A vehicle emergency control method and system under driver abnormal behavior

Abstract

The present application relates to the technical field of driving safety, and particularly relates to a vehicle emergency control method and system under abnormal behavior of a driver, which acquires source domain monitoring data collected by a target vehicle, then generates target domain monitoring data matched with the source domain monitoring data based on sample data, then inputs the source domain monitoring data and the target domain monitoring data into a preset behavior judgment model to obtain driver behavior data output by the preset behavior judgment model, and finally performs emergency control on the target vehicle based on the driver behavior data. The present application generates matched target domain data based on source domain monitoring data through a cross-domain data generation and unified model training mechanism, and uses a unified preset behavior judgment model for behavior recognition, thereby getting rid of the dependence on sensors such as cameras and radars, so that old vehicle models can realize driver abnormal monitoring and emergency control without adding new hardware, and effectively solve the problems of high hardware requirement and poor compatibility of traditional solutions.

Description

Technical Field

[0001] This invention relates to the field of driving safety technology, and in particular to a vehicle emergency control method and system for abnormal driver behavior. Background Technology

[0002] With the increasing intelligence of automobiles, driver abnormal behavior monitoring and emergency control technology has become an important research direction for improving driving safety. Drivers may make operational errors due to distraction (such as using a mobile phone), fatigue, sudden illness, or health problems, which can lead to traffic accidents. Therefore, real-time monitoring of driver status and taking proactive control measures (such as automatic deceleration, lane keeping, and emergency braking) when abnormalities occur are of paramount importance for preventing accidents and protecting life and property.

[0003] However, existing driver abnormal behavior monitoring and emergency control systems typically require the integration of multiple hardware devices, such as cameras, millimeter-wave radar, inertial measurement units (IMUs), and steering wheel torque sensors, to collect relevant data. While this multi-sensor fusion architecture can improve monitoring accuracy, it places high demands on vehicle hardware configurations, severely limiting the large-scale application of driver abnormal behavior monitoring technology and preventing many older vehicles from enjoying the protection of advanced safety technologies.

[0004] Therefore, there is a need for an emergency control method and system for abnormal driver behavior that does not require large-scale hardware modifications and has strong compatibility. Summary of the Invention

[0005] Therefore, the present invention provides a vehicle emergency control method and system for abnormal driver behavior, in order to solve the problems of high hardware requirements and poor compatibility of existing safety systems.

[0006] This invention provides a vehicle emergency control method for abnormal driver behavior, comprising:

[0007] Acquire source domain monitoring data collected from the target vehicle;

[0008] Target domain monitoring data is generated based on sample data and matched with source domain monitoring data. The target domain monitoring data consists of data types that cannot be collected by the target vehicle.

[0009] The source domain monitoring data and the target domain monitoring data are input into the preset behavior judgment model to obtain the driver behavior data output by the preset behavior judgment model.

[0010] Emergency control of the target vehicle is carried out based on driver behavior data.

[0011] In a preferred implementation: the source domain monitoring data includes source domain time-series data, and the target domain monitoring data includes target domain time-series data; target domain monitoring data matching the source domain monitoring data is generated based on sample data, including:

[0012] The source domain time series data is input into the preset knowledge transfer model to obtain the target domain time series data output by the preset knowledge transfer model. The preset knowledge transfer model is trained based on sample data and is used to generate time series data that matches the input time series data but is of a different type according to the time series characteristics of the time series data.

[0013] In a preferred implementation: the pre-defined knowledge transfer model includes an encoder, a domain adaptation layer, and a decoder connected in sequence, wherein:

[0014] The encoder is used to extract the temporal features of the source domain temporal data and generate the first context vector;

[0015] The domain adaptation layer consists of multiple sequentially connected feedforward neural network layers, which are used to adjust the feature distribution of the first context vector to obtain the second context vector;

[0016] The decoder is used to generate target domain time-series data based on the second context vector.

[0017] In a preferred implementation: before step 1, which involves emergency control of the target vehicle based on driver behavior data, the method further includes:

[0018] The source domain monitoring data is input into the preset lightweight model to obtain the driver behavior data output by the preset lightweight model. The preset lightweight model is obtained by knowledge distillation training on the preset behavior judgment model.

[0019] In a preferred implementation: the sample data includes source domain sample data and sample behavior data corresponding to the source domain sample data; before the step of inputting the source domain monitoring data into a preset lightweight model to obtain the driver behavior data output by the preset lightweight model, the method further includes training the preset lightweight model, specifically including:

[0020] Obtain the initial lightweight model;

[0021] Historical source domain monitoring data was used as the first training data.

[0022] Obtain the historical driver behavior data corresponding to the first training data as the output result of the teacher model corresponding to the first training data;

[0023] Calculate the similarity between the first training data and the source domain sample data, and select the sample behavior data corresponding to the source domain sample data with the highest similarity as the true label corresponding to the first training data;

[0024] The first training data is input into the initial lightweight model to obtain the student model output corresponding to the first training data;

[0025] Based on the true labels corresponding to the first training data, the output results of the teacher model, and the output results of the student model, a first loss function is established.

[0026] The initial lightweight model is optimized based on the first loss function to obtain the trained preset lightweight model.

[0027] In one preferred implementation: the sample data also includes target domain sample data corresponding to the source domain sample data; training the preset lightweight model further includes:

[0028] Select source domain sample data with similarity higher than a preset threshold as the second training data;

[0029] The sample behavior data corresponding to the second training data is used as the true label corresponding to the second training data.

[0030] The second training data and the target domain sample data corresponding to the second training data are input into the preset behavior judgment model to obtain the teacher model output result corresponding to the second training data;

[0031] The second training data is input into the initial lightweight model to obtain the student model output corresponding to the second training data;

[0032] A second loss function is established based on the true labels corresponding to the second training data, the output results of the teacher model, and the output results of the student model.

[0033] The initial lightweight model is optimized based on the second loss function to obtain the trained preset lightweight model.

[0034] In one preferred implementation: training the pre-defined lightweight model also includes:

[0035] Source domain sample data representing abnormal driver behavior was selected as the third training data;

[0036] The sample behavior data corresponding to the third training data is used as the true label corresponding to the third training data.

[0037] The third training data and the target domain sample data corresponding to the third training data are input into the preset behavior judgment model to obtain the teacher model output result corresponding to the third training data;

[0038] The third training data is input into the initial lightweight model to obtain the student model output corresponding to the third training data;

[0039] A third loss function is established based on the true labels corresponding to the third training data, the output results of the teacher model, and the output results of the student model.

[0040] The initial lightweight model is optimized based on the third loss function to obtain the trained preset lightweight model.

[0041] In one preferred implementation: training the pre-defined lightweight model also includes:

[0042] The first loss function, the second loss function, and the third loss function are weighted and summed to obtain the total loss function.

[0043] The initial lightweight model is optimized based on the total loss function to obtain the pre-trained lightweight model.

[0044] In a preferred implementation: emergency control of the target vehicle is performed based on driver behavior data, including:

[0045] Determine the level of behavioral risk based on driver behavior data;

[0046] Emergency control measures are implemented on the target vehicle based on the level of behavioral risk.

[0047] The present invention also provides a vehicle emergency control system for abnormal driver behavior, comprising:

[0048] The data acquisition module is used to acquire source domain monitoring data collected from the target vehicle;

[0049] The data filling module is used to generate target domain monitoring data that matches the source domain monitoring data based on sample data. The target domain monitoring data is the type of data that the target vehicle cannot collect.

[0050] The behavior judgment module is used to input source domain monitoring data and target domain monitoring data into the preset behavior judgment model to obtain driver behavior data output by the preset behavior judgment model.

[0051] The emergency control module is used to perform emergency control on the target vehicle based on driver behavior data.

[0052] The beneficial effects of using the above embodiments are:

[0053] This invention provides a vehicle emergency control method and system for abnormal driver behavior. It first acquires source domain monitoring data collected from the target vehicle, then generates target domain monitoring data matching the source domain monitoring data based on sample data. Next, the source domain monitoring data and target domain monitoring data are input into a preset behavior judgment model to obtain driver behavior data output by the model. Finally, emergency control is performed on the target vehicle based on the driver behavior data. This invention utilizes a cross-domain data generation and unified model training mechanism to generate matching target domain data based on source domain monitoring data and uses a unified preset behavior judgment model for behavior recognition. This eliminates reliance on sensors such as cameras and radar, enabling older vehicle models to achieve abnormal driver behavior monitoring and emergency control without additional hardware, effectively solving the problems of high hardware requirements and poor compatibility in traditional solutions. Attached Figure Description

[0054] Figure 1 A flowchart of the vehicle emergency control method under abnormal driver behavior provided by the present invention;

[0055] Figure 2 for Figure 1 A detailed step diagram of step S106 is shown below;

[0056] Figure 3 for Figure 1 Another specific step diagram for step S106;

[0057] Figure 4 for Figure 1 Another specific step diagram of step S106;

[0058] Figure 5 The present invention provides a system structure diagram of a vehicle emergency control system for abnormal driver behavior. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Combination Figure 1 As shown, a specific embodiment of the present invention discloses a vehicle emergency control method under abnormal driver behavior, comprising:

[0061] S101. Obtain source domain monitoring data collected by the target vehicle;

[0062] S102. Generate target domain monitoring data that matches source domain monitoring data based on sample data, wherein the target domain monitoring data is the type of data that the target vehicle cannot collect;

[0063] S103. Input the source domain monitoring data and the target domain monitoring data into the preset behavior judgment model to obtain the driver behavior data output by the preset behavior judgment model;

[0064] S104. Based on driver behavior data, implement emergency control on the target vehicle.

[0065] In the above process, source domain monitoring data refers to the types of data that the target vehicle can collect, target domain monitoring data refers to the types of data that the target vehicle cannot collect, and sample data refers to pre-prepared, accurate data. In this invention, given the source domain monitoring data, the patterns represented by the sample data can be used to complete the target domain monitoring data. The preset behavior judgment model is a unified model for judging driver behavior. Through this scheme, when facing the safety control of some older vehicle models, it is possible to allow the vehicle model to collect only the data it can collect (i.e., source domain target data), and then automatically complete these data with target domain feature data to generate complete and reasonable data, thereby using the data from the preset behavior judgment model for behavior monitoring. It is understood that the above steps S102 and S103 can be executed in the cloud.

[0066] This embodiment generates matching target domain data based on source domain monitoring data through a cross-domain data generation and unified model training mechanism, and uses a unified preset behavior judgment model for behavior recognition. This eliminates the dependence on sensors such as cameras and radar, enabling older vehicle models to achieve abnormal driver monitoring and emergency control without the need for additional hardware. This effectively solves the problems of high hardware requirements and poor compatibility in traditional solutions.

[0067] Specifically, the target domain monitoring data can be implemented using any existing technology. For example, the most frequent values ​​in the sample data can be used as the target domain monitoring data to complete the data. Alternatively, the source domain monitoring data and the data in the sample data that are of the same type as the source domain monitoring data can be matched to select the data entry that is most similar to the source domain monitoring data, and the data entry that is of a different type from the source domain monitoring data can be used as the target domain monitoring data.

[0068] The present invention also provides a more preferred scheme for supplementing target domain monitoring data. In a new embodiment, the source domain monitoring data includes source domain time-series data, and the target domain monitoring data includes target domain time-series data. Based on this, the above step S102, generating target domain monitoring data matching the source domain monitoring data based on sample data, specifically includes:

[0069] The source domain time series data is input into the preset knowledge transfer model to obtain the target domain time series data output by the preset knowledge transfer model. The preset knowledge transfer model is trained based on sample data and is used to generate time series data that matches the input time series data but is of a different type according to the time series characteristics of the time series data.

[0070] In the above process, the pre-defined knowledge transfer model is trained using labeled sample data (complete data, including source and target domains) from other vehicle models or scenarios, transferring the learned temporal features to the target domain (older vehicle models with missing data). By capturing the long-term dependencies between the source and target domain temporal data through the pre-defined knowledge transfer model (e.g., a driver's accelerator pedal action is inevitably accompanied by a forward scan, and lane departure rate is determined by steering angle and vehicle speed), the generated data exhibits temporal coherence, avoiding the "behavioral discontinuities" caused by traditional random completion. The model enforces physical consistency between the source and target domain data through comparative learning (e.g., the covariance relationship between the rate of change of acceleration and the change of steering angle), ensuring that the generated target domain data conforms to vehicle dynamics laws without relying on a large amount of labeled data from real-world scenarios.

[0071] Understandably, abnormal driver behaviors (such as sudden deceleration due to illness) often exhibit temporal evolution characteristics. Time-series data can completely capture the start, development, and end of an action, while discrete data (such as single-frame heart rate values) easily loses crucial dynamic information. By analyzing the temporal correlation between source domain time-series data (such as steering wheel angle sequences) and target domain time-series data (such as eye movement trajectories), an implicit logical chain between driver intent and vehicle control can be established (such as "frequent phone use – steering wheel deviating from lane – emergency braking"), thereby improving the causal reasoning ability of behavior recognition.

[0072] Compared to other solutions, this embodiment improves data utilization efficiency by reusing limited data through generative models, and supports cross-domain generalization through temporal knowledge transfer. While ensuring the physical rationality of the generated data, it maximizes the utilization of existing data value. This not only solves the problem of behavior monitoring blind spots caused by the lack of sensors in traditional solutions, but also avoids the logical paradox risks that may be caused by pure data generation methods. It provides a reusable technical paradigm for the low-cost and high-precision implementation of driver status monitoring systems.

[0073] It is conceivable that the aforementioned pre-defined knowledge transfer model can be implemented using any existing machine learning model. However, this invention further provides a feasible embodiment. In a new embodiment, the pre-defined knowledge transfer model includes an encoder, a domain adaptation layer, and a decoder connected in sequence, wherein:

[0074] The encoder is used to extract the temporal features of the source domain temporal data and generate the first context vector;

[0075] The domain adaptation layer consists of multiple sequentially connected feedforward neural network layers, which are used to adjust the feature distribution of the first context vector to obtain the second context vector;

[0076] The decoder is used to generate target domain time-series data based on the second context vector.

[0077] In the above, the encoder is preferably implemented based on a Transformer architecture (such as Temporal FusionTransformer) or an LSTM network, capturing long-term dependencies in the source domain temporal data (such as acceleration trends over 10 consecutive seconds) through a multi-head attention mechanism. The input source domain temporal data (such as throttle opening sequences) is mapped to a first context vector, preserving temporal dynamic features (such as action acceleration) and implicit behavioral patterns (such as driver predictive actions) during the encoding process. The domain adaptation layer employs a multi-layer feedforward neural network stacked structure to force consistency in the feature space mapping between the target domain and the source domain, for example, eliminating cross-domain data distribution differences (such as domain offset between driver behavior data and vehicle dynamics parameters), ensuring that the generated target domain temporal data is physically strongly correlated with the source domain (such as the temporal coupling between rapid acceleration and gaze shift). The decoder module can also be implemented based on the inverse process of Transformer or the temporal prediction capability of LSTM. It can reconstruct the target domain temporal data (such as the driver's pupil diameter change curve) from the adjusted second context vector, generate a temporal sequence that conforms to the distribution of target domain data (such as the sampling rate of biological signals) while maintaining temporal continuity, and support multimodal output (such as generating eye movement trajectory and gesture action at the same time).

[0078] It is conceivable that the above embodiments are most effective when dealing with time-series data compensation. However, in practice, the target domain data often includes other non-time-series data. In this case, a combination of the various completion methods mentioned above can be used to complete the data.

[0079] Furthermore, if the above methods are deployed locally, the vehicle's infotainment system needs sufficiently high hardware computing power; if executed in the cloud, the vehicle's communication environment needs to be good. However, in reality, both conditions may not be met simultaneously when the vehicle is in motion. Therefore, to resolve the contradiction between insufficient computing power for local deployment and reliance on communication in the cloud in traditional solutions, the invention also provides a preferred embodiment. This embodiment further establishes a lightweight model based on a preset behavior judgment model to facilitate local deployment and reduce the hardware and environmental requirements for security testing.

[0080] Specifically, in a preferred embodiment, before step S104, which performs emergency control on the target vehicle based on driver behavior data, step S105 is further included, which involves inputting source domain monitoring data into a preset lightweight model to obtain driver behavior data output by the preset lightweight model. The preset lightweight model is obtained by knowledge distillation training on a preset behavior judgment model.

[0081] The above method utilizes knowledge distillation technology to train a lightweight model (i.e., a student model) based on a pre-set behavior judgment model (teacher model). This compresses the model size with almost no loss of the original model's performance, reducing inference latency and meeting the real-time operation requirements of automotive-grade chips (such as MCUs). Furthermore, compared to the pre-set behavior judgment model, the lightweight model only requires source domain monitoring data to make behavior judgments, making it more consistent with the vehicle's actual situation and resulting in faster inference speeds.

[0082] Knowledge distillation is a technique that compresses models through knowledge transfer between models. Essentially, it implicitly transfers the knowledge of a complex teacher model (a pre-defined behavior judgment model) to a student model (a lightweight model). Unlike traditional fine-tuning, knowledge distillation does not rely on a large amount of labeled data. Instead, it guides the student model's learning by mining the feature representations of the intermediate layers of the teacher model.

[0083] The general process of knowledge distillation is as follows: The same data is simultaneously input into both the teacher and student models. Based on the outputs of the teacher and student models and the corresponding ground truth labels, a loss function can be established to complete the training. Specifically, after obtaining the teacher model's output, it needs to be softmax normalized based on a preset temperature to obtain a soft label. Then, the student model's output is softmax normalized based on the same preset temperature to obtain a soft prediction. A soft loss function is calculated based on the soft label and the soft prediction. Next, the student model's output itself is used as a hard prediction, and the ground truth labels are used as hard labels. A hard loss function is calculated based on the hard labels and the hard prediction. Finally, the soft and hard loss functions are fused at a specific ratio to obtain a final loss function for training the student model, completing the knowledge distillation from the teacher model to the student model.

[0084] In this embodiment, achieving knowledge distillation of the lightweight model through a predictive behavior judgment model faces two challenges. First, in existing technologies, both the teacher and student models input the same data, while in this embodiment, the preset behavior judgment model and the lightweight model input different data. Second, building the lightweight model requires training with user source domain monitoring data to make it meaningful and enable personalized local deployment. However, as mentioned earlier, this system can only collect real source domain monitoring data, while the corresponding target domain monitoring data and driver status data are based on algorithmic reasoning and prediction. Therefore, during knowledge distillation, the system cannot determine the true labels corresponding to the source domain monitoring data.

[0085] Therefore, the present invention further provides a feasible knowledge distillation method that can solve the above problems. Specifically, in a preferred embodiment, the sample data includes source domain sample data and sample behavior data corresponding to the source domain sample data. The source domain sample data is sample data of the same type as the source domain monitoring data, and the sample behavior data is the actual driver behavior data corresponding to each sample data. Before step S105, inputting the source domain monitoring data into the preset lightweight model to obtain the driver behavior data output by the preset lightweight model, the vehicle emergency control method under abnormal driver behavior further includes step S106, training the preset lightweight model, and combining... Figure 2 As shown, this step specifically includes:

[0086] S201. Obtain the initial lightweight model;

[0087] S202. Obtain historical source domain monitoring data as the first training data;

[0088] S203. Obtain the historical driver behavior data corresponding to the first training data as the output result of the teacher model corresponding to the first training data.

[0089] S204. Calculate the similarity between the first training data and the source domain sample data, and select the sample behavior data corresponding to the source domain sample data with the highest similarity as the true label corresponding to the first training data.

[0090] S205. Input the first training data into the initial lightweight model to obtain the student model output result corresponding to the first training data;

[0091] S206. Based on the true labels corresponding to the first training data, the output results of the teacher model and the output results of the student model, establish the first loss function;

[0092] S207. Optimize the initial lightweight model according to the first loss function to obtain the trained preset lightweight model.

[0093] In the above process, the historical source domain monitoring data and historical driver behavior data are all existing data collected or predicted in the past. A lightweight model is trained using this historical source domain monitoring data to achieve personalization.

[0094] In this model, the first training data is input into the initial lightweight model, which naturally yields the corresponding student model output. The historical driver behavior data corresponding to the first training data is itself generated by the predictive behavior model, so it can be directly used as the teacher model output corresponding to the first training data, thus resolving the inconsistency between the teacher and student model inputs.

[0095] Furthermore, in reality, driver abnormal behavior data is mostly similar, and abnormal behaviors constitute a minority of all behavior types. For most normal behaviors, as long as the lightweight model determines that they are still within the scope of normal behavior, even if the specific behavior type is incorrectly evaluated, it will not affect the final control effect. Therefore, this embodiment also matches source domain sample data most similar to the first training data and approximates its sample behavior data as the true label, solving the problem of missing true labels corresponding to the first training data and achieving knowledge distillation.

[0096] Furthermore, the historical behavior data itself is derived from the inferred target domain monitoring data, which indicates that the output of the teacher model corresponding to the first training data may not be accurate enough. Therefore, it is best to use more realistic data for further training to improve the realism.

[0097] Combination Figure 3 As shown, in a preferred embodiment, the sample data further includes target domain sample data corresponding to the source domain sample data (the target domain sample data is the same type of sample data as the target domain monitoring data). Step S106, training the preset lightweight model, further includes:

[0098] S301. Select source domain sample data with similarity higher than a preset threshold as the second training data;

[0099] S302. Use the sample behavior data corresponding to the second training data as the true label corresponding to the second training data.

[0100] S303. Input the second training data and the target domain sample data corresponding to the second training data into the preset behavior judgment model to obtain the teacher model output result corresponding to the second training data.

[0101] S304. Input the second training data into the initial lightweight model to obtain the student model output corresponding to the second training data;

[0102] S305. Based on the true labels corresponding to the second training data, the output results of the teacher model and the output results of the student model, establish a second loss function;

[0103] S306. Optimize the initial lightweight model according to the second loss function to obtain the trained preset lightweight model.

[0104] Similarly, based on the similarity of different drivers' behaviors and the tolerance of the present invention for deviations in judging normal behavior, this embodiment further uses sample data as second training data to provide more realistic training data for the preset lightweight model in order to improve the model's accuracy.

[0105] In this model, the sample behavior data corresponding to the second training data naturally corresponds to its true labels. After the second training data is input into the initial lightweight model, the corresponding student model output result is naturally obtained. Furthermore, the second training data is essentially source domain sample data, which itself has corresponding real target domain sample data. Therefore, it can be input into the preset behavior judgment model to obtain the teacher model output result, thereby achieving knowledge distillation.

[0106] Furthermore, for the same reason, it's conceivable that in reality, driver behavior data representing abnormal behavior is ultimately a minority of data. The historical source domain monitoring data mentioned earlier as the first training data (since users mostly drive normally, the collected historical source domain monitoring data likely represents normal states) and the source domain sample data mentioned as the second training data (sample data similar to normal state data) may both lack data representing abnormal behavior. Therefore, to improve the generalization of the lightweight model, this invention also provides an embodiment. Combined with... Figure 4 As shown, in a new embodiment, step S106, training the preset lightweight model, further includes:

[0107] S401. Select source domain sample data that characterizes abnormal driver behavior as the third training data.

[0108] S402. Use the sample behavior data corresponding to the third training data as the true label corresponding to the third training data.

[0109] S403. Input the third training data and the target domain sample data corresponding to the third training data into the preset behavior judgment model to obtain the teacher model output result corresponding to the third training data.

[0110] S404. Input the third training data into the initial lightweight model to obtain the student model output corresponding to the third training data;

[0111] S405. Based on the true labels corresponding to the third training data, the output results of the teacher model and the output results of the student model, establish a third loss function;

[0112] S406. Optimize the initial lightweight model according to the third loss function to obtain the trained preset lightweight model.

[0113] In this embodiment, source domain sample data representing abnormal driver behavior was selected as the third training data, providing more diverse and realistic training samples for the initial lightweight model, enabling the lightweight model to learn the judgment logic of abnormal behavior, thereby improving its generalization.

[0114] In this model, the sample behavior data corresponding to the third training data naturally corresponds to its true labels. After the third training data is input into the initial lightweight model, the corresponding student model output result is naturally obtained. Furthermore, the third training data is essentially source domain sample data, which itself has corresponding true target domain sample data. Therefore, it can be input into the preset behavior judgment model to obtain the teacher model output result, thereby achieving knowledge distillation.

[0115] It is understood that the three optimization processes described in the above three embodiments can be performed separately, that is, the initial lightweight model can be optimized by backpropagation using the first loss function, the second loss function and the third loss function respectively, or the three loss functions can be fused and backpropagation optimization can be performed simultaneously to improve training efficiency.

[0116] Specifically, in a preferred embodiment, step S106, training the preset lightweight model, further includes:

[0117] The first loss function, the second loss function, and the third loss function are weighted and summed to obtain the total loss function.

[0118] The initial lightweight model is optimized based on the total loss function to obtain the pre-trained lightweight model.

[0119] The weight ratios of the first, second, and third loss functions can be reasonably adjusted based on the proportion of the target domain monitoring data.

[0120] Furthermore, in a preferred embodiment, step S104, which involves emergency control of the target vehicle based on driver behavior data, specifically includes:

[0121] Determine the level of behavioral risk based on driver behavior data;

[0122] Emergency control measures are implemented on the target vehicle based on the level of behavioral risk.

[0123] This embodiment achieves efficient coordination between monitoring abnormal driver behavior and emergency control by constructing a dynamic hierarchical response mechanism. For example, in the risk level determination stage, driver behavior data (such as distraction index and fatigue level score) output by a preset behavior judgment model can be combined with a dynamic threshold algorithm to classify risks into three levels: low (0-30 points), medium (31-70 points), and high (71-100 points). These levels trigger a first-level warning (such as voice prompts), a second-level intervention (such as reducing power), and a third-level emergency control (such as emergency braking), respectively, ensuring that the intervention intensity is accurately matched with the risk level.

[0124] Combination Figure 5 As shown, the present invention also provides a vehicle emergency control system for abnormal driver behavior, comprising:

[0125] The data acquisition module 510 is used to acquire source domain monitoring data collected by the target vehicle;

[0126] The data filling module 520 is used to generate target domain monitoring data that matches the source domain monitoring data based on the sample data. The target domain monitoring data is the type of data that the target vehicle cannot collect.

[0127] The behavior judgment module 530 is used to input source domain monitoring data and target domain monitoring data into a preset behavior judgment model to obtain driver behavior data output by the preset behavior judgment model.

[0128] The emergency control module 540 is used to perform emergency control on the target vehicle based on driver behavior data.

[0129] Furthermore, in one embodiment, the vehicle emergency control system for abnormal driver behavior further includes:

[0130] The data learning module is used to train a pre-defined lightweight model;

[0131] The rapid judgment module is used to input source domain monitoring data into a preset lightweight model to obtain driver behavior data output by the preset lightweight model. The preset lightweight model is obtained by knowledge distillation training on a preset behavior judgment model.

[0132] It should be noted that the corresponding systems provided in the above embodiments can implement the technical solutions described in the above method embodiments. The specific implementation principles of the above modules or units can be found in the corresponding content in the above method embodiments, and will not be repeated here.

[0133] This invention provides a vehicle emergency control method and system for abnormal driver behavior. It first acquires source domain monitoring data collected from the target vehicle, then generates target domain monitoring data matching the source domain monitoring data based on sample data. Next, the source domain monitoring data and target domain monitoring data are input into a preset behavior judgment model to obtain driver behavior data output by the model. Finally, emergency control is performed on the target vehicle based on the driver behavior data. This invention utilizes a cross-domain data generation and unified model training mechanism to generate matching target domain data based on source domain monitoring data and uses a unified preset behavior judgment model for behavior recognition. This eliminates reliance on sensors such as cameras and radar, enabling older vehicle models to achieve abnormal driver behavior monitoring and emergency control without additional hardware, effectively solving the problems of high hardware requirements and poor compatibility in traditional solutions.

[0134] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0135] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for emergency control of a vehicle in case of abnormal behavior of a driver, characterized in that, The method comprises the following steps: acquiring source domain monitoring data collected by a target vehicle, the source domain monitoring data comprising source domain time series data; inputting the source domain time series data into a preset knowledge transfer model to obtain target domain time series data output by the preset knowledge transfer model, wherein the preset knowledge transfer model is trained based on sample data and is used to generate time series data of different categories that matches the input time series data according to the time series characteristics of the time series data; the sample data is accurate data prepared in advance; the target domain monitoring data is a data category that cannot be collected by the target vehicle, and the target domain monitoring data comprises target domain time series data; inputting the source domain monitoring data and the target domain monitoring data into a preset behavior judgment model to obtain driver behavior data output by the preset behavior judgment model; performing emergency control on the target vehicle based on the driver behavior data.

2. The method of claim 1, wherein The preset knowledge transfer model comprises an encoder, a domain adaptation layer and a decoder connected in sequence, wherein: the encoder is used to extract time series characteristics of the source domain time series data and generate a first context vector; the domain adaptation layer comprises a plurality of feedforward neural network layers connected in sequence and is used to adjust the feature distribution of the first context vector to obtain a second context vector; the decoder is used to generate the target domain time series data according to the second context vector.

3. The driver unusual behavior under vehicle emergency control method according to claim 1, characterized in that, Before the step of performing emergency control on the target vehicle based on the driver behavior data, the method further comprises: inputting the source domain monitoring data into a preset lightweight model to obtain driver behavior data output by the preset lightweight model, wherein the preset lightweight model is obtained by performing knowledge distillation training on the preset behavior judgment model.

4. The method of claim 3, wherein The sample data comprises source domain sample data and sample behavior data corresponding to the source domain sample data; before the step of inputting the source domain monitoring data into the preset lightweight model to obtain driver behavior data output by the preset lightweight model, the method further comprises training the preset lightweight model, specifically comprising: acquiring an initial lightweight model; acquiring historical source domain monitoring data as first training data; acquiring historical driver behavior data corresponding to the first training data as an output result of a teacher model corresponding to the first training data; calculating the similarity of the first training data and the source domain sample data, and selecting sample behavior data corresponding to the source domain sample data with the highest similarity as a real label corresponding to the first training data; inputting the first training data into the initial lightweight model to obtain a student model output result corresponding to the first training data; establishing a first loss function according to the real label corresponding to the first training data, the teacher model output result and the student model output result; optimizing the initial lightweight model according to the first loss function to obtain a trained preset lightweight model.

5. The method for vehicle emergency control in the event of a driver abnormal behavior according to claim 3, characterized in that, The sample data comprises source domain sample data and sample behavior data corresponding to the source domain sample data, and the sample data further comprises target domain sample data corresponding to the source domain sample data; before the step of inputting the source domain monitoring data into the preset lightweight model to obtain driver behavior data output by the preset lightweight model, the method further comprises training the preset lightweight model, specifically comprising: select source domain sample data with a similarity higher than a preset threshold as second training data; sample behavior data corresponding to the second training data as a true label corresponding to the second training data; input the second training data and target domain sample data corresponding to the second training data into the preset behavior judgment model to obtain a teacher model output result corresponding to the second training data; input the second training data into the initial lightweight model to obtain a student model output result corresponding to the second training data; establish a second loss function according to the true label corresponding to the second training data, the teacher model output result and the student model output result; optimize the initial lightweight model according to the second loss function to obtain a preset lightweight model trained.

6. The driver unusual behavior under the vehicle emergency control method according to claim 3, characterized in that, The sample data includes source domain sample data and sample behavior data corresponding to the source domain sample data, and the sample data further includes target domain sample data corresponding to the source domain sample data. Before the step of inputting the source domain monitoring data into the preset lightweight model to obtain driver behavior data output by the preset lightweight model, the method further includes training the preset lightweight model, specifically including: select source domain sample data representing abnormal behavior of a driver as third training data; sample behavior data corresponding to the third training data as a true label corresponding to the third training data; input the third training data and target domain sample data corresponding to the third training data into the preset behavior judgment model to obtain a teacher model output result corresponding to the third training data; input the third training data into the initial lightweight model to obtain a student model output result corresponding to the third training data; establish a third loss function according to the true label corresponding to the third training data, the teacher model output result and the student model output result; optimize the initial lightweight model according to the third loss function to obtain a preset lightweight model trained.

7. The driver unusual behavior under the vehicle emergency control method according to claim 3, characterized in that, The sample data includes source domain sample data and sample behavior data corresponding to the source domain sample data, and the sample data further includes target domain sample data corresponding to the source domain sample data. Before the step of inputting the source domain monitoring data into the preset lightweight model to obtain driver behavior data output by the preset lightweight model, the method further includes training the preset lightweight model, specifically including: obtain an initial lightweight model; obtain historical source domain monitoring data as first training data; obtain historical driver behavior data corresponding to the first training data as a teacher model output result corresponding to the first training data; calculate a similarity of the first training data and source domain sample data, and select sample behavior data corresponding to source domain sample data with the highest similarity as a true label corresponding to the first training data; input the first training data into the initial lightweight model to obtain a student model output result corresponding to the first training data; establish a first loss function according to the true label corresponding to the first training data, the teacher model output result and the student model output result; select source domain sample data with a similarity higher than a preset threshold as second training data; sample behavior data corresponding to the second training data as a true label corresponding to the second training data; The second training data and the target domain sample data corresponding to the second training data are input into the preset behavior judgment model to obtain a teacher model output result corresponding to the second training data; The second training data are input into the initial lightweight model to obtain a student model output result corresponding to the second training data; The second loss function is established according to the real label corresponding to the second training data, the teacher model output result and the student model output result; The source domain sample data representing the abnormal behavior of the driver are selected as the third training data; The sample behavior data corresponding to the third training data is used as the real label corresponding to the third training data; The third training data and the target domain sample data corresponding to the third training data are input into the preset behavior judgment model to obtain a teacher model output result corresponding to the third training data; The third training data are input into the initial lightweight model to obtain a student model output result corresponding to the third training data; The third loss function is established according to the real label corresponding to the third training data, the teacher model output result and the student model output result; The first loss function, the second loss function and the third loss function are weighted and summed to obtain a total loss function; The initial lightweight model is optimized according to the total loss function to obtain a trained preset lightweight model.

8. The driver unusual behavior under the vehicle emergency control method according to claim 1, characterized in that, Based on the driver behavior data, the target vehicle is subjected to emergency control, including: According to the driver behavior data, the behavior risk level is judged; According to the behavior risk level, the target vehicle is subjected to emergency control.

9. A vehicle emergency control system under abnormal behavior of a driver, characterized by, It includes: The data acquisition module is used to acquire the source domain monitoring data collected by the target vehicle, and the source domain monitoring data includes source domain time series data; The data filling module is used to input the source domain time series data into the preset knowledge transfer model to obtain the target domain time series data output by the preset knowledge transfer model, wherein the preset knowledge transfer model is trained based on sample data and is used to generate time series data of different categories matched with the input time series data according to the time series characteristics of the time series data; the sample data is accurate data prepared in advance; the target domain monitoring data is a data category that cannot be collected by the target vehicle, and the target domain monitoring data includes target domain time series data; The behavior judgment module is used to input the source domain monitoring data and the target domain monitoring data into the preset behavior judgment model to obtain the driver behavior data output by the preset behavior judgment model; The emergency control module is used to control the target vehicle based on the driver behavior data.

Images