Facial feature recognition-based safe driving early warning system and method

Abstract

The invention belongs to the technical field of facial recognition, and particularly relates to a safe driving early warning system and method based on facial feature recognition, and the method comprises the following steps: obtaining and building a driver three-dimensional model, a vehicle model and an external environment model, and generating a correlation model database according to the operation of the obtained models in a simulation system; according to the three model combination operation monitoring data in the association model database, obtaining influence parameters on the driver three-dimensional model in different states, and generating a dynamic data set; and constructing a static structure line and a dynamic structure line based on the driver three-dimensional model, associating the driver characteristic data obtained during illumination compensation, and respectively comparing model data with dynamic data. According to the invention, the method can improve the recognition robustness through the three-dimensional dynamic model, reduces the environment interference through the mixed illumination compensation algorithm, achieves the precise risk assessment through multi-mode melting, and reduces the interference of a driver caused by false alarm.

Description

Technical Field

[0001] The present invention belongs to the technical field of face recognition, and particularly relates to a safe driving warning system and method based on face feature recognition. Background Art

[0002] A safe driving warning system and method based on face feature recognition is a system that uses computer vision and artificial intelligence technologies to judge the driving state of a driver by real-time monitoring of the driver's face features (such as eyes, mouth, head posture, etc.), and issues an alarm when detecting fatigue, distraction or other dangerous behaviors. This system aims to improve driving safety and reduce traffic accidents caused by human factors.

[0003] Problems existing in the prior art:

[0004] In the existing face recognition process, especially in the driving state, the accuracy of face feature recognition in complex environments is insufficient, such as changes in lighting, occlusion, etc., and most human face static models cannot adapt to dynamic driving scenarios, such as muscle deformation, environmental interaction, etc. Therefore, during driving, the recognition of the driver's state is not accurate enough, and inaccurate recognition during the recognition process is likely to cause false alarms of information, affecting the driver in a normal driving state. Summary of the Invention

[0005] The purpose of the present invention is to provide a safe driving warning system and method based on face feature recognition, which can improve the recognition robustness through a three-dimensional dynamic model, reduce environmental interference through a hybrid light compensation algorithm, and achieve accurate risk assessment through multi-modal fusion, reducing the interference to the driver caused by false alarms.

[0006] The technical solutions adopted by the present invention are specifically as follows:

[0007] A safe driving warning method based on face feature recognition includes the following steps:

[0008] Obtain and establish a driver three-dimensional model, a vehicle model, and an external environment model, and generate an associated model database by running in a simulation system according to the obtained models;

[0009] Run the monitoring data according to the combination of the three models in the associated model database, obtain the influence parameters on the driver three-dimensional model in different states, and generate a dynamic data set;

[0010] Construct static and dynamic contour lines based on the driver three-dimensional model, and associate light compensation;

[0011] Attach the static contour line to the driver three-dimensional model, associate light compensation, and generate model data and record it in the associated model database when the monitoring parameters in the simulation system change, generating a control group for the corresponding dynamic data set;

[0012] Attach the dynamic wireframe to the driver's three-dimensional model, associate with light compensation, and compare the monitoring parameters in the simulation system with the control group to generate dynamic data and enter it into the dynamic data set;

[0013] The driver feature data obtained in real time is respectively compared with the model data and the dynamic data;

[0014] Determine whether it coincides with the model data or the dynamic data;

[0015] Obtain the deviation according to the comparison between the model data and the dynamic data, determine the risk level according to the deviation, and verify the driver's state;

[0016] Determine the risk level according to the obtained driver's state, and trigger the corresponding level warning system.

[0017] The driver's three-dimensional model, vehicle model, and external environment model respectively include a static model and a dynamic model. The static model is the established reference model, and the dynamic model is established according to the obtained model data. The dynamic model includes any independent motion model of the driver's three-dimensional model, vehicle model, and external environment model, as well as the relative motion model of any combination;

[0018] The associated model database is a collection of the static model and the dynamic model, as well as a collection of the associated data of any model corresponding to the driver's three-dimensional model, vehicle model, and external environment model.

[0019] The dynamic data set at least includes a collection of the associated data of the driver's three-dimensional model. The associated data includes, but is not limited to, the static and dynamic position information of the driver's three-dimensional model, the relative coordinate position information between multiple models, and the light information;

[0020] Among them, the static wireframe of the driver's three-dimensional model includes, but is not limited to, the reference contour lines of the eye sockets, lips, chin, and side face based on the face;

[0021] The dynamic wireframe includes, but is not limited to, the dynamic contour lines of the lip opening and closing, eye socket contraction, wrinkle generation, and side face contraction based on the face.

[0022] The driver feature data changes according to time, integrates multi-modal facial features, judges the change trend between the driver's static contour and dynamic contour, and determines whether the relationship between the change trend and the model data and dynamic data exceeds the threshold during the change process. According to the deviation size exceeding the threshold, the driver's state and risk level are judged.

[0023] A safe driving warning system based on facial feature recognition, including:

[0024] The image acquisition and preprocessing module acquires images of the driver's face and the area around the face through in-vehicle image acquisition devices, and at least one infrared camera and a hybrid light compensation module are provided.

[0025] The three-dimensional dynamic modeling module generates a driver's facial muscle dynamics model and extracts static and dynamic contour line features.

[0026] The multi-modal feature fusion module integrates facial expressions, physiological parameters, and environmental sensor data to measure the deviation between the dynamic contour line and the reference model.

[0027] The risk assessment and warning module evaluates the risk level based on an incremental learning model of the dynamic data set and triggers voice, vibration, or emergency braking responses according to the classification threshold.

[0028] The dynamic data set stores and runs a dynamic data system for storing dynamic data of personalized driving features and data parameters of any combination of the driver's three-dimensional model, vehicle model, and external environment model.

[0029] The image acquisition and preprocessing module further includes at least two cameras with cross perspectives, and the hybrid light compensation module uses the Retinex algorithm combined with CAN bus data to achieve hybrid light compensation.

[0030] The Hausdorff distance is used to measure the deviation between the dynamic contour line and the reference model.

[0031] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the method described in any one of the foregoing is implemented.

[0032] According to another aspect of the embodiments of the present invention, there is also provided a computer program product including a computer program, and when the computer program is executed by a processor, the method described in any one of the foregoing is implemented.

[0033] The technical effects achieved by the present invention are as follows:

[0034] In the present invention, by dynamically adjusting parameters, the facial recognition error caused by the light change inside and outside the vehicle is eliminated, the extraction accuracy of static / dynamic contour lines is improved, and combined with CAN bus data, the real-time optimization of light parameters is realized to ensure the robustness in complex driving scenarios.

[0035] The present invention constructs static and dynamic construction lines based on a driver's three-dimensional model, attaches them to the driver's three-dimensional model respectively, generates model data and dynamic data respectively, realizes personalized simulation and comparison of different drivers' driving states through a control group generated by the two, uploads the corresponding data to the corresponding associated model database and dynamic data set, constructs characteristic data for the same driver, and determines the risk level by judging whether it coincides with the model data or dynamic data, obtaining the deviation according to the comparison between the model data and the dynamic data, verifying the driver's state, and thus formulating vehicle early warning, remote early warning, remote takeover, autonomous driving takeover, alarm, first aid early warning, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a schematic flowchart of a safety driving early warning method based on facial feature recognition according to the present invention;

[0037] Figure 2 is a quantitative flowchart of an embodiment in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] In order to make the objectives and advantages of the present invention clearer, the present invention will be specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific implementation manners of the present invention, and does not strictly limit the scope of protection specifically claimed by the present invention.

[0039] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0040] According to an embodiment of the present invention, a method embodiment of a safety driving early warning method based on facial feature recognition is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0041] As Figure 1 shown, a safety driving early warning method based on facial feature recognition includes the following steps:

[0042] S1. Obtain and establish a three-dimensional driver model, a vehicle model, and an external environment model, and generate an associated model database by running in a simulation system based on the obtained models.

[0043] S2. Monitor data by running combinations of the three models in the associated model database, obtain influence parameters on the three-dimensional driver model in different states, and generate a dynamic data set.

[0044] S3. Construct static and dynamic contour lines based on the three-dimensional driver model, and associate light compensation.

[0045] S4. Attach the static contour lines to the three-dimensional driver model, associate light compensation, and generate model data and enter it into the associated model database when the monitoring parameters in the simulation system change, generating a control group corresponding to the dynamic data set.

[0046] S5. Attach the dynamic contour lines to the three-dimensional driver model, associate light compensation, and compare the monitoring parameters in the simulation system with the control group to generate dynamic data and enter it into the dynamic data set.

[0047] S6. Obtain the driver's characteristic data in real time, and compare the model data with the dynamic data respectively.

[0048] S7. Determine whether it coincides with the model data or the dynamic data.

[0049] S8. Obtain the deviation based on the comparison between the model data and the dynamic data, determine the risk level according to the deviation, and verify the driver's state.

[0050] S9. Determine the risk level according to the obtained driver's state, and trigger the corresponding level warning system.

[0051] Furthermore, it should be noted that in steps S1 and S2, the simulation system establishes a three-dimensional space simulation model through a computer or a cloud computer system, as well as a scene simulation model during the vehicle operation process, and obtains an external environment model under different weather conditions through big data. The computer system includes at least one set of physical engines, especially a physical engine based on light. Through the established three-dimensional driver model, vehicle model, and external environment model, the change in the illumination intensity caused by light passing through each model, as well as the scattering and refraction of light, are realized. Especially in the vehicle model, it is necessary to further access the light transmittance and reflectance of the window glass to achieve a detailed monitoring of the light change in the simulation system. When gases such as carbon monoxide and carbon dioxide are generated inside the vehicle model, the influence of the gas concentration on the light is obtained to further calculate the light intensity.

[0052] Further, during the simulation process, various behavioral states of the driver's three-dimensional model in static and dynamic states are simulated to obtain states such as mental state and health state. A static state with a normal mental state as the benchmark is set, and states such as fatigue state and inattentiveness are set as super-threshold states in the static state. Through the driver's action capture, real-time generation of states is achieved to obtain the facial features of different drivers in different states. According to the facial features in the static state obtained in real time, it is judged whether it is close to or coincides with the super-threshold state. If it coincides, it is determined as dangerous, and the above states are recorded in the associated model database.

[0053] Furthermore, during the simulation process, when it is set to the dynamic state, a dynamic state with a normal mental state as the benchmark is set, and states such as fatigue state and inattentiveness are set as super-threshold states in the dynamic state. Through the driver's action simulation capture, real-time generation of states is achieved to obtain the facial features of different drivers in different states. According to the facial features in the static state obtained in real time, it is judged whether it is close to or coincides with the super-threshold state. If it coincides, it is determined as dangerous, and the above states are recorded in the dynamic dataset.

[0054] There are three model combination methods in the associated model database, that is, the driver's dynamic state includes but is not limited to: driving state, the driver's state relative to the external environment model, and the driver's state relative to the vehicle model.

[0055] Optionally, the driving state is the driving behavior state and mental state of the driver during driving;

[0056] Optionally, the driver's state relative to the external environment model is the state during the relative movement with the oncoming vehicle, and also includes the illumination of the oncoming vehicle's lights, the illumination of the rear vehicle's lights, the honking of the horn, etc.;

[0057] Optionally, the driver's state relative to the vehicle model is the non-driving action behavior in the driving state.

[0058] In the above steps, the devices used for simulation operation include action capture devices, cameras, infrared sensors, gas sensors, light sensors, sound sensors, etc.

[0059] In step S3, the static contour lines of the driver's three-dimensional model include but are not limited to the reference contour lines of the eye sockets, lips, chin, and side face based on the face, which are the static states with the reference;

[0060] The dynamic contour lines include but are not limited to the dynamic contour lines of the lip opening and closing, eye socket contraction, wrinkle generation, and side face contraction based on the face, which are the super-threshold states;

[0061] Obtaining the contour line is used to achieve the basic judgment of the above two states, and by comparing the driver's state obtained in real time, the two states are compared.

[0062] For example, by scanning the vehicle and the illumination conditions of various external environments obtained in real time, the state of the driver in the cab is simulated, such as simulating the impact of different lights on the driver, the impact of different environments on the driver, and the impact on the driver during different external risk judgments, so as to achieve the prediction of the driver's judgment and state, so as to achieve targeted prediction and early warning in different environments, and improve driving safety.

[0063] The above two states are respectively applied to the judgment basis of the driver's state in non-driving state and driving state. The contour line of the face uses dlib to detect 68 facial key points, generate the reference contour lines of the eye sockets, lips, chin, and side face, predict the facial deformation in the next 300ms through LSTM, combine the optical flow method to track the wrinkle generation rate, and further fuse the ToF depth data and RGB images to establish a three-dimensional facial model with millimeter-level accuracy.

[0064] In step S3, the illumination compensation uses a hybrid illumination compensation module. The hybrid illumination compensation module uses the Retinex algorithm combined with CAN bus data to achieve hybrid illumination compensation, and uses an improved Retinex algorithm to eliminate the influence of ambient light. The formula is:

[0065] Icorrected = log(Iraw) - γ · log(Iillumination);

[0066] Among them, Iraw is the pixel value of the original image, which is the facial image data of the driver directly collected by the on-vehicle camera;

[0067] Iillumination is the ambient light intensity, which is calculated from the data such as the headlight state and rain sensor obtained through the vehicle CAN bus;

[0068] γ is the dynamic adjustment parameter of illumination compensation, which is adaptively adjusted according to vehicle environment data (such as vehicle speed, windshield light transmittance), and the range is usually 0.8 - 1.2;

[0069] Icorrected is the pixel value of the corrected image, which is the facial feature data after eliminating the interference of ambient light and is used for subsequent three-dimensional modeling and risk assessment;

[0070] According to the above formula: by dynamically adjusting the parameters, the facial recognition error caused by the change of light inside and outside the vehicle (such as backlight, night) is eliminated, the extraction accuracy of static / dynamic contour lines is improved, and combined with CAN bus data (such as rain sensor, headlight state), the real-time optimization of illumination parameters is realized to ensure the robustness in complex driving scenarios.

[0071] When in the fatigue detection state: The corrected image can more accurately capture dynamic contour changes such as orbital contraction and wrinkle generation;

[0072] When in the disease warning state: Through the analysis of the facial temperature gradient under stable lighting conditions, assist in judging the risk of sudden diseases.

[0073] In steps S4 and S5, by constructing static and dynamic contour lines based on the driver's three-dimensional model in step S3 and attaching them to the driver's three-dimensional model respectively, model data and dynamic data are generated respectively. Through the control group generated by the two, personalized simulation and comparison of different drivers' driving states are realized, and the corresponding data is uploaded to the corresponding associated model database and dynamic data set to construct characteristic data for the same driver. By judging whether it coincides with the model data or dynamic data, the deviation is obtained through the comparison of the model data and dynamic data, and the risk level is determined according to the deviation to verify the driver's state, so as to formulate vehicle warnings, remote warnings, remote takeover, autonomous driving takeover, alarms, first aid warnings, etc.

[0074] As an alternative embodiment, the driver's three-dimensional model, vehicle model, and external environment model each include a static model and a dynamic model. The static model is a reference model established, and the dynamic model is established according to the obtained model data. The dynamic model includes any independent motion model of the driver's three-dimensional model, vehicle model, and external environment model, as well as a relative motion model of any combination;

[0075] The associated model database is a collection of static and dynamic models, as well as a collection of associated data of any model corresponding to the driver's three-dimensional model, vehicle model, and external environment model.

[0076] Among them, the independent motion model preferably represents the dynamic state between the driver's three-dimensional model and any vehicle model and external environment model, that is, when the driver is in a dynamic driving state, the relationship between his own actions and the vehicle model and / or external environment model, and the relationship includes but is not limited to driving behavior, lighting, sound, harmful gas concentration, etc. Set the above model as one of the independent motion models, and there are also independent motion models centered on the vehicle model or external environment model respectively, formed by the combination with other models.

[0077] The reference model established by the static model is used to compare the behavioral actions of the dynamic model. The static model needs to establish corresponding static models according to different drivers, as well as dynamic models corresponding to the static models, to form a set of associated data and enter it into the associated model database. By establishing the associated model database, it is possible to establish corresponding safe driving warning systems for different drivers, so as to realize the lightweight of the database, store the associated model data of multiple corresponding drivers in the central server or cloud, reduce the operation pressure of the in-vehicle safe driving warning system, and improve the performance.

[0078] As an alternative embodiment, please refer to Figure 2 , the dynamic data set at least includes a set of associated data of the driver's three-dimensional model. The associated data includes, but is not limited to, the static and dynamic position information of the driver's three-dimensional model, the relative coordinate position information between multiple models, and the lighting information.

[0079] Optionally, the dynamic data set also includes, but is not limited to, key physiological parameters such as heart rate, body temperature (especially facial temperature and the change of facial temperature), blood oxygen, medical history, sleep quality, CO / CO2 concentration, etc. The multi-modal feature fusion module fuses the data in the dynamic data set to realize the collaborative monitoring of the driver's state through the association of multiple data, so as to improve the monitoring accuracy. The acquisition methods of the above multiple data include using watches / wristbands, etc.

[0080] Furthermore, the associated data is the position information of the driver relative to the vehicle model, such as the acquisition of the driver's driving posture, whether it conforms to the optimal driving state of the driver, and the judgment and correction of the driver's sitting posture for the field of vision.

[0081] Even further, the associated data also includes the relationship between the vehicle model and the external environment model. By monitoring the position change between the vehicle driving state and the external environment, the vehicle driving path can be monitored.

[0082] As an alternative embodiment, the driver characteristic data changes according to time, integrates multi-modal facial features, judges the change trend between the static contour and the dynamic contour of the driver, and judges the driver's state and risk level according to whether the relationship between the change trend and the model data and the dynamic data exceeds the threshold during the change process and the size of the deviation exceeding the threshold.

[0083] Optionally, the risk levels include low risk level, medium risk level and high risk level. The judgment method is expressed by the following formula:

[0084]

[0085] D is the deviation between the dynamic contour line and the reference model, T high and Tlow is a preset high and low risk threshold.

[0086] Through the above threshold evaluation model, the efficiency of calculation is realized, and dynamic data sets are used for training to dynamically optimize the threshold to ensure classification accuracy.

[0087] For example:

[0088]

[0089]

[0090] Normalize features with different dimensions to a unified range:

[0091]

[0092] For example, for the determination of PERCLOS fatigue, the judgment formula is used:

[0093]

[0094] Perform weighted fusion of multiple variables:

[0095]

[0096] Among them, R is the comprehensive risk value, w i is the feature weight (such as fatigue weight 0.4, distraction weight 0.3), f i is the normalized feature value.

[0097] The risk level is thus divided:

[0098]

[0099] Use incremental learning to dynamically update the threshold based on historical data, such as using the sliding window method:

[0100] T new = α·T old (1 - α)·mean of new data;

[0101] α represents the learning rate.

[0102] A safe driving warning system based on facial feature recognition, comprising:

[0103] An image acquisition and preprocessing module, which acquires the driver's face and the images around the face through in-vehicle image acquisition devices, and at least one infrared camera and a hybrid light compensation module are set;

[0104] A three-dimensional dynamic modeling module, which uses structured light and ToF technology to generate a driver's facial muscle dynamics model and extracts static contour and dynamic contour features;

[0105] A multi-modal feature fusion module that integrates facial expressions, physiological parameters, and environmental sensor data to measure the deviation between the dynamic contour and the reference model;

[0106] A risk assessment and early warning module that evaluates the risk level based on an incremental learning model of the dynamic data set and triggers voice, vibration, or emergency braking responses according to the classification threshold.

[0107] As an alternative embodiment, the dynamic data set stores and runs a dynamic data system for storing the dynamic data of personalized driving features and the data parameters of any combination of the driver's three-dimensional model, vehicle model, and external environment model.

[0108] As an alternative embodiment, the image acquisition and preprocessing module further includes at least two cameras with cross-angled views for obtaining the status images of two different sides of the driver.

[0109] As an alternative embodiment, the Hausdorff distance is used to measure the deviation between the dynamic contour and the reference model. The Hausdorff distance is a commonly used shape matching metric in computer vision that can effectively quantify the deviation between the dynamic contour lines (such as lip opening and closing, orbital contraction) and the reference model.

[0110] It should be noted that the above-mentioned modules can be implemented by software or hardware. For the latter, it can be achieved in the following ways, but not limited to: all the above modules are located in the same processor; or, the above-mentioned modules are separately located in different processors in any combination form.

[0111] According to another aspect of the embodiments of the present invention, an electronic device is further provided. The electronic device includes a memory and a processor; the memory is used to store programs; the processor executes the programs to implement the method in any one of the foregoing.

[0112] According to another aspect of the embodiments of the present invention, a computer-readable storage medium is further provided. The storage medium stores a computer program, and when the computer program is executed by a processor, the method in any one of the foregoing is implemented.

[0113] According to another aspect of the embodiments of the present invention, a computer program product is further provided, including a computer program, and when the computer program is executed by a processor, the method in any one of the foregoing is implemented.

[0114] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. The structures, devices, and operation methods not specifically described and explained in the present invention, unless otherwise specified and limited, are implemented according to the conventional means in the art.

Claims

1. A safe driving warning method based on facial feature recognition, characterized in that: The steps include: Acquire and establish a three-dimensional model of the driver, a vehicle model, and an external environment model, and generate a related model database by running the acquired models in a simulation system; According to the combined operation monitoring data of the three models in the associated model database, the influencing parameters of the driver's three-dimensional model under different conditions are obtained, and a dynamic data set is generated; Construct static and dynamic alignments based on the driver's 3D model and associate them with illumination compensation; Attach the static line to the three-dimensional model of the driver, associate the illumination compensation, and monitor the parameter changes in the simulation system to generate model data and enter it into the associated model database to generate a control group corresponding to the dynamic data set; Attach the dynamic configuration line to the driver's 3D model, associate the illumination compensation, and compare the monitoring parameters in the simulation system with the control group to generate dynamic data and enter the dynamic data set; Driver characteristic data acquired in real time, respectively compared with model data and dynamic data; Determine whether it coincides with model data or dynamic data; Obtain deviations by comparing model data with dynamic data, determine risk levels based on the deviations, and verify driver status; The risk level is determined based on the acquired driver status, and the corresponding level warning system is triggered.

2. The method for safe driving warning based on facial feature recognition according to claim 1, characterized in that: The driver three-dimensional model, vehicle model, and external environment model respectively include a static model and a dynamic model, wherein the static model is an established reference model, and a dynamic model is established according to the acquired model data, wherein the dynamic model includes any independent motion model of the driver three-dimensional model, vehicle model, and external environment model, and any combined relative motion model; The associated model database is a collection of static models and dynamic models, as well as a collection of model associated data corresponding to any one of the driver three-dimensional model, vehicle model, and external environment model.

3. The method for safe driving warning based on facial feature recognition according to claim 1, characterized in that: The dynamic data set at least includes a set of driver 3D model associated data, including but not limited to static and dynamic position information of the driver 3D model, relative coordinate position information between multiple models, and lighting information; The static contours of the three-dimensional model of the driver include but are not limited to the reference contours based on the eye sockets, lips, chin, and side face of the face; The dynamic contour lines include, but are not limited to, dynamic contour lines based on the opening and closing of lips, contraction of eye sockets, generation of wrinkles, and contraction of side faces.

4. The method for safe driving warning based on facial feature recognition according to claim 1, characterized in that: The driver characteristic data changes over time, integrates multimodal facial features, determines the changing trend between the driver's static profile and dynamic profile, determines whether the relationship between the changing trend and the model data and dynamic data exceeds a threshold during the changing process, and determines the driver's status and risk level based on the size of the deviation exceeding the threshold.

5. A safe driving warning system based on facial feature recognition, running any method in claims 1 to 4, characterized in that: include: An image acquisition and preprocessing module, which acquires images of the driver's face and surrounding areas through an on-board image acquisition device, and is provided with at least one infrared camera and a mixed illumination compensation module; A three-dimensional dynamic modeling module, which generates a facial muscle dynamics model of the driver and extracts static and dynamic line features; Multimodal feature fusion module, which integrates facial expressions, physiological parameters and environmental sensor data to measure the deviation of dynamic configuration from the baseline model; The risk assessment and warning module evaluates the risk level based on an incremental learning model of a dynamic data set and triggers voice, vibration or emergency braking responses according to the classification threshold.

6. The facial feature recognition-based safe driving warning system according to claim 5, characterized in that: The dynamic data set stores and runs a dynamic data system for storing dynamic data of personalized driving characteristics, as well as data parameters of any combination of a driver's three-dimensional model, a vehicle model, and an external environment model.

7. The facial feature recognition-based safe driving warning system according to claim 5, characterized in that: The image acquisition and preprocessing module further comprises at least two cameras with intersecting viewing angles, and the mixed illumination compensation module uses a Retinex algorithm in combination with CAN bus data to achieve mixed illumination compensation.

8. The facial feature recognition-based safe driving warning system according to claim 5, characterized in that: The Hausdorff distance is used to measure the deviation of the dynamic configuration from the benchmark model.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images