Driving risk identification model training method, driving risk identification method and device

Abstract

The invention is suitable for the technical field of artificial intelligence, and provides a driving risk identification model training method and a driving risk identification method and device, and the method comprises the steps: obtaining to-be-processed data of a plurality of drivers; determining a corresponding event according to the to-be-processed data; performing textualization processing on the identifier of the driver and the event corresponding to the driver to obtain a text processing result; a to-be-trained driving risk identification model is trained according to the text processing result, a trained driving risk identification model is obtained, the to-be-trained driving risk identification model is a large language model, and the trained driving risk identification model is used for outputting driving risks. The model trained in the mode is more accurate.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and in particular relates to a driving risk identification model training method, a driving risk identification method, a device, a vehicle-mounted device, and a computer-readable storage medium. Background Art

[0002] With the increase in social traffic flow, the importance of road safety has become increasingly prominent, and the driver's driving behavior is an important factor affecting road safety.

[0003] At present, the driving risk of the driver is usually identified through a recommendation system, and then the driver is reminded according to the identified driving risk. Specifically, the recommendation system extracts features related to driving risk from the current driving data, and then predicts the corresponding driving risk according to the extracted features.

[0004] However, when the above method is used to predict driving risks, the accuracy of the prediction results obtained is low. SUMMARY OF THE INVENTION Technical issues

[0005] One of the purposes of the embodiments of the present application is to provide a driving risk identification model training method, which can solve the problem of low accuracy of driving risks predicted by existing models. Technical Solutions

[0006] The technical solution adopted in the embodiment of the present application is:

[0007] In a first aspect, an embodiment of the present application provides a driving risk identification model training method, comprising:

[0008] Acquiring data to be processed of a plurality of drivers, wherein the data to be processed includes vehicle data and / or non-vehicle data;

[0009] Determine a corresponding event according to the data to be processed;

[0010] Performing text processing on the driver's identification and the event corresponding to the driver to obtain a text processing result;

[0011] The driving risk identification model to be trained is trained according to the text processing result to obtain a trained driving risk identification model, wherein the driving risk identification model to be trained is a large language model, and the trained driving risk identification model is used to output driving risks.

[0012] In a second aspect, a driving risk identification method is provided, comprising:

[0013] Acquiring driver data to be processed, where the data to be processed includes vehicle data and / or non-vehicle data;

[0014] Determine a corresponding event according to the data to be processed to obtain a real-time event;

[0015] Acquire historical events and historical alarm information, wherein the historical events are events determined before the real-time events are determined;

[0016] Performing text processing on the real-time events, the historical events and the historical alarm information to obtain a text to be detected;

[0017] The text to be detected is used as the trained driving risk identification model as described in the first aspect to obtain the driving risk output by the trained driving risk identification model.

[0018] In a third aspect, a driving risk identification device is provided, comprising:

[0019] A to-be-processed data acquisition module, used to acquire the driver's to-be-processed data, wherein the to-be-processed data includes vehicle data and / or non-vehicle data;

[0020] A real-time event determination module, used to determine a corresponding event according to the data to be processed to obtain a real-time event;

[0021] A historical event acquisition module, used to acquire historical events and historical alarm information, wherein the historical events are events determined before the real-time events are determined;

[0022] A module for determining text to be detected, used for processing the real-time event, the historical event and the historical alarm information into text form to obtain the text to be detected;

[0023] The driving risk determination module is used to use the text to be detected as the trained driving risk identification model as described in the first aspect to obtain the driving risk output by the trained driving risk identification model.

[0024] In a fourth aspect, a vehicle-mounted device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the computer program.

[0025] In a fifth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0026] In a sixth aspect, an embodiment of the present application provides a computer program product. When the computer program product runs on a charging management device, the charging management device executes the method described in the first aspect above. Beneficial Effects

[0027] The beneficial effect of the embodiment of the present application is that: in the embodiment of the present application, since the data to be processed includes vehicle data and / or non-vehicle data, and the vehicle data and / or non-vehicle data will affect the driving risk, therefore, determining the event corresponding to the data to be processed is conducive to obtaining information related to the driving risk, so that after the driving risk identification model is trained according to the text-processed event and the driver's identification, it is conducive to improving the accuracy of the driving risk output by the driving risk identification model. In addition, since the driving risk identification model to be trained is a large language model, and the large language model has logical reasoning ability, therefore, after the driving risk identification model to be trained is trained according to the text processing result, the trained driving risk identification model can recognize the logical relationship of the input text and output the corresponding driving risk according to the recognition result. That is, since the trained driving risk identification model obtains the driving risk not only according to the input text, but also according to the logical relationship of the input text obtained by its reasoning, the accuracy of the obtained driving risk is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or exemplary technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 It is a flowchart of a driving risk identification model training method provided by an embodiment of the present application;

[0030] Figure 2 It is a flowchart of a driving risk identification method provided by an embodiment of the present application;

[0031] Figure 3 This is a structural block diagram of a driving risk identification device provided by an embodiment of the present application;

[0032] Figure 4 It is a structural schematic diagram of a vehicle-mounted device provided in one embodiment of the present application. Embodiments of the present invention

[0033] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0034] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0035] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0036] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0037] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the phrases "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. appearing in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways.

[0038] With the development of economy, there are more and more vehicles on the road, and the increase of vehicles will increase the probability of traffic accidents. During the driving process, experienced drivers usually have a lower probability of traffic accidents, because experienced drivers usually predict driving risks.

[0039] When relying on the driver to predict driving risks, there may be a situation where the driving risks are not predicted in time. For example, when the driver is not paying attention during the driving process, it is difficult to predict the driving risks in time. In order to improve the timeliness of the driver's knowledge of driving risks, the driving risks can be predicted through the recommendation system, and then the driver can be reminded based on the predicted driving risks. Since the recommendation system predicts driving risks by extracting features related to driving risks from the acquired driving data, the driving risks obtained by the recommendation system are only related to the driving data itself. However, since the driving data itself can reflect limited information, when the driving risks are predicted only based on the features extracted from the driving data itself, the accuracy of the driving risks obtained is low.

[0040] In order to improve the accuracy of the obtained driving risk, an embodiment of the present application provides a driving risk identification model training method.

[0041] In this method, the corresponding event is determined based on the acquired data to be processed, and the event and the driver's identification are text-processed. Then, the driving risk identification model to be trained constructed according to the language large model is trained based on the obtained text-processing results.

[0042] The driving risk identification model training method provided in the embodiment of the present application is described below with reference to the accompanying drawings.

[0043] Figure 1 A flow chart of a driving risk identification model training method provided by an embodiment of the present application is shown, and the method can be applied to electronic devices, as described in detail as follows:

[0044] S11, obtaining data to be processed of multiple drivers, where the data to be processed includes vehicle data and / or non-vehicle data.

[0045] Among them, each driver may correspond to one or more groups of data to be processed, and each group of data to be processed represents the relevant data of a driver in a driving process, which includes one or more vehicle data, and / or, includes one or more non-vehicle data. For example, assuming that driver A drives vehicle 1 in the first period and obtains a group of data to be processed 1, drives vehicle 2 in the second period and obtains a group of data to be processed 2, and drives vehicle 3 in the third period and obtains another group of data to be processed 3, then driver A corresponds to three groups of data to be processed in the first period, the second period and the third period.

[0046] In this application, vehicle data refers to data generated by the vehicle itself, such as the vehicle's speed, acceleration, etc., and non-vehicle data refers to data not generated by the vehicle itself, such as video data and audio data collected by additional cameras installed on the vehicle.

[0047] S12: Determine a corresponding event according to the data to be processed.

[0048] Specifically, corresponding events are determined for each group of data to be processed. The number of events may be equal to 1 or may be greater than 1. These events are all associated with the driver.

[0049] In the embodiment of the present application, the judgment conditions corresponding to different events are predetermined. After obtaining the data to be processed, it is analyzed whether the data to be processed matches a predetermined judgment condition. If so, it is determined that the event corresponding to the data to be processed is the event corresponding to the matching judgment condition.

[0050] S13, performing text processing on the identification of the driver and the event corresponding to the driver to obtain a text processing result.

[0051] The text processing here includes text vectorization, which is used to convert text into numerical features so that the subsequent driving risk identification model to be trained can process it.

[0052] Of course, text processing includes not only text vectorization, but also data cleaning, text standardization and other processing, which will not be elaborated here.

[0053] S14, training the driving risk identification model to be trained according to the above text processing results to obtain a trained driving risk identification model, wherein the above driving risk identification model to be trained is a large language model, and the above trained driving risk identification model is used to output driving risks.

[0054] Specifically, the text processing result is used as the input of the driving risk identification model to be trained to obtain the driving risk output by the driving risk identification model to be trained, and the loss value of the driving risk identification model to be trained is calculated according to the output driving risk. The parameters of the driving risk identification model to be trained are adjusted according to the loss value, and the above steps are repeated until the calculated loss value meets the iteration stop condition, and the iteration is stopped to obtain the trained driving risk identification model.

[0055] In the embodiment of the present application, since the data to be processed includes vehicle data and / or non-vehicle data, and the vehicle data and / or non-vehicle data can affect driving risk, determining the event corresponding to the data to be processed is conducive to obtaining information related to driving risk, so that after the driving risk identification model is trained according to the text-processed event and the driver's identification, it is conducive to improving the accuracy of the driving risk output by the driving risk identification model. Since the driving risk identification model to be trained is a large language model, and the large language model has logical reasoning ability, after the driving risk identification model to be trained is trained according to the text processing result, the trained driving risk identification model can recognize the logical relationship of the input text and output the corresponding driving risk according to the recognition result. That is, since the trained driving risk identification model obtains the driving risk not only according to the input text, but also according to the logical relationship of the input text obtained by its reasoning, the accuracy of the obtained driving risk is further improved.

[0056] In some embodiments, the data to be processed includes vehicle data, and the vehicle data includes acceleration and / or angular velocity, wherein the acceleration and angular velocity can be acquired from a six-axis sensor of the vehicle, and the S12 includes:

[0057] A1. When the data of the vehicle includes acceleration, calculate the acceleration change rate of the acceleration, and determine whether the vehicle has a sudden deceleration event or a sudden acceleration event based on the acceleration change rate.

[0058] The acceleration rate of change is the rate of change of acceleration over time. Assuming that acceleration is a function that changes over time, the acceleration rate of change can be obtained by taking the time derivative of the function.

[0059] In the embodiment of the present application, after the acceleration change rate is obtained according to the acceleration calculation, the acceleration change rate is compared with the preset rapid deceleration threshold value. If the acceleration change rate indicates a decrease in value greater than the rapid deceleration threshold value, it is determined that the vehicle has a rapid deceleration event. If the acceleration change rate is compared with the preset rapid acceleration threshold value, if the acceleration change rate indicates an increase in value greater than the rapid acceleration threshold value, it is determined that the vehicle has a rapid acceleration event. Optionally, the absolute value of the rapid acceleration threshold value and the absolute value of the rapid deceleration threshold value may be equal or unequal.

[0060] Optionally, considering that different drivers have different driving habits, when judging whether a vehicle has a sudden acceleration event (or a sudden deceleration event), the driver's historical vehicle acceleration change rate and the corresponding traffic accident occurrence situation can be obtained, and the currently calculated acceleration change rate and the acquired historical acceleration change rate and the corresponding traffic accident occurrence situation are combined to judge whether the vehicle has a sudden acceleration event (or a sudden deceleration event). For example, when the currently calculated acceleration change rate is greater than the sudden acceleration threshold, but not greater than the average value of the historical acceleration change rate, and the driver does not have a traffic accident when driving at the historical average value of the acceleration change rate, it is also determined that there is no sudden acceleration event. Since the historical acceleration change rate and the corresponding traffic accident occurrence situation are combined, it is beneficial to improve the accuracy of the judgment result obtained.

[0061] A2. When the data of the vehicle includes angular velocity, calculate the angular velocity change rate of the angular velocity, and determine whether the vehicle has an abnormal posture event based on the angular velocity change rate.

[0062] The angular velocity change rate refers to the rate of change of the angular velocity over time, which is also called angular acceleration. Assuming that the angular velocity is a function that changes over time, the acceleration change rate can be obtained by taking the time derivative of the function.

[0063] In an embodiment of the present application, after the angular velocity change rate is obtained based on the angular velocity calculation, the angular velocity change rate is compared with a preset angular velocity change rate range. If the angular velocity change rate is not within the angular velocity change rate range, it is determined that the vehicle has an abnormal posture event.

[0064] Optionally, considering that different drivers have different driving habits, when judging whether the vehicle has an abnormal posture event, the driver's historical vehicle angular velocity change rate and the corresponding traffic accident situation can be obtained, and the currently calculated angular velocity change rate and the obtained historical angular velocity change rate and the corresponding traffic accident situation are combined to judge whether the vehicle has an abnormal posture event. Since the historical angular velocity change rate and the corresponding traffic accident situation are combined, it is beneficial to improve the accuracy of the obtained judgment result.

[0065] Optionally, after the acceleration (or angular velocity) is collected, the collected acceleration (or angular velocity) may be filtered to remove noise and improve the accuracy of the acceleration (or angular velocity) involved in subsequent calculations.

[0066] In some embodiments, the data to be processed includes non-vehicle data, and the non-vehicle data includes video data in front of the vehicle. In this case, the S12 includes:

[0067] Determine whether the vehicle has an abnormal lane crossing event based on video data in front of the vehicle, wherein when the vehicle has an abnormal lane crossing event, the vehicle crosses the lane alternately on both sides of the lane.

[0068] The video data in front of the vehicle can be collected by a front camera installed on the vehicle for collecting the video data in front of the vehicle.

[0069] The video data in front of the vehicle includes information about the environment in front of the vehicle. For example, when the vehicle is traveling on a road with lane lines, the video data in front of the vehicle includes information about the lane lines. Of course, the video data in front of the vehicle may also include information about the road surface boundary of the road, which is not limited here.

[0070] In the embodiment of the present application, whether the vehicle has an abnormal event of crossing the lane line can be determined based on the relative position of the vehicle and the lane line. Specifically, the relative position of the vehicle and the lane line can be determined based on the position of the lane line in the video frame corresponding to the video data and the installation position of the front camera on the vehicle.

[0071] Optionally, the determining whether the vehicle has an abnormal event of crossing the line and controlling the vehicle according to the video data in front of the vehicle includes:

[0072] B1. Detect lane lines based on the video data in front of the vehicle.

[0073] Specifically, the presence of a lane line in the video data may be detected based on the color and / or features of the lane line (a lane line in a short distance is a straight line). For example, the position of the lane line may be detected by a pre-trained computer vision deep learning model.

[0074] B2. Calculating a position change of the vehicle relative to the lane line according to the position of the detected lane line in consecutive adjacent video frames corresponding to the video data.

[0075] Specifically, considering that the installation position of the front camera is unchanged relative to the vehicle, when the vehicle does not move, the position of the same lane line in the video frame captured by the front camera is unchanged, and when the vehicle moves, the position of the same lane line in the video frame captured by the front camera changes. Therefore, the relative position relationship between the vehicle and the lane line can be calculated based on the coordinate information of the lane line detected in the adjacent video frames and the internal and external parameters of the front camera, and the relative position relationship reflects the movement trajectory of the vehicle.

[0076] Optionally, considering that some abnormal values ​​may be obtained during the detection of lane lines, that is, the coordinate information of abnormal lane lines may be obtained, therefore, in order to improve the accuracy of the obtained position change of the vehicle relative to the lane line, the motion trajectory of the vehicle may be smoothed, that is, the relative position relationship between the vehicle and the lane line may be smoothed. The smoothing process here includes smoothing process using Kalman filtering technology and smoothing process using geometric methods (such as moving average, spline interpolation method), which are not limited here.

[0077] B3. When the change in the position of the vehicle relative to the lane line indicates that the vehicle crosses the lane line alternately on both sides of the lane line, it is determined that the vehicle has the abnormal lane control event.

[0078] Specifically, it is determined whether the vehicle has an abnormal event of crossing the lane line according to the change in the position of the vehicle relative to the lane line corresponding to the time sequence change. For example, assuming that at time t1, the vehicle is on the left side of lane line 1, and at time (t1+1), the vehicle is on the right side of lane line 1, at time (t1+2), the vehicle is on the left side of lane line 1, and at time (t1+3), the vehicle is on the right side of lane line 1, then it is determined that the vehicle has an abnormal event of crossing the lane line.

[0079] Optionally, it is possible to determine whether the vehicle has an abnormal lane crossing control event by determining whether the position change of the vehicle relative to the lane line (i.e., the vehicle's motion trajectory) is an "S"-shaped lane crossing change. That is, if it is determined that the vehicle's motion trajectory is an "S" shape, and the "S" shape is on the lane line, it is determined that the vehicle has an abnormal lane crossing control event.

[0080] Since the video data in front of the vehicle is analyzed for abnormal events of crossing the line and the existence of abnormal events will increase the probability of traffic accidents, that is, there is a certain driving risk. Therefore, in the embodiment of the present application, analyzing the abnormal events of crossing the line is beneficial to improving the accuracy of the driving risk identification model obtained through subsequent training.

[0081] In some embodiments, the data to be processed includes non-vehicle data, and the non-vehicle data includes video data in front of the vehicle. In this case, the S12 includes:

[0082] Determine whether the vehicle has an abnormal roadside vehicle control event based on the video data in front of the vehicle. When the above roadside vehicle control abnormal event occurs, the distance between the above vehicle and the road boundary exceeds the preset boundary threshold.

[0083] Specifically, whether there is a road boundary in the video data can be detected based on the difference between the road boundary and the non-road boundary of the lane (such as color, material, etc.). For example, the position of the road boundary can be detected by a pre-trained computer vision deep learning model.

[0084] In an embodiment of the present application, after the road boundary is detected, the relative position relationship between the vehicle and the road boundary can be calculated based on the coordinate information of the road boundary detected in the adjacent video frames corresponding to the video data and the internal and external parameters of the front camera of the vehicle. The distance between the vehicle and the road boundary is determined based on the relative position relationship, and the distance is compared with a preset distance threshold. If it is greater than the preset distance threshold, it indicates that there is an abnormal roadside vehicle control event.

[0085] Optionally, considering that some abnormal values ​​may be obtained during the detection of the road surface boundary, that is, the coordinate information of the abnormal road surface boundary is obtained, therefore, in order to improve the accuracy of the obtained position change of the vehicle relative to the road surface boundary, the relative position relationship between the vehicle and the road surface boundary can be smoothed. The smoothing process here includes smoothing process performed by using Kalman filtering technology, and also includes smoothing process performed by using geometric methods (such as moving average, spline interpolation method), which are not limited here.

[0086] Since the video data in front of the vehicle is analyzed for abnormal roadside control events, the existence of abnormal roadside control events will increase the probability of traffic accidents, that is, there is a certain driving risk. Therefore, in the embodiment of the present application, analyzing abnormal roadside control events is beneficial to improving the accuracy of the driving risk identification model obtained through subsequent training.

[0087] In some embodiments, the data to be processed includes non-vehicle data, and the non-vehicle data includes video data in front of the vehicle. In this case, the S12 includes:

[0088] Determine whether the vehicle ahead of the vehicle has an abnormal event of not decelerating based on the video data in front of the vehicle, wherein when the vehicle ahead of the vehicle has an abnormal event of not decelerating, the vehicle ahead of the vehicle has no deceleration behavior but has a collision event.

[0089] Specifically, after the lane line is detected in the video data, the lane line where the vehicle is located and the position of the vehicle in front of the vehicle and also in the lane line (assuming it is called the front vehicle) can be determined. For example, the position of the front vehicle in the video frame corresponding to the video data and the position of the lane line in the video frame corresponding to the video data can be detected according to the computer vision deep learning model, and then the actual position of the front vehicle is calculated according to the position of the front vehicle in the video frame corresponding to the video data and the internal and external parameters of the front camera. After obtaining the actual position, it can be determined whether the front vehicle has no deceleration behavior according to the change of the actual position of the front vehicle, or according to the change of the distance between the actual position of the front vehicle and the vehicle. When judging that the front vehicle has no deceleration behavior, it is also necessary to detect whether the front vehicle has a collision event, such as judging whether the front vehicle has a collision event by analyzing the motion trajectory of the front vehicle, or judging whether the front vehicle has a collision event by analyzing the distance between the front vehicle and other objects. If it is judged that the front vehicle has a collision event but no deceleration behavior, it is determined that the front vehicle of the vehicle has an abnormal event of not decelerating.

[0090] Optionally, the determining, based on the video data in front of the vehicle, whether there is an abnormal event of a vehicle preceding the vehicle not decelerating includes:

[0091] C1. Calculating the rate of change of the speed of the vehicle ahead of the vehicle based on the video data ahead of the vehicle.

[0092] Specifically, after calculating multiple actual positions of the front vehicle based on the position of the front vehicle in the video frame corresponding to the video data and the internal and external parameters of the front camera, for each actual position, calculate the difference between it and the adjacent actual position to obtain the actual position difference, calculate the speed of the front vehicle corresponding to the time length based on the actual position difference and the time length corresponding to the two actual positions for calculating the actual position difference, and then calculate the rate of change of the speed of the front vehicle based on the at least two calculated speeds of the front vehicle and the time length corresponding to the at least two speeds.

[0093] C2. judging whether the vehicle ahead of the vehicle has decelerated based on the rate of change of the speed of the vehicle ahead of the vehicle.

[0094] Specifically, when it is determined that the rate of change of the speed is less than 0, it is determined that the vehicle ahead of the vehicle has decelerated, otherwise, it is determined that the vehicle ahead of the vehicle has not decelerated.

[0095] C3. When it is determined that the vehicle preceding the vehicle does not have any deceleration behavior and a collision occurs, it is determined that the vehicle preceding the vehicle has the above-mentioned non-deceleration abnormal event.

[0096] Since the video data in front of the vehicle is analyzed for abnormal events of no deceleration, the existence of abnormal events of no deceleration will increase the probability of traffic accidents, that is, there is a certain driving risk. Therefore, in the embodiment of the present application, performing abnormal event analysis of no deceleration is beneficial to improving the accuracy of the driving risk identification model obtained by subsequent training.

[0097] In some embodiments, the data to be processed includes non-vehicle data, and the non-vehicle data includes video data in front of the vehicle. In this case, the S12 includes:

[0098] Determine whether the vehicle has an abnormal vehicle control event without yielding based on video data in front of the vehicle, wherein, when the vehicle has an abnormal vehicle control event without yielding, the vehicle in front of the vehicle has inserted into the lane where the vehicle is located, and there is a risk of collision between the vehicle and the vehicle in front of the vehicle but the vehicle has not made any evasive action.

[0099] Specifically, the position of the vehicle and the lane line in the video frame corresponding to the video data in front of the vehicle is detected based on the video data. When there is a vehicle in front of the vehicle and in a lane other than the same lane as the vehicle, it is determined based on the trajectory of the existing vehicle whether the existing vehicle has the behavior of inserting into the lane where the vehicle is located. If so, it is determined whether there is a risk of collision between the vehicle and the existing vehicle, and whether the vehicle has no evasive action when there is a risk of collision. The evasive action here includes at least one of the following: starting emergency braking to slow down or stop the vehicle, starting steering assistance to help the driver avoid obstacles more effectively, and starting the lane keeping assist function to stabilize the vehicle in the lane to prevent loss of control due to avoidance.

[0100] Optionally, the determining whether the vehicle has an abnormal vehicle control event of not yielding according to the video data in front of the vehicle includes:

[0101] D1. Determine a trajectory of a candidate inserting vehicle based on video data in front of the vehicle, wherein the candidate inserting vehicle is a vehicle in front of the vehicle and in a lane adjacent to the lane of the vehicle.

[0102] Specifically, when it is determined based on the video data in front of the vehicle that there is a vehicle in front of the vehicle and in a lane adjacent to the lane of the vehicle (i.e., the candidate inserted vehicle mentioned above), the actual position of the candidate inserted vehicle at different times is calculated based on the position of the candidate inserted vehicle in the video frame corresponding to the video data and the internal and external parameters of the front camera, and then the trajectory of the candidate inserted vehicle is generated based on the actual position at different times.

[0103] Optionally, outliers of the generated trajectory are removed to achieve smoothing of the generated trajectory and improve the accuracy of the trajectory of the candidate insertion vehicle generated finally.

[0104] D2. Determine, based on the trajectory of the candidate inserting vehicle, whether the candidate inserting vehicle has an inserting behavior of inserting into the lane where the vehicle is located.

[0105] Specifically, it is determined according to the trajectory of the candidate inserting vehicle whether the candidate inserting vehicle travels from an adjacent lane to the lane where the vehicle is located. If so, it is determined that the candidate inserting vehicle has an inserting behavior.

[0106] D3. When the candidate inserting vehicle has an insertion behavior of inserting into the lane where the vehicle is located, determine whether there is a collision risk between the vehicle and the candidate inserting vehicle.

[0107] Specifically, the motion trajectory of the vehicle can be generated to predict whether the motion trajectory of the vehicle and the trajectory of the candidate inserting vehicle will intersect at a future time. If so, it is determined that there is a risk of collision between the vehicle and the candidate inserting vehicle. Specifically, when performing trajectory prediction, the possible trajectory of the vehicle in the next few seconds can be predicted based on the current position, speed, acceleration, steering angle and other information of the vehicle, combined with the kinematic model.

[0108] D4. When there is a risk of collision between the vehicle and the candidate inserting vehicle, determine whether the vehicle has an evasive action, and when it is determined that there is no evasive action, determine that the vehicle has the abnormal vehicle control event of not avoiding.

[0109] Among them, the avoidance action may include initiating emergency braking and initiating steering assistance.

[0110] When the avoidance action includes the initiation of emergency braking, the following methods can be used to determine whether the emergency braking is activated: (1) Determine whether the sudden deceleration of the vehicle exceeds the preset deceleration threshold. If so, it indicates that the probability of emergency braking activation is high (the probability of emergency braking activation is greater than 60%). (2) Determine whether the brake lights light up quickly and continuously. If so, it indicates that the probability of emergency braking activation is high. (3) Determine whether the acceleration of the vehicle suddenly changes from a positive value to a negative value, and the deceleration is large. If so, it indicates that the probability of emergency braking activation is high.

[0111] When the avoidance action includes starting the steering assist, it can be determined whether the steering assist is started in the following manner: detecting whether the function switch of the steering assist is turned on.

[0112] Of course, the avoidance action may also include other actions, which are not listed here one by one.

[0113] Since the video data in front of the vehicle is analyzed for abnormal events of not yielding and controlling the vehicle, the existence of abnormal events of not yielding and controlling the vehicle will increase the probability of traffic accidents, that is, there is a certain driving risk. Therefore, in the embodiment of the present application, analyzing the abnormal events of not yielding and controlling the vehicle is beneficial to improving the accuracy of the driving risk identification model obtained through subsequent training.

[0114] It should be pointed out that among the above-mentioned abnormal events of vehicle control over the line, abnormal event of vehicle control on the roadside, abnormal event of not slowing down, and abnormal event of vehicle control without avoiding, only one of the events may be detected, or multiple events may be detected, which is not limited here.

[0115] In some embodiments, the data to be processed includes non-vehicle data, and the non-vehicle data includes video data in the cockpit, and the S12 includes:

[0116] E1. Detecting the driver's facial key points based on the video data in the cockpit, where the facial key points include feature points corresponding to the eyes.

[0117] The video data in the cockpit can be obtained by installing an in-vehicle camera in the vehicle with the lens facing the driver.

[0118] It should be pointed out that the above-mentioned facial key points may also include feature points corresponding to the nose and mouth.

[0119] Specifically, detecting facial key points refers to determining the two-dimensional coordinates of facial key points of a video frame corresponding to the video data in the cockpit.

[0120] E2. Determine the driver's head posture and sight direction based on the facial key points.

[0121] Specifically, the two-dimensional coordinates of the determined facial key points are mapped to three-dimensional coordinates, and then the rotation and translation vectors of the in-vehicle camera are calculated based on the three-dimensional coordinates and their two-dimensional projections on the video frame. Finally, the Euler angles are calculated based on the rotation and translation vectors to obtain the driver's head posture.

[0122] In the embodiment of the present application, the sight direction can be calculated based on feature points such as the corner of the eye and the pupil position.

[0123] E3. Determine whether the driver's line of sight deviates from a normal driving line of sight according to the driver's head posture and the line of sight direction.

[0124] Specifically, the normal driving line of sight direction is determined based on the driver's head posture, and the normal driving line of sight direction is compared with the driver's line of sight direction determined based on feature points such as the driver's eye corners and pupil positions. If the two deviate greatly, it is determined that the driver's line of sight direction has deviated from the normal driving line of sight direction.

[0125] In the embodiment of the present application, since the driver relies on his line of sight to drive the vehicle, analyzing whether the driver's line of sight deviates from the normal driving line of sight is conducive to the subsequent training of a more accurate driving risk identification model.

[0126] Optionally, considering that in addition to deviating from the normal driving line of sight, the driver may also close his eyes, yawn, tilt his head left or right, etc., all of which will affect the driving risk, it is also possible to determine whether these events occur based on the acquired facial key points.

[0127] Among them, whether the driver's eyes are in an open or closed state can be determined by analyzing the key points of the eyes in continuous video frames.

[0128] Among them, whether the driver yawns can be determined by analyzing the key points of the mouth in continuous video frames.

[0129] Among them, whether the driver has a left or right head tilt event can be determined by analyzing the driver's head posture in continuous video frames.

[0130] In some embodiments, considering that the probability of a traffic accident occurring when the driver is in a drowsy state is greater than the probability of a traffic accident occurring when the driver is not in a drowsy state, in addition to analyzing the events listed above, it is also possible to analyze whether the driver has a drowsy expression event, and train a driving risk recognition model based on the analysis results of whether a drowsy expression event has occurred.

[0131] Specifically, a deep learning model for detecting the driver's sleepy expression can be pre-trained, and the driver's facial expressions in the video data in the cockpit can be recognized based on the deep learning model. If the driver is recognized to have a sleepy expression and the duration of the sleepy expression is greater than a preset sleepiness duration threshold, it is determined that the driver has a sleepy expression event.

[0132] In some embodiments, the data to be processed includes non-vehicle data, the non-vehicle data includes voice data, and S12 includes:

[0133] F1. Detecting whether the voice data meets a preset voice condition, where the voice condition includes that the voice data is the driver's voice data.

[0134] The voice data here can be collected by a microphone installed in the cockpit. When the voice condition includes that the voice data is the voice data of the driver, the audio features of the driver can be pre-recorded. After the voice data is collected from the microphone, the features of the voice data are determined, and the features of the voice data are compared with the audio features of the driver. If the two match, it is determined that the collected voice data meets the preset voice condition.

[0135] Optionally, considering that when there is a passenger with aggressive speech next to the driver, the aggressive speech may stimulate the driver to make irrational driving behavior, and at this time, the driving risk of the vehicle is also high, therefore, the above-mentioned voice condition may also include: the voice data contains specified words. It should be pointed out that the voice data at this time is not necessarily the driver's voice data. That is, when the voice condition includes: the voice data contains specified words, it is determined whether the collected voice data meets the preset voice condition by analyzing whether the collected voice data contains the specified words. In this judgment process, there is no need to pre-record the driver's audio features.

[0136] F2. When the voice data satisfies the voice conditions, determine that a speaking event exists.

[0137] In the embodiment of the present application, considering that a speaking event may have an impact on driving risk, it is determined whether a speaking event occurs so that a driving risk identification model can be subsequently trained based on the determination result, which is beneficial to improving the accuracy of the trained driving risk identification model.

[0138] In some embodiments, before the textual processing of the driver's identification and the event corresponding to the driver is performed, the method further includes:

[0139] The category information of the driver is obtained, where the category information is used to indicate the category to which the driving habit of the driver belongs.

[0140] Correspondingly, the textual processing of the driver's identification and the event corresponding to the driver includes:

[0141] The identification of the driver, the event corresponding to the driver and the category information of the driver are processed in text form.

[0142] Specifically, multiple categories are preset, and different categories correspond to different category information (the category information includes the category name and may also include category description information). The category information includes: speeding type, not wearing a seat belt type, not obeying traffic lights type, stable driving type, sudden acceleration type, sudden deceleration type, etc. It should be pointed out that the category information corresponding to the same driver may be greater than 1. For example, the same driver may belong to both the speeding type and the traffic lights non-observance type.

[0143] Since the category information is used to indicate the category to which the above-mentioned driver's driving habits belong, and driving habits are usually difficult to change, and driving habits will affect driving risks, therefore, the driver's category information is also included when training the driving risk identification model, which is beneficial to improving the accuracy of the trained driving risk identification model.

[0144] In some embodiments, the above S14 includes:

[0145] According to the above text processing results, low-rank adaptive training is performed on the driving risk identification model to be trained to obtain the above trained driving risk identification model.

[0146] Among them, Low-Rank Adaptation (LoRA) training is a technique for fine-tuning large language models. It fine-tunes the original model by training only low-rank matrices and then injecting these parameters into the original model. This method reduces computing requirements and makes training resources much smaller than directly training the original model, which is conducive to improving training speed.

[0147] In an embodiment of the present application, the rank r of the two matrices (assuming matrix A and matrix B) involved in LoRA training can be determined randomly, or determined according to the width and depth of the original model (i.e., the driving risk identification model to be trained) and the complexity of the driving risk identification task. Since the expressive power of LoRA training is related to r, when r is determined according to the width and depth of the original model and the complexity of the driving risk identification task, it is beneficial to improve the accuracy of the determined r.

[0148] After the trained driving risk identification model is obtained according to the above method, the trained driving risk identification model can be used to identify driving risks.

[0149] Figure 2 A flow chart of a driving risk identification method provided by an embodiment of the present application is shown, and is described in detail as follows:

[0150] S21, obtaining the driver's data to be processed, where the data to be processed includes vehicle data and / or non-vehicle data.

[0151] Among them, the data included in the vehicle data here is similar to the data included in the vehicle data above. Similarly, the data included in the non-vehicle data is similar to the data included in the non-vehicle data above, and will not be repeated here.

[0152] In the embodiment of the present application, when the driver is driving the vehicle, the driver's data to be processed can be obtained so as to estimate the driving risk later based on the data to be processed. That is, in the embodiment of the present application, the obtained data to be processed is the data to be processed obtained by the same driver in the same driving event.

[0153] S22, determining corresponding events according to the data to be processed, and obtaining real-time events.

[0154] The process of determining the real-time event is similar to the process of determining the corresponding event according to the data to be processed, and will not be repeated here.

[0155] S23, obtaining historical events and historical alarm information, wherein the historical events are events determined before the real-time events are determined.

[0156] Specifically, the event determined before the real-time event is determined this time is taken as a historical event and obtained. It should be noted that the historical events here include historical events corresponding to multiple drivers. For example, before the real-time event is determined, the driver in S21 has a determined event, and the historical events here include the event determined by the driver before the real-time event is determined.

[0157] In an embodiment of the present application, historical alarm information corresponds to driving events. For example, assuming that the driver has a driving event in the time period (t1-t2), and there is alarm information in the time period (t1-t2), then the driving event corresponds to the existing alarm information.

[0158] In an embodiment of the present application, historical alarm information can be obtained based on feedback from the driver or other staff, or can be obtained from reported alarm information.

[0159] S24, text processing is performed on the real-time events, the historical events and the historical alarm information to obtain a text to be detected.

[0160] S25, using the text to be detected as the trained driving risk identification model as described above, to obtain the driving risk output by the trained driving risk identification model.

[0161] In the embodiment of the present application, since the input of the trained driving risk identification model includes real-time events, historical events and historical alarm information, that is, the input of the trained driving risk identification model includes both real-time information and historical information, therefore, when there are fewer real-time events, the output driving risk is mainly affected by the historical information, and when there are more real-time events, the real-time events will also affect the output driving risk, that is, the output driving risk is more matched with the driver himself, thereby improving the accuracy of the output driving risk.

[0162] In some embodiments, before S24, the method further includes:

[0163] The driver's category information is obtained, where the category information is used to indicate the category to which the driver's driving habits belong.

[0164] Correspondingly, the above S24 includes:

[0165] The driver's category information, the real-time events, the historical events and the historical alarm information are processed into text to obtain a text to be detected.

[0166] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0167] Corresponding to the driving risk identification method described in the above embodiment, Figure 3 A structural block diagram of a driving risk identification device provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0168] Reference Figure 3 The driving risk identification device 3 includes: a to-be-processed data acquisition module 31, a real-time event determination module 32, a historical event acquisition module 33, a to-be-detected text determination module 34, and a driving risk determination module 35. Among them:

[0169] The to-be-processed data acquisition module 31 is used to acquire the driver's to-be-processed data, where the to-be-processed data includes vehicle data and / or non-vehicle data.

[0170] The real-time event determination module 32 is used to determine the corresponding event according to the above-mentioned data to be processed to obtain the real-time event.

[0171] The historical event acquisition module 33 is used to acquire historical events and historical alarm information. The historical events are events determined before the real-time events are determined.

[0172] The module 34 for determining the text to be detected is used to process the real-time events, the historical events and the historical alarm information into text form to obtain the text to be detected.

[0173] The driving risk determination module 35 is used to use the above-mentioned text to be detected as the trained driving risk identification model as described in any one of claims 1 to 10 to obtain the driving risk output by the above-mentioned trained driving risk identification model.

[0174] In the embodiment of the present application, since the input of the trained driving risk identification model includes real-time events, historical events and historical alarm information, that is, the input of the trained driving risk identification model includes both real-time information and historical information, therefore, when there are fewer real-time events, the output driving risk is mainly affected by the historical information, and when there are more real-time events, the real-time events will also affect the output driving risk, that is, the output driving risk is more matched with the driver himself, thereby improving the accuracy of the output driving risk.

[0175] In some embodiments, the driving risk identification device 3 includes:

[0176] The category information acquisition module is used to acquire the category information of the driver, where the category information is used to indicate the category to which the driving habit of the driver belongs.

[0177] Correspondingly, the to-be-detected text determination module 34 is specifically used for:

[0178] The driver's category information, the real-time events, the historical events and the historical alarm information are processed into text to obtain a text to be detected.

[0179] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0180] Figure 4 This is a schematic diagram of the structure of a vehicle-mounted device provided in one embodiment of the present application. Figure 4 As shown, the vehicle-mounted device 4 of this embodiment includes: at least one processor 40 ( Figure 4 Only one processor is shown in the figure), a memory 41, and a computer program 42 stored in the memory 41 and executable on the at least one processor 40, wherein the processor 40 implements the steps in any of the above-mentioned method embodiments when executing the computer program 42.

[0181] The vehicle-mounted device 4 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The vehicle-mounted device may include, but is not limited to, a processor 40 and a memory 41. Those skilled in the art will appreciate that Figure 4 It is only an example of the vehicle-mounted device 4 and does not constitute a limitation on the vehicle-mounted device 4. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0182] The processor 40 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0183] In some embodiments, the memory 41 may be an internal storage unit of the vehicle-mounted device 4, such as a hard disk or memory of the vehicle-mounted device 4. In other embodiments, the memory 41 may also be an external storage device of the vehicle-mounted device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the vehicle-mounted device 4. Further, the memory 41 may also include both an internal storage unit and an external storage device of the vehicle-mounted device 4. The memory 41 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory 41 may also be used to temporarily store data that has been output or is to be output.

[0184] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0185] An embodiment of the present application also provides a network device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor implements the steps in any of the above-mentioned method embodiments when executing the computer program.

[0186] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0187] An embodiment of the present application provides a computer program product. When the computer program product is run on a vehicle-mounted device, the vehicle-mounted device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0188] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera / vehicle-mounted device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0189] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0190] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0191] In the embodiments provided in the present application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0192] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0193] It should be noted that the information collection process (such as face image collection process, fingerprint information collection process, etc.) / feature extraction process involved in this application is performed with the user's knowledge and permission, that is, the information collection process / feature extraction process complies with the requirements of laws and regulations and does not constitute an act that harms the public interest.

[0194] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A driving risk identification model training method, characterized in that: include: Acquiring data to be processed of a plurality of drivers, wherein the data to be processed includes vehicle data and / or non-vehicle data; Determine a corresponding event according to the data to be processed; Performing text processing on the driver's identification and the event corresponding to the driver to obtain a text processing result; The driving risk identification model to be trained is trained according to the text processing result to obtain a trained driving risk identification model, wherein the driving risk identification model to be trained is a large language model, and the trained driving risk identification model is used to output driving risks.

2. The driving risk identification model training method according to claim 1, characterized in that: The data to be processed includes vehicle data, the vehicle data includes acceleration and / or angular velocity, and determining a corresponding event according to the data to be processed includes: In a case where the vehicle data includes acceleration, calculating the acceleration change rate of the acceleration, and determining whether the vehicle has a sudden deceleration event or a sudden acceleration event according to the acceleration change rate; In a case where the data of the vehicle includes an angular velocity, an angular velocity change rate of the angular velocity is calculated, and whether the vehicle has an abnormal posture event is determined according to the angular velocity change rate.

3. The driving risk identification model training method according to claim 1, characterized in that: The data to be processed includes non-vehicle data, and the non-vehicle data includes video data in front of the vehicle. The determining a corresponding event according to the data to be processed includes: Determine whether the vehicle has an abnormal lane crossing control event according to the video data in front of the vehicle, wherein when the abnormal lane crossing control event occurs, the vehicle crosses the lane alternately on both sides of the lane line; and / or, Determining whether the vehicle has a roadside vehicle control abnormal event according to the video data in front of the vehicle, wherein when the roadside vehicle control abnormal event exists, the distance between the vehicle and the road boundary exceeds a preset boundary threshold; and / or, Determine whether the vehicle ahead of the vehicle has an abnormal event of not decelerating according to the video data in front of the vehicle, wherein when the vehicle ahead of the vehicle has an abnormal event of not decelerating, the vehicle ahead of the vehicle has no deceleration behavior but has a collision event; and / or, Determine whether the vehicle has an abnormal vehicle control event without yielding based on video data in front of the vehicle, wherein, when the vehicle has an abnormal vehicle control event without yielding, the vehicle in front of the vehicle has inserted into the lane where the vehicle is located, and there is a risk of collision between the vehicle and the vehicle in front of the vehicle but the vehicle has not taken any evasive action.

4. The driving risk identification model training method according to claim 3, characterized in that: The determining, based on the video data in front of the vehicle, whether the vehicle has an abnormal event of crossing the line and controlling the vehicle, comprises: Detecting lane lines based on video data in front of the vehicle; Calculating a position change of the vehicle relative to the lane line according to the position of the detected lane line in consecutive adjacent video frames corresponding to the video data; When the change in the position of the vehicle relative to the lane line indicates that the vehicle crosses the lane line alternately on both sides of the lane line, it is determined that the vehicle has the lane-crossing abnormal event.

5. The driving risk identification model training method according to claim 3, characterized in that: The determining, based on the video data in front of the vehicle, whether the vehicle ahead of the vehicle has an abnormal event of not decelerating, comprises: Calculating a rate of change of a speed of a preceding vehicle of the vehicle based on video data in front of the vehicle; Determining whether the vehicle ahead of the vehicle has decelerated according to the rate of change of the speed of the vehicle ahead of the vehicle; When it is determined that the vehicle preceding the vehicle has no deceleration behavior and a collision occurs, it is determined that the vehicle preceding the vehicle has the non-deceleration abnormal event.

6. The driving risk identification model training method according to claim 3, characterized in that: The determining, based on the video data in front of the vehicle, whether the vehicle has an abnormal vehicle control event of not avoiding the vehicle, includes: Determining a trajectory of a candidate insertion vehicle based on video data in front of the vehicle, wherein the candidate insertion vehicle is a vehicle in front of the vehicle and in a lane adjacent to a lane of the vehicle; Determining, according to the trajectory of the candidate inserting vehicle, whether the candidate inserting vehicle has an inserting behavior of inserting into the lane where the vehicle is located; In the case where the candidate inserting vehicle has an inserting behavior of inserting into the lane where the vehicle is located, determining whether there is a collision risk between the vehicle and the candidate inserting vehicle; In the case where there is a risk of collision between the vehicle and the candidate inserting vehicle, it is determined whether the vehicle has an evasive action, and if it is determined that there is no evasive action, it is determined that the vehicle has the non-avoidance vehicle control abnormal event.

7. The driving risk identification model training method according to claim 1, characterized in that: The data to be processed includes non-vehicle data, and the non-vehicle data includes video data in the cockpit. The determining a corresponding event according to the data to be processed includes: Detecting key points of the driver's face based on the video data in the cockpit, wherein the key points of the face include feature points corresponding to the eyes; Determining the driver's head posture and sight direction according to the facial key points; Determine whether the driver's line of sight deviates from a normal driving line of sight direction according to the driver's head posture and the line of sight direction.

8. The driving risk identification model training method according to claim 1, characterized in that: The data to be processed includes non-vehicle data, and the non-vehicle data includes voice data. The determining of a corresponding event according to the data to be processed includes: detecting whether the voice data meets a preset voice condition, the voice condition including that the voice data is voice data of a driver; In a case where the voice data satisfies the voice condition, it is determined that a speaking event exists.

9. The driving risk identification model training method according to any one of claims 1 to 8, characterized in that: Before the textual processing of the driver's identification and the event corresponding to the driver is performed, the method further includes: Acquiring category information of the driver, where the category information is used to indicate the category to which the driver's driving habits belong; The textual processing of the driver's identification and the event corresponding to the driver includes: The identification of the driver, the event corresponding to the driver, and the category information of the driver are processed in text form.

10. The driving risk identification model training method according to any one of claims 1 to 8, characterized in that: The step of training the driving risk identification model to be trained according to the text processing result to obtain the trained driving risk identification model includes: Low-rank adaptive training is performed on the driving risk identification model to be trained according to the text processing result to obtain the trained driving risk identification model.

11. A driving risk identification method, characterized in that: include: Acquiring driver data to be processed, where the data to be processed includes vehicle data and / or non-vehicle data; Determine a corresponding event according to the data to be processed to obtain a real-time event; Acquire historical events and historical alarm information, wherein the historical events are events determined before the real-time events are determined; Performing text processing on the real-time events, the historical events and the historical alarm information to obtain a text to be detected; The text to be detected is used as the trained driving risk identification model as described in any one of claims 1 to 10 to obtain the driving risk output by the trained driving risk identification model.

12. A driving risk identification device, characterized in that: include: A to-be-processed data acquisition module, used to acquire the driver's to-be-processed data, wherein the to-be-processed data includes vehicle data and / or non-vehicle data; A real-time event determination module, used to determine a corresponding event according to the data to be processed to obtain a real-time event; A historical event acquisition module, used to acquire historical events and historical alarm information, wherein the historical events are events determined before the real-time events are determined; A module for determining text to be detected, used for processing the real-time event, the historical event and the historical alarm information into text form to obtain the text to be detected; A driving risk determination module is used to use the text to be detected as a trained driving risk identification model as described in any one of claims 1 to 10 to obtain the driving risk output by the trained driving risk identification model.

13. A vehicle-mounted device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 10 is implemented.

14. 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 10 is implemented.

15. 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 10 is implemented.

Images