Driving behavior evaluation method, device, equipment, medium and product

Abstract

The application provides a driving behavior evaluation method, device, equipment, computer readable storage medium and computer program product. The method comprises: obtaining target driving data; the target driving data comprises target vehicle driving state data, target vehicle driving environment data, target driver operation information and target driver physiological state information; processing the target driving data based on a target student model to identify non-standard driving behavior of the target driver; the target student model is obtained by distillation training based on a teacher model in the cloud and deployed into the target vehicle; when the non-standard driving behavior is identified, obtaining driving suggestions for the non-standard driving behavior; obtaining interaction habit information of the target driver, determining a corresponding information display mode according to the interaction habit information, and outputting information of the non-standard driving behavior and the driving suggestions through the information display mode. The method improves the identification efficiency and accuracy of non-standard driving behavior.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a driving behavior evaluation method, apparatus, device, computer-readable storage medium, and computer program product. Background Technology

[0002] With the rapid development of intelligent and connected vehicle technologies, driving safety has become a major focus of the automotive industry. To effectively prevent traffic accidents and protect the lives of drivers and passengers, real-time monitoring of drivers' unsafe driving behaviors has become particularly important.

[0003] In recent years, deep learning models have made significant progress in the automotive field. However, on the one hand, high-performance models often involve large computational costs and do not take into account the limitations of memory and computing power in vehicle-side devices; on the other hand, although smaller models are more suitable for edge computing environments, their accuracy in recognizing irregular driving behaviors is usually lower than that of larger models.

[0004] Therefore, how to improve the recognition accuracy of small models while ensuring real-time performance has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a driving behavior evaluation method, apparatus, device, computer-readable storage medium, and computer program product, which can improve the recognition accuracy of small models while ensuring real-time performance.

[0006] In a first aspect, embodiments of this application provide a driving behavior evaluation method, including:

[0007] Acquire target driving data; wherein, the target driving data includes target vehicle driving status data, target vehicle driving environment data, target driver operation information, and target driver physiological status information;

[0008] The target driving data is processed based on the target student model to identify the non-standard driving behavior of the target driver; wherein the target student model is trained in the cloud based on the teacher model distillation and deployed to the target vehicle;

[0009] When irregular driving behavior is detected, driving suggestions are obtained for the irregular driving behavior;

[0010] The system acquires the target driver's interaction habit information, determines the corresponding information display mode based on the interaction habit information, and outputs information about the non-standard driving behavior and driving suggestions to the target driver through the information display mode.

[0011] Secondly, embodiments of this application provide a driving behavior evaluation device, comprising:

[0012] The acquisition module is used to acquire target driving data; wherein, the target driving data includes target vehicle driving status data, target vehicle driving environment data, target driver operation information, and target driver physiological status information;

[0013] The processing module is used to process the target driving data based on the target student model in order to identify the non-standard driving behavior of the target driver; wherein the target student model is trained in the cloud based on the teacher model distillation and deployed to the target vehicle;

[0014] The acquisition module is also used to acquire driving suggestions for the non-standard driving behavior when non-standard driving behavior is identified, and to acquire the interaction habit information of the target driver.

[0015] The processing module is also used to determine the corresponding information display mode based on the interaction habit information, and output the information on the non-standard driving behavior and the driving suggestions to the target driver through the information display mode.

[0016] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the driving behavior evaluation method provided in the first aspect of embodiments of this application.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the driving behavior evaluation method provided in the first aspect of embodiments of this application.

[0018] Fifthly, embodiments of this application provide a computer program product having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the driving behavior evaluation method provided in the first aspect of embodiments of this application.

[0019] The technical solution provided in this application can train a target student model in the cloud based on the teacher model distillation, so that the target student model's recognition results for the target driver's irregular driving behavior are close to those of the teacher model, thus improving the accuracy of the recognition results. At the same time, since the target student model has a small parameter scale, the requirements for the computing power, storage and other hardware of the deployment device are low. The target student model trained by distillation can be directly deployed on the vehicle, so that the vehicle can use the target student model to calculate the target driving data, improving the real-time performance of the target driving data processing and thus improving the recognition efficiency. In addition, when irregular driving behavior of the target driver is detected, a personalized information display mode can be used to push driving suggestions, which is highly practical and improves the driving experience. Attached Figure Description

[0020] Figure 1 A flowchart illustrating a driving behavior evaluation method provided in an embodiment of this application;

[0021] Figure 2 A schematic flowchart illustrating the target student model training process provided in this application embodiment;

[0022] Figure 3 Another flowchart illustrating the target student model training process provided in this application embodiment;

[0023] Figure 4 A schematic diagram of the structure of the driving behavior evaluation device provided in the embodiments of this application;

[0024] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application. Those skilled in the art can make adjustments as needed to suit specific application scenarios. Furthermore, it should be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not the entire structure.

[0026] Figure 1 This is a schematic flowchart of a driving behavior evaluation method provided in an embodiment of this application. Figure 1 As shown, the method may include:

[0027] S101. Obtain target driving data.

[0028] The target driving data includes target vehicle driving status data, target vehicle driving environment data, target driver operation information, and target driver physiological status information.

[0029] Target vehicle driving status data refers to data related to the vehicle's state during operation. This data may include, but is not limited to, vehicle speed, acceleration, braking frequency, engine speed, gear position, steering wheel angle, and vehicle stability data (such as yaw rate and lateral acceleration). This data can be collected through corresponding onboard sensors. Target vehicle driving environment data refers to environmental information surrounding the target vehicle. This data can be obtained through visual sensors (such as cameras, radar, and lidar). It may include, but is not limited to, information about surrounding vehicles, pedestrians, road signs, traffic lights, weather conditions, and lighting conditions. Target driver operation information refers to the driver's actions while driving the vehicle. This data may include, but is not limited to, turn signal usage, high beam usage, accelerator pedal depth, brake pedal depth, and steering wheel input. The physiological state information of the target driver refers to the data related to the physiological state of the target driver during the driving process. The physiological state information of the target driver can be obtained through corresponding sensors (such as eye trackers, electroencephalogram sensors and / or infrared sensors). The physiological state information of the target driver may include, but is not limited to, information such as the fatigue state, the degree of inattention and the degree of emotional fluctuation of the target driver.

[0030] After obtaining the target driving data, operations such as data cleaning, data standardization, and data normalization can be performed on the target driving data to ensure the quality and consistency of the data input into the target student model.

[0031] Data cleaning refers to checking the consistency of the target driving data and removing invalid and missing values. Data standardization refers to converting data with different dimensions in the target driving data to the same dimension. Data normalization refers to scaling different target driving data to the same range to facilitate processing by the target student model.

[0032] S102. Process the target driving data based on the target student model to identify the non-standard driving behavior of the target driver.

[0033] The network structure of the target student model is smaller than that of the teacher model, which is a pre-trained large model. In this embodiment, the target student model can be trained in the cloud. That is, the cloud can train the target student model through distillation of the teacher model, and then deploy the distilled target student model on the vehicle (i.e., the target vehicle). The lightweight nature of the target student model is more suitable for the resource-constrained vehicle, enabling the processing of target driving data to be realized on the vehicle, improving recognition efficiency and meeting the needs of real-time applications.

[0034] After acquiring the target driving data, the target driving data is analyzed and processed using a target student model to identify the target driver's non-standard driving behaviors. These non-standard driving behaviors may include: changing lanes without using turn signals, failing to turn off high beams when meeting oncoming traffic, rapid acceleration, sudden braking, frequent lane changes, speeding, fatigued driving, inattentiveness, illegal overtaking, illegal U-turns, and / or running red lights.

[0035] S103. When irregular driving behavior is detected, obtain driving suggestions for the irregular driving behavior.

[0036] When irregular driving behavior is detected, corresponding driving suggestions can be generated for the target driver. For example, if it is detected that the target driver did not use the turn signal when changing lanes, the following driving suggestion can be generated: "It is recommended that you turn on the turn signal in advance before changing lanes, and check the rearview mirror and blind spots, and change lanes only after confirming that it is safe to do so."

[0037] Optionally, the above-mentioned driving suggestions for obtaining non-standard driving behavior may include: obtaining the driving behavior characteristics of the target driver; and retrieving driving suggestions for non-standard driving behavior from a pre-established driving behavior knowledge base based on the driving behavior characteristics.

[0038] In this embodiment, a driving behavior knowledge base can be established beforehand. This knowledge base is a database containing rich driving behavior rules and knowledge, covering safe driving suggestions, error correction measures, and methods for cultivating good driving habits for different driving behavior characteristics in various driving scenarios. The driving behavior knowledge base supports dynamic updates, continuously learning and optimizing the accuracy and practicality of driving suggestions. Specifically, expert advice and historical accident case analysis results can be collected to construct a driving behavior knowledge base containing various driving scenarios and their corresponding safety suggestions. Furthermore, when non-standard driving behavior is identified in a target driver, the driver's driving behavior characteristics can be obtained. These characteristics characterize the target driver's driving style, such as aggressive, conservative, or impulsive. These driving behavior characteristics can be obtained by analyzing the target driver's historical driving data. Driving suggestions targeting non-standard driving behavior are retrieved from the driving behavior knowledge base based on the target driver's driving behavior characteristics to provide personalized driving suggestions to the target driver. For example, matching driving suggestions can be provided for aggressive and conservative drivers.

[0039] S104. Obtain the target driver's interaction habit information, determine the corresponding information display mode based on the interaction habit information, and output information on non-standard driving behavior and driving suggestions to the target driver through the information display mode.

[0040] The interaction habit information can be used to characterize the interaction mode that the target driver prefers. For example, if the target driver frequently uses the in-vehicle display screen, it indicates that their interaction habit is more inclined towards screen display mode; if the target driver frequently uses voice commands, it indicates that their interaction habit is more inclined towards voice interaction mode. In the embodiments of this application, the target driver's interaction habit information can be obtained by analyzing the target driver's historical operation records.

[0041] After obtaining driving suggestions, the system can determine the appropriate information display mode for the target driver based on their interaction habits. This means selecting the most suitable method (such as voice prompts or screen displays) to provide personalized driving suggestions. For example, if the target driver prefers to see rather than hear, the driving suggestions can be conveyed through a screen display.

[0042] The driving behavior evaluation method provided in this application can train a target student model in the cloud based on a teacher model distillation, making the target student model's recognition results for the target driver's non-standard driving behavior close to those of the teacher model, thus improving the accuracy of the recognition results. At the same time, since the target student model has a small parameter scale, the requirements for the computing power, storage, and other hardware of the deployment device are low. The target student model trained by distillation is directly deployed on the vehicle, enabling the vehicle to perform calculations on the target driving data through the target student model, improving the real-time performance of the target driving data processing, thereby improving the recognition efficiency. In addition, when non-standard driving behavior of the target driver is identified, a personalized information display mode can be used to push driving suggestions, which is highly practical and improves the driving experience.

[0043] In one embodiment, optionally, if the non-standard driving behavior of the target driver is detected, it can also be determined whether the non-standard driving behavior will lead to a driving safety hazard. If it is determined that there is a driving safety hazard, the corresponding emergency intervention mechanism is triggered.

[0044] In this embodiment, non-standard driving behavior can be assessed to determine whether it will lead to driving safety hazards. If it is predicted that driving safety hazards may occur (such as excessive driver fatigue, serious distraction, or an impending collision), the target vehicle can automatically trigger an emergency intervention mechanism, such as issuing a strong alarm or having the target vehicle's own intelligent driving system temporarily take over, in order to ensure driving safety.

[0045] In one embodiment, alternatively, such as Figure 2 As shown, the target student model can be trained in the cloud through the following process:

[0046] S201. Obtain training data.

[0047] The training data includes sample driving data and corresponding sample recognition results. The sample driving data, used for training the target student model, originates from historical driving records, simulated driving environments, or driving data from specific test scenarios. This sample driving data includes sample vehicle driving status data, sample vehicle driving environment data, sample driver operation information, and sample driver physiological state information. After obtaining the sample driving data, non-standard driving behaviors can be labeled to obtain sample recognition results. These results are real-world labeled data used to supervise the training of the target student model. The teacher model is used to identify non-standard driving behaviors of drivers in the sample driving data.

[0048] S202. Input the sample driving data into the initial student model and teacher model.

[0049] In this context, the initial student model is a model that has not yet been trained or optimized and is ready to learn knowledge from the teacher model. In deep learning, the student model is typically a simplified version designed to mimic the behavior of a more complex and accurate teacher model.

[0050] The teacher model has a larger network size than the initial student model. The teacher model is a neural network model that has been trained on a large amount of data and has achieved a certain level of performance, enabling it to accurately identify improper driving behavior in driving data. In applications such as knowledge distillation, the teacher model transfers the knowledge it has learned to the initial student model.

[0051] S203. Obtain the first intermediate layer features and the first recognition result of the teacher model.

[0052] The sample driving data is input into the initial student model and teacher model. The input sample driving data is processed by the input layer and intermediate layer of each model to extract features. Based on the extracted features, the final recognition result is output through the output layer.

[0053] The first intermediate layer features of the teacher model are the features extracted from the intermediate layers of the teacher model's network structure, and the first recognition result is the recognition result of the teacher model for irregular driving behavior obtained based on sample driving data. For example, the teacher model can output the probability of one or more specific irregular driving behaviors. When the probability exceeds a preset threshold, it indicates that the driver in the sample driving data has the corresponding irregular driving behavior.

[0054] S204. The first recognition result is used as a soft label, the sample recognition result is used as a hard label, and the first intermediate layer features are used as feature-level knowledge.

[0055] S205. Distill the initial student model based on soft labels, hard labels, and feature-level knowledge to obtain the target student model.

[0056] The process of distilling the initial student model using hard labels is essentially a supervised training process based on the sample recognition results. Soft labels represent the first recognition result output by the already trained teacher model. Using these soft labels for supervised training of the initial student model, this knowledge distillation process guides the initial student model to learn knowledge from the teacher model, thereby improving its recognition accuracy. Simultaneously, the first intermediate layer features output from the teacher model's intermediate layers are used to guide the initial student model's training, allowing it to learn richer feature representations, thus improving the overall distillation performance of the initial student model.

[0057] Optionally, such as Figure 3 As shown, the above S205 may include:

[0058] S301. Construct a first loss function based on the soft label and the second recognition result.

[0059] After inputting sample driving data into the initial student model and the teacher model, we can also obtain the features extracted by the intermediate layer of the initial student model, i.e., the second intermediate layer features, and the second recognition result output by the output layer of the initial student model. Based on the soft label output by the teacher model and the second recognition result of the initial student model, we construct a first loss function. The first loss function can be the KL divergence, expressed as follows:

[0060]

[0061] Among them, D KL Let P(x) represent the KL divergence, P(x) represent the soft label output by the teacher model, Q(x) represent the second recognition result output by the initial student model, and x represent the input sample driving data.

[0062] S302. Construct a second loss function based on the hard label and the second recognition result.

[0063] Hard labels refer to the ground truth values ​​of drivers' non-standard driving behaviors in the sample driving data, i.e., the true labeled data. The second loss function can be the cross-entropy loss function, as shown below:

[0064]

[0065] Where H represents the cross-entropy loss function, N is the number of samples, B is the number of classes, and y i,b p represents the true label of the i-th sample belonging to the b-th category. i,bThis represents the initial student model's predicted probability that the i-th sample belongs to the b-th category (i.e., the second identification result).

[0066] S303. Construct a third loss function based on the features of the first intermediate layer and the features of the second intermediate layer.

[0067] The first intermediate layer features output from the teacher model are used as knowledge to supervise the learning of the intermediate layers of the initial student model, allowing the initial student model to more closely mimic the feature extraction methods of the teacher model. The third loss function can be a mean squared error function, minimizing the feature differences between it and the intermediate layer features of the initial student model. The third loss function can be expressed as follows:

[0068]

[0069] Among them, L i Let y represent the third loss function. i This represents the first intermediate layer features of the teacher model. This represents the second intermediate layer features of the initial student model.

[0070] S304. Based on the first loss function, the second loss function, and the third loss function, perform distillation training on the initial student model to obtain the target student model.

[0071] For example, a total loss function can be constructed based on the first, second, and third loss functions. The initial student model is then trained using distillation based on this total loss function. If the loss value of the total loss function is greater than a preset convergence threshold, the model parameters of the initial student model are adjusted, and the initial student model continues to be trained based on the above distillation training process until the loss value of the total loss function is less than or equal to the preset convergence threshold, thus obtaining a trained target student model. The constructed total loss function serves as the joint optimization objective to supervise the learning of the initial student model. The total loss function can be expressed as follows:

[0072] Total_loss=Hard_loss+α*Soft_loss+β*Feature_loss;

[0073] Where Total_loss represents the total loss function, Soft_loss represents the loss between the soft label and the second recognition result output by the initial student model (i.e., the first loss function), Hard_loss represents the loss between the hard label and the second recognition result output by the initial student model (i.e., the second loss function), Feature_loss represents the loss between the first intermediate layer feature output by the teacher model and the second intermediate layer feature output by the initial student model (i.e., the third loss function), and α and β are hyperparameters used to control the proportion of Soft_loss and Feature_loss in Total_loss. The specific values ​​of the hyperparameters can be set based on actual needs.

[0074] In this embodiment, hard labels, soft labels, and feature-level knowledge are combined to supervise the distillation training of the initial student model. That is, in addition to the initial student model being supervised by the hard labels in the training data, the recognition results of the teacher model on the sample driving data are also used to supervise the training of the initial student model. At the same time, the feature-level knowledge output by the intermediate layer of the teacher model is also used to supervise the training of the initial student model. This not only makes the output distribution of the initial student model close to that of the teacher model, but also allows the initial student model to imitate the feature extraction method of the teacher model, thereby improving the ability of the trained target student model to recognize irregular driving behavior and improving the recognition accuracy of the target student model.

[0075] Figure 4 This is a schematic diagram of a driving behavior evaluation device provided in an embodiment of this application. Figure 4 As shown, the device may include an acquisition module 401 and a processing module 402.

[0076] Specifically, the acquisition module 401 is used to acquire target driving data; wherein, the target driving data includes target vehicle driving status data, target vehicle driving environment data, target driver operation information, and target driver physiological status information;

[0077] The processing module 402 is used to process the target driving data based on the target student model in order to identify the non-standard driving behavior of the target driver; wherein, the target student model is trained in the cloud based on the teacher model distillation and deployed to the target vehicle;

[0078] The acquisition module 401 is also used to acquire driving suggestions for irregular driving behavior when irregular driving behavior is identified, and to acquire the target driver's interaction habit information.

[0079] The processing module 402 is also used to determine the corresponding information display mode based on the interaction habit information, and output information on irregular driving behavior and driving suggestions to the target driver through the information display mode.

[0080] Based on the above embodiments, optionally, the cloud-based training process for the target student model includes: acquiring training data; wherein the training data includes sample driving data and sample recognition results corresponding to the sample driving data, the sample driving data including sample vehicle driving status data, sample vehicle driving environment data, sample driver operation information, and sample driver physiological state information; a teacher model is used to identify the driver's non-standard driving behavior in the sample driving data; the sample driving data is input into the initial student model and the teacher model; the first intermediate layer features and the first recognition result of the teacher model are acquired; the first recognition result is used as a soft label, the sample recognition result is used as a hard label, and the first intermediate layer features are used as feature-level knowledge; the initial student model is distilled and trained according to the soft label, hard label, and feature-level knowledge to obtain the target student model.

[0081] Based on the above embodiments, optionally, the cloud-based training process for the target student model further includes: obtaining the second intermediate layer features and the second recognition result of the initial student model; constructing a first loss function based on the soft label and the second recognition result; constructing a second loss function based on the hard label and the second recognition result; constructing a third loss function based on the first intermediate layer features and the second intermediate layer features; and performing distillation training on the initial student model based on the first loss function, the second loss function, and the third loss function to obtain the target student model.

[0082] Based on the above embodiments, optionally, the cloud-based training process for the target student model further includes: constructing a total loss function based on the first loss function, the second loss function, and the third loss function; and performing distillation training on the initial student model based on the total loss function to obtain the target student model.

[0083] Optionally, based on the above embodiments, the acquisition module 401 is further configured to acquire the driving behavior characteristics of the target driver; and retrieve driving suggestions for non-standard driving behavior from a pre-established driving behavior knowledge base based on the driving behavior characteristics.

[0084] Based on the above embodiments, optionally, the processing module 402 is further configured to determine whether irregular driving behavior will lead to driving safety hazards; and if it is determined that there are driving safety hazards, to trigger the corresponding emergency intervention mechanism.

[0085] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 5 As shown, the device includes a processor 50, a memory 51, an input device 52, and an output device 53; the number of processors 50 in the device can be one or more. Figure 5Taking a processor 50 as an example; the processor 50, memory 51, input device 52, and output device 53 in this device can be connected via a bus or other means. Figure 5 Taking the example of a connection between China and Israel via a bus.

[0086] The memory 51, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the driving behavior evaluation method in the embodiments of this application (e.g., acquisition module 401 and processing module 402 in the driving behavior evaluation device). The processor 50 executes various functional applications and data processing of the device by running the software programs, instructions, and modules stored in the memory 51, thereby implementing the aforementioned driving behavior evaluation method.

[0087] The memory 51 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created during the driving behavior evaluation process. Furthermore, the memory 51 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid-state storage device. In some instances, the memory 51 may further include memory remotely located relative to the processor 50, which can be connected to a device / terminal / server via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0088] Input device 52 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the electronic device. Output device 53 may include display devices such as a display screen.

[0089] In one embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, the computer program performing the following steps when executed by a processor:

[0090] Acquire target driving data; wherein, the target driving data includes target vehicle driving status data, target vehicle driving environment data, target driver operation information, and target driver physiological status information;

[0091] The target driving data is processed based on the target student model to identify the non-standard driving behavior of the target driver; wherein the target student model is trained in the cloud based on the teacher model distillation and deployed to the target vehicle;

[0092] When irregular driving behavior is detected, driving suggestions are obtained for the irregular driving behavior;

[0093] The system acquires the target driver's interaction habit information, determines the corresponding information display mode based on the interaction habit information, and outputs information about the non-standard driving behavior and driving suggestions to the target driver through the information display mode.

[0094] In one embodiment, a computer program product is also provided, on which a computer program is stored, which, when executed by a processor, performs the following steps:

[0095] Acquire target driving data; wherein, the target driving data includes target vehicle driving status data, target vehicle driving environment data, target driver operation information, and target driver physiological status information;

[0096] The target driving data is processed based on the target student model to identify the non-standard driving behavior of the target driver; wherein the target student model is trained in the cloud based on the teacher model distillation and deployed to the target vehicle;

[0097] When irregular driving behavior is detected, driving suggestions are obtained for the irregular driving behavior;

[0098] The system acquires the target driver's interaction habit information, determines the corresponding information display mode based on the interaction habit information, and outputs information about the non-standard driving behavior and driving suggestions to the target driver through the information display mode.

[0099] The driving behavior evaluation device, electronic device, computer-readable storage medium, and computer program product provided in the above embodiments can execute the driving behavior evaluation method provided in any embodiment of this application, and have the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in the above embodiments can be found in the driving behavior evaluation method provided in any embodiment of this application.

[0100] Based on the above description of the implementation methods, those skilled in the art can clearly understand that this application can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0101] It is worth noting that the units and modules included in the above embodiments are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of this application.

[0102] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

Claims

1. A method for evaluating driving behavior, characterized in that, include: Acquire target driving data; wherein, the target driving data includes target vehicle driving status data, target vehicle driving environment data, target driver operation information, and target driver physiological status information; The target driving data is processed based on the target student model to identify the non-standard driving behavior of the target driver; wherein the target student model is trained in the cloud based on the teacher model distillation and deployed to the target vehicle; When irregular driving behavior is detected, driving suggestions are obtained for the irregular driving behavior; The system acquires the interaction habit information of the target driver, determines the corresponding information display mode based on the interaction habit information, and outputs information about the non-standard driving behavior and driving suggestions to the target driver through the information display mode. The training process of the target student model includes: Acquire training data; wherein the training data includes sample driving data and the sample recognition results corresponding to the sample driving data, the sample driving data includes sample vehicle driving status data, sample vehicle driving environment data, sample driver operation information, and sample driver physiological state information; the teacher model is used to identify the driver's non-standard driving behavior in the sample driving data. The sample driving data is input into the initial student model and teacher model; Obtain the first intermediate layer features and the first recognition result of the teacher model; The first identification result is used as a soft label, the sample identification result is used as a hard label, and the first intermediate layer feature is used as feature-level knowledge. Obtain the second intermediate layer features and the second recognition result of the initial student model; A first loss function is constructed based on the soft label and the second recognition result; A second loss function is constructed based on the hard label and the second recognition result; A third loss function is constructed based on the features of the first intermediate layer and the features of the second intermediate layer; The initial student model is distilled and trained based on the first loss function, the second loss function, and the third loss function to obtain the target student model.

2. The method according to claim 1, characterized in that, The step of distilling the initial student model according to the first loss function, the second loss function, and the third loss function to obtain the target student model includes: A total loss function is constructed based on the first loss function, the second loss function, and the third loss function. The initial student model is then trained by distillation based on the total loss function to obtain the target student model.

3. The method according to any one of claims 1 to 2, characterized in that, The process of obtaining driving suggestions for the aforementioned non-standard driving behavior includes: Obtain the driving behavior characteristics of the target driver; Based on the driving behavior characteristics, driving suggestions for the non-standard driving behavior are retrieved from a pre-established driving behavior knowledge base.

4. The method according to any one of claims 1 to 2, characterized in that, Also includes: Determine whether the aforementioned irregular driving behavior will lead to a driving safety hazard; If a driving safety hazard is identified, the corresponding emergency intervention mechanism will be triggered.

5. A driving behavior evaluation device, characterized in that, include: The acquisition module is used to acquire target driving data; wherein, the target driving data includes target vehicle driving status data, target vehicle driving environment data, target driver operation information, and target driver physiological status information; The processing module is used to process the target driving data based on the target student model in order to identify the non-standard driving behavior of the target driver; wherein the target student model is trained in the cloud based on the teacher model distillation and deployed to the target vehicle; The acquisition module is also used to acquire driving suggestions for the non-standard driving behavior when non-standard driving behavior is identified, and to acquire the interaction habit information of the target driver. The processing module is also used to determine the corresponding information display mode based on the interaction habit information, and output the information on the non-standard driving behavior and the driving suggestions to the target driver through the information display mode; The training process of the target student model includes: Acquire training data; wherein the training data includes sample driving data and the sample recognition results corresponding to the sample driving data, the sample driving data includes sample vehicle driving status data, sample vehicle driving environment data, sample driver operation information, and sample driver physiological state information; the teacher model is used to identify the driver's non-standard driving behavior in the sample driving data. The sample driving data is input into the initial student model and teacher model; Obtain the first intermediate layer features and the first recognition result of the teacher model; The first identification result is used as a soft label, the sample identification result is used as a hard label, and the first intermediate layer feature is used as feature-level knowledge. Obtain the second intermediate layer features and the second recognition result of the initial student model; A first loss function is constructed based on the soft label and the second recognition result; A second loss function is constructed based on the hard label and the second recognition result; A third loss function is constructed based on the features of the first intermediate layer and the features of the second intermediate layer; The initial student model is distilled and trained based on the first loss function, the second loss function, and the third loss function to obtain the target student model.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

8. A computer program product having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images