses the lane dividing line and enters the target lane can key warnings or interventions be provided, thereby effectively avoiding traffic accidents and ensuring the smoothness and safety of road traffic. Therefore, 0.5 seconds and 0.6 seconds are selected as the optimal sliding window widths respectively to improve the effectiveness of real-time detection of cognitive distraction.

[0031] Preferably, as an improvement, the characterization parameters include eye movement characteristic parameters and vehicle characteristic parameters;

[0032] Eye movement characteristic parameters include: pupil area and pupil area change rate used to characterize the driver's pupil changes, gaze frequency and average gaze time used to characterize the driver's gaze, scanning frequency, average scanning time and average scanning amplitude used to characterize the driver's scanning, and rearview mirror gaze times and average single rearview mirror gaze time used to characterize the driver's rearview mirror inspection behavior;

[0033] The vehicle characteristic parameters include: the turn signal opening rate and the relative opening time of the turn signal used to characterize the usage of the turn signal, the steering wheel angle and steering wheel angle entropy used to characterize the steering wheel control situation, the vehicle speed, the vehicle speed standard deviation and the longitudinal angular velocity used to characterize the longitudinal running state of the vehicle, and the lateral acceleration, yaw angular velocity, lateral offset distance and maximum lateral acceleration used to characterize the lateral running state of the vehicle.

[0034] The beneficial effects of this improvement are: eye movement characteristics can directly reflect the driver's visual attention and cognitive load, while vehicle characteristics indirectly reflect the driver's driving behavior and vehicle status. The combination of the two provides a richer and more accurate source of information for cognitive distraction detection; pupil area and pupil area change rate can reflect the driver's psychological tension; gaze frequency and average gaze time can reveal the driver's visual attention allocation; and vehicle characteristic parameters such as speed standard deviation and lateral offset distance can intuitively show the stability of vehicle operation. Only by comprehensively analyzing these parameters can we more accurately determine whether the driver has cognitive distraction behavior; different drivers and driving scenarios may show different changes in characteristic parameters. This improvement enhances the adaptability of the detection system by introducing diverse characterization parameters.

[0035] Preferably, as an improvement, S3 also includes performing a similarity comparison between multiple key parameters in the intention stage or the execution stage, dividing the single or multiple key parameters reflecting the same information of the driver into a single key parameter group, and selecting one key parameter from each of the multiple key parameter groups as a representative key parameter as the final intention key parameter set or execution key parameter set.

[0036] The beneficial effect of this improvement is that through similarity comparison and grouping, single or multiple key parameters that reflect the same information of the driver can be merged into a key parameter group. This can reduce data redundancy and make subsequent data processing and analysis more efficient; selecting a representative key parameter from each of the multiple key parameter groups as the final parameter set can significantly reduce the amount of data that needs to be processed. This can not only reduce the consumption of computing resources, but also improve the execution efficiency of the detection process, allowing the system to respond more quickly to changes in the driver's driving state; by selecting representative parameters from each group, it can be ensured that the final parameter set can fully and accurately reflect the driver's driving state, thereby enhancing the robustness of the detection.

[0037] Preferably, as an improvement, the intention key parameter set includes: pupil area change rate, average fixation time, average glance time, average glance amplitude, vehicle speed mean, vehicle speed standard deviation and lateral offset distance;

[0038] The key execution parameter set includes: pupil area change rate, average fixation time, average glance time, average glance amplitude, steering wheel angle standard deviation, vehicle speed mean, longitudinal acceleration standard deviation, lateral acceleration standard deviation and yaw rate standard deviation.

[0039] The beneficial effect of this improvement is that the selection of representative key parameters involves multiple factors, such as the depth of information acquisition, sensitivity to task relevance, degree of reflection of individual differences and stability of data, ensuring that the model can more comprehensively and accurately detect the driver's cognitive distraction state at the intention stage and execution stage.

[0040] Preferably, as an improvement, in S1, a simulated driving test is used to obtain lane-changing driving data of drivers in a base state and a cognitive distraction state, and the driver is used to perform secondary tasks during autonomous lane-changing driving to simulate cognitive distraction behavior; the secondary tasks include a conversation for simulating distraction caused by interactive question-and-answer between the driver and others and a continuous subtraction operation for simulating distraction caused by spontaneous thinking of the driver.

[0041] The beneficial effect of this improvement is that in real life, drivers' cognitive distractions can be divided into two categories: one is caused by interactive communication with others, such as chatting with passengers or answering phone calls, which can easily distract the driver and increase the risk of traffic accidents; the other is caused by the driver's own thinking activities, such as immersing himself in thinking about work and family affairs. This solution designs two sub-tasks to simulate the above common types of distraction by increasing the driver's cognitive load and exposing him to a cognitive distraction environment; so that the simulation results have a high practical application value; and in the simulated driving test, experimental variables such as the difficulty, duration, and timing of the sub-task can be easily controlled to study the impact of these variables on the driver's driving behavior and cognitive state. This control helps to have a deeper understanding of the generation mechanism and influencing factors of distraction behavior.

[0042] Preferably, as an improvement, the S1 also includes an environmental vehicle set up on the simulated driving road to induce the driver's autonomous lane change behavior, the environmental vehicle including a low-speed vehicle located in the current lane of the driving vehicle and a high-speed vehicle located in the target lane for the driving vehicle to change lanes.

[0043] The beneficial effect of this improvement is that if the vehicle in front of the driver in the current lane is traveling at a slower speed than the driving vehicle, and the vehicle in the adjacent lane is traveling at a faster speed, the vehicle in front of the current lane blocks the driver's path to maintain a stable driving speed. Only by changing lanes to the adjacent lane can the driver's smooth driving needs be met; the speed and position design of the environment vehicle can greatly stimulate the subject's lane-changing intention, and the lane-changing driving data obtained through autonomous lane-changing driving is more in line with the actual driving situation.

[0044] Preferably, as an improvement, S4 includes:

[0045] S401, training sample selection, the intention key parameter set and the execution key parameter set are used as the original data set, and the original data set is divided into a training set and a test set;

[0046] S402, data preprocessing, using the Z-score method adopted in this paper to standardize the data;

[0047] S403, model training, establishing a cognitive distraction discrimination model based on the random forest algorithm, using the training set data to fit the cognitive distraction discrimination model, and the random forest will be trained by constructing multiple decision trees; during the training process, the random forest will sample the data multiple times and generate different decision trees, and each tree will randomly select and split some features;

[0048] S404, model evaluation, the evaluation indicators of the cognitive distraction discrimination model include accuracy, precision, recall and F1 score;

[0049] S405: Real-time detection of cognitive distraction, testing samples of the test set.

[0050] The beneficial effects of this improvement are: Z-score ensures the consistency and comparability of the input of the cognitive distraction discrimination model, which helps to improve the generalization ability of the cognitive distraction discrimination model. This integrated learning method in S403 can significantly improve the classification accuracy and stability of the cognitive distraction discrimination model. The random forest algorithm performs well in processing high-dimensional data and complex classification problems, and can accurately identify the cognitive distraction of drivers in different states; the random forest algorithm provides a wealth of parameter adjustment options, such as the number of decision trees, the randomness of feature selection, the threshold of splitting, etc. These parameters can be flexibly adjusted and optimized according to actual data and needs to achieve the best classification effect. This flexibility enables the cognitive distraction discrimination model to adapt to complex lane change driving scenarios and different driving groups. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flow chart of an embodiment of the present invention.

[0052] Figure 2 Design a diagram for the test scenario.

[0053] Figure 3 Flowchart for cognitive distraction detection.

[0054] Figure 4 This is a line graph showing the relationship between the sliding window width and accuracy in the intention stage.

[0055] Figure 5 It is a line graph showing the relationship between the sliding window width and accuracy during the execution phase. DETAILED DESCRIPTION

[0056] The following is further described in detail through specific implementation methods:

[0057] Example

[0058] Basically as attached Figure 1 As shown, a lane-changing behavior cognitive distraction detection method based on real-time driving includes the following steps:

[0059] S1. sequentially collecting lane-changing driving data in a base state and a cognitive distraction state, wherein the base state is lane-changing driving without cognitive distraction behavior, and the cognitive distraction state is lane-changing driving with cognitive distraction behavior, and the lane-changing driving data includes eye movement data and vehicle operation status data;

[0060] S2. Dividing the lane-changing driving into an intention phase and an execution phase according to the node time, extracting characterization parameters for quantifying the driver's cognitive distraction behavior from the lane-changing driving data, extracting multiple characterization parameters of the intention phase as an intention characterization parameter set, and extracting multiple characterization parameters of the execution phase as an execution characterization parameter set;

[0061] The node moments include the intention phase start time, the execution phase start time and the execution phase end time. The intention phase start time is the moment when the driver first pays attention to the rearview mirror or turns on the turn signal for the first time. The execution phase start time is also the intention phase end time, which is the moment when the vehicle first starts to make a large lateral displacement or the first starts to turn the steering wheel significantly. The execution phase end time is the moment when the vehicle ends the large lateral displacement or the steering wheel returns to the center position.

[0062] S3. Use a statistical difference test method to verify the differences in representation parameters of the driving groups in the intention stage or the execution stage under the base state and the cognitive distraction state, and select the representation parameters with probability values ​​less than the preset significance level as the key parameters for quantifying the cognitive distraction behavior of the driver in the intention stage or the execution stage; then multiple key parameters selected from the intention representation parameter set are used as the intention key parameter set, and multiple key parameters selected from the execution representation parameter set are used as the execution key parameter set.

[0063] S4. constructing a cognitive distraction discrimination model for identifying cognitive distraction behaviors of drivers, and training the cognitive distraction discrimination model using a key parameter set;

[0064] S5. Identify whether the current real-time driving data meets the node time, extract the key parameters of the current intention stage or execution stage and input them into the trained cognitive distraction discrimination model to obtain the binary decision result of the distraction state or the base state.

[0065] In S1, in order to ensure that sufficient and effective driving data of drivers performing lane change behaviors in a state of cognitive distraction is obtained, this embodiment adopts a simulated driving design experiment to collect driving data, and the experiment is set up as follows:

[0066] The driving simulator SimLab, a highly integrated and advanced simulation platform, is used to simulate the vibration of the vehicle engine and the driving experience of vehicle acceleration, deceleration, bumps, turns, etc. in typical road scenarios, and can collect the driver's physiological and psychological data such as brain electricity, skin electricity, eye movement, driving behavior data, and vehicle operation status data in real time. This experiment is based on the two-dimensional scene edited and constructed by SiLab static and dynamic traffic scenes. At the same time, the DG3 glasses-type eye tracking system is used to collect the driver's eye movement data.

[0067] As attached Figure 2 As shown in the figure, the scenario design of this test is: a two-way two-lane straight highway with a total length of 10km. The first 4km of the section is the driver's adaptation stage, the middle 1km, that is, between 4km and 5km, is the early exposure stage where the driver is immersed in the distraction task, and the last 5km is the test stage. Seeking the expected speed and avoiding slow vehicles ahead, especially large trucks, are the main factors for drivers to change lanes.

[0068] To further enable the driver to change lanes autonomously, environmental vehicles are set up in the adaptation phase and the test phase;

[0069] The environmental vehicles in the adaptation stage include: on the right lane, that is, the current lane where the driven vehicle is located, a large truck with a driving speed of 70km / h is set at 100m and 400m in front of the starting position; on the left lane, a car with a driving speed of 90km / h is set at 150m and 200m in front of the starting position. The starting position is the starting point of the driven vehicle.

[0070] The environmental vehicles in the test phase include: on the right lane, i.e. the current lane where the driven vehicle is located, there are large trucks with a driving speed of 70km / h set at 5600m, 6000m, 6400m, 6800m and 7200m ahead of the starting position; on the left lane, there are cars with driving speeds of 60km / h, 90km / h, 95km / h, 105km / h and 115km / h set at 5600m, 5650m, 5680m, 5700m and 5720m ahead of the starting position respectively.

[0071] During the adaptation phase, the driver first overtakes all vehicles at a speed of personal preference based on his or her judgment of the road conditions. After overtaking all vehicles, the test organizer directs the driver to the right lane and adjusts the speed to a constant speed of 100 km / h. When the driver reaches the early exposure phase, the instruction directs the driver to adjust the speed to a constant speed of 80 km / h again, and informs the driver of the secondary task and that he or she can drive according to his or her preferences in the subsequent test phase. During the test phase, the driver performs the secondary task and changes lanes autonomously.

[0072] The tasks included conversation and continuous subtraction operations; conversation simulated the distraction caused by the driver's interactive question-and-answer session with others. The content of the conversation mainly involved some common sense questions and generally applicable topics. The difficulty of the questions was equal to ensure that all subjects could understand and participate in them, avoid questions that were too simple or too complex, and ensure that all subjects felt comfortable during the conversation and did not increase cognitive load due to difficult questions. Continuous subtraction operations simulated the distraction caused by the driver's spontaneous thinking, referred to as calculation. When the driver drove to the early exposure stage, the test organizer would give the driver a three-digit number, such as 788. During the test stage, the driver calculated the continuous subtraction operation from 788 to 3 until the driving task was completed. During this period, the test organizer would not interfere with the driver.

[0073] This test selected people of different genders and different driving proficiency as the test population, so that the final set of key parameters is more comprehensive; the unskilled driving group refers to the driving group with a driving mileage of less than 40,000 kilometers, with a total of 18 people, including 10 males and 8 females, with an average age of 23.49 years old and an average driving mileage of 22,500 kilometers; the skilled driving group refers to the driving mileage of more than 40,000 kilometers, with a total of 15 people, including 8 males and 7 females, with an average age of 38.01 years old and an average driving mileage of 365,900 kilometers.

[0074] The driving data obtained by the test group driving in the adaptation phase reflects the driving performance of different drivers in the base state; the driving data obtained by the test group driving and performing the conversation sub-task in the test phase reflects the driving performance of different drivers in the conversation state; the driving data obtained by the test group driving and performing the calculation sub-task in the test phase reflects the driving performance of different drivers in the calculation state. This embodiment divides the cognitive distraction state into a conversation state and a calculation state, and collects lane change driving data in each of the two cognitive distraction states, so that the key parameter set finally obtained is more representative.

[0075] This embodiment collects eye movement data and vehicle operation status data during driving and removes abnormal data, that is, individual values ​​in the sample set are significantly different from most data values. The Laida criterion method (3σ) is used to remove abnormal data in the sample data;

[0076] Assume that the total number of experimental sample data is X, and a certain measurement data X i The average of all measurements The difference is greater than 3 times the standard deviation σ, which indicates that the measured data X i Should be eliminated, the following formula should be followed:

[0077]

[0078] The results of multiple tests show that the random variable follows a normal distribution and the measured values ​​are in The probability of being within the interval is 99.73%, and the probability of appearing outside the range is only 0.27%. Therefore, if there is data outside the range in the test data, it should be eliminated.

[0079] In S2, the lane change phase is divided into the intention phase and the execution phase. In the intention phase, the driver mainly focuses on ensuring a safe distance from the surrounding vehicles, collecting and evaluating the traffic conditions of the target lane, and making judgments and decisions on lane change. In the execution phase, the driver's main driving task is to manipulate the steering wheel, accelerator and brake pedals, control the lateral and longitudinal movement of the vehicle, and change the vehicle from the current lane to the target lane until the vehicle is driving smoothly in the target lane.

[0080] The node times used to divide the intention stage and the execution stage include: the intention stage start time T0, the execution stage start time T1 (also the intention stage end time) and the execution stage end time T2. The data between T0 and T1 are the intention stage samples, and the data between T1 and T2 are the execution stage samples.

[0081] The initial intention phase window is set to 10 seconds. Assuming that within 10 seconds before the execution phase, the frequency of the driver's attention to the rearview mirror is n (n = 1, 2, 3, ..., n), the time of the first attention to the rearview mirror is taken as the base point, and the difference between this time and the start time of the execution phase is calculated, which is defined as the first intention time window width candidate value. At the same time, the time when the driver first turns on the turn signal is recorded, and the difference between this time and the start time of the execution phase is calculated, and it is recorded as the second intention time window width candidate value. Finally, the first intention time window width candidate value and the second intention time window width candidate value are compared, and the maximum value is selected as the final intention time window width value. T1 and T2 can be determined jointly according to the lateral position of the vehicle combined with the steering wheel angle data. T1 is the time point when the vehicle first starts to make a large lateral displacement or the first time to turn the steering wheel significantly, that is, it starts to drive from the current lane to the target lane; T2 The end time of the execution phase is the time when the vehicle ends the large lateral displacement or the steering wheel returns to the center, that is, it has been driving on the target lane and keeps moving forward steadily.

[0082] Extracting characterization parameters from the eye movement data and vehicle operation status data collected in the experiment, wherein the characterization parameters include eye movement characteristic parameters and vehicle characteristic parameters;

[0083] Eye movement characteristic parameters include: pupil area (mm) used to characterize the driver's pupil changes 2) and pupil area change rate (%), gaze frequency (times / s) and average gaze time (s) for characterizing gaze conditions, glance frequency (times / s), average glance time (ms) and average glance amplitude (°) for characterizing glance conditions, and the number of rearview mirror gazes (times) and average single rearview mirror gaze time (s) for rearview mirror inspection behavior.

[0084] Pupillary changes are regulated by the autonomic nervous system, especially the interaction between the vagus nerve and the sympathetic nerve. In the situation of cognitive distraction, the driver may experience pupil dilation, that is, the process of enlargement of pupil area and pupil diameter. The pupil area can reflect the driver's attention level and cognitive load. The pupil area index may not be universal enough due to the heterogeneity of drivers, while the pupil area change rate can more accurately capture the physiological changes of the driver under the state of cognitive distraction.

[0085] The pupil area change rate calculation formula is as follows:

[0086]

[0087] In the formula, is the pupil area change rate, PA i is the measured value of pupil area, and PA0 is the initial value of pupil area (average value in the 2-min resting state).

[0088] The vehicle characteristic parameters include: turn signal opening rate (%) and turn signal relative opening time (s) used to characterize the use of turn signals, steering wheel angle (°) and steering wheel angle entropy used to characterize the steering wheel control situation, vehicle speed (km / h), vehicle speed standard deviation (km / h) and longitudinal angular velocity (m / s 2 ), used to characterize the lateral acceleration of the vehicle's lateral running state (m / s 2 ), yaw rate (rsd / s 2 ), lateral offset distance (m) and maximum lateral acceleration (m / s 2 ).

[0089] The turn signal activation rate refers to the percentage of the number of times the turn signal is activated to the total number of lane changes; the relative turn signal activation time accurately quantifies the time difference between the moment the driver activates the turn signal and the actual start of the lane change action; the steering wheel angle refers to the offset angle of the steering wheel relative to the return position at a specific moment; the steering wheel angle entropy parameter measures the randomness and uncertainty of the driver's use of the steering wheel during driving, and is obtained by calculating the information entropy of the steering wheel angle value.

[0090] In S3, according to the difference in performance of the driving group in the base state and the distracted state during the intention stage or the execution stage, the representation parameters with significant difference performance are used as the key parameters of this stage;

[0091] This embodiment uses the standard deviation of a single factor to analyze the differences of each characterization parameter under the three driving conditions, and applies the independent sample T in the statistical difference test method to further test the statistical significance of each difference characterization parameter. The probability value of the characterization parameter under the three driving conditions, i.e., the P value, is calculated, and the significance level value is preset to 0.05. When P<0.05, the characterization parameter has a significant difference and can be used as a key parameter in the lane change stage.

[0092] The specific method of characterizing parameter difference analysis is as follows, for example, for the intention stage:

[0093] (1) Pupil area (mm) 2 )

[0094] The pupil area data of drivers in different driving states are as follows: in the conversation state, the average pupil area is 1552.871mm 2 , the standard deviation of pupil area is 396.385; under the calculation state, the mean pupil area is 1558.778mm 2 , the standard deviation of pupil area is 559.330; in the base state, the mean pupil area is 1408.320mm 2 , the standard deviation of pupil area is 427.802.

[0095] The differences in pupil area were further tested: the F value of all drivers was 3.322, the P value was 0.037, and multiple comparisons P 交谈&基态 =0.028, P 计算&基态 =0.025.

[0096] (2) Pupil area change rate (%)

[0097] The data of pupil area change rate of drivers under different driving conditions are as follows: in the conversation state, the mean pupil area change rate is 32.682% and the standard deviation of pupil area change rate is 9.930; in the calculation state, the mean pupil area change rate is 34.587% and the standard deviation of pupil area change rate is 8.516; in the base state, the mean pupil area change rate is 20.182% and the standard deviation of pupil area change rate is 10.292.

[0098] The differences in pupil area were further tested: the F value of all drivers was 65.184, the P value was 0.000, and multiple comparisons P 交谈&基态 =0.000, P 计算&基态 =0.000.

[0099] (3) Gaze frequency (times / s)

[0100] In the conversation state, the mean is 1.700 and the standard deviation is 0.335; in the calculation state, the mean is 1.640 and the standard deviation is 0.460; in the base state, the mean is 2.146 and the standard deviation is 0.434.

[0101] Further differences were tested: the F value for all drivers was 44.423, the P value was 0.000, and multiple comparisons P 交谈&基态 =0.000, P 计算&基态 =0.000.

[0102] (4) Average fixation time (s)

[0103] In the conversation state, the mean is 0.525 and the standard deviation is 0.109; in the calculation state, the mean is 0.569 and the standard deviation is 0.141; in the base state, the mean is 0.405 and the standard deviation is 0.094.

[0104] Further differences were tested: the F value for all drivers was 52.291, the P value was 0.000, and multiple comparisons P 交谈&基态 =0.000, P 计算&基态 =0.000.

[0105] (5) Scanning frequency (times / s)

[0106] In the conversation state, the mean is 0.191 and the standard deviation is 0.259; in the calculation state, the mean is 1.115 and the standard deviation is 0.194; in the base state, the mean is 1.326 and the standard deviation is 0.329.

[0107] Further differences were tested: the F value for all drivers was 15.413, the P value was 0.000, and multiple comparisons P 交谈&基态 =0.000, P 计算&基态 =0.000.

[0108] (6) Average scanning time (ms)

[0109] In the conversation state, the mean is 67.823 and the standard deviation is 20.709; in the calculation state, the mean is 68.631 and the standard deviation is 20.307; in the base state, the mean is 59.605 and the standard deviation is 17.119.

[0110] Further differences were tested: the F value for all drivers was 6.305, the P value was 0.002, and multiple comparisons P 交谈&基态 =0.004, P 计算&基态 =0.002.

[0111] (7) Average scanning amplitude (°)

[0112] In the conversation state, the mean is 11.325 and the standard deviation is 3.995; in the calculation state, the mean is 11.213 and the standard deviation is 4.409; in the base state, the mean is 7.538 and the standard deviation is 2.724.

[0113] Further differences were tested: the F value of all drivers was 31.579, the P value was 0.000, and multiple comparisons P 交谈&基态 =0.000, P 计算&基态 =0.000.

[0114] The difference analysis method of the remaining representation parameters in the intention stage is the same as the above method; and the difference analysis of the representation parameters in the execution stage is the same as that in the intention stage, which will not be repeated here. The representation parameters with a difference test P value < 0.05 are selected as key parameters.

[0115] Based on the differences in the performance of the various representation parameters in the intention stage and the execution stage in S3, the representation parameters with significant differences, namely the key parameters, are selected to perform cognitive distraction detection on real-time driving in the intention stage and the execution stage in turn.

[0116] As attached Figure 3 As shown, the specific method of cognitive distraction detection is as follows:

[0117] Step 1: Detect the start time of the trigger intention phase;

[0118] Specifically, step 1.1: real-time monitoring of driving data, and identification of whether the lane change behavior intention phase start time has been reached according to the node time, that is, whether there is the first behavior of paying attention to the rearview mirror or turning on the turn signal for the first time; step 1.2: if the conditions are identified, it is determined that the current driving has entered the lane change behavior intention phase, and the intention phase detection process is entered;

[0119] Step 2: Detection process in the intention phase, lane change driving data acquisition and cognitive distraction detection in the intention phase;

[0120] Specifically: Step 2.1: Continuously acquire driving data to form a time series; Step 2.2: Determine whether the time window width of the current data sequence reaches the sliding window width; if not, continue to acquire data; if reached, proceed to the next step; Step 2.3: Extract key parameters of the intention stage from the sliding window. Step 2.4: Use the cognitive distraction discrimination model to determine whether these key parameters indicate the presence of cognitive distraction behavior; if there is cognitive distraction behavior, output the detection result and stop the cognitive distraction detection process; if there is no cognitive distraction behavior, move the sliding window to the next time window, and re-extract key parameters for detection in combination with the current sliding time window and the historical sliding time window until cognitive distraction behavior is identified or the intention stage ends;

[0121] Step 3: Continuous monitoring during the intention phase;

[0122] Specifically, step 3.1: in the intention phase, the data acquisition and distraction detection steps in step 2 are continuously looped until cognitive distraction behavior or lane change driving data is identified to trigger the start of the execution phase, that is, the moment when the vehicle first starts to make a large lateral displacement or first starts to turn the steering wheel significantly;

[0123] Step 4: Trigger the start time detection of the execution phase;

[0124] Specifically, step 4.1: according to the node time, that is, the time when the vehicle first starts to make a large lateral displacement or first starts to turn the steering wheel significantly, identify whether the execution phase start time of the lane change behavior has been reached; step 4.2: if the conditions are identified to be met, it is determined that the current driving enters the execution phase of the lane change behavior, and enters the detection process of the execution phase;

[0125] Step 5: Execution phase data acquisition and distraction detection;

[0126] Specifically, step 5.1: continuously acquire lane-changing driving data in the execution phase to form a time series; step 5.2: determine whether the time window width of the current data sequence reaches the sliding window width; if not, continue to acquire data; if reached, proceed to the next step; step 5.3: extract key parameters of the execution phase from the sliding window; step 5.4: use the cognitive distraction discrimination model to determine whether these key parameters indicate the presence of cognitive distraction behavior; if there is cognitive distraction behavior, output the detection result and stop cognitive distraction detection; if there is no cognitive distraction behavior, move the sliding window to the next time window, combine the current sliding time window and the historical sliding time window to re-extract key parameters for detection until distraction behavior is identified or the execution phase ends;

[0127] Step 6: Continuous monitoring during the execution phase

[0128] Specifically, step 6.1: in the execution phase, the data acquisition and distraction detection steps in step 5 are continuously looped until cognitive distraction behavior or lane change driving data is identified to trigger the execution phase end time, that is, the time when the vehicle ends the large lateral displacement or the steering wheel returns to the center position;

[0129] Step 7: End the test

[0130] Specifically: Step 7.1: When it is detected that the execution phase deadline condition is met or cognitive distraction behavior is detected, stop the entire detection process; Step 7.2: Take corresponding safety measures or issue a warning to the driver based on the detection results.

[0131] Among them, according to the distribution of the intention time window of the driving group in S3, the width of the initial sliding time window is determined to be in the range of [0.5, 2.5] seconds.

[0132] Because the sliding window slides on the data through a window of fixed size, the data in the window is processed each time, and the sliding window is fixed in a single stage, some key parameters obtained by the fixed width of the sliding window have similar data characteristics. For example, the number of gazes and the average gaze time in the intention stage show high correlation or similarity. The corresponding gaze frequency can be calculated from the average gaze time of the fixed time window width. The average gaze time and gaze frequency reflect the same information and status of the driver.

[0133] A similarity comparison is performed between multiple key parameters in the intention stage or the execution stage, and single or multiple key parameters that reflect the same information of the driver are divided into a single key parameter group. A key parameter is selected from each of the multiple key parameter groups as a representative key parameter as the final intention key parameter set or execution key parameter set. That is, when selecting key parameters for detecting driver cognitive distraction behavior from the characterization parameters, in order to reduce the data dimension, reduce the amount of calculation and improve the generalization ability of the model, when two or more characterization parameters with significant differences are similar, one of them is selected to represent the driver's current state as the key parameter. For example, the average gaze time and gaze frequency are both significantly different in the intention stage, but the two are similar in characterizing the driver's information. In this case, only the average gaze time is selected as the representative key parameter between the two and added to the intention key parameter set.

[0134] The selection criteria for representative key parameters involve a variety of factors, such as the depth of information acquisition, sensitivity to task relevance, degree of reflection of individual differences, and stability of data.

[0135] In summary, the set of key parameters of intention includes: pupil area change rate %, average fixation time s, average glance time ms, average glance amplitude °, mean vehicle speed km / h, standard deviation of vehicle speed km / h and lateral offset distance m;

[0136] The key parameter set for execution includes: pupil area change rate (%), average fixation time (s), average glance time (ms), average glance amplitude (°), steering wheel angle standard deviation (°), vehicle speed mean (km / h), longitudinal acceleration standard deviation (m / s) 2 , lateral acceleration standard deviation m / s 2 And the standard deviation of yaw rate rad / s.

[0137] In the specific method of cognitive distraction detection, a random forest algorithm is used to establish a cognitive distraction discrimination model; in S4, the specific steps of establishing the driver cognitive distraction discrimination model based on the random forest algorithm include:

[0138] S401. Training sample selection: The key parameter set of the driving population obtained in the simulation test is selected as the original data set, and the original data set is divided into a training set and a test set with a division ratio of 7:3. When dividing the data set, the proportion of each category in the original data set needs to be considered to ensure that the proportion of each category of samples in the divided training set and test set is consistent with the original data set. In model training, the training set is further divided into a training subset and a validation subset with a ratio of 8:2.

[0139] S402, data preprocessing, because the feature parameters are diversified data, in order to avoid the interference of model learning effect caused by training different types of data together, and to eliminate the influence of individual differences in driver's eye movement parameters on cognitive distraction recognition, this paper adopts the Z-score method to standardize the data to improve the consistency and comparability of the data.

[0140] S403, model training, training the model based on the random forest algorithm, using the training set data to fit the model, and the random forest will be trained by building multiple decision trees. During the training process, the random forest will sample the data multiple times and generate different decision trees, and each tree will randomly select and split some features.

[0141] S404, model evaluation, the evaluation indicators of the model include accuracy, precision, recall and F1 score;

[0142] S405: Real-time detection of cognitive distraction, testing samples of the test set.

[0143] Among them, the selection of sliding window width value has an important impact on distraction detection. If the sliding window width value is too small, the data in a single time window is insufficient to determine the distraction state, and the recognition accuracy is low. If the sliding window width value is too large, it may cause a delay in the response time of distraction behavior, reducing the real-time performance of detection. Therefore, choosing an appropriate sliding window width value is the key to ensuring the accuracy and real-time performance of distraction detection. The optimal sliding time window is calculated through model evaluation.

[0144] Based on the initial sliding window width value range, the detection results under different sliding window lengths in the intention stage are shown in the attached figure. Figure 4 As shown in the figure, it can be seen that there are significant differences in the accuracy of cognitive distraction detection under different sliding windows, but there is no obvious regularity. When the optimal width of the sliding window is 1.2 seconds, the model has the highest accuracy, reaching 96.6%. However, considering the accuracy and timeliness of distraction detection, the optimal sliding window width in the intention stage is determined to be 0.5 seconds. When the sliding window width is 0.5 seconds, the accuracy reaches 96.1%, indicating that the sliding window width of 0.5 seconds can effectively capture the key behavioral and physiological characteristics related to cognitive distraction.

[0145] Using the same method as the previous one, the distraction detection accuracy under different sliding window width values ​​in the execution phase was obtained, as shown in the attached figure. Figure 5 As shown in the figure, it can be seen that there are significant differences in the accuracy of cognitive distraction detection in the execution stage under different sliding windows, and there is no obvious regularity. When the optimal width of the sliding window is 1.3 seconds, the model has the highest accuracy, reaching 99.2%. Taking into account the accuracy and timeliness of distraction detection, the optimal sliding window width of the intention stage is determined to be 0.6 seconds. When the sliding window width is 0.6 seconds, the accuracy reaches 97.2%, indicating that the sliding window width of 0.6 seconds can effectively capture the key behavioral and physiological characteristics related to cognitive distraction.

[0146] The optimal sliding window width value was selected to train the model. The evaluation showed that the accuracy of the intention phase was 0.961, the precision was 0.976, the recall was 0.967, and the F1 was 0.972. All scores were greater than 0.9, indicating that the model had a good recognition effect for cognitive distraction detection. The accuracy of the execution phase was 0.972, the precision was 0.973, the recall was 0.990, and the F1 was 0.981. All scores were greater than 0.9, indicating that the model had a good recognition effect for cognitive distraction detection.

[0147] Because it is beneficial to be able to identify the driver's distracted state before he crosses the lane dividing line and enters the target lane, it can provide key warnings or interventions, thereby effectively avoiding traffic accidents and ensuring smooth and safe road traffic. In order to more significantly demonstrate the effectiveness of the real-time detection results of this embodiment, detecting the cognitive distraction behavior before the driver crosses the lane dividing line and enters the target lane is regarded as a successful detection.

[0148] The results of the cognitive distraction samples were analyzed in this way. 91.67% of the samples were successfully detected in the intention stage, 95% of the samples were successfully detected before crossing the line, 3.33% of the samples were successfully detected after crossing the line in the execution stage, and 1.67% of the samples failed to detect. Overall, the accuracy rate of detecting driver cognitive distraction before changing lanes and crossing the line reached 95%.

[0149] 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 lane-changing behavior cognitive distraction detection method based on real-time driving, characterized in that: The following steps are involved: S1. sequentially collecting lane-changing driving data in a base state and a cognitive distraction state, wherein the base state is lane-changing driving without cognitive distraction behavior, and the cognitive distraction state is lane-changing driving with cognitive distraction behavior, and the lane-changing driving data includes eye movement data and vehicle operation status data; S2. Dividing the lane-changing driving into an intention phase and an execution phase according to the node time, extracting characterization parameters for quantifying the driver's cognitive distraction behavior from the lane-changing driving data, extracting multiple characterization parameters of the intention phase as an intention characterization parameter set, and extracting multiple characterization parameters of the execution phase as an execution characterization parameter set; The node moments include the intention phase start time, the execution phase start time and the execution phase end time. The intention phase start time is the moment when the driver first pays attention to the rearview mirror or turns on the turn signal for the first time. The execution phase start time is also the intention phase end time, which is the moment when the vehicle first starts to make a large lateral displacement or the first starts to turn the steering wheel significantly. The execution phase end time is the moment when the vehicle ends the large lateral displacement or the steering wheel returns to the center position. S3. Calculate the probability values ​​of the characterization parameters of the driving group in the intention stage or the execution stage under the base state and the cognitive distraction state by using a statistical difference test method, and select the key parameters used to quantify the cognitive distraction behavior of the driver in the intention stage or the execution stage from the characterization parameters. The key parameters are the characterization parameters whose probability values ​​are less than the preset significance level values; multiple key parameters selected from the intention characterization parameter set are used as the intention key parameter set, and multiple key parameters selected from the execution characterization parameter set are used as the execution key parameter set; S4. constructing a cognitive distraction discrimination model for identifying cognitive distraction behaviors of drivers, and training the cognitive distraction discrimination model using a key parameter set; S5. Identify whether the current real-time driving data meets the node time, extract the key parameters of the current intention stage or execution stage and input them into the trained cognitive distraction discrimination model to obtain the binary decision result of the distraction state or the base state.

2. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 1, characterized in that: The S5 cognitive distraction detection process includes: Step 1: Trigger the intention phase start time detection, detect whether the real-time driving triggers the intention phase start time, if triggered, determine that the current driving enters the intention phase, and enter the intention phase detection process; Step 2: Detection process of the intention stage: continuously obtain lane change driving data of the intention stage to form a time series until the width of the time series reaches the preset sliding window width, extract the key parameters of the intention stage from the sliding window, input the cognitive distraction discrimination model, and judge whether there is cognitive distraction behavior in the current driving. If there is cognitive distraction behavior, exit the cognitive distraction detection process; if there is no cognitive distraction behavior, move the sliding window to the next time window, and re-extract the key parameters for detection by combining the current sliding time window and the historical sliding time window; Step 3: Continuous monitoring during the intention phase, and continuous looping of step 2 during the intention phase until cognitive distraction behavior or lane change driving data is identified to trigger the start of the execution phase; Step 4: Trigger the execution phase start time detection, detect whether the real-time driving triggers the execution phase start time, if triggered, determine that the current driving enters the execution phase, and enter the execution phase detection process; Step 5: Detection process of the execution phase: continuously obtain the lane change driving data of the execution phase to form a time series until the width of the time series reaches the width of the sliding window, extract the key parameters of the execution phase from the sliding window, input the cognitive distraction discrimination model, and determine whether there is cognitive distraction behavior in the current driving. If there is cognitive distraction behavior, exit the cognitive distraction detection process. If there is no cognitive distraction behavior, move the sliding window to the next time window, and re-extract the key parameters for detection by combining the current sliding time window and the historical sliding time window; Step 6: Continuous monitoring during the execution phase. Step 5 is continuously repeated during the execution phase until cognitive distraction behavior or lane change driving data is identified, triggering the execution phase deadline.

3. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 2, characterized in that: The width of the sliding window is selected in the range of [0.5, 2.5] seconds.

4. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 3, characterized in that: From the sliding window width selection range, the optimal sliding window width for the intention stage is selected as 0.5 seconds, and the optimal sliding window width for the execution stage is selected as 0.6 seconds.

5. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 1, characterized in that: The characterization parameters include eye movement characteristic parameters and vehicle characteristic parameters; Eye movement characteristic parameters include: pupil area and pupil area change rate used to characterize the driver's pupil changes, gaze frequency and average gaze time used to characterize the driver's gaze, scanning frequency, average scanning time and average scanning amplitude used to characterize the driver's scanning, and rearview mirror gaze times and average single rearview mirror gaze time used to characterize the driver's rearview mirror inspection behavior; The vehicle characteristic parameters include: the turn signal opening rate and the relative opening time of the turn signal used to characterize the usage of the turn signal, the steering wheel angle and steering wheel angle entropy used to characterize the steering wheel control situation, the vehicle speed, the vehicle speed standard deviation and the longitudinal angular velocity used to characterize the longitudinal running state of the vehicle, and the lateral acceleration, yaw angular velocity, lateral offset distance and maximum lateral acceleration used to characterize the lateral running state of the vehicle.

6. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 1, characterized in that: S3 also includes performing a similarity comparison between multiple key parameters in the intention stage or the execution stage, dividing the single or multiple key parameters that reflect the same information of the driver into a single key parameter group, and selecting a key parameter from each of the multiple key parameter groups as a representative key parameter as the final intention key parameter set or execution key parameter set.

7. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 6, characterized in that: The key parameter set of intention includes: pupil area change rate, average fixation time, average glance time, average glance amplitude, vehicle speed mean, vehicle speed standard deviation and lateral offset distance; The key execution parameter set includes: pupil area change rate, average fixation time, average glance time, average glance amplitude, steering wheel angle standard deviation, vehicle speed mean, longitudinal acceleration standard deviation, lateral acceleration standard deviation and yaw rate standard deviation.

8. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 1, characterized in that: In S1, a simulated driving test is used to obtain lane-changing driving data of drivers in a base state and a cognitive distraction state, and the driver is used to perform secondary tasks during autonomous lane-changing driving to simulate cognitive distraction behavior; the secondary tasks include a conversation for simulating distraction caused by interactive question-and-answer between the driver and others and a continuous subtraction operation for simulating distraction caused by spontaneous thinking of the driver.

9. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 8, characterized in that: The S1 also includes setting up environmental vehicles on the simulated driving road to induce the driver's autonomous lane change behavior, and the environmental vehicles include a low-speed vehicle located in the current lane of the driving vehicle and a high-speed vehicle located in the target lane for the driving vehicle to change lanes.

10. The method for detecting lane-changing behavior cognitive distraction based on real-time driving according to claim 1, characterized in that: The S4 includes: S401, training sample selection, the intention key parameter set and the execution key parameter set are used as the original data set, and the original data set is divided into a training set and a test set; S402, data preprocessing, using the Z-score method adopted in this paper to standardize the data; S403, model training, establishing a cognitive distraction discrimination model based on the random forest algorithm, using the training set data to fit the cognitive distraction discrimination model, and the random forest will be trained by constructing multiple decision trees; during the training process, the random forest will sample the data multiple times and generate different decision trees, and each tree will randomly select and split some features; S404, model evaluation, the evaluation indicators of the cognitive distraction discrimination model include accuracy, precision, recall and F1 score; S405: Real-time detection of cognitive distraction, testing samples of the test set.

Images