An automatic driving right smooth handover method based on driver fatigue state perception

Abstract

The application is suitable for the field of automatic driving technology, and provides an automatic driving right smooth handover method based on driver fatigue state perception, which comprises the following steps: collecting vehicle driving data and lane line images, and performing pretreatment and parameter definition; determining the driver fatigue state in a sliding window based on lane position standard deviation, number of lane deviation per unit time, vehicle speed standard deviation and vehicle speed mutation number; dynamically setting the steering wheel input torque threshold and takeover time condition according to the fatigue state; detecting the input torque applied to the steering wheel by the driver in real time, combining the fatigue state and the road condition, selecting three smooth handover strategies of fast, ordinary or slow, and realizing the smooth transition of the steering torque through a sigmoid gradual function. The method solves the problems of difficult takeover under the fatigue state, false triggering under the sober state and abruptness in the handover process, and improves the safety, smoothness and environmental adaptability of the driving right handover.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, and in particular relates to a method for smooth handover of autonomous driving rights based on driver fatigue perception. Background Technology

[0002] With the evolution of autonomous driving technology, vehicle systems are now capable of autonomous navigation and control in limited scenarios, thus partially reducing the driver's workload. In this process, the handover mechanism of driving control becomes a critical factor affecting system safety and reliability. Especially when switching from autonomous driving to manual driving, achieving a safe, smooth, and immediate transfer of control is a core issue for improving driving safety and the driving experience.

[0003] Currently, most common driving control transfer schemes rely on preset fixed thresholds to determine the driver's intention to take over. A typical approach is to detect the torque or steering angle signal applied by the driver to the steering wheel; once the measured value exceeds a set threshold, it is determined that the driver intends to take over and a transfer of control is triggered. While this method achieves a transfer of control to some extent, its judgment criteria are relatively simplistic and fail to fully incorporate dynamic factors related to people, vehicles, and the environment.

[0004] However, such handover methods based on fixed thresholds still have significant limitations: First, they do not fully consider individual differences and real-time changes in driver status, especially since fatigue can lead to decreased reaction speed and operational precision, and fixed thresholds can easily cause takeover delays or false triggers, increasing driving risks; second, existing methods lack a smooth transition mechanism for the handover process, and the switch of control is often abrupt, easily causing fluctuations in steering wheel torque and vehicle trajectory oscillations, affecting ride comfort; third, most current strategies fail to dynamically adjust to real-time road conditions, making it difficult to adapt to handover needs of varying urgency in complex traffic environments, thus limiting the system's environmental adaptability. Therefore, there is an urgent need for an intelligent driving control handover method that can comprehensively consider driver status, handover smoothness, and environmental scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a method for smooth handover of autonomous driving rights based on driver fatigue perception, aiming to solve the problems mentioned in the background art.

[0006] The present invention is implemented as follows: a method for smooth handover of autonomous driving control based on driver fatigue perception includes the following steps:

[0007] Step 1: Data Acquisition and Preprocessing;

[0008] Collect vehicle driving data and lane line images, perform filtering preprocessing on the collected data, and define parameters;

[0009] Step 2: Calculation of detection indicators;

[0010] Based on the preprocessed data, detection indicators used to characterize the driver's control ability are calculated within the set detection time window.

[0011] Step 3: Determine driver fatigue status;

[0012] Set thresholds for each detection indicator, compare the magnitude of each detection indicator with the corresponding threshold, count the number of abnormal indicators, and determine the driver's fatigue state based on the number of abnormal indicators.

[0013] Step 4: Threshold setting;

[0014] Based on the determined driver fatigue level, set the corresponding steering wheel input torque threshold.

[0015] Step 5: Driver handover identification;

[0016] The system collects the input torque applied by the driver to the steering wheel in real time. When the input torque exceeds the steering wheel input torque threshold and the duration reaches the time threshold, it is determined that the driver has manually taken over the vehicle.

[0017] Step 6: Smooth transfer of driving rights;

[0018] After determining that the driver has manually taken over the vehicle, a time-based sigmoid transition strategy is adopted based on the current road conditions and the driver's fatigue state to smoothly transition the steering wheel torque from the system torque to the driver's input torque, thereby achieving a smooth transfer of driving control.

[0019] In a further technical solution, step 1 includes the following specific steps:

[0020] Data Acquisition: Real-time vehicle speed is acquired via CAN bus. and acceleration Lane line image data is collected through the vehicle-mounted forward-facing camera. The camera acquisition frequency is set to 20 Hz, and the CAN bus data acquisition frequency is synchronized at 20 Hz.

[0021] Preprocessing: A fixed detection time window of 60 seconds is set, and index calculations and fatigue level determination are performed in units of this window, with a window sliding step of 30 seconds; Kalman filtering is applied to the collected data; the Kalman filtering method for the collected vehicle speed and acceleration is specifically divided into the following steps:

[0022] First, define the state variables. :

[0023]

[0024] in, for The actual vehicle speed at any given time, in km / h; for The actual acceleration at time t, in m / s² 2 ; For discrete time steps, corresponding to the sensor sampling times, when the sampling frequency is 20 Hz, the discrete time step is... sampling interval It is 0.05 s;

[0025] At the same time, construct the state equations:

[0026]

[0027] Wherein, the state transition matrix Process noise vector It follows a pattern with a mean of 0 and a covariance matrix of . Gaussian white noise represents non-uniform disturbances during vehicle movement. Vehicle speed process noise variance Take 6.5 (km / h) 2 Acceleration process noise variance Take 0.05 (m / s) 2 ) 2 ;

[0028] Furthermore, the observation equation is constructed as follows:

[0029]

[0030] Among them, the observation vector , for Vehicle speed observation value at any time for Acceleration observations at different times; observation matrix ; Observation noise vector It follows a pattern with a mean of 0 and a covariance matrix of . Gaussian white noise, Vehicle speed sensor noise variance Take 1.0 (m / s) 2 Accelerometer noise variance Take 0.15 (m / s) 2 ) 2 ;

[0031] Secondly, filter initialization is performed; for the initial state vector , and These are the first vehicle speed and acceleration observations collected by the sensors, respectively. When the vehicle starts from a standstill, both are zero; an initial error covariance matrix is ​​set. ;

[0032] Next, state prediction, i.e., prior estimation, is performed; the formulas for state prediction and error covariance prediction are as follows:

[0033]

[0034]

[0035] in, for Prior state estimation at time step; for Posterior time estimation; for Time-prior error covariance; for Posterior error covariance at time step;

[0036] Subsequently, the Kalman gain was calculated. :

[0037]

[0038] Next, a state update, i.e., posterior estimation, is performed; the formulas for state update and error covariance update are as follows:

[0039]

[0040]

[0041] Finally, output the results. , and These are the filtered vehicle speed and acceleration values, respectively. The filtered state variables, for Posterior error covariance at time step It is the identity matrix;

[0042] The parameters are defined as follows:

[0043] Lane centerline: The left and right lane lines are extracted through image processing, and the line connecting the midpoints of the two lane lines is calculated as the lane centerline; Lane position: The horizontal offset distance of the vehicle's center of gravity from the lane centerline. The unit is cm. A leftward offset is recorded as a positive value, and a rightward offset as a negative value. Sudden change in vehicle speed: defined as an absolute acceleration value greater than 1.5 m / s². 2 The situation was a sudden change in vehicle speed.

[0044] A further technical solution involves extracting the left and right lane lines of a road through image processing in step 1. The specific steps are as follows:

[0045] First, the acquired RGB image is converted to grayscale, and Gaussian blur is used to remove noise;

[0046] Secondly, the Roberts operator edge detection algorithm is used to extract edge features from the image and generate a binary edge image;

[0047] Next, define the region of interest (ROI) in the lower half of the image, use a mask to filter out irrelevant regions, and output an edge image that only contains potential lane line regions.

[0048] Subsequently, the Hough transform is applied to detect line segments in the edge map and the coordinates of the line segment endpoints are obtained;

[0049] Next, the slope of each line segment is calculated, and the lines are classified as left or right lane lines based on the sign of the slope, and then averaged and fitted.

[0050] Finally, output the coordinates of the left and right lane lines.

[0051] In a further technical solution, step 2 includes the following specific steps:

[0052] Based on the data and parameter definitions collected in step 1, four detection indicators are calculated within the set 60-second detection time window. The specific calculation method is as follows:

[0053] Standard deviation of lane position Extract the lane position data collected within the detection window and denote it as the sample set. ; Calculate the average of the sample set ,in The number of times data is collected within the detection window, i.e. Calculate the standard deviation of lane position. ;

[0054] Number of times deviating from the lane per unit time When the absolute value of the offset distance between the vehicle's center of gravity and the lane centerline is ≥30cm, it is determined as a lane departure; the number of lane departures occurring within the monitoring window is counted. The unit is times per minute;

[0055] Standard deviation of vehicle speed Extract the vehicle speed data collected within the detection window and denote it as the sample set. Calculate the average of the sample set. Standard deviation of vehicle speed ;

[0056] Number of sudden speed changes Statistical detection window acceleration >1.5 m / s 2 The number of times is The unit is times per minute.

[0057] In a further technical solution, step 3 includes the following specific steps:

[0058] Lane position standard deviation threshold Set to 8 cm, that is when When the distance is ≤8 cm, the lane position is relatively stable. When the distance is greater than 8 cm, the lane position is abnormal; the threshold for the number of lane departures per unit time. Set to 2 times / minute, that is, when Lane departure frequency is normal when it is ≤2 times / minute. Lane departure frequency is abnormal when it exceeds 2 times / minute; vehicle speed standard deviation threshold. Set to 4km / h, that is when At speeds ≤4 km / h, vehicle speed control is stable. Vehicle speed control malfunctions at speeds >4 km / h; threshold for sudden speed changes. Set to 3 times / minute, that is when When the speed is ≤3 times / minute, the vehicle speed adjustment is relatively smooth. The vehicle speed adjustment is abnormal when it is >3 times / minute;

[0059] Compare the relationship between the four detection indicators and their corresponding thresholds, and count the number of abnormal indicators. ,according to Values ​​used to determine driver fatigue status: when At that time, it was determined that the driver was conscious and had normal control over the vehicle; when At that time, it was determined that the driver was in a state of mild fatigue, and his ability to control the vehicle was reduced; when When the driver is determined to be in a state of severe fatigue, their ability to control the vehicle is significantly reduced; when or At this point, maintain the current fatigue state and proceed to the next sliding window detection.

[0060] A further technical solution involves setting corresponding steering wheel input torque thresholds in step 4 based on different driver fatigue states. The corresponding relationships are as follows: the steering wheel input torque threshold for a conscious state is set to 1.5 N·m; the steering wheel input torque threshold for a mildly fatigued state is set to 1.0 N·m; and the steering wheel input torque threshold for a severely fatigued state is set to 0.6 N·m.

[0061] In a further technical solution, step 5 includes the following steps:

[0062] First, the torque sensor built into the vehicle's steering system collects the input torque applied by the driver to the steering wheel in real time. The data is collected at a frequency of 20 Hz and transmitted to the vehicle controller via the CAN bus. Next, the collected data is filtered. Then, driver transfer recognition is performed: when… And duration Reaching the time threshold When the automatic driving system is disengaged, driving control is transferred to the driver; the specific conditions for this determination are as follows:

[0063] Awake state: =1.5 N·m, =0.3 s, that is ≥1.5 N·m and If the time is ≥0.3 s, it is determined that the driver has manually taken over the vehicle; mild fatigue state: =1.0 N·m, =0.5 s, that is ≥1.0 N·m and If the time is ≥0.5 s, it is determined that the driver has manually taken over the vehicle; severe fatigue state: =0.6 N·m, =0.8 s, that is ≥0.6 N·m and If the time is ≥0.8 s, it is determined that the driver has manually taken over the vehicle.

[0064] In a further technical solution, step 6 includes the following specific steps:

[0065] First, the vehicle's onboard sensors and forward-facing camera assess the current road conditions to determine the urgency of handing over driving control. Then, based on the urgency of the handover and the driver's fatigue level, a corresponding transition strategy is selected to achieve a smooth change in steering wheel torque, allowing the vehicle to smoothly switch from autonomous driving mode to manual driving mode. The specific method is as follows:

[0066] Driving urgency assessment: The system uses a forward-facing camera and lidar to capture road information ahead, including distance to obstacles, vehicle density, and visibility. A safe distance threshold of 50 m, a vehicle density threshold of 5 vehicles / 100 m, and a visibility threshold of 100 m are set. If at least one of the following conditions occurs: the distance to an obstacle is less than the safe distance threshold, the vehicle density is greater than the vehicle density threshold, or the visibility is lower than the visibility threshold, the situation is deemed urgent, indicating a high-risk road condition. If none of these three conditions occur, the situation is deemed not urgent, indicating a stable road condition.

[0067] Time-based smooth handover strategy: A time-based sigmoid gradient method is used to achieve a smooth transition of steering torque, and the transition time is set accordingly. The transition types are divided into three categories based on their differences: fast transition, normal transition, and slow transition. The transition duration of a fast transition is... =2 s; Transition duration for normal transition =4 s; Transition duration for a slow transition =6 s; the time-based sigmoid gradient function is as follows:

[0068]

[0069]

[0070] In the formula, For driver control weights, Increasing gradually from 0 to 1, Indicates fully autonomous driving. This indicates complete manual control; This is the steepness coefficient; the larger the value, the steeper the curve. Here, it is set to 12 / T. The current transition time is in seconds. , is the center point of the sigmoid function, i.e., the midpoint of the transition; This is the total steering torque; The system torque applied to the steering wheel by the steering system.

[0071] The present invention provides a method for smooth handover of autonomous driving control based on driver fatigue perception, which has the following advantages:

[0072] (1) Improve the safety and reliability of handover: Dynamically adjust the handover trigger threshold (torque and time) based on the driver's real-time fatigue status. Lower the trigger threshold when the driver is fatigued to avoid untimely handover; raise the threshold when the driver is alert to prevent false triggering, thereby significantly enhancing the safety tolerance of the handover process.

[0073] (2) Achieving a smooth and natural transition of driving control: The torque gradient strategy based on the sigmoid function is adopted to make the control control continuously and smoothly transferred between the system and the driver, effectively avoiding sudden changes in direction and vehicle vibration, and greatly improving ride comfort and operation smoothness.

[0074] (3) Enhance environmental adaptability: Integrate the urgency of road conditions (such as vehicle distance, density, and visibility) with the driver's state to make comprehensive decisions. In emergencies, rapid handover ensures safety, and in non-emergency situations, select the optimal handover rhythm based on fatigue level, enabling the system to have intelligent response capabilities in dynamic environments.

[0075] (4) Provides a low-cost and highly reliable fatigue perception solution: No additional biosensors are required. The driver's status is indirectly assessed by using the vehicle's existing sensor data (lane keeping, vehicle speed stability). The method is low-cost and easy to integrate, providing key driver status information for the handover system.

[0076] (5) Optimize human-machine collaboration efficiency: By deeply integrating driver status, environmental information and handover control, the system can make more humane and reasonable handover decisions, enhance human-machine mutual trust and collaboration efficiency, and improve the overall availability and user experience of the autonomous driving system. Attached Figure Description

[0077] Figure 1 This is a schematic diagram of a method for smooth handover of autonomous driving rights based on driver fatigue perception, provided in an embodiment of the present invention.

[0078] Figure 2 Weight parameters for rapid transition The sigmoid gradient function of steering wheel torque (where a is the weighting parameter) The sigmoid gradient function, where b is the total steering torque. (sigmoid gradient function);

[0079] Figure 3 Weight parameters for normal transition The sigmoid gradient function of steering wheel torque (where a is the weighting parameter) The sigmoid gradient function, where b is the total steering torque. (sigmoid gradient function);

[0080] Figure 4 Weight parameters for slow transition The sigmoid gradient function of steering wheel torque (where a is the weighting parameter) The sigmoid gradient function, where b is the total steering torque. (sigmoid gradient function). Detailed Implementation

[0081] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0082] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0083] like Figure 1As shown, an embodiment of the present invention provides a method for smooth handover of autonomous driving rights based on driver fatigue perception, comprising the following steps:

[0084] Step 1: Collect vehicle driving data and lane line images, and perform preprocessing and parameter definition;

[0085] Data Acquisition: Real-time vehicle speed is acquired via CAN bus. acceleration Data such as lane line images are collected via a vehicle-mounted forward-facing camera. The camera's acquisition frequency is set to 20 Hz, and the CAN bus data acquisition frequency is synchronized at 20 Hz to ensure data timing consistency.

[0086] Preprocessing: A fixed detection time window of 60 seconds is set. Index calculations and fatigue level determination are performed within this window, with a window sliding step of 30 seconds, meaning the fatigue level results are updated every 30 seconds. Kalman filtering is applied to the collected data to ensure accuracy and real-time performance. The Kalman filtering method for the collected vehicle speed and acceleration involves the following steps:

[0087] First, define the state variables. :

[0088]

[0089] in, for The actual vehicle speed at any given time, in km / h; for The actual acceleration at time t, in m / s² 2 ; For discrete time steps, corresponding to the sensor sampling times, when the sampling frequency is 20 Hz, the discrete time step is... sampling interval It is 0.05 s.

[0090] At the same time, construct the state equations:

[0091]

[0092] Wherein, the state transition matrix Process noise vector It follows a pattern with a mean of 0 and a covariance matrix of . Gaussian white noise represents non-uniform disturbances during vehicle movement. Vehicle speed process noise variance Take 6.5 (km / h) 2 Acceleration process noise variance Take 0.05 (m / s)2 ) 2 .

[0093] Furthermore, the observation equation is constructed as follows:

[0094]

[0095] Among them, the observation vector , for Vehicle speed observation value at any time for Acceleration observations at different times; observation matrix ; Observation noise vector It follows a pattern with a mean of 0 and a covariance matrix of . Gaussian white noise, Vehicle speed sensor noise variance Take 1.0 (m / s) 2 Accelerometer noise variance Take 0.15 (m / s) 2 ) 2 .

[0096] Next, filter initialization is performed. For the initial state vector... , and These are the first vehicle speed and acceleration observations collected by the sensors, respectively. When the vehicle starts from a standstill, both are zero. An initial error covariance matrix is ​​set. .

[0097] Next, state prediction, i.e., prior estimation, is performed. The formulas for state prediction and error covariance prediction are as follows:

[0098]

[0099]

[0100] in, for Prior state estimation at time step; for Posterior time estimation; for Time-prior error covariance; for The covariance of the posterior error at time step.

[0101] Subsequently, the Kalman gain was calculated. :

[0102]

[0103] Next, a state update, i.e., posterior estimation, is performed. The formulas for state update and error covariance update are as follows:

[0104]

[0105]

[0106] Finally, output the results. , and These are the filtered vehicle speed and acceleration values, respectively. The filtered state variables, for Posterior error covariance at time step It is an identity matrix.

[0107] The parameters are defined as follows:

[0108] Lane centerline: The left and right lane lines are extracted through image processing, and the line connecting the midpoints of the two lane lines is calculated as the lane centerline; Lane position: The horizontal offset distance of the vehicle's center of gravity from the lane centerline. The unit is cm. A leftward offset is recorded as a positive value, and a rightward offset as a negative value. Sudden change in vehicle speed: defined as an absolute acceleration value greater than 1.5 m / s². 2 The situation involves a sudden change in vehicle speed. The method for extracting the left and right lane markings using image processing involves the following steps:

[0109] First, the acquired RGB image is converted to grayscale, and Gaussian blur is used to remove noise.

[0110] Secondly, the Roberts operator edge detection algorithm is used to extract edge features from the image and generate a binary edge image.

[0111] Next, define the region of interest (ROI) in the lower half of the image, use a mask to filter out irrelevant regions, and output an edge image containing only the potential lane line regions.

[0112] Subsequently, the Hough transform is applied to detect line segments in the edge map and the coordinates of the line segment endpoints are obtained.

[0113] Next, the slope of each line segment is calculated, and the lines are classified as left or right lane lines based on the sign of the slope, and then averaged and fitted.

[0114] Finally, output the coordinates of the left and right lane lines.

[0115] Step 2: Calculation of detection indicators;

[0116] Based on the collected data and parameter definitions, four detection indicators were calculated within the set 60-second detection time window. The specific calculation method is as follows:

[0117] Standard deviation of lane position Extract the lane position data collected within the detection window and denote it as the sample set. Calculate the average of the sample set. ,in The number of times data is collected within the detection window, i.e. Next. Calculate the standard deviation of lane position. The standard deviation of lane position reflects the stability of a vehicle's position within the lane; a higher value indicates a weaker lane-keeping ability.

[0118] Number of times deviating from the lane per unit time A lane departure is defined as an instance where the absolute value of the deviation between the vehicle's center of gravity and the lane centerline is ≥30cm. The number of lane departures recorded within the monitoring window is the statistical measure. The unit is times per minute.

[0119] Standard deviation of vehicle speed Extract the vehicle speed data collected within the detection window and denote it as the sample set. Calculate the average of the sample set. Standard deviation of vehicle speed The standard deviation of vehicle speed reflects the stability of vehicle speed control; the larger the value, the more drastic the speed fluctuations.

[0120] Number of sudden speed changes Statistical detection window acceleration >1.5 m / s 2 The number of times is The unit is times per minute. The number of sudden speed changes reflects the driver's precision in adjusting the vehicle speed.

[0121] Step 3: Setting detection thresholds and determining driver fatigue status;

[0122] Lane position standard deviation threshold Set to 8 cm, that is when When the distance is ≤8 cm, the lane position is relatively stable. A lane departure distance greater than 8 cm indicates an abnormal lane position. The threshold for the number of lane departures per unit time. Set to 2 times / minute, that is, when Lane departure frequency is normal when it is ≤2 times / minute. Lane departure frequency is abnormal when it exceeds 2 times per minute. Vehicle speed standard deviation threshold. Set to 4km / h, that is when At speeds ≤4 km / h, vehicle speed control is stable. Vehicle speed control malfunctions at speeds >4 km / h. Threshold for the number of sudden speed changes. Set to 3 times / minute, that is when When the speed is ≤3 times / minute, the vehicle speed adjustment is relatively smooth. The vehicle speed adjustment was abnormal when it was >3 times per minute.

[0123] Compare the relationship between the four detection indicators and their corresponding thresholds, and count the number of abnormal indicators. ( (Choose 0, 1, 2, 3, 4), according to Values ​​used to determine driver fatigue status: when At that time, it was determined that the driver was conscious and had normal control over the vehicle; when At this time, the driver is judged to be in a state of mild fatigue, at which point the driver's attention is slightly reduced and the ability to control the vehicle is slightly weakened; when At this point, the driver is determined to be in a state of severe fatigue, at which point the driver's ability to control the vehicle is significantly reduced. or At this point, maintain the current fatigue state and proceed to the next sliding window detection.

[0124] Step 4: Set the steering wheel input torque threshold;

[0125] Set corresponding steering wheel input torque thresholds based on different driver fatigue levels. The corresponding relationships are as follows: the steering wheel input torque threshold for a conscious state is set to 1.5 N·m; the steering wheel input torque threshold for a mildly fatigued state is set to 1.0 N·m; and the steering wheel input torque threshold for a severely fatigued state is set to 0.6 N·m.

[0126] Step 5: Driver handover identification;

[0127] First, the torque sensor built into the vehicle's steering system collects the input torque applied by the driver to the steering wheel in real time. The data is collected at a frequency of 20 Hz and transmitted to the vehicle controller via the CAN bus. Next, the collected data is filtered. Then, driver transfer recognition is performed: when… And duration Reaching the time threshold When the automatic driving system is disengaged, driving control is transferred to the driver. Specific conditions for this determination are as follows:

[0128] Awake state: =1.5 N·m, =0.3 s, that is ≥1.5 N·m and If the time is ≥0.3 s, it is determined that the driver has manually taken over the vehicle. Mild fatigue state: =1.0 N·m, =0.5 s, that is ≥1.0 N·m and If the time is ≥0.5 s, it is determined that the driver has manually taken over the vehicle. Severe fatigue state: =0.6 N·m, =0.8 s, that is ≥0.6 N·m and If the time is ≥0.8 s, it is determined that the driver has manually taken over the vehicle.

[0129] Step 6: Smooth transfer of driving rights;

[0130] After determining that the driver has manually taken over the vehicle, a smooth transfer of driving control is performed. First, onboard sensors and a forward-facing camera assess the current road conditions to determine the urgency of the driving control transfer. Then, based on the urgency of the transfer and the driver's fatigue level, a corresponding transition strategy is selected to achieve a smooth change in steering wheel torque, ensuring a seamless transition from autonomous driving mode to manual driving mode. The specific method is as follows:

[0131] Driving urgency assessment: This involves using onboard forward-facing cameras and LiDAR to capture road information ahead, including distance to obstacles, vehicle density, and visibility. A safe distance threshold of 50 m, a vehicle density threshold of 5 vehicles / 100 m, and a visibility threshold of 100 m are set. If at least one of the following conditions occurs: the distance to an obstacle is less than the safe distance threshold, the vehicle density is greater than the vehicle density threshold, or the visibility is lower than the visibility threshold, the situation is deemed urgent, indicating a high-risk road condition. If none of these three conditions occur, the situation is deemed not urgent, indicating a stable road condition.

[0132] Time-based smooth handover strategy: A time-based sigmoid gradient method is used to achieve a smooth transition of steering torque, and the transition time is set accordingly. The transition types are divided into three categories based on their differences: fast transition, normal transition, and slow transition. The transition duration of a fast transition is... =2 s; Transition duration for normal transition =4 s; Transition duration for a slow transition =6 s. The time-based sigmoid gradient function is as follows:

[0133]

[0134]

[0135] In the formula, For driver control weights, Increasing gradually from 0 to 1, Indicates fully autonomous driving. This indicates complete manual control; This is the steepness coefficient; the larger the value, the steeper the curve. Here, it is set to 12 / T. The current transition time is in seconds. , is the center point of the sigmoid function, i.e., the midpoint of the transition; This is the total steering torque; The system torque applied to the steering wheel by the steering system.

[0136] Figure 2 The weighted parameter indicates whether the transfer of driving rights is urgent or not, and the driver is conscious. The diagram illustrates the sigmoid progression of steering wheel torque over transition time. The driver's input torque is set to 2 N·m, and the system torque to 0.5 N·m. Figure a shows that the driver's control weight transitions from 0 to 1 over 2 seconds via a sigmoid progression, with the driver's control weight and the steering system control weight each accounting for 0.5 at 1 second. Figure b shows that the steering wheel torque transitions from the system torque of 0.5 N·m to the driver's input torque of 2 N·m over 2 seconds via a sigmoid progression. This rapid transition allows for quick vehicle takeover when the driver is in a good condition, but requires greater operational force to avoid accidental triggering.

[0137] Figure 3 This indicates the weighting parameters when the transfer of driving rights is not urgent and the driver is slightly fatigued. The diagram illustrates the sigmoid transition relationship between steering wheel torque and transition time. The driver's input torque is set to 1.5 N·m, and the system torque to 0.5 N·m. Figure a shows that the driver's control weight transitions from 0 to 1 over 4 seconds via a sigmoid transition, with the driver's control weight and the steering system control weight each accounting for 0.5 at 2 seconds. Figure b shows that the steering wheel torque transitions from the system torque of 0.5 N·m to the driver's input torque of 1.5 N·m over 4 seconds via a sigmoid transition. This normal transition balances the take-off speed and the take-off threshold.

[0138] Figure 4 This indicates the weighting parameters when the transfer of driving rights is not urgent and the driver is severely fatigued. The diagram illustrates the sigmoid transition relationship between steering wheel torque and transition time. The driver's input torque is set to 1 N·m, and the system torque to 0.5 N·m. Figure a shows that the driver's control weight transitions from 0 to 1 over 6 seconds via a sigmoid transition, with the driver's control weight and the steering system control weight each accounting for 0.5 at 3 seconds. Figure b shows that the steering wheel torque transitions from the system torque of 0.5 N·m to the driver's input torque of 1 N·m over 6 seconds via a sigmoid transition. This slow transition allows for triggering takeover with a smaller steering torque when the driver is highly fatigued, reducing the risk of difficulty in takeover due to fatigue.

[0139] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for smooth handover of autonomous driving rights based on driver fatigue perception, characterized in that, Includes the following steps: Step 1: Data Acquisition and Preprocessing; Collect vehicle driving data and lane line images, perform filtering preprocessing on the collected data, and define parameters; Step 2: Calculation of detection indicators; Based on the preprocessed data, detection indicators used to characterize the driver's control ability are calculated within the set detection time window. Step 3: Determine driver fatigue status; Set thresholds for each detection indicator, compare the magnitude of each detection indicator with the corresponding threshold, count the number of abnormal indicators, and determine the driver's fatigue state based on the number of abnormal indicators. Step 4: Threshold setting; Based on the determined driver fatigue level, set the corresponding steering wheel input torque threshold. Step 5: Driver handover identification; The system collects the input torque applied by the driver to the steering wheel in real time. When the input torque exceeds the steering wheel input torque threshold and the duration reaches the time threshold, it is determined that the driver has manually taken over the vehicle. Step 6: Smooth transfer of driving rights; After determining that the driver has manually taken over the vehicle, a time-based sigmoid transition strategy is adopted based on the current road conditions and the driver's fatigue state to smoothly transition the steering wheel torque from the system torque to the driver's input torque, thereby achieving a smooth transfer of driving control.

2. The method for smooth handover of autonomous driving rights based on driver fatigue perception according to claim 1, characterized in that, Step 1 includes the following specific steps: Data Acquisition: Real-time vehicle speed is acquired via CAN bus. and acceleration Lane line image data is collected through the vehicle-mounted forward-facing camera. The camera acquisition frequency is set to 20 Hz, and the CAN bus data acquisition frequency is synchronized at 20 Hz. Preprocessing: A fixed detection time window of 60 seconds is set, and index calculations and fatigue level determination are performed in units of this window, with a window sliding step of 30 seconds; Kalman filtering is applied to the collected data; the Kalman filtering method for the collected vehicle speed and acceleration is specifically divided into the following steps: First, define the state variables. : in, for The actual vehicle speed at any given time, in km / h; for The actual acceleration at time t, in m / s² 2 ; For discrete time steps, corresponding to the sensor sampling times, when the sampling frequency is 20 Hz, the discrete time step is... sampling interval It is 0.05 s; At the same time, construct the state equations: Wherein, the state transition matrix Process noise vector It follows a pattern with a mean of 0 and a covariance matrix of . Gaussian white noise represents non-uniform disturbances during vehicle movement. Vehicle speed process noise variance Take 6.5 (km / h) 2 Acceleration process noise variance Take 0.05 (m / s) 2 ) 2 ; Furthermore, the observation equation is constructed as follows: Among them, the observation vector , for Vehicle speed observation value at any time for Acceleration observations at different times; observation matrix ; Observation noise vector It follows a pattern with a mean of 0 and a covariance matrix of . Gaussian white noise, Vehicle speed sensor noise variance Take 1.0 (m / s) 2 Accelerometer noise variance Take 0.15 (m / s) 2 ) 2 ; Secondly, filter initialization is performed; for the initial state vector , and These are the first vehicle speed and acceleration observations collected by the sensors, respectively. When the vehicle starts from a standstill, both are zero; an initial error covariance matrix is ​​set. ; Next, state prediction, i.e., prior estimation, is performed; the formulas for state prediction and error covariance prediction are as follows: in, for Prior state estimation at time step; for Posterior time estimation; for Time-prior error covariance; for Posterior error covariance at time step; Subsequently, the Kalman gain was calculated. : Next, a state update, i.e., posterior estimation, is performed; the formulas for state update and error covariance update are as follows: Finally, output the results. , and These are the filtered vehicle speed and acceleration values, respectively. The filtered state variables, for Posterior error covariance at time step It is the identity matrix; The parameters are defined as follows: Lane centerline: The left and right lane lines are extracted through image processing, and the line connecting the midpoints of the two lane lines is calculated as the lane centerline; Lane position: The horizontal offset distance of the vehicle's center of gravity from the lane centerline. The unit is cm. A leftward offset is recorded as a positive value, and a rightward offset as a negative value. Sudden change in vehicle speed: defined as an absolute acceleration value greater than 1.5 m / s². 2 The situation was a sudden change in vehicle speed.

3. The method for smooth handover of autonomous driving rights based on driver fatigue perception according to claim 2, characterized in that, In step 1, the method for extracting the left and right lane lines of the road through image processing includes the following steps: First, the acquired RGB image is converted to grayscale, and Gaussian blur is used to remove noise; Secondly, the Roberts operator edge detection algorithm is used to extract edge features from the image and generate a binary edge image; Next, define the region of interest (ROI) in the lower half of the image, use a mask to filter out irrelevant regions, and output an edge image that only contains potential lane line regions. Subsequently, the Hough transform is applied to detect line segments in the edge map and the coordinates of the line segment endpoints are obtained; Next, the slope of each line segment is calculated, and the lines are classified as left or right lane lines based on the sign of the slope, and then averaged and fitted. Finally, output the coordinates of the left and right lane lines.

4. The method for smooth handover of autonomous driving rights based on driver fatigue perception according to claim 2, characterized in that, Step 2 includes the following specific steps: Based on the data and parameter definitions collected in step 1, four detection indicators are calculated within the set 60-second detection time window. The specific calculation method is as follows: Standard deviation of lane position Extract the lane position data collected within the detection window and denote it as the sample set. ; Calculate the average of the sample set ,in The number of times data is collected within the detection window, i.e. Calculate the standard deviation of lane position. ; Number of times deviating from the lane per unit time When the absolute value of the offset distance between the vehicle's center of gravity and the lane centerline is ≥30 cm, it is considered a lane departure. The number of times a lane departure occurs within the monitoring window is counted. The unit is times per minute; Standard deviation of vehicle speed Extract the vehicle speed data collected within the detection window and denote it as the sample set. Calculate the average of the sample set. Standard deviation of vehicle speed ; Number of sudden speed changes Statistical detection window acceleration >1.5 m / s 2 The number of times is The unit is times per minute.

5. The method for smooth handover of autonomous driving rights based on driver fatigue perception according to claim 4, characterized in that, Step 3 includes the following specific steps: Lane position standard deviation threshold Set to 8 cm, that is when When the distance is ≤8 cm, the lane position is relatively stable. When the distance is greater than 8 cm, the lane position is abnormal; Lane departure threshold per unit time Set to 2 times / minute, that is, when Lane departure frequency is normal when it is ≤2 times / minute. Lane departure frequency is abnormal when it exceeds 2 times / minute; vehicle speed standard deviation threshold. Set to 4km / h, that is when At speeds ≤4 km / h, vehicle speed control is stable. Vehicle speed control malfunctions at speeds >4 km / h; threshold for sudden speed changes. Set to 3 times / minute, that is when When the speed is ≤3 times / minute, the vehicle speed adjustment is relatively smooth. The vehicle speed adjustment is abnormal when it is >3 times / minute; Compare the relationship between the four detection indicators and their corresponding thresholds, and count the number of abnormal indicators. ,according to Values ​​used to determine driver fatigue status: when At that time, it was determined that the driver was conscious and had normal control over the vehicle; when At that time, it was determined that the driver was in a state of mild fatigue, and his ability to control the vehicle was reduced; when When the driver is determined to be in a state of severe fatigue, their ability to control the vehicle is significantly reduced; when or At this point, maintain the current fatigue state and proceed to the next sliding window detection.

6. The method for smooth handover of autonomous driving rights based on driver fatigue perception according to claim 5, characterized in that, In step 4, a corresponding steering wheel input torque threshold is set according to different driver fatigue states. The corresponding relationships are as follows: the steering wheel input torque threshold for a conscious state is set to 1.5 N·m; the steering wheel input torque threshold for a mildly fatigued state is set to 1.0 N·m; and the steering wheel input torque threshold for a severely fatigued state is set to 0.6 N·m.

7. The method for smooth handover of autonomous driving rights based on driver fatigue perception according to claim 6, characterized in that, Step 5 includes the following steps: First, the torque sensor built into the vehicle's steering system collects the input torque applied by the driver to the steering wheel in real time. The data is collected at a frequency of 20 Hz and transmitted to the vehicle controller via the CAN bus. Next, the collected data is filtered. Then, driver transfer recognition is performed: when… And duration Reaching the time threshold When the automatic driving system is disengaged, driving control is transferred to the driver; the specific conditions for this determination are as follows: Awake state: =1.5 N·m, =0.3 s, that is ≥1.5 N·m and If the time is ≥0.3 s, it is determined that the driver has manually taken over the vehicle; mild fatigue state: =1.0 N·m, =0.5 s, that is ≥1.0 N·m and If the time is ≥0.5 s, it is determined that the driver has manually taken over the vehicle; severe fatigue state: =0.6 N·m, =0.8 s, that is ≥0.6 N·m and If the time is ≥0.8 s, it is determined that the driver has manually taken over the vehicle.

8. The method for smooth handover of autonomous driving rights based on driver fatigue perception according to claim 7, characterized in that, Step 6 includes the following specific steps: First, the vehicle's onboard sensors and forward-facing camera assess the current road conditions to determine the urgency of handing over driving control. Then, based on the urgency of the handover and the driver's fatigue level, a corresponding transition strategy is selected to achieve a smooth change in steering wheel torque, allowing the vehicle to smoothly switch from autonomous driving mode to manual driving mode. The specific method is as follows: Driving urgency assessment: The system uses a forward-facing camera and lidar to capture road information ahead, including distance to obstacles, vehicle density, and visibility. A safe distance threshold of 50 m, a vehicle density threshold of 5 vehicles / 100 m, and a visibility threshold of 100 m are set. If at least one of the following conditions occurs: the distance to an obstacle is less than the safe distance threshold, the vehicle density is greater than the vehicle density threshold, or the visibility is lower than the visibility threshold, the situation is deemed urgent, indicating a high-risk road condition. If none of these three conditions occur, the situation is deemed not urgent, indicating a stable road condition. Time-based smooth handover strategy: A time-based sigmoid gradient method is used to achieve a smooth transition of steering torque, and the transition time is set accordingly. The transition types are divided into three categories based on their differences: fast transition, normal transition, and slow transition. The transition duration of a fast transition is... =2 s; Transition duration for normal transition =4 s; Transition duration for a slow transition =6 s; the time-based sigmoid gradient function is as follows: In the formula, For driver control weights, Increasing gradually from 0 to 1, Indicates fully autonomous driving. This indicates complete manual control; This is the steepness coefficient; the larger the value, the steeper the curve. Here, it is set to 12 / T. The current transition time is in seconds. , is the center point of the sigmoid function, i.e., the midpoint of the transition; This is the total steering torque; The system torque applied to the steering wheel by the steering system.

Images