Driver identity recognition method, device, equipment and medium

Abstract

The invention discloses a driver identity recognition method and device, equipment and a medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining the driving behavior data of a current driver in real time in the operation process of a target vehicle; performing time sequence analysis and frequency domain analysis on the driving behavior data to extract time sequence features and frequency domain features, and performing target driving event identification on the driving behavior data to extract event features; inputting the time sequence characteristics into a first preset network model to obtain a first output result, and inputting the frequency domain characteristics and the event characteristics into a second preset network model to obtain a second output result; and fusing the first output result and the second output result to obtain a first fusion result, and performing identity recognition on the current driver based on the first fusion result. According to the invention, the driver identity can be identified more intelligently and accurately on the premise of protecting the privacy of the driver.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a driver identification method, device, equipment and medium. Background Art

[0002] In the modern taxi and ride-hailing industries, driver replacement is a long-standing safety hazard that urgently needs to be addressed. Replacement drivers often lack legal driving qualifications, and their irregular driving behavior can easily lead to passenger accidents while exposing operating companies to severe legal risks and financial losses.

[0003] Currently, the traditional solutions to the shift replacement problem in the industry mainly include the following two methods:

[0004] Video surveillance recognition: This method uses onboard cameras to monitor the driver's face for identification. However, this approach not only risks violating the driver's privacy but also faces challenges such as high equipment costs and complex management. For example, the installation and maintenance of a large number of cameras requires significant investment, while storing and analyzing massive amounts of video data places high demands on both technology and manpower.

[0005] Password or card verification: Authentication is performed by the driver entering a password or swiping a driver's card. However, these methods have obvious flaws. Passwords can be easily leaked, forged, or lent, and driver cards can be misused by others, making it difficult to effectively guarantee the accuracy and reliability of identity verification.

[0006] In summary, how to identify the driver's identity more intelligently and accurately while protecting privacy is a problem that needs to be solved. Summary of the Invention

[0007] In view of this, the purpose of the present invention is to provide a driver identification method, device, equipment and medium that can more intelligently and accurately identify the driver's identity while protecting privacy. The specific solution is as follows:

[0008] In a first aspect, the present application discloses a driver identification method, comprising:

[0009] Acquire the current driver's driving behavior data in real time during the operation of the target vehicle;

[0010] performing time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features, and identifying target driving events on the driving behavior data to extract event features;

[0011] Inputting the time series features into a first preset network model to obtain a first output result, and inputting the frequency domain features and the event features into a second preset network model to obtain a second output result;

[0012] The first output result and the second output result are fused to obtain a first fusion result, and the identity of the current driver is identified based on the first fusion result.

[0013] Optionally, performing time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features respectively includes:

[0014] The first preset network model is used to perform time series analysis on the driving behavior data to extract time series features, and the Fourier transform function is used to perform frequency domain analysis on the driving behavior data to extract frequency domain features; wherein, the first preset network model is a long short-term memory network model, and the second preset network model is a Transformer model.

[0015] Optionally, the driving behavior data includes acceleration, braking force, and steering wheel angle; the target driving event includes a sudden acceleration event, a sudden braking event, and a sudden turn event;

[0016] Accordingly, the identifying of target driving events from the driving behavior data to extract event features includes:

[0017] If the acceleration exceeds a preset acceleration threshold, it is determined that a sudden acceleration event currently exists, and event features corresponding to the sudden acceleration event are extracted;

[0018] If the braking force exceeds a preset braking force threshold, it is determined that the sudden braking event currently occurs, and event features corresponding to the sudden braking event are extracted;

[0019] If the steering wheel steering angle exceeds a preset steering angle threshold, it is determined that the sharp turn event currently exists, and event features corresponding to the sharp turn event are extracted.

[0020] Optionally, the fusing the first output result and the second output result to obtain a first fusion result includes:

[0021] Normalizing the first output result and the second output result respectively using a preset normalization function to obtain corresponding first processing results and second processing results;

[0022] The first processing result and the second processing result are vector-joined using a target weight parameter to obtain a first fusion result.

[0023] Optionally, identifying the current driver based on the first fusion result includes:

[0024] Inputting the first fusion result into a pre-trained fully connected network to output an identity probability using the fully connected network; wherein the fully connected network is constructed based on a preset weight matrix, a preset bias term, and a preset activation function, and the identity probability is a probability that the current driver is a legal driver associated with the target vehicle;

[0025] If the identity probability is higher than a preset probability threshold, the current driver is determined to be the legal driver; otherwise, the current driver is determined not to be the legal driver and an abnormality alarm is issued.

[0026] Optionally, identifying the current driver based on the first fusion result includes:

[0027] Determining a legitimate driver associated with the target vehicle, and determining a second fusion result corresponding to the historical driving behavior data of the legitimate driver;

[0028] Calculating the similarity between the first fusion result and the second fusion result;

[0029] If the similarity is lower than a preset similarity threshold, it is determined that the current driver is not the legal driver and an abnormality alarm is issued; otherwise, it is determined that the current driver is the legal driver.

[0030] Optionally, identifying the current driver based on the first fusion result includes:

[0031] Inputting the first fusion result into a pre-trained fully connected network to output an identity probability using the fully connected network; wherein the fully connected network is constructed based on a preset weight matrix, a preset bias term, and a preset activation function, and the identity probability is a probability that the current driver is a legal driver associated with the target vehicle;

[0032] If the identity probability is higher than a preset probability threshold, determining that the current driver is the legal driver;

[0033] If the identity probability is not higher than a preset probability threshold, determining a second fusion result corresponding to the historical driving behavior data of the legal driver, and calculating the similarity between the first fusion result and the second fusion result;

[0034] If the similarity is lower than a preset similarity threshold, it is determined that the current driver is not the legal driver and an abnormality alarm is issued; otherwise, it is determined that the current driver is the legal driver.

[0035] In a second aspect, the present application discloses a driver identification device, comprising:

[0036] A data acquisition module is used to obtain the current driver's driving behavior data in real time during the operation of the target vehicle;

[0037] a feature extraction module, configured to perform time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features, and identify target driving events on the driving behavior data to extract event features;

[0038] a model processing module, configured to input the time series features into a first preset network model to obtain a first output result, and input the frequency domain features and the event features into a second preset network model to obtain a second output result;

[0039] An identity recognition module is used to fuse the first output result and the second output result to obtain a first fusion result, and to perform identity recognition on the current driver based on the first fusion result.

[0040] In a third aspect, the present application discloses an electronic device, comprising:

[0041] Memory, used to store computer programs;

[0042] A processor is used to execute the computer program to implement the steps of the driver identification method disclosed above.

[0043] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the steps of the aforementioned driver identification method are implemented.

[0044] It can be seen that the present application obtains the driving behavior data of the current driver in real time during the operation of the target vehicle; performs time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features, and identifies target driving events on the driving behavior data to extract event features; inputs the time series features into a first preset network model to obtain a first output result, and inputs the frequency domain features and the event features into a second preset network model to obtain a second output result; fuses the first output result and the second output result to obtain a first fusion result, and identifies the current driver based on the first fusion result.

[0045] Beneficial effects: The present application obtains the driving behavior data of the current driver in real time during the operation of the target vehicle, and simultaneously performs time series analysis, frequency domain analysis and target driving event recognition on the driving behavior data to extract corresponding time series features, frequency domain features and event features for subsequent identity recognition. Compared with single-dimensional features, the present application can more comprehensively characterize the driver's unique behavior pattern by extracting multi-dimensional features, significantly improving the accuracy and robustness of identity recognition. Furthermore, the present application uses a first preset network model and a second preset network model to process the above-mentioned extracted multi-dimensional features in parallel, specifically using the first preset network model to process the time series features, and using the second preset network model to process the frequency domain features and event features, and then fuses the output results of the two network models to obtain a first fusion result, and finally performs identity recognition on the current driver based on the first fusion result. That is, the present application performs identity recognition by fusing the output results of the first preset network model and the second preset network model, which can comprehensively consider time series features and complex behavior features, enhance the model's recognition ability for complex driving behaviors, avoid errors caused by a single model, and ensure accurate judgment of the driver's identity. In addition, this application does not need to rely on means involving driver privacy such as video surveillance, but only realizes identity recognition through their driving behavior data, thus protecting the driver's privacy. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0047] Figure 1 A flowchart of a driver identification method disclosed in this application;

[0048] Figure 2 This is a system architecture diagram for driver identification disclosed in this application;

[0049] Figure 3 A schematic diagram of feature extraction disclosed in this application;

[0050] Figure 4 This is a flowchart of a specific driver identification method disclosed in this application;

[0051] Figure 5 A schematic diagram of a result fusion process disclosed in this application;

[0052] Figure 6 A flowchart of a driver identity verification process disclosed in this application;

[0053] Figure 7 This is a schematic structural diagram of a driver identification device disclosed in this application;

[0054] Figure 8 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] Currently, the traditional solutions to the shift replacement problem in the industry mainly include video surveillance identification and password or card verification. Among them, video surveillance identification not only has the risk of infringing the driver's privacy, but also faces difficulties such as high equipment costs and complex post-management. In password or card verification, the password is easily leaked, forged or lent, and the driver's card may also be misused by others, making it difficult to effectively guarantee the accuracy and reliability of identity verification. To this end, the embodiments of the present application disclose a driver identity recognition method, device, equipment and medium, which can more intelligently and accurately identify the driver's identity while protecting privacy.

[0057] See also Figure 1 As shown, the embodiment of the present application discloses a driver identification method, which includes:

[0058] Step S11: Acquire the current driver's driving behavior data in real time during the operation of the target vehicle.

[0059] In this embodiment, the current driver's driving behavior data is acquired in real time during the operation of the target vehicle. In a specific embodiment, the current driver's driving behavior data can be collected from vehicle-mounted sensors, and the driving behavior data specifically includes but is not limited to acceleration, braking force, steering wheel angle, vehicle speed, GPS location, and other data.

[0060] It is understandable that if Figure 2The vehicle-mounted sensors shown may specifically include, but are not limited to, accelerometers, brake sensors, steering wheel speed sensors, GPS modules, and gyroscopes. The accelerometer is used to monitor the vehicle's acceleration and deceleration in real time. This sensor can capture changes in acceleration during different driving phases (e.g., acceleration, braking, and constant speed), providing important data for assessing the driver's driving habits. The brake sensor is used to monitor braking force and frequency, analyzing the driver's braking habits and methods. By monitoring braking force and reaction time, it can identify irregular driver behaviors such as sudden braking or acceleration. The steering wheel speed sensor is used to detect the speed and frequency of the driver's steering wheel operation. This sensor can identify the driver's steering habits, including sharp turns and frequent lane changes. The GPS module provides real-time vehicle location data and records the vehicle's driving trajectory. Combined with GPS location data, the system can further analyze the vehicle's geographic environment and regional characteristics of driving behavior. The gyroscope detects the vehicle's rotation angle to monitor steering operations and lane changes. Combined with acceleration and braking data, the gyroscope can more accurately reflect the driver's driving pattern.

[0061] In a specific implementation, data from vehicle-mounted sensors can be read regularly to ensure real-time and continuous data collection. The collection frequency needs to be adjusted according to the performance requirements of different sensors. Generally, the collection frequency of acceleration sensors and brake sensors should be above 100 Hz, while the collection frequency of GPS modules is generally 1 Hz.

[0062] It should also be pointed out that in addition to obtaining acceleration, braking force, and steering wheel angle from corresponding sensors, they can also be obtained by parsing vehicle CAN (Controller Area Network) data.

[0063] Furthermore, the driving behavior data collected by the on-board sensors is transmitted to the cloud server in real time through the wireless communication module for processing and storage, ensuring the timeliness and accuracy of the data. Among them, the wireless communication module supports communication protocols such as 4G / 5G or Wi-Fi, which can ensure the stability of data transmission in different environments. In addition, the cloud server in this application can store a large amount of driving behavior data and provide data support for model training. The server is also responsible for real-time analysis and storage of the data.

[0064] like Figure 2As shown, the embodiments of the present application integrate a computing unit, such as an NPU (Neural-network Processing Unit), into the vehicle-mounted device to process driving behavior data and perform driving behavior analysis in real time, thereby verifying and identifying the driver's identity in real time and providing feedback. In other words, the driver identification method of the present application can be applied to the computing unit.

[0065] Step S12: performing time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features, and identifying target driving events on the driving behavior data to extract event features.

[0066] In this embodiment, after acquiring the current driver's driving behavior data, the data is simultaneously subjected to time series analysis, frequency domain analysis, and target driving event recognition to extract corresponding time series features, frequency domain features, and event features for subsequent identity recognition. Compared to single-dimensional features, this application extracts multi-dimensional features to more comprehensively characterize the driver's unique behavior patterns, significantly improving the accuracy and robustness of identity recognition.

[0067] It is understandable that this embodiment is provided with a feature extraction module for extracting features that are helpful for identifying the driver's identity from the original data, specifically Figure 3 As shown in the figure, the feature extraction module can be divided into three submodules: time series feature extraction module, frequency domain feature extraction module and event feature extraction module. Among them, the time series feature extraction module mainly processes the collected continuous driving behavior data and extracts behavioral features in the time series, such as dynamic behaviors such as acceleration, braking, and steering. These features can capture the driver's behavior patterns and driving habits, helping the system to identify the driver's identity. The frequency domain feature extraction module is mainly used to analyze the frequency domain features of driving behavior data. The frequency domain features help capture the periodicity and high-frequency changes in driving behavior, further improving the system's recognition accuracy of driver behavior. In particular, when capturing high-frequency changes such as the driver's acceleration and deceleration, braking frequency, and steering frequency, the frequency domain features provide effective support. The event feature extraction module focuses on extracting target driving events during the driving process. These events are an important basis for judging the driver's behavior and can further determine whether the driver is a legal driver.

[0068] Step S13: inputting the time series features into a first preset network model to obtain a first output result, and inputting the frequency domain features and the event features into a second preset network model to obtain a second output result.

[0069] In an embodiment of the present application, a first preset network model and a second preset network model are used to process the multi-dimensional features extracted above in parallel. Specifically, the first preset network model is used to process the time series features to obtain a first output result, and the second preset network model is used to process the frequency domain features and event features to obtain a second output result.

[0070] Step S14: fusing the first output result and the second output result to obtain a first fusion result, and performing identity recognition on the current driver based on the first fusion result.

[0071] In this embodiment, the output results of the first preset network model and the second preset network model are fused to obtain a first fusion result, and finally, the current driver's identity is identified based on the first fusion result. That is, by fusing the output results of the first preset network model and the second preset network model for identity identification, the present application can comprehensively consider temporal characteristics and complex behavioral characteristics, enhance the model's ability to identify complex driving behaviors, avoid errors caused by a single model, and ensure accurate judgment of the driver's identity. In addition, the present application does not rely on methods that involve driver privacy, such as video surveillance, but rather achieves identity identification solely through their driving behavior data, thus protecting the driver's privacy.

[0072] It is understandable that the purpose of this application to identify the current driver is to determine whether the current driver is a legal driver. If the current driver is not a legal driver, it means that a driver replacement situation has occurred. Therefore, an abnormal alarm signal will be issued and the management platform will be notified, and then it will be handled manually. Figure 2 As shown in .

[0073] As can be seen, the present application obtains the current driver's driving behavior data in real time during the operation of the target vehicle, and simultaneously performs time series analysis, frequency domain analysis, and target driving event recognition on the driving behavior data to extract corresponding time series features, frequency domain features, and event features for subsequent identity recognition. Compared with single-dimensional features, the present application can more comprehensively characterize the driver's unique behavior pattern by extracting multi-dimensional features, significantly improving the accuracy and robustness of identity recognition. Furthermore, the present application uses a first preset network model and a second preset network model to process the above-mentioned extracted multi-dimensional features in parallel. Specifically, the first preset network model is used to process the time series features, and the second preset network model is used to process the frequency domain features and event features. The output results of the two network models are then fused to obtain a first fusion result. Finally, the current driver's identity is recognized based on the first fusion result. That is, the present application performs identity recognition by fusing the output results of the first preset network model and the second preset network model. It can comprehensively consider time series features and complex behavior features, enhance the model's recognition ability for complex driving behaviors, avoid errors caused by a single model, and ensure accurate judgment of the driver's identity. In addition, this application does not need to rely on means involving driver privacy such as video surveillance, but only realizes identity recognition through their driving behavior data, thus protecting the driver's privacy.

[0074] See also Figure 4 As shown, the embodiment of the present application discloses a specific driver identification method. Compared with the previous embodiment, this embodiment further illustrates and optimizes the technical solution. Specifically, it includes:

[0075] Step S21: Acquire the current driver's driving behavior data in real time during the operation of the target vehicle.

[0076] Step S22: Using a first preset network model to perform time series analysis on the driving behavior data to extract time series features, using a Fourier transform function to perform frequency domain analysis on the driving behavior data to extract frequency domain features, and identifying target driving events on the driving behavior data to extract event features.

[0077] In this embodiment, a first preset network model can be used to extract temporal features of driving behavior data. Specifically, the first preset network model can be a long short-term memory (LSTM) network model. Specifically, the temporal feature extraction module uses the LSTM network model to process continuous driving behavior data, extracting time series features, primarily analyzing dynamic behaviors such as acceleration, braking, and steering. By processing temporal data such as acceleration, braking force, and steering using the LSTM model, the LSTM can capture long-term dependencies in the data and is particularly well-suited for processing time-dependent driving behaviors. This allows the extraction of characteristic data such as acceleration and deceleration patterns, braking force and duration, and the frequency and amplitude of steering operations.

[0078] Furthermore, the frequency domain feature extraction module can specifically use Fourier transforms to perform frequency domain analysis on acceleration, braking force, vehicle speed, steering, and other data. By extracting frequency domain features, the system can capture periodic changes and high-frequency fluctuations in driving behavior. This embodiment converts time domain signals into frequency domain signals and uses Fourier transforms to analyze the frequency domain features of key driving behaviors, such as steering frequency and acceleration / deceleration frequency. Through frequency domain analysis, the system can identify whether the driver has frequent changes in driving behavior, helping to determine the driver's driving habits, thereby extracting characteristic values ​​such as steering frequency, acceleration / deceleration frequency, and braking frequency.

[0079] Furthermore, the driving behavior data includes acceleration, braking force, and steering wheel steering angle; the target driving events include sudden acceleration events, sudden braking events, and sharp turn events; accordingly, the identification of target driving events on the driving behavior data to extract event features includes: if the acceleration exceeds a preset acceleration threshold, it is determined that the sudden acceleration event currently exists, and the event features corresponding to the sudden acceleration event are extracted; if the braking force exceeds the preset braking force threshold, it is determined that the sudden braking event currently exists, and the event features corresponding to the sudden braking event are extracted; if the steering wheel steering angle exceeds the preset steering angle threshold, it is determined that the sharp turn event currently exists, and the event features corresponding to the sharp turn event are extracted.

[0080] Specifically, the event feature extraction module is primarily responsible for identifying and extracting target driving events from the driving process, such as sudden acceleration, sudden braking, and sharp turns. These events are crucial for determining driver behavior. In a specific implementation, thresholds can be set to detect rapid changes in acceleration, braking force, and steering angle, identifying abnormal behavior events such as sudden acceleration, sudden braking, and sharp turns. The extracted event features are then passed as input to the subsequent network model for identification.

[0081] Specifically, when the acceleration exceeds the preset acceleration threshold, it is determined that a sudden acceleration event currently exists, and the event features corresponding to the sudden acceleration event are extracted; when the braking force exceeds the preset braking force threshold, it is determined that a sudden braking event currently exists, and the event features corresponding to the sudden braking event are extracted; when the steering wheel steering angle exceeds the preset steering angle threshold, it is determined that a sharp turn event currently exists, and the event features corresponding to the sharp turn event are extracted.

[0082] Step S23: inputting the time series features into a first preset network model to obtain a first output result, and inputting the frequency domain features and the event features into a second preset network model to obtain a second output result.

[0083] In this embodiment, the first preset network model is specifically a long short-term memory network model (LSTM model), and the second preset network model is specifically a Transformer model. Specifically, this application inputs the extracted multi-dimensional feature data into a machine learning model for identity verification, specifically using a hybrid LSTM and Transformer model to process the data, and then uses the model's output to determine whether the current driver is a legitimate driver.

[0084] The LSTM model is designed to capture long-term dependencies in time series data and is particularly well-suited for analyzing the temporal characteristics of driving behavior (e.g., continuous operations like acceleration and deceleration). Its input is the temporal features processed by the temporal feature extraction module. The LSTM model learns the driver's behavioral patterns during driving (e.g., acceleration, deceleration, braking, steering, etc.) and outputs results related to the driver's identity. It can identify long-term behavioral characteristics associated with the driver and determine whether they match the behavioral patterns of a known legitimate driver.

[0085] The Transformer model, through its multi-head attention mechanism, is able to better capture complex behavioral features, especially with higher sensitivity when processing diverse types of driving behaviors. The Transformer model takes data from the frequency domain feature extraction module and the event feature extraction module as input. Using the multi-head attention mechanism, it weights different features to extract complex patterns in driving behavior. This allows it to analyze complex driving behavior patterns, increasing sensitivity to complex or irregular driving behaviors, thereby improving identity verification accuracy.

[0086] It should be noted that there are two levels of LSTM processing in the entire execution process of the embodiment of the present application, but the purposes are different. The task of the first-level LSTM is to extract abstract time series feature vectors from the original time series data, which can be understood as compressing / encoding the original, high-dimensional, noisy sensor data stream into a low-dimensional, meaningful feature vector that can represent the time series pattern of driving behavior during this period; its input is the original or simply filtered sensor time series data, and the output is a time series feature vector (for example, 128 dimensions). The task of the second-level LSTM is to learn the mapping relationship between these time series patterns and specific driver identities based on the time series feature vectors extracted by the first level. It receives time series features that have been refined by the first-level LSTM and represent driving habits, that is, its input is the time series feature vector output by the first level, and the output is the predicted contribution of the LSTM branch to the driver's identity.

[0087] Step S24: using a preset normalization function to perform normalization processing on the first output result and the second output result respectively, so as to obtain corresponding first processing results and second processing results.

[0088] In this embodiment, the LSTM model obtains the first output result by processing the time series features, and is recorded as ,in, represents the set of real numbers, is the LSTM hidden layer dimension, for example, 128 dimensions. This vector captures the long-term dynamic patterns of driving behavior, such as acceleration habits and changes in braking force. This embodiment uses a multi-layer LSTM (e.g., 3 layers) and dropout (ratio = 0.2) to prevent overfitting.

[0089] The Transformer model processes frequency domain features (such as steering frequency, vehicle speed spectrum) and event features (such as emergency brake signs) through a multi-head attention mechanism, and outputs a global context feature vector , and record it as the second output result, where is the embedding dimension, for example, it can be 64 dimensions. The vector reflects the complex correlation of driving behavior, such as the pattern of sharp turns and braking linkage.

[0090] In this embodiment, Figure 5 As shown in the figure, the output results of the LSTM model and the Transformer model need to be layer normalized to eliminate the feature scale differences, which can be expressed as follows:

[0091] ;

[0092] ;

[0093] in, and are the first processing result and the second processing result after normalization, respectively, and LayerNorm represents the normalization function.

[0094] Step S25: performing vector splicing on the first processing result and the second processing result using the target weight parameter to obtain a first fusion result, and performing identity recognition on the current driver based on the first fusion result.

[0095] In this embodiment, Figure 5 As shown, it is also necessary to use the learnable target weight parameters and Dynamically adjust the importance of the two branches (i.e., the LSTM branch and the Transformer branch), perform dynamic weight allocation, and then perform vector splicing on the first processing result and the second processing result to obtain the first fusion result, as follows:

[0096] ;

[0097] in, is the fused feature vector, i.e. the first fusion result; 、 ,and , and The initial value of is set to 0.5 and optimized by back propagation; Represents a vector concatenation operation, and the feature dimension after fusion is d1+d2.

[0098] That is, this application fuses the output results of the LSTM model and the Transformer model, and adopts a weighted fusion method. The system can comprehensively consider the timing characteristics and complex behavior characteristics to perform more accurate identity authentication, thereby enhancing the model's ability to recognize complex driving behaviors, avoiding the errors caused by a single model, and ensuring accurate judgment of the driver's identity.

[0099] In a first specific embodiment, the identity identification of the current driver based on the first fusion result includes: inputting the first fusion result into a pre-trained fully-connected network to output an identity probability using the fully-connected network; wherein the fully-connected network is constructed based on a preset weight matrix, a preset bias term and a preset activation function, and the identity probability is the probability that the current driver is a legal driver associated with the target vehicle; if the identity probability is higher than a preset probability threshold, the current driver is determined to be the legal driver; otherwise, the current driver is determined not to be the legal driver, and an abnormality alarm is issued.

[0100] That is, if Figure 5As shown in FIG, when the identity of the current driver is identified based on the first fusion result, the first fusion result can be directly input into the pre-trained fully connected network to use the fully connected network to output the identity probability. The identity probability is also the probability that the current driver is the legal driver associated with the target vehicle, as follows:

[0101] ;

[0102] Among them, p is the output identity probability, which means the probability that the current driver is a legal driver. ,The closer p is to 1, the higher the possibility of being a legal driver; represents the preset weight matrix, which is used to map the fusion features to the output space and perform linear transformation on the fusion features; b is the preset bias term, which is used to adjust the flexibility of the model by adding an offset based on the weighted sum; is the activation function, namely the Sigmoid function, to compress the result after linear transformation to the interval [0,1]. The formula is:

[0103] .

[0104] If the identity probability exceeds a preset threshold (e.g., 0.9), the driver is deemed legitimate. Otherwise, the driver is deemed unlawful and an abnormality alarm is issued. In other words, when the system detects abnormal behavior, it immediately issues an alarm and sends the information to the management platform for human intervention. Through a connection with the management platform, the alarm information is delivered to relevant personnel via SMS, email, app notifications, and other means. The alarm information includes a detailed description of the abnormal behavior, the time of occurrence, and the vehicle's location. This ensures that if shift substitution or illegal driving is detected, relevant personnel can intervene promptly to prevent safety hazards.

[0105] It's easy to understand that the core function of a fully connected network is to map high-dimensional fusion features into a low-dimensional space to complete the classification task. The fully connected layer learns W and b to find the feature combination that best distinguishes legal from illegal drivers. The training process can be summarized as follows:

[0106] First, we need to prepare training data. Each training sample is a fused feature vector (i.e., a vector of time series, frequency domain, and event features extracted by LSTM and Transformer) ), and each training sample has a corresponding binary classification label y∈{0,1}, where y=1 represents driving data of a legal driver; y=0 represents driving data of an illegal driver. The above training data is then input into a fully connected network, and the network outputs the identity probability p. This application can use a binary cross-entropy loss function to train the fully connected network. The training goal is to make the predicted identity probability p approach the true label y. During the training process, the weight matrix W and the bias term b can be optimized using a gradient descent algorithm until the loss function converges, resulting in a trained fully connected network.

[0107] In a second specific embodiment, the identity identification of the current driver based on the first fusion result includes: determining a legal driver associated with the target vehicle, and determining a second fusion result corresponding to the historical driving behavior data of the legal driver; calculating the similarity between the first fusion result and the second fusion result; if the similarity is lower than a preset similarity threshold, it is determined that the current driver is not the legal driver, and an abnormality alarm is issued; otherwise, the current driver is determined to be the legal driver.

[0108] That is, in addition to using a trained fully connected network to directly output identity probabilities, the present application can also determine whether the current driver is a legal driver by calculating a similarity score. Specifically, the legal driver associated with the target vehicle can be determined first, and the second fusion result corresponding to the legal driver's historical driving behavior data can be determined. The second fusion result is obtained in the same manner as the first fusion result. Then, by calculating the similarity between the first fusion result and the second fusion result, if the similarity is lower than the preset similarity threshold, it means that there is a significant difference between the driving behavior data of the current driver and the driving behavior data of the legal driver. Therefore, it is determined that the current driver is not a legal driver, and an abnormal alarm is issued to notify the management platform. Otherwise, the current driver is determined to be a legal driver. Specifically, the similarity can be calculated using methods such as Euclidean distance and cosine similarity.

[0109] In a third specific embodiment, the identity identification of the current driver based on the first fusion result includes: inputting the first fusion result into a pre-trained fully connected network to use the fully connected network to output an identity probability; wherein the fully connected network is constructed based on a preset weight matrix, a preset bias term and a preset activation function, and the identity probability is the probability that the current driver is a legal driver associated with the target vehicle; if the identity probability is higher than a preset probability threshold, the current driver is determined to be the legal driver; if the identity probability is not higher than the preset probability threshold, the second fusion result corresponding to the historical driving behavior data of the legal driver is determined, and the similarity between the first fusion result and the second fusion result is calculated; if the similarity is lower than the preset similarity threshold, the current driver is determined not to be the legal driver, and an abnormality alarm is issued, otherwise the current driver is determined to be the legal driver.

[0110] In other words, the present application can simultaneously use a fully connected network and similarity calculation to jointly determine whether the current driver is a legitimate driver. First, the first fusion result is input into a pre-trained fully connected network, which then uses the fully connected network to output an identity probability. When the identity probability is above a preset probability threshold, the current driver is directly determined to be a legitimate driver. When the identity probability is not above the preset probability threshold, a second fusion result corresponding to the legitimate driver's historical driving behavior data is determined, and the similarity between the first and second fusion results is calculated. The similarity is then compared with the corresponding preset similarity threshold to determine whether the current driver is legitimate. Through this scheme, identity probability calculation is an efficient and real-time first judgment, directly using the trained model to determine identity, while similarity is a precise and safe second judgment. When the identity probability is low, it is verified by the legitimate driver's historical driving behavior data to ensure alarm accuracy. If only p is used, the system is sensitive to sudden interference and has a high false alarm rate. If only similarity is used, the real-time computational overhead is high. Therefore, by combining the fully connected network and similarity calculation, the present application can more accurately identify the driver's identity.

[0111] Figure 6This is a flowchart of a driver identity verification process disclosed in this application. When the driver starts an operation task, he will first enter his driver ID or swipe a card to log in to the vehicle system, thereby identifying which legal driver should be operating. After the operation begins, during the driving process, the vehicle system will continuously collect real-time driving behavior data and process the data into features that can be used for model reasoning through the feature extraction module, that is, extract time series features, frequency domain features and event features. Then, an LSTM and Transformer hybrid model is used for analysis to output the result of whether the driver is a legal driver. If the driver's identity matches, it will pause for a period of time (for example, after 10 minutes) and then restart the identity recognition process; if it does not match, it means that the driver is an illegal driver, and then an alarm will be triggered and the management platform will be notified.

[0112] For a more specific processing procedure of the above step S21, reference may be made to the corresponding contents disclosed in the above embodiments, which will not be described again here.

[0113] It can be seen that this application combines the time series modeling capabilities of the LSTM model with the global attention mechanism of the Transformer model to solve the problem of traditional single models' inadequate capture of complex driving behavior characteristics. This application can accurately verify the driver's identity by analyzing the driver's driving behavior data (such as acceleration, braking, steering, speed, etc.), and promptly issue an alarm or manual intervention when substitution or illegal driving is discovered. It can effectively eliminate the safety hazards caused by substitution drivers and ensure the safety of passengers and vehicles, thereby effectively improving vehicle safety and management efficiency. Compared with traditional facial recognition, password verification and other methods, this application can provide more reliable identity authentication without infringing on the driver's privacy.

[0114] See also Figure 7 As shown, the embodiment of the present application discloses a driver identification device, which includes:

[0115] The data acquisition module 11 is used to obtain the driving behavior data of the current driver in real time during the operation of the target vehicle;

[0116] a feature extraction module 12 for performing time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features, and identifying target driving events on the driving behavior data to extract event features;

[0117] A model processing module 13 is configured to input the time series features into a first preset network model to obtain a first output result, and input the frequency domain features and the event features into a second preset network model to obtain a second output result;

[0118] The identity recognition module 14 is configured to fuse the first output result and the second output result to obtain a first fusion result, and perform identity recognition on the current driver based on the first fusion result.

[0119] As can be seen, the present application obtains the current driver's driving behavior data in real time during the operation of the target vehicle, and simultaneously performs time series analysis, frequency domain analysis, and target driving event recognition on the driving behavior data to extract corresponding time series features, frequency domain features, and event features for subsequent identity recognition. Compared with single-dimensional features, the present application can more comprehensively characterize the driver's unique behavior pattern by extracting multi-dimensional features, significantly improving the accuracy and robustness of identity recognition. Furthermore, the present application uses a first preset network model and a second preset network model to process the above-mentioned extracted multi-dimensional features in parallel. Specifically, the first preset network model is used to process the time series features, and the second preset network model is used to process the frequency domain features and event features. The output results of the two network models are then fused to obtain a first fusion result. Finally, the current driver's identity is recognized based on the first fusion result. That is, the present application performs identity recognition by fusing the output results of the first preset network model and the second preset network model. It can comprehensively consider time series features and complex behavior features, enhance the model's recognition ability for complex driving behaviors, avoid errors caused by a single model, and ensure accurate judgment of the driver's identity. In addition, this application does not need to rely on means involving driver privacy such as video surveillance, but only realizes identity recognition through their driving behavior data, thus protecting the driver's privacy.

[0120] In some specific embodiments, the feature extraction module 12 specifically includes:

[0121] A time-frequency domain feature extraction unit is configured to perform time series analysis on the driving behavior data using the first preset network model to extract time series features, and to perform frequency domain analysis on the driving behavior data using a Fourier transform function to extract frequency domain features; wherein the first preset network model is a long short-term memory network model, and the second preset network model is a Transformer model.

[0122] In some specific embodiments, the driving behavior data includes acceleration, braking force, and steering wheel angle; the target driving event includes a sudden acceleration event, a sudden braking event, and a sudden turn event;

[0123] Accordingly, the feature extraction module 12 specifically includes:

[0124] a first event feature extraction unit, configured to determine that a sudden acceleration event currently exists if the acceleration exceeds a preset acceleration threshold, and extract event features corresponding to the sudden acceleration event;

[0125] a second event feature extraction unit, configured to determine that a sudden braking event currently exists if the braking force exceeds a preset braking force threshold, and extract event features corresponding to the sudden braking event;

[0126] The third event feature extraction unit is configured to determine that a sharp turn event currently exists if the steering wheel steering angle exceeds a preset steering angle threshold, and to extract event features corresponding to the sharp turn event.

[0127] In some specific embodiments, the identity recognition module 14 may specifically include:

[0128] a normalization unit, configured to perform normalization processing on the first output result and the second output result respectively using a preset normalization function to obtain corresponding first processing results and second processing results;

[0129] A splicing unit is used to perform vector splicing on the first processing result and the second processing result using a target weight parameter to obtain a first fusion result.

[0130] In some specific embodiments, the identity recognition module 14 may specifically include:

[0131] a first identity probability acquisition unit, configured to input the first fusion result into a pre-trained fully connected network, so as to output an identity probability using the pre-trained fully connected network; wherein the pre-trained fully connected network is constructed based on a preset weight matrix, a preset bias term, and a preset activation function, and the identity probability is a probability that the current driver is a legal driver associated with the target vehicle;

[0132] The first determination unit is configured to determine that the current driver is the legal driver if the identity probability is higher than a preset probability threshold, otherwise determine that the current driver is not the legal driver and issue an abnormality alarm.

[0133] In some specific embodiments, the identity recognition module 14 may specifically include:

[0134] a result acquisition unit, configured to determine a legal driver associated with the target vehicle and determine a second fusion result corresponding to the historical driving behavior data of the legal driver;

[0135] a first calculation unit, configured to calculate a similarity between the first fusion result and the second fusion result;

[0136] The second determination unit is configured to determine that the current driver is not the legal driver and issue an abnormality alarm if the similarity is lower than a preset similarity threshold; otherwise, determine that the current driver is the legal driver.

[0137] In some specific embodiments, the identity recognition module 14 may specifically include:

[0138] a second identity probability acquisition unit, configured to input the first fusion result into a pre-trained fully connected network, so as to output an identity probability using the pre-trained fully connected network; wherein the pre-trained fully connected network is constructed based on a preset weight matrix, a preset bias term, and a preset activation function, and the identity probability is a probability that the current driver is a legal driver associated with the target vehicle;

[0139] a third determination unit, configured to determine that the current driver is the legal driver if the identity probability is higher than a preset probability threshold;

[0140] a second calculation unit, configured to determine a second fusion result corresponding to the historical driving behavior data of the legal driver if the identity probability is not higher than a preset probability threshold, and calculate a similarity between the first fusion result and the second fusion result;

[0141] The fourth determination unit is configured to determine that the current driver is not the legal driver and issue an abnormality alarm if the similarity is lower than a preset similarity threshold; otherwise, determine that the current driver is the legal driver.

[0142] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Specifically, the device may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the driver identification method performed by the electronic device as disclosed in any of the aforementioned embodiments.

[0143] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0144] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 21 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0145] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon include an operating system 221, a computer program 222 and data 223, etc. The storage method can be temporary storage or permanent storage.

[0146] The operating system 221 is used to manage and control the hardware devices and computer programs 222 on the electronic device 20, enabling the processor 21 to calculate and process the massive amount of data 223 in the memory 22. It can be run under Windows, Unix, Linux, or other operating systems. In addition to computer programs capable of performing the driver identification method disclosed in any of the aforementioned embodiments and executed by the electronic device 20, the computer programs 222 may further include computer programs capable of performing other specific tasks. Data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0147] Furthermore, an embodiment of the present application also discloses a computer-readable storage medium, in which a computer program is stored. When the computer program is loaded and executed by a processor, the driver identification method steps disclosed in any of the aforementioned embodiments are implemented.

[0148] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0149] Those skilled in the art may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0150] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a compact disc read-only memory (CD-ROM), or any other form of storage medium known in the art.

[0151] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0152] The above is a detailed introduction to the driver identity recognition method, device, equipment and storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A driver identification method, characterized in that: include: Acquire the current driver's driving behavior data in real time during the operation of the target vehicle; performing time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features, and identifying target driving events on the driving behavior data to extract event features; Inputting the time series features into a first preset network model to obtain a first output result, and inputting the frequency domain features and the event features into a second preset network model to obtain a second output result; The first output result and the second output result are fused to obtain a first fusion result, and the identity of the current driver is identified based on the first fusion result.

2. The driver identification method according to claim 1, characterized in that: The performing time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features respectively includes: The first preset network model is used to perform time series analysis on the driving behavior data to extract time series features, and the Fourier transform function is used to perform frequency domain analysis on the driving behavior data to extract frequency domain features; wherein, the first preset network model is a long short-term memory network model, and the second preset network model is a Transformer model.

3. The driver identification method according to claim 1, characterized in that: The driving behavior data includes acceleration, braking force, and steering wheel angle; the target driving events include sudden acceleration events, sudden braking events, and sharp turning events; Accordingly, the identifying of target driving events from the driving behavior data to extract event features includes: If the acceleration exceeds a preset acceleration threshold, it is determined that a sudden acceleration event currently exists, and event features corresponding to the sudden acceleration event are extracted; If the braking force exceeds a preset braking force threshold, it is determined that the sudden braking event currently occurs, and event features corresponding to the sudden braking event are extracted; If the steering wheel steering angle exceeds a preset steering angle threshold, it is determined that the sharp turn event currently exists, and event features corresponding to the sharp turn event are extracted.

4. The driver identification method according to claim 1, characterized in that: The fusing the first output result and the second output result to obtain a first fusion result includes: Normalizing the first output result and the second output result respectively using a preset normalization function to obtain corresponding first processing results and second processing results; The first processing result and the second processing result are vector-joined using a target weight parameter to obtain a first fusion result.

5. The driver identification method according to any one of claims 1 to 4, characterized in that: The identifying the current driver based on the first fusion result includes: Inputting the first fusion result into a pre-trained fully connected network to output an identity probability using the fully connected network; wherein the fully connected network is constructed based on a preset weight matrix, a preset bias term, and a preset activation function, and the identity probability is a probability that the current driver is a legal driver associated with the target vehicle; If the identity probability is higher than a preset probability threshold, the current driver is determined to be the legal driver; otherwise, the current driver is determined not to be the legal driver and an abnormality alarm is issued.

6. The driver identification method according to any one of claims 1 to 4, characterized in that: The identifying the current driver based on the first fusion result includes: Determining a legitimate driver associated with the target vehicle, and determining a second fusion result corresponding to the historical driving behavior data of the legitimate driver; Calculating the similarity between the first fusion result and the second fusion result; If the similarity is lower than a preset similarity threshold, it is determined that the current driver is not the legal driver and an abnormality alarm is issued; otherwise, it is determined that the current driver is the legal driver.

7. The driver identification method according to any one of claims 1 to 4, characterized in that: The identifying the current driver based on the first fusion result includes: Inputting the first fusion result into a pre-trained fully connected network to output an identity probability using the fully connected network; wherein the fully connected network is constructed based on a preset weight matrix, a preset bias term, and a preset activation function, and the identity probability is a probability that the current driver is a legal driver associated with the target vehicle; If the identity probability is higher than a preset probability threshold, determining that the current driver is the legal driver; If the identity probability is not higher than a preset probability threshold, determining a second fusion result corresponding to the historical driving behavior data of the legal driver, and calculating the similarity between the first fusion result and the second fusion result; If the similarity is lower than a preset similarity threshold, it is determined that the current driver is not the legal driver and an abnormality alarm is issued; otherwise, it is determined that the current driver is the legal driver.

8. A driver identification device, characterized in that: include: A data acquisition module is used to obtain the current driver's driving behavior data in real time during the operation of the target vehicle; a feature extraction module, configured to perform time series analysis and frequency domain analysis on the driving behavior data to extract time series features and frequency domain features, and identify target driving events on the driving behavior data to extract event features; a model processing module, configured to input the time series features into a first preset network model to obtain a first output result, and input the frequency domain features and the event features into a second preset network model to obtain a second output result; An identity recognition module is used to fuse the first output result and the second output result to obtain a first fusion result, and to perform identity recognition on the current driver based on the first fusion result.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the driver identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Used to store a computer program; wherein, when the computer program is executed by a processor, the steps of the driver identification method according to any one of claims 1 to 7 are implemented.

Images