A method, device and system for judging fatigue driving

Abstract

This application discloses a method, device, and system for assessing fatigued driving, relating to the field of vehicle driving monitoring. The method includes acquiring multiple normal driving trajectories of vehicles on a target road segment; using a dynamic motion primitive model to spatiotemporally align the multiple normal driving trajectories, obtaining aligned normal driving trajectories; based on the aligned normal driving trajectories, using a Gaussian mixture model for probability modeling to obtain a probability density function; based on the probability density function, using Gaussian mixture regression to generate a representative expected trajectory of normal driving for vehicles on the target road segment; determining the deviation between the real-time driving trajectory of the vehicle to be assessed and the representative expected trajectory on the target road segment; if the deviation exceeds a dynamic judgment threshold, determining that the vehicle to be assessed is in a suspected state of fatigued driving. This application can adaptively learn the distribution of normal driving trajectories on specific road segments and accurately identify fatigued driving behavior.

Description

Technical Field

[0001] This application relates to the field of vehicle driving monitoring, and in particular to a method, device and system for judging fatigued driving. Background Technology

[0002] With the continuous development of intelligent transportation systems and road safety technologies, fatigued driving, as one of the major causes of traffic accidents, has received increasing attention from academia and industry. Traditional fatigued driving detection methods mainly rely on onboard sensors (such as steering wheel operation signals, eye tracking, heart rate monitoring, etc.) or driver physiological state recognition technologies. While these methods are effective in specific scenarios, they generally suffer from limitations such as high installation costs, strong invasiveness, susceptibility to individual differences, and difficulty in deployment in large-scale road networks.

[0003] In recent years, non-contact driving behavior analysis technology based on video surveillance has gradually emerged, and indirectly judging the driver's state by analyzing the characteristics of the vehicle's driving trajectory on the road has become a promising new direction. In existing technologies, some studies attempt to model vehicle trajectories using fixed cameras or in-vehicle video and assess driving anomalies through indicators such as trajectory deviation and lateral offset frequency. However, these methods mostly rely on preset rules or simple statistical thresholds, lacking the ability to finely model "normal driving" behavior patterns and struggling to adapt to complex road geometry and traffic flow changes. Furthermore, most schemes do not fully consider road segment specificity—that is, the impact of different road alignments, speed limits, traffic densities, and other factors on the distribution of normal trajectories—leading to high false alarm rates and poor generalization ability.

[0004] Against this backdrop, there is an urgent need for a fatigue driving assessment method that can adaptively learn the distribution of normal driving trajectories on specific road sections and thereby achieve high-precision abnormal trajectory identification. Summary of the Invention

[0005] The purpose of this application is to provide a method, device and system for judging fatigued driving, which can adaptively learn the distribution of normal driving trajectory on a specific road segment and identify fatigued driving behavior with high accuracy.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for assessing driver fatigue, including: Obtain multiple normal driving trajectories of vehicles on the target road segment; A dynamic movement primitive (DMP) model is used to perform spatiotemporal alignment on multiple normal driving trajectories to obtain the aligned multiple normal driving trajectories. Based on the aligned multiple normal driving trajectories, a Gaussian Mixture Model (GMM) is used for probability modeling to obtain the probability density function. Based on the probability density function, Gaussian Mixture Regression (GMR) is used to generate representative expected trajectories of vehicles traveling normally on the target road segment. Determine the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory; If the deviation is greater than the dynamic judgment threshold, the vehicle under evaluation is determined to be in a state of suspected fatigue driving.

[0007] Secondly, this application provides a fatigue driving assessment device, including: an aerial video acquisition device and a processor; The aerial video acquisition device is used to acquire multiple normal driving trajectories of vehicles on the target road segment, as well as the real-time driving trajectory of the vehicles to be evaluated on the target road segment. The processor is used to perform spatiotemporal alignment of multiple normal driving trajectories using a dynamic motion primitive model to obtain aligned normal driving trajectories; based on the aligned normal driving trajectories, a Gaussian mixture model is used for probability modeling to obtain a probability density function; based on the probability density function, a Gaussian mixture regression is used to generate a representative expected trajectory of normal driving of vehicles on the target road segment; the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory is determined; if the deviation is greater than a dynamic judgment threshold, the vehicle to be evaluated is determined to be in a suspected state of fatigued driving.

[0008] Thirdly, this application provides a fatigue driving assessment system, including: The sample acquisition module is used to acquire multiple normal driving trajectories of vehicles on the target road segment; The spatiotemporal alignment module is used to perform spatiotemporal alignment of multiple normal driving trajectories using a dynamic motion primitive model, thereby obtaining multiple aligned normal driving trajectories. The probability modeling module is used to perform probability modeling based on multiple aligned normal driving trajectories using a Gaussian mixture model to obtain the probability density function; The expected trajectory generation module is used to generate a representative expected trajectory of vehicles traveling normally on the target road segment based on the probability density function and Gaussian mixture regression. The deviation calculation module is used to determine the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory. The determination module is used to determine that the vehicle under evaluation is in a suspected state of fatigued driving if the deviation is greater than the dynamic determination threshold.

[0009] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a fatigue driving assessment method, device, and system. It utilizes a Gaussian mixture model to learn the multimodal distribution of normal driving trajectories. The Gaussian mixture model is entirely driven by historical data of the target road segment. Through data-driven probabilistic modeling, it accurately characterizes the statistical distribution boundary of "normal driving" under a specific road segment, significantly improving the discrimination accuracy. Furthermore, the representative expected trajectory is generated by the normal driving behavior of the target road segment itself, naturally adapting to the characteristics of that road segment. Thus, the representative expected trajectory can be used to identify fatigue driving behavior with high accuracy. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a fatigue driving assessment method provided in this application embodiment; Figure 2 This is a schematic diagram of high-altitude perspective trajectory acquisition provided in an embodiment of this application; Figure 3 This is a schematic diagram of the trajectory spatiotemporal sequence provided in an embodiment of this application; Figure 4 A schematic diagram illustrating the process of constructing a representative expected trajectory based on a Gaussian mixture model / Gaussian mixture regression, as provided in an embodiment of this application. Figure 5 A more detailed flowchart of a fatigue driving assessment method provided in this application embodiment is shown below. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0014] In one exemplary embodiment, such as Figure 1 As shown, a method for judging fatigued driving is provided, including the following steps 101 to 106.

[0015] Step 101: Obtain multiple normal driving trajectories of vehicles on the target road segment.

[0016] Step 102: Using a dynamic motion primitive model, multiple normal driving trajectories are spatiotemporally aligned to obtain the aligned multiple normal driving trajectories.

[0017] Step 103: Based on the aligned multiple normal driving trajectories, use a Gaussian mixture model to perform probability modeling and obtain the probability density function.

[0018] Step 104: Based on the probability density function, Gaussian mixture regression is used to generate a representative expected trajectory of vehicles traveling normally on the target road segment.

[0019] Step 105: Determine the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory.

[0020] Step 106: If the deviation is greater than the dynamic judgment threshold, the vehicle under evaluation is determined to be in a state of suspected fatigue driving.

[0021] In another exemplary embodiment of this application, step 101 described above may be replaced by steps 201 to 202.

[0022] Step 201: Using aerial video acquisition equipment deployed above the target road segment, continuously capture the driving process of multiple vehicles passing through the target road segment, and use target detection algorithm and multi-target tracking algorithm to extract the driving trajectory to obtain the original spatiotemporal trajectory sequence.

[0023] Road segment video data acquisition: Deploy high-altitude video acquisition equipment (such as hovering drones) above the target road segment to continuously capture the driving process of N vehicles passing through the road segment at a fixed frame rate (e.g., 30fps). Figure 2 As shown. The driving trajectory of each vehicle is extracted using object detection and multi-object tracking algorithms to obtain the original spatiotemporal trajectory sequence (e.g., Figure 3 (as shown) ; in Indicates the first a car, The number of points on its trajectory, ( x , y () represents pixel coordinates. For timestamps.

[0024] Step 202: Filter driving trajectories under non-fatigue driving conditions from the original spatiotemporal trajectory sequence to obtain multiple normal driving trajectories of vehicles on the target road segment.

[0025] The collected K trajectories and their corresponding vehicle motion videos are manually interpreted to remove obviously abnormal trajectories, retaining the set of M trajectories judged as "non-fatigue driving" (i.e., normal). .

[0026] In another exemplary embodiment of this application, to eliminate the problem of inconsistent trajectory duration and phase caused by differences in speed and start and end times of different vehicles, a dynamic motion primitive model is used to standardize and align each normal trajectory. Therefore, step 102 can be replaced by steps 301 to 304.

[0027] Step 301: Randomly select one trajectory from multiple normal driving trajectories as a reference trajectory.

[0028] Randomly select one of them as a reference trajectory. This is used for subsequent alignment references.

[0029] Step 302: Perform time normalization on the reference trajectory to obtain the time-normalized reference trajectory.

[0030] For reference trajectory Time normalization is performed to [0,1], denoted as .

[0031] Step 303: Define each normal driving trajectory other than the reference trajectory among multiple normal driving trajectories as the trajectory to be aligned, and establish a dynamic motion primitive model for the x and y components of each trajectory to be aligned.

[0032] For each trajectory to be aligned Dynamic motion primitive models are established for its x and y components respectively: ; Where z represents the position (x or y). Let z represent the first and second derivatives, respectively. K , D The spring-damping coefficient, For the target destination, It is a nonlinear forcing term. Scaling factor For phase variables.

[0033] Step 304: Using the dynamic motion primitive model of the x and y components of each trajectory to be aligned, adjust the start and end points of each trajectory to be aligned to be consistent with the time-normalized reference trajectory, and generate multiple aligned normal driving trajectories.

[0034] Leveraging the trajectory generalization capability of DMP, The starting and ending points are forcibly adjusted to be consistent with Consistency (i.e.) ， Generate aligned trajectories All trajectories have the same number of sampling points S+1 and a uniform time axis.

[0035] In another exemplary embodiment of this application, step 103 described above may be replaced by steps 401 to 402.

[0036] Step 401: Align the multiple normal driving tracks Stacking the data by time steps forms the training dataset: ;in, For the training dataset, The sampling point number, To align multiple normal driving trajectories at sampling points The coordinates below, This is the sequence number of the normal driving trajectory. The number of normal driving trajectories. This is the sequence number of the largest sampling point.

[0037] Step 402: Use a Gaussian mixture model to analyze the joint variables. The distribution is modeled, and C Gaussian components are learned to obtain the probability density function; where, Represents the x-coordinate, Represents the vertical axis.

[0038] For example, the probability density function of a Gaussian mixture model is: ; In the formula, Let be the probability density function. For the first The weighting parameters of each Gaussian component. For the first The mean vector and covariance matrix of each Gaussian component. It is a one-dimensional Gaussian probability density function. The number of Gaussian components.

[0039] In another exemplary embodiment of this application, step 104 is used to construct a representative expected trajectory of the road segment, and step 104 can be replaced by the following steps 501 to 502.

[0040] Step 501: At each sampling point, based on the probability density function, Gaussian mixture regression is used to generate the expected trajectory for each sampling point as follows: ; In the formula, Sampling points The expected trajectory below, Let x be the x-coordinate of the desired trajectory. Let y be the ordinate of the desired trajectory. Sampling points The conditional mean of (x,y); For posterior weights.

[0041] ; .

[0042] in, For the first Each Gaussian component at the sampling point The mean of the time components below, For the first Each Gaussian component at the sampling point The time variance below, For the first The weighting parameters of each Gaussian component. For the first Each Gaussian component at the sampling point The mean of the time components below, For the first Each Gaussian component at the sampling point The time variance.

[0043] Step 502: Combining the expected trajectories from all sampling points, the representative expected trajectory for normal vehicle travel on the target road segment is obtained as follows: In the formula, This represents the expected trajectory.

[0044] Figure 4 The process of constructing the representative expected trajectory is explained. Figure 4 Part (a) in the text represents the input trajectory. Figure 4 Part (b) in the text represents the 6 Gaussian components (i.e., C=6). Figure 4 Part (c) in the diagram represents the Gaussian mixture regression process, and the green trajectory in the middle is the representative expected trajectory.

[0045] In another exemplary embodiment of this application, step 105 described above is used to calculate the real-time trajectory deviation. Step 105 can be replaced by steps 601 to 604.

[0046] Step 601: Using a dynamic motion primitive model, the real-time driving trajectory of the vehicle to be evaluated is spatiotemporally aligned with the representative expected trajectory to obtain the aligned real-time driving trajectory.

[0047] Newly collected vehicle trajectories to be evaluated First, DMP alignment is performed to obtain... .

[0048] Step 602: Using the formula Calculate the spatial deviation between the aligned real-time driving trajectory and the representative expected trajectory; where, For spatial bias, As the first weight, As the second weight, The coordinates of the aligned real-time driving trajectory. The coordinates represent the expected trajectory. =0.7, =0.3, which can be manually adjusted according to the curvature of the curve.

[0049] Step 603: Using the formula Calculate the speed consistency deviation between the aligned real-time driving trajectory and the representative expected trajectory; where, This is due to deviations in speed consistency. The instantaneous velocity at sampling point s of the aligned real-time driving trajectory. , To determine the position of the aligned real-time driving trajectory at sampling point s+1, The position of the aligned real-time driving trajectory at sampling point s; The velocity of the representative expected trajectory at sampling point s, , The representative expected trajectory at the sampling points The location, The representative expected trajectory at the sampling points The location.

[0050] Step 604: Based on the spatial deviation and the velocity consistency deviation, use the formula... The comprehensive deviation is determined and used as the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory; where, To account for the overall deviation, , and For weight parameters, This represents the Fraser distance between the aligned real-time driving trajectory and the representative expected trajectory. Used to evaluate the similarity between two trajectories. , and It is considered a set value.

[0051] In another exemplary embodiment of this application, in order to better match the characteristics of the road segment, the method may further include: triggering an early warning when it is determined that the vehicle to be evaluated is in a suspected state of fatigued driving; calculating the daily early warning rate of the target road segment; if the early warning rate exceeds p times the average early warning rate for n consecutive days, adjusting the dynamic judgment threshold or redetermining the representative expected trajectory.

[0052] The daily warning rate is equal to the ratio of the number of warnings issued each day to the total number of vehicles detected. n can be 2.

[0053] Reference Figure 5 This application collects historical driving trajectories of multiple vehicles on a target road segment using high-altitude video (such as drone hovering monitoring), filters out non-fatigue driving (i.e., normal) trajectory samples, aligns the trajectories using a dynamic motion primitive model, and then uses a Gaussian mixture model to probabilistically model the spatial-temporal distribution of the normal trajectories. Finally, it combines Gaussian mixture regression to generate a representative expected trajectory for the road segment. In practical application, the real-time collected vehicle trajectories are matched with this representative trajectory. When the trajectory deviation exceeds a dynamic judgment threshold, it is judged as suspected fatigue driving. This method does not rely on onboard equipment and has advantages such as non-intrusiveness, strong scalability, and adaptability to road segment characteristics. It effectively solves the problems of coarse normal behavior modeling, high misjudgment rate, and high deployment cost in existing technologies, providing a new technical path for proactive traffic safety early warning in large-scale road networks.

[0054] The core of this application lies in: standardizing and aligning the normal driving trajectories of vehicles collected by multi-source high-altitude video equipment; using one normal trajectory as a reference, modeling other trajectories using a dynamic motion primitive model (DMP); and achieving spatiotemporal alignment by adjusting the start and end points to match the reference trajectory and utilizing the trajectory generalization capability of the DMP. Subsequently, based on the aligned normal trajectory samples, probabilistic modeling is performed using a Gaussian mixture model (Gaussian mixture regression), and a road segment-specific representative expected trajectory is generated by combining Gaussian mixture regression. Finally, by comparing whether the deviation between the real-time trajectory and the expected trajectory exceeds a dynamic threshold, it is determined whether it is suspected fatigue driving. The key point of the invention is to construct a road segment-specific normal trajectory probabilistic model in a data-driven manner to achieve non-contact, highly adaptable automatic fatigue driving identification.

[0055] The beneficial effects of this application are as follows: 1. High accuracy and low false alarm rate: It can effectively distinguish between normal driving variations and real fatigue behavior.

[0056] Advantages: Traditional methods based on fixed thresholds or simple rules (such as lateral offset > 0.5m) are prone to misjudging trajectory fluctuations caused by differences in vehicle speed, road curvature, or temporary disturbances (such as obstacle avoidance) as fatigue driving. This invention, however, uses data-driven probabilistic modeling to accurately characterize the statistical distribution boundary of "normal driving" on specific road sections, significantly improving judgment accuracy and reducing false alarm rates.

[0057] Technology Sources: GMM / GMR Modeling: Gaussian mixture models are used to learn the multimodal distribution of normal trajectories (e.g., the lateral behavior of curve entrances and exits is different), and smooth desired trajectories are generated through GMR; Empirical Threshold Setting: Thresholds are set based on expert experience to match the evaluation criteria with the characteristics of road segments.

[0058] Cause: GMM can capture the inherent variability of normal driving behavior (such as lane preferences of different drivers), while dynamic thresholds ensure that only those that truly deviate from the trajectory of the "normal behavior cloud" are marked as abnormal, thus avoiding misjudging reasonable variations as fatigue.

[0059] 2. Strong robustness and generalization ability: Applicable to different vehicle speeds, vehicle types and traffic conditions.

[0060] Advantages: Existing video analysis methods often suffer from inconsistent trajectory lengths and sampling densities due to varying vehicle speeds, making standardized modeling difficult. This invention can standardize the trajectories of vehicles at any speed, achieving stable evaluation across different scenarios.

[0061] Technical source: DMP-based trajectory alignment: using dynamic motion primitives to map trajectories with different start and end times and different speeds to a unified normalized time axis s∈[0,1], and force the start / end point alignment.

[0062] Reason for its existence: The core advantage of DMP lies in decoupling trajectory shape from execution time. Regardless of whether the train is fast or slow, after alignment, they are represented on the same spatiotemporal grid, making subsequent modeling and deviation calculations comparable and fundamentally solving the problem of heterogeneous multi-source trajectories.

[0063] 3. Retain key dynamic features and take into account both spatial and temporal anomaly detection.

[0064] Advantages: Although time normalization was performed, this invention did not lose speed-related abnormal information and can still detect behaviors such as unconscious deceleration and speed fluctuations caused by fatigue.

[0065] Technical source: Comprehensive deviation index design: not only calculating spatial position deviation It also explicitly introduces the normalized velocity curve bias. DMP generalization preserves local dynamics: the trajectory reconstructed by DMP implicitly contains the original acceleration characteristics.

[0066] Cause: By calculating the distance between adjacent points as "relative velocity" in the normalized time domain and comparing it with the expected velocity distribution, the abnormal patterns of velocity dynamics are effectively captured, making up for the shortcomings of pure spatial methods and realizing the fusion and discrimination of multi-dimensional fatigue features.

[0067] 4. Non-contact, low-cost, and easy to deploy, suitable for large-scale road network monitoring.

[0068] Advantages: Compared to solutions that rely on vehicle-mounted sensors (such as eye trackers and steering wheel torque), this invention only requires high-altitude video (such as drones or fixed cameras), without the need to modify vehicles, resulting in low deployment costs and the ability to monitor multiple lanes and multiple vehicles simultaneously.

[0069] Technology source: High-altitude video acquisition + manual annotation of normal trajectories: Constructing a "non-intrusive" data acquisition and modeling paradigm; Automated process: From trajectory extraction, alignment, modeling to evaluation, everything can be done automatically.

[0070] Reasons for this: The vision-centric architecture naturally supports non-contact monitoring. Combined with the flexible deployment capabilities of drones, it can quickly cover accident-prone road sections, making it particularly suitable for proactive safety warnings in key areas such as highways and mountain curves.

[0071] 5. Road segment adaptability: The model automatically adjusts according to the road geometry.

[0072] Advantages: The same algorithm can maintain high accuracy on different road sections such as straight roads, curves, and slopes, without the need for manual reconfiguration of rules.

[0073] Technical source: GMM is built based on actual collected normal trajectories: the model is entirely driven by historical data of this road segment; empirical thresholds are based on local data distribution.

[0074] Cause of this: The representative expected trajectory and judgment threshold are generated by the normal driving behavior of the road segment itself, which is naturally adapted to the characteristics of the road segment such as alignment, speed limit, and sight distance, so as to achieve "one model for one place" and avoid the failure of the general model under complex road conditions.

[0075] Based on the same inventive concept, this application also provides a fatigue driving assessment device for implementing the above-described method. The solution provided by this device is similar to the solution described in the above-described method; therefore, the specific limitations of one or more fatigue driving assessment device embodiments provided below can be found in the limitations of the fatigue driving assessment method described above, and will not be repeated here.

[0076] In an exemplary embodiment, a fatigue driving assessment device is provided, comprising: an aerial video acquisition device and a processor. The aerial video acquisition device is used to acquire multiple normal driving trajectories of vehicles on a target road segment, as well as the real-time driving trajectory of the vehicle to be assessed on the target road segment. The processor is used to perform spatiotemporal alignment of the multiple normal driving trajectories using a dynamic motion primitive model to obtain aligned multiple normal driving trajectories; based on the aligned multiple normal driving trajectories, a Gaussian mixture model is used for probability modeling to obtain a probability density function; based on the probability density function, a Gaussian mixture regression is used to generate a representative expected trajectory of normal driving of the vehicle on the target road segment; the deviation between the real-time driving trajectory of the vehicle to be assessed on the target road segment and the representative expected trajectory is determined; if the deviation is greater than a dynamic judgment threshold, the vehicle to be assessed is determined to be in a suspected fatigue driving state.

[0077] In an exemplary embodiment, this application also provides a fatigue driving assessment system for implementing the above method. The system includes: a sample acquisition module, a spatiotemporal alignment module, a probability modeling module, a desired trajectory generation module, a deviation calculation module, and a judgment module.

[0078] The sample acquisition module is used to acquire multiple normal driving trajectories of vehicles on the target road segment.

[0079] The spatiotemporal alignment module is used to perform spatiotemporal alignment of multiple normal driving trajectories using a dynamic motion primitive model, thereby obtaining multiple aligned normal driving trajectories.

[0080] The probability modeling module is used to perform probability modeling based on multiple aligned normal driving trajectories using a Gaussian mixture model to obtain the probability density function.

[0081] The expected trajectory generation module is used to generate representative expected trajectories of vehicles traveling normally on the target road segment based on the probability density function and Gaussian mixture regression.

[0082] The deviation calculation module is used to determine the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory.

[0083] The determination module is used to determine that the vehicle under evaluation is in a suspected state of fatigued driving if the deviation is greater than the dynamic determination threshold.

[0084] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0085] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for assessing driver fatigue, characterized in that, include: Obtain multiple normal driving trajectories of vehicles on the target road segment; A dynamic motion primitive model is used to perform spatiotemporal alignment of multiple normal driving trajectories to obtain the aligned multiple normal driving trajectories; Based on the aligned multiple normal driving trajectories, a Gaussian mixture model is used for probability modeling to obtain the probability density function. Based on the probability density function, Gaussian mixture regression is used to generate representative expected trajectories of vehicles traveling normally on the target road segment; Determine the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory; If the deviation is greater than the dynamic judgment threshold, the vehicle under evaluation is determined to be in a state of suspected fatigue driving.

2. The fatigue driving assessment method according to claim 1, characterized in that, Obtain multiple normal driving trajectories of vehicles on the target road segment, specifically including: By using aerial video acquisition equipment deployed above the target road segment, the driving process of multiple vehicles passing through the target road segment is continuously captured, and the driving trajectory is extracted using target detection algorithm and multi-target tracking algorithm to obtain the original spatiotemporal trajectory sequence; By filtering driving trajectories under non-fatigue driving conditions from the original spatiotemporal trajectory sequence, multiple normal driving trajectories of vehicles on the target road segment are obtained.

3. The fatigue driving assessment method according to claim 1, characterized in that, A dynamic motion primitive model is used to perform spatiotemporal alignment of multiple normal driving trajectories, resulting in aligned normal driving trajectories, specifically including: One trajectory is randomly selected from multiple normal driving trajectories as a reference trajectory; The reference trajectory is time-normalized to obtain the time-normalized reference trajectory; Each normal driving trajectory other than the reference trajectory among multiple normal driving trajectories is defined as the trajectory to be aligned, and a dynamic motion primitive model is established for the x and y components of each trajectory to be aligned. By using the dynamic motion primitive model of the x and y components of each trajectory to be aligned, the starting point and ending point of each trajectory to be aligned are adjusted to be consistent with the time-normalized reference trajectory, thereby generating multiple aligned normal driving trajectories.

4. The fatigue driving assessment method according to claim 1, characterized in that, Based on the aligned multiple normal driving trajectories, a Gaussian mixture model is used for probability modeling to obtain the probability density function, which specifically includes: The aligned multiple normal driving trajectories are stacked according to time steps to form a training dataset: ;in, For the training dataset, The sampling point number, To align multiple normal driving trajectories at sampling points The coordinates below, This is the sequence number of the normal driving trajectory. The number of normal driving trajectories. The maximum sampling point number; Use a Gaussian mixture model to analyze the joint variables. The distribution is modeled to obtain the probability density function; where, Represents the x-coordinate, Represents the vertical axis.

5. The fatigue driving assessment method according to claim 1 or 4, characterized in that, The expression for the probability density function is: ； In the formula, Let be the probability density function. The sampling point number, Represents the x-coordinate, Represents the ordinate, For the first The weighting parameters of each Gaussian component. and For the first The mean vector and covariance matrix of each Gaussian component. It is a one-dimensional Gaussian probability density function. The number of Gaussian components.

6. The fatigue driving assessment method according to claim 1, characterized in that, Based on the probability density function, Gaussian mixture regression is used to generate representative expected trajectories of vehicles traveling normally on the target road segment, specifically including: At each sampling point, the expected trajectory is generated using Gaussian mixture regression based on the probability density function: In the formula, Sampling points The expected trajectory below, Let x be the x-coordinate of the desired trajectory. Let y be the ordinate of the desired trajectory. The number of Gaussian components. Sampling points Down( x , y The conditional mean of ) For posterior weights, , For the first The weighting parameters of each Gaussian component. For the first Each Gaussian component at the sampling point The mean of the time components below, For the first Each Gaussian component at the sampling point The time variance below, It is a one-dimensional Gaussian probability density function. For the first The weighting parameters of each Gaussian component. For the first Each Gaussian component at the sampling point The mean of the time components below, For the first Each Gaussian component at the sampling point The time variance below; Based on the expected trajectories from all sampling points, the representative expected trajectory for normal vehicle travel on the target road segment is obtained as follows: In the formula, As a representative expected trajectory, This is the sequence number of the largest sampling point.

7. The fatigue driving assessment method according to claim 1, characterized in that, Determining the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory specifically includes: A dynamic motion primitive model is used to spatiotemporally align the real-time driving trajectory of the vehicle under evaluation with the representative expected trajectory to obtain the aligned real-time driving trajectory. Using formula Calculate the spatial deviation between the aligned real-time driving trajectory and the representative expected trajectory; where, For spatial bias, As the first weight, As the second weight, The maximum sampling point number, The coordinates of the aligned real-time driving trajectory. The coordinates of the representative expected trajectory; Using formula Calculate the speed consistency deviation between the aligned real-time driving trajectory and the representative expected trajectory; where, For speed consistency deviation, The maximum sampling point number; The instantaneous velocity at sampling point s of the aligned real-time driving trajectory. , To determine the position of the aligned real-time driving trajectory at sampling point s+1, The position of the aligned real-time driving trajectory at sampling point s; The velocity of the representative expected trajectory at sampling point s, , The representative expected trajectory at the sampling points The location, The representative expected trajectory at the sampling points The location; Based on the spatial deviation and the velocity consistency deviation, using the formula The comprehensive deviation is determined and used as the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory; where, To account for the overall deviation, and For weight parameters, This represents the Fraser distance between the aligned real-time driving trajectory and the representative expected trajectory.

8. The fatigue driving assessment method according to claim 1, characterized in that, Also includes: When the vehicle under evaluation is determined to be in a state of suspected fatigued driving, an early warning is triggered; Calculate the daily warning rate for the target road section; If the warning rate exceeds p times the average warning rate for n consecutive days, the dynamic judgment threshold will be adjusted, or the representative expected trajectory will be redefined.

9. A fatigue driving assessment device, characterized in that, include: Aerial video acquisition equipment and processor; The aerial video acquisition device is used to acquire multiple normal driving trajectories of vehicles on the target road segment, as well as the real-time driving trajectory of the vehicles to be evaluated on the target road segment. The processor is used to perform spatiotemporal alignment of multiple normal driving trajectories using a dynamic motion primitive model to obtain multiple aligned normal driving trajectories; based on the multiple aligned normal driving trajectories, a Gaussian mixture model is used for probability modeling to obtain the probability density function; Based on the probability density function, Gaussian mixture regression is used to generate a representative expected trajectory of a vehicle driving normally on the target road segment; the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory is determined; if the deviation is greater than the dynamic judgment threshold, the vehicle to be evaluated is determined to be in a suspected state of fatigued driving.

10. A fatigue driving assessment system, characterized in that, include: The sample acquisition module is used to acquire multiple normal driving trajectories of vehicles on the target road segment; The spatiotemporal alignment module is used to perform spatiotemporal alignment of multiple normal driving trajectories using a dynamic motion primitive model, thereby obtaining multiple aligned normal driving trajectories. The probability modeling module is used to perform probability modeling based on multiple aligned normal driving trajectories using a Gaussian mixture model to obtain the probability density function; The expected trajectory generation module is used to generate a representative expected trajectory of vehicles traveling normally on the target road segment based on the probability density function and Gaussian mixture regression. The deviation calculation module is used to determine the deviation between the real-time driving trajectory of the vehicle to be evaluated on the target road segment and the representative expected trajectory. The determination module is used to determine that the vehicle under evaluation is in a suspected state of fatigued driving if the deviation is greater than the dynamic determination threshold.

Images