An automatic driving control right transfer method considering driver manipulation characteristics

Abstract

The application is suitable for the field of automatic driving technology, and provides an automatic driving control right transfer method considering driver operation characteristics, including the following steps: data collection and preprocessing; calculating external driving risk score and driver state evaluation score respectively; calculating final risk value based on the external driving risk score and the driver state evaluation score; and designing and realizing automatic driving "man-machine" control right switching active warning based on the final risk value and considering threshold hysteresis characteristics. The method can realize individualized adjustment of takeover time and mode, and effectively solves the control right switching problem when the automatic driving system actively leaves the auxiliary driving state and initiates manual takeover in the man-machine co-driving mode.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automatic driving, and particularly relates to an automatic driving control right transfer method considering driver operation characteristics. BACKGROUND

[0002] In recent years, automatic driving technology has made breakthrough progress, and automatic driving systems have shown significant advantages in improving driving efficiency and reducing human load. However, due to factors such as perception accuracy, decision algorithm and environmental complexity, current automatic driving systems cannot completely replace human drivers in all scenarios. To achieve a smooth transition from traditional human-controlled driving to fully automatic driving, human-machine co-driving systems are widely used as a transitional form.

[0003] The human-machine co-driving system refers to a driving mode in which the automatic driving system and the human driver cooperate to complete the driving task. Under the current technical level, when the system faces complex traffic situations that are difficult to identify or cannot be handled, the driver needs to take over the control right of the vehicle in time. Under this background, how to ensure the seamless switching and smooth transition of the driving control right between the system and the driver has become a key problem in the design of the human-machine co-driving system.

[0004] Currently, the risk assessment research of automatic driving control right switching still highly depends on traditional parameters such as vehicle state and road environment, such as vehicle speed, vehicle distance and road curvature. Although this method can reflect part of the objective risk, it has obvious limitations: on the one hand, most researches focus on structured indicators, and lack effective modeling of unstructured and nonlinear risk signals, such as the implicit intention conveyed by other vehicles passing through lights and sounding horns, resulting in insufficient description of complex traffic situations; on the other hand, although artificial intelligence algorithms such as deep learning have made progress in environmental perception, they have not yet systematically integrated driver personalized characteristics into the risk assessment system, nor have they constructed a switching mechanism with redundancy and hierarchical prompts. The existing methods mostly use a single threshold to trigger the control right switching, which has low flexibility and poor fault tolerance, and it is difficult to ensure the smoothness and safety of the driving right switching. SUMMARY

[0005] The purpose of the embodiment of the application is to provide an automatic driving control right transfer method considering driver operation characteristics, which aims to solve the problems raised in the background.

[0006] The embodiment of the application is implemented as follows: an automatic driving control right transfer method considering driver operation characteristics, comprising the following steps:

[0007] Step 1: data collection and preprocessing;

[0008] Step 2: driving risk quantification;

[0009] The process involves calculating two separate components: external driving risk score and driver state assessment score. The external driving risk score is obtained using the analytic hierarchy process (AHP) and entropy weight method. The driver state score is assessed based on driving habits, blinking frequency, and yawning frequency. A comprehensive driving hazard coefficient score is calculated based on both the external driving risk score and the driver state assessment score to obtain the final risk value. ;

[0010] Step 3: Early warning scheme selection;

[0011] Based on the final risk value Considering the threshold hysteresis characteristics, we designed and implemented an active early warning system for the switching of human-machine control in autonomous driving based on big data feature analysis of driving behavior.

[0012] A further technical solution involves using the environmental perception system of the intelligent connected vehicle in step 1 to acquire driver control behavior data, road traffic scene data, and vehicle driving status data, and storing and preprocessing the above data to output feature data for driving risk quantification.

[0013] In a further technical solution, the environmental perception system includes radar, cameras, in-vehicle CAN bus, and V2X devices.

[0014] A further technical solution, the specific calculation method for step 2 is as follows:

[0015] In the Analytic Hierarchy Process (AHP), the target layer is defined as external driving risk assessment, and the criterion layer is defined as... It is divided into two layers, the first layer is the vehicle status. and external environment ,Right now The second layer contains seven indicators, including vehicle speed. Vehicle sideslip angle Lane deviance TTC collision time Frequency of hazard warning lights of surrounding vehicles Distance from road construction signs and the area enclosed by traffic cones ,Right now , ;

[0016] against A number of drivers conducted a continuous 1-hour driving experiment in a fixed urban road scenario, collecting the above-mentioned indicator information at a sampling frequency of 5Hz. The obtained data underwent Kalman filtering to achieve noise suppression and data smoothing. The original data matrix for a single sampling period of the seven evaluation indicators is as follows: The result after aligning, normalizing, and normalizing the original data is: At the same time, collect this Based on the risk preference of drivers for the aforementioned indicators, a judgment matrix was constructed. After passing a consistency test, the weights of the first and second layers of indicators in the criterion layer were obtained using the eigenvalue method. and ;

[0017] The entropy values ​​of each evaluation index are obtained using the entropy weight method, as shown in equation (1).

[0018] (1);

[0019] In the formula , This indicates the relative contribution of the sample to this indicator. For the first The information entropy of an indicator is used to measure the uncertainty of the information contained in that indicator. For the first The coefficient of difference of each evaluation indicator The index weights are derived from the entropy weight method;

[0020] The final weights of the sub-indicators in the analytic hierarchy process are: The index weights obtained by the entropy weight method are: Assuming the overall weight of each driving evaluation indicator is: Then its external driving risk score for:

[0021] (2);

[0022] Combining the results of the objective entropy weighting method and the analytic hierarchy process, the constraint equation is shown in equation (3). Solving this equation using the Lagrange multiplication method yields the final weight values. :

[0023] (3);

[0024] Secondly, the driver's driving state is taken into account as an influencing factor, and the model is constructed from two dimensions: long-term driving ability and short-term reaction level.

[0025] Long-term driving ability is quantified by the driver's driving experience; short-term reaction level is measured by the driver's drowsiness level, calculated using a combination of blinking frequency and yawning frequency to determine the driver's drowsiness level. The calculation formula is shown in equation (4):

[0026] (4);

[0027] in, blink frequency of each driver when he is sober and has normal driving ability, real-time blink frequency of each driver during the test, allocation weight coefficient of blink frequency and yawning frequency, nonlinear adjustment parameter, and peak value and standard deviation of yawning frequency when tired, blink frequency characteristic function, yawning frequency characteristic function, actual yawning frequency of each driver during the test;

[0028] The result of the driving risk final score which integrates vehicle status and external environment and considers driver status is shown in equation (5):

[0029] (5);

[0030] wherein, final risk value; Hanning window function, length of Hanning window set according to the average reaction time of the driver during the experiment; sampling point; The higher the score, the greater the degree of driving risk considered by the driver.

[0031] Further technical solutions, in the step 2, the frequency of surrounding vehicle danger warning light defined as the frequency of other vehicles turning on the danger warning flash light within the sampling period , i.e. ; the area surrounded by traffic cones defined as the effective area of the polygonal region formed by the traffic cones, with the geometric center of the region as the center and expanded by 0.5 m radius for quantification, if only one traffic cone is identified, the area is the area of the circle with the traffic cone as the center and 1 m as the radius.

[0032] Further technical solutions, in the step 3, the danger type is divided into potential risk and high-risk real-time risk; when , it is high-risk real-time risk, and when For potential risks, the system guides the driver's attention back to the driving task from non-driving related tasks through a reminder; for high-risk real-time risks, the driver must immediately take over vehicle control; for potential risks, when the driver actively steps on the accelerator pedal and the opening exceeds 5%, or actively rotates the steering wheel and the turning angle exceeds 5°, indicating that the driver does not respond to the reminder signal, the system continues to maintain the automatic driving state; for high-risk real-time risks, the system needs to remind the driver while taking defensive driving actions to give the driver time to take over the vehicle;

[0033] If , the steering wheel vibrates, the interior red ambient light flashes, and a voice warning is given, and automatic emergency braking intervention is taken; if , the steering wheel vibrates, and the interior red ambient light flashes; if , the steering wheel vibrates and the seat vibrates; to avoid frequent switching of the early warning scheme, the threshold is set to have a hysteresis interval, i.e. when is near the boundary of the interval, the system will maintain the early warning state of the previous level until the score continuously stays in the new interval for more than 3 seconds, triggering the early warning scheme switching, effectively suppressing false triggering caused by signal fluctuations, and improving the stability of the early warning output.

[0034] The automatic driving control transfer method considering the driving characteristics of the driver provided by the embodiment of the application adjusts the takeover time and mode individually by accurately depicting the driving style and behavior habits of the driver, effectively solving the control switching problem when the automatic driving system actively leaves the assisted driving state and initiates manual takeover in the man-machine co-driving mode. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The flowchart of the automatic driving control transfer method considering the driving characteristics of the driver provided by the embodiment of the application;

[0036] Figure 2 The external driving risk indicator diagram;

[0037] Figure 3 The traffic cone area diagram. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages of the application clearer, the application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application.

[0039] The specific implementation of the application is described in detail below in combination with specific embodiments.

[0040] like Figure 1 As shown, an embodiment of the present invention provides a method for transferring autonomous driving control that takes into account the driver's handling characteristics, comprising the following steps:

[0041] Step 1: Data collection and preprocessing;

[0042] By utilizing the environmental perception system (such as radar, cameras, in-vehicle CAN bus and V2X equipment) on intelligent connected vehicles, driver control behavior data, road traffic scene data and vehicle driving status data are obtained, and the above data are stored and preprocessed to output feature data for driving risk quantification.

[0043] Step 2: Quantify driving risks;

[0044] To calculate the overall driving hazard coefficient score, this step requires calculating both the external driving risk score and the driver state assessment score. The external driving risk score can be obtained using the analytic hierarchy process (AHP) and the entropy weight method; the driver state score can be assessed based on driving habits, blinking frequency, and yawning frequency. The specific calculation methods are as follows:

[0045] In the Analytic Hierarchy Process (AHP), the target layer is defined as external driving risk assessment, and the criterion layer is defined as... It is divided into two layers, the first layer being the vehicle status. and external environment ,Right now The second layer contains seven indicators, including vehicle speed. Vehicle sideslip angle Lane deviance TTC collision time Frequency of hazard warning lights of surrounding vehicles Distance from road construction signs Area enclosed by traffic cones ,Right now , Specific indicators are as follows: Figure 2 As shown.

[0046] Among them, the frequency of hazard warning lights of surrounding vehicles Defined as identifying other vehicles within a sampling period, with the vehicle's center of mass as the center and a radius of 5m. Frequency of activating hazard warning lights ,Right now Area enclosed by traffic cones Defined as the effective area of ​​the polygonal region formed by traffic cones, quantified by taking the geometric center of this region as the center and expanding outwards by a radius of 0.5 m, such as... Figure 3The distance from the geometric center of the cones to the farthest cone (L in the figure). If the traffic cone only identifies one, its area is the area of a circle with the traffic cone as the center and 1 m as the radius.

[0047] For professional drivers, a 1-hour driving experiment was conducted in a fixed urban road scene. The above index information was collected by V2X equipment, CAN bus signals, radar, and internal and external cameras, with a sampling frequency of 5 Hz. The resulting data was processed by Kalman filtering to achieve noise suppression and data smoothing. The original data matrix of a single sampling period of the seven evaluation indexes is . The results of the alignment, normalization and normalization of the original data are . At the same time, the risk preference degree of the drivers for the above indexes was collected to construct a judgment matrix, and the weights of the first and second layer indexes of the criterion layer were obtained by eigenvalue method after consistency test and .

[0048] The entropy value of each evaluation index is obtained by entropy weight method as shown in formula (1).

[0049] (1);

[0050] In the formula, , , represents the relative contribution of the sample under the index, is the information entropy of the first index, which measures the information uncertainty contained in the index, is the difference coefficient of the first evaluation index, is the index weight obtained by entropy weight method.

[0051] The sub-index weight of the analytic hierarchy process is finally , the index weight obtained by entropy weight method is , and assuming that the comprehensive weight of each evaluation index of driving is , the external driving risk score is:

[0052] (2);

[0053] The deviation between subjective and objective should be closer to be more reasonable. The results of objective entropy weight method and analytic hierarchy process are combined to obtain the constraint equation as shown in formula (3), and the Lagrange is used to solve the equation to obtain the optimal weight value .

[0054] (3);

[0055] Secondly, the driving state of the driver is considered as an influencing factor, which is modeled from two dimensions of long-term driving ability and short-term reaction level.

[0056] Among them, the long-term driving ability is quantified by the driver's driving age, which is generally positively correlated with the driving risk response ability. The short-term reaction level is measured by the driver's drowsiness level. In order to improve the accuracy of drowsiness level discrimination, the joint features of blink frequency and yawning frequency are used to calculate the driver's drowsiness level , and the calculation formula is shown in equation (4).

[0057] (4);

[0058] Among them, is the blink frequency of each driver when he is awake and has normal driving ability, is the real-time blink frequency of each driver at the test, is the deployment weight coefficient of blink frequency and yawning frequency, is a non-linear adjustment parameter, and are the peak value and standard deviation of yawning frequency when tired, is the blink frequency feature function, is the yawning frequency feature function, is the actual yawning frequency of each driver at the test.

[0059] The final score of the driving risk considering the vehicle state and the external environment and the driver's state is shown in equation (5).

[0060] (5);

[0061] Among them, is the final risk value; is the Hanning window function, the length of the Hanning window is set according to the average reaction time of the driver during the experiment; is the sampling point. The higher the score, the greater the degree of driving risk that the driver considers.

[0062] Step 3: Early warning scheme selection.

[0063] Based on the comprehensive score of the driving risk coefficient calculated in step 2, i.e. the final risk value , and considering the threshold hysteresis characteristics, an automatic driving "man-machine" control right switching active early warning method based on driving behavior big data feature analysis is designed and implemented.

[0064] The danger type is divided into potential risk and high real-time risk. When , it is high real-time risk, and when , it is potential risk. The potential risk means that the danger is likely to occur or the autonomous driving system cannot accurately identify the current danger situation, at which time the system guides the driver's attention back to the driving task from the non-driving related task through the reminding mode. The high real-time risk means that a highly urgent and immediate threat driving situation has been identified, and the driver must immediately take over the vehicle control to deal with the possible danger. For the potential risk, when the driver actively steps on the accelerator pedal and the opening exceeds 5% or actively rotates the steering wheel and the turning angle exceeds 5°, it indicates that the driver does not respond to the reminder signal, and the system continues to maintain the autonomous driving state. For the high real-time risk, the system needs to remind the driver while taking defensive driving actions to reserve time for the driver to take over the vehicle.

[0065] If , the steering wheel vibrates, the red ambient light in the vehicle flashes, and the voice warning is given, and the automatic emergency braking intervention is taken; if , the steering wheel vibrates, and the red ambient light in the vehicle flashes; if , the steering wheel vibrates and the seat vibrates. In order to avoid frequent switching of the early warning scheme, the threshold is set as an interval with hysteresis characteristics. That is, when is near the boundary of the interval, the system will maintain the early warning state of the previous level until the score continuously stays in the new interval for more than 3 seconds, and the early warning scheme switching can be triggered, so as to effectively suppress the false triggering caused by signal fluctuation and improve the stability of the early warning output.

[0066] The above only describes the preferred embodiments of the present application and is not used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for transferring control of automated driving that considers driver handling characteristics, characterized in that, Includes the following steps: Step 1: Data collection and preprocessing; Step 2: Quantify driving risks; The process involves calculating two separate components: external driving risk score and driver state assessment score. The external driving risk score is obtained using the analytic hierarchy process (AHP) and entropy weight method. The driver state score is assessed based on driving habits, blinking frequency, and yawning frequency. A comprehensive driving hazard coefficient score is calculated based on both the external driving risk score and the driver state assessment score to obtain the final risk value. ; Step 3: Early warning scheme selection; Based on the final risk value Meanwhile, considering the threshold hysteresis characteristics, we designed and implemented an active early warning system for the switching of human-machine control in autonomous driving based on the big data feature analysis of driving behavior. The specific calculation method for step 2 is as follows: In the Analytic Hierarchy Process (AHP), the target layer is defined as external driving risk assessment, and the criterion layer is defined as... It is divided into two layers, the first layer being the vehicle status. and external environment ,Right now The second layer contains seven indicators, including vehicle speed. Vehicle sideslip angle Lane deviance TTC collision time Frequency of hazard warning lights of surrounding vehicles Distance from road construction signs and the area enclosed by traffic cones ,Right now , ; against A number of drivers conducted a 1-hour driving experiment in a fixed urban road scenario, collecting the above-mentioned indicator information at a sampling frequency of 5Hz. The obtained data underwent Kalman filtering to achieve noise suppression and data smoothing. The original data matrix for a single sampling period of the seven evaluation indicators is as follows: The result after aligning, normalizing, and normalizing the original data is: At the same time, collect this Based on the risk preference of drivers for the aforementioned indicators, a judgment matrix was constructed. After passing a consistency test, the weights of the first and second layers of indicators in the criterion layer were obtained using the eigenvalue method. and ; The entropy values ​​of each evaluation index are obtained using the entropy weight method, as shown in equation (1): （1）； In the formula, , This indicates the relative contribution of the sample to this indicator. For the first The information entropy of an indicator is used to measure the uncertainty of the information contained in that indicator. For the first The coefficient of difference of each evaluation indicator The index weights are derived from the entropy weight method. The final weights of the sub-indicators in the analytic hierarchy process are: The index weights obtained by the entropy weight method are: Assuming the overall weight of each driving evaluation indicator is: Then its external driving risk score for: （2）； Combining the results of the objective entropy weighting method and the analytic hierarchy process, the constraint equation is shown in equation (3). Solving this equation using the Lagrange multiplication method yields the final weight values. : （3）； Secondly, the driver's driving state is taken into account as an influencing factor, and the model is constructed from two dimensions: long-term driving ability and short-term reaction level. Long-term driving ability is quantified by the driver's driving experience; short-term reaction level is measured by the driver's drowsiness level, calculated using a combination of blinking frequency and yawning frequency to determine the driver's drowsiness level. The calculation formula is shown in equation (4): （4）； in, The blinking frequency of each driver when they are awake and have normal driving ability. This refers to the real-time blinking frequency of each driver during the test. The weighting coefficients for adjusting blinking frequency and yawning frequency. It is a non-linear adjustment parameter. and These represent the peak and standard deviation of yawning frequency when fatigued. The blink frequency characteristic function, The characteristic function of yawning frequency. The actual yawning frequency of each driver during the test; The final score for driving risk, taking into account both vehicle status and external environment, as well as driver status, is shown in Equation (5): （5）； in, This is the final risk value; Let Hanning window be the function, and the length of the Hanning window be... The settings were based on the average reaction time reported by the drivers during the experiment; For sampling points; The higher the score, the greater the driver's perception of the driving hazard.

2. The autonomous driving control transfer method considering driver operation characteristics according to claim 1, characterized in that, In step 1, the environmental perception system of the intelligent connected vehicle is used to acquire driver control behavior data, road traffic scene data, and vehicle driving status data. The above data is stored and preprocessed, and feature data is output for driving risk quantification.

3. The autonomous driving control transfer method considering driver handling characteristics according to claim 2, characterized in that, The environmental perception system includes radar, cameras, in-vehicle CAN bus, and V2X devices.

4. The autonomous driving control transfer method considering driver operation characteristics according to claim 1, characterized in that, In step 2, the frequency of hazard warning lights of surrounding vehicles Defined as identifying other vehicles within a sampling period, with the vehicle's center of mass as the center and a radius of 5m. Frequency of activating hazard warning lights ,Right now Area enclosed by traffic cones Defined as the effective area of ​​the polygonal region formed by the traffic cone, it is quantified with the geometric center of the region as the center and an outer radius of 0.5m. If only one traffic cone is identified, its area is the area of ​​a circle with the traffic cone as the center and a radius of 1m.

5. The autonomous driving control transfer method considering driver operation characteristics according to claim 1, characterized in that, In step 3, the hazard types are divided into two categories: potential risks and high-risk real-time risks; when At that time, it is considered a high-risk real-time risk. At any given time, this is considered a potential risk. For potential risks, the system uses reminders to redirect the driver's attention from non-driving tasks back to driving tasks. For high-risk real-time risks, the driver is required to immediately take control of the vehicle. For potential risks, if the system detects that the driver has actively pressed the accelerator pedal with an opening exceeding 5%, or actively rotated the steering wheel with an angle exceeding 5°, it indicates that the driver has not responded to the reminder signal, and the system continues to maintain automatic driving mode. For high-risk real-time risks, the system needs to remind the driver while taking defensive driving actions, allowing time for the driver to take over the vehicle. like The steering wheel vibrates, the red ambient lighting inside the car flashes, and a voice warning is issued, and automatic emergency braking is initiated; if The steering wheel vibrates, and the red ambient lighting inside the car flashes; if Steering wheel vibration and seat vibration; to avoid frequent switching of warning schemes, the threshold is set to an interval with hysteresis characteristics, that is, when When the score is near the boundary of the interval, the system will maintain the previous level of warning status until the score remains in the new interval for more than 3 seconds before the warning scheme can be switched.

Images