Fatigue state recognition method and system based on body pressure data of driver

Abstract

The invention is suitable for the technical field of safe driving of a driver, and provides a fatigue state recognition method and system based on body pressure data of the driver, and the method comprises the following steps: obtaining pressure data in a driving process; performing standardization processing on the pressure data, and performing window separation and label generation; constructing an LSTM model based on the processed pressure data, and training the model; adjusting a discard rate of a Dropout layer in the LSTM model, and adjusting a probability threshold value of a sigmoid function of a full connection layer in the LSTM model; and outputting the driving state category. The method can effectively capture the long-distance dependency relationship in the hip pressure data based on the LSTM model, accurately judges the physiological fatigue state of the driver, simplifies the data processing flow, reduces the calculation complexity, achieves the quick and accurate recognition of the physiological fatigue state of the driver compared with the recognition through a camera, and improves the recognition efficiency. And the method has good real-time performance and robustness.

Description

Technical Field

[0001] The present invention belongs to the technical field of driver safety driving, and in particular relates to a method and system for identifying a fatigue state based on driver body pressure data. Background Art

[0002] Driver fatigue is a significant factor affecting traffic safety, and accurately identifying a driver's fatigue state is crucial for preventing traffic accidents. Traditional fatigue detection methods rely primarily on cameras to capture visual features such as the driver's facial expressions and eye movements, analyzing these features to determine fatigue status. However, camera-based methods have limitations. For example, in low light conditions, when the driver is wearing sunglasses, or when obstructions are present, the camera may not accurately capture key features. Furthermore, camera-based detection methods are sensitive to environmental conditions such as strong light, shadows, or complex backgrounds, all of which can affect detection accuracy.

[0003] In contrast, physical force recognition methods based on driver body pressure data offer a non-invasive, real-time solution. By installing a pressure sensor on the seat, the driver's body pressure distribution data can be acquired. This data can reflect changes in the driver's sitting posture, physical activity, and physiological state. Body pressure data offers the following advantages: First, the data collection process is unaffected by environmental factors such as light and obstructions; second, it can continuously monitor the driver's physiological state, providing continuous time series data; and third, the sensor is easy to install and has no significant impact on the driver's driving behavior.

[0004] In terms of technical implementation, fatigue recognition methods based on body pressure data typically employ deep learning models, such as long short-term memory (LSTM) networks. LSTMs can effectively process time series data and capture the dynamic characteristics of body pressure changes. By extracting characteristics from body pressure data (such as the frequency, amplitude, and duration of pressure changes), LSTM models can learn patterns in driver fatigue, enabling accurate fatigue identification. Compared to camera-based methods, methods based on body pressure data are more robust in complex environments and are more user-friendly for driver privacy.

[0005] Therefore, the driver physiological fatigue recognition method based on body pressure data is a new technology with potential, which can effectively make up for the shortcomings of the traditional camera method. It is of great significance to further optimize the sensor design, improve the data processing algorithm, and explore the integration with other fatigue detection methods to achieve a more efficient and accurate fatigue recognition system. Summary of the Invention

[0006] The purpose of the embodiments of the present invention is to provide a method for identifying a fatigue state based on driver's body pressure data, aiming to solve the problems raised in the above-mentioned background technology.

[0007] The embodiment of the present invention is implemented as follows: a method for identifying a fatigue state based on driver body pressure data comprises the following steps:

[0008] Obtain stress data during driving;

[0009] Standardize the pressure data, perform window separation and label generation;

[0010] Build an LSTM model based on the processed pressure data and train the model;

[0011] Adjust the dropout rate of the Dropout layer in the LSTM model and the probability threshold of the sigmoid function of the fully connected layer in the LSTM model;

[0012] Output driving status category.

[0013] Another object of an embodiment of the present invention is to provide a fatigue state recognition system based on driver body pressure data, which is used to implement a fatigue state recognition method based on driver body pressure data, including:

[0014] A data acquisition module, used to obtain pressure data during driving;

[0015] The data processing module standardizes the pressure data, performs window separation and label generation;

[0016] Model building module, used to build LSTM model based on processed pressure data and train the model;

[0017] The model adjustment module is used to adjust the dropout rate of the Dropout layer in the LSTM model and the probability threshold of the sigmoid function of the fully connected layer in the LSTM model;

[0018] The output module is used to output the driving status category.

[0019] The fatigue state recognition method based on driver body pressure data provided by an embodiment of the present invention can effectively capture the long-distance dependency in hip pressure data based on the LSTM model, accurately judge the driver's physiological fatigue state, and learn the dependency between time steps through its internal gating mechanism, which plays a key role in understanding the driver's behavior pattern; the LSTM model has a certain robustness to the noise data in the driver's pressure data time series. By learning the overall pattern of the time series, it can ignore the outliers of individual time steps, which is very important for processing sensor data in the real world; the LSTM model has strong scalability, so in addition to hip pressure data, other sensor data can also be added. Compared with using a camera for recognition, the computational complexity is relatively small when processing hip pressure data, and training and reasoning can be performed quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a method for identifying a fatigue state based on driver's body pressure data provided by an embodiment of the present invention;

[0021] Figure 2 A structural block diagram of a fatigue state recognition system based on driver body pressure data provided by an embodiment of the present invention;

[0022] Figure 3 A diagram of body pressure data for fatigue driving provided by an embodiment of the present invention;

[0023] Figure 4 This is a body pressure data diagram of a normal driving state provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0026] like Figure 1 FIG. 1 is a flowchart of a method for identifying a fatigue state based on driver's body pressure data according to an embodiment of the present invention, comprising the following steps:

[0027] S1. Obtain pressure data during driving:

[0028] Using a simulation device to provide a realistic driving environment, nine pressure film sensors were evenly arranged on a 50cm×50cm driver's seat, with a spacing of approximately 5-10cm between each sensor. A 10Hz data acquisition frequency was used to obtain the following pressure data on the driver's buttocks during driving:

[0029] ;

[0030] in, is the pressure data corresponding to the first sensor at the i-th time node.

[0031] S2. Standardize the pressure data, perform window separation and label generation, specifically including the following processes:

[0032] S2.1. Process the pressure data using mean-variance standardization:

[0033] ;

[0034] in, is the original pressure data, i represents the i-th time node, and j represents the j-th sensor; is the mean of the pressure data of all time nodes of the j-th pressure sensor, is the standard deviation of the pressure at all time nodes of the j-th pressure sensor;

[0035] The processed pressure data matrix is ​​as follows:

[0036] ;

[0037] in, is the pressure data corresponding to the jth sensor at the i-th time node;

[0038] S2.2, using 2h as the time node, divide the pressure data matrix Z into multiple matrices , each matrix is ​​as follows:

[0039] ;

[0040] in, represents the lth time window, is the pressure data corresponding to the jth sensor at the tth time node;

[0041] S2.3. For each matrix Assign a label If the pressure data corresponding to multiple sensors As time goes by, if the driver's performance changes from a long-term steady state to a frequent fluctuation, then the driver is considered to be physically fatigued, and the label ,otherwise .

[0042] S3. Build an LSTM model based on the processed pressure data and train the model. Specifically, build a two-layer LSTM model. The dimension of the input layer should match the feature dimension of the pressure data. Add a Dropout layer between the LSTM layers to prevent overfitting. Add a fully connected layer after the LSTM layer to output the final classification (fatigue or not). Use the training set to train the LSTM model. Select the appropriate optimizer Adam and the cross entropy loss function for binary classification. During the training process, use the validation set to verify the model and adjust the model's hyperparameters to obtain the best model performance. The LSTM model learns each matrix through the following steps. Situation:

[0043] S3.1. Each matrix As input to the model, the shape is ,in is the number of time steps, is the number of sensors;

[0044] S3.2. The first LSTM layer processes the pressure data at each time step and extracts time series features:

[0045] S3.2.1, the forget gate is calculated as follows:

[0046] ;

[0047] in The output of the forget gate determines which information needs to be discarded from the cell state; The sigmoid activation function limits the output to between 0 and 1; is the weight matrix of the forget gate; is the bias term of the forget gate; is the hidden state of the previous time step; is the input pressure data of the current time step;

[0048] S3.2.2, the input gate is calculated as follows:

[0049] ;

[0050] ;

[0051] in, The output of the input gate determines what information needs to be stored in the cell state; For candidate cell states, new information; is the weight matrix of the input gate; is the weight matrix of the candidate cell state; is the bias term of the input gate; is the bias term of the candidate cell state;

[0052] S3.2.3. The cell state update calculation is as follows:

[0053] ;

[0054] in, is the cell state at the current time step;

[0055] S3.2.4, the output gate is calculated as follows:

[0056] ;

[0057] ;

[0058] in, The output of the output gate determines what information is output from the cell state; is the hidden state of the current time step; is the weight matrix of the output gate; is the bias term of the output gate;

[0059] S3.2.5, Dropout layer is calculated as follows:

[0060] ;

[0061] in, is the hidden state after applying Dropout; is the Dropout function, which randomly discards the output of some neurons. is the dropout rate of the Dropout layer;

[0062] S3.3, the second layer LSTM learns higher-level features and returns the output of the last time step. The input data is , perform the same calculation as step S3.2 to obtain the output of the second layer LSTM ;

[0063] S3.4. Output layer calculation: output of the second LSTM layer (the hidden state at the last time step) as the input to the output layer:

[0064] ;

[0065] in, is the final output, indicating the probability of fatigue driving; is the weight matrix of the output layer; is the bias term of the output layer;

[0066] S4. Adjust the dropout rate of the Dropout layer in the LSTM model and the probability threshold of the sigmoid function of the fully connected layer in the LSTM model. The specific process includes the following:

[0067] The dropout rate in the S4.1 and Dropout layers primarily addresses the problem of data overfitting. Set the initial dropout rate to 0.2. If the model performs well on the training set but poorly on the validation set, this indicates overfitting. Increase the dropout rate. Otherwise, decrease it. Adjust the dropout rate in steps of 0.1. Use cross-validation to select the optimal dropout rate parameter.

[0068] S4.2. Use the precision-recall curve to find the optimal probability threshold of the sigmoid function. The probability threshold that maximizes the precision and keeps the recall within an acceptable range is selected as the optimal probability threshold.

[0069] S5. Output the driving status category.

[0070] like Figure 2 FIG. 1 is a block diagram of a fatigue state recognition system based on driver body pressure data provided by an embodiment of the present invention, comprising:

[0071] The data acquisition module 100 is used to acquire pressure data during driving;

[0072] The data processing module 200 performs standardization processing on the pressure data, performs window separation and label generation;

[0073] A model building module 300 is used to build an LSTM model based on the processed pressure data and train the model;

[0074] A model adjustment module 400 is used to adjust the dropout rate of the Dropout layer in the LSTM model and adjust the probability threshold of the sigmoid function of the fully connected layer in the LSTM model;

[0075] The output module 500 is configured to output the driving status category.

[0076] Effect verification:

[0077] Hardware: RX-M0404M pressure film sensor;

[0078] Software: Python 3.8, CUDA 11.8;

[0079] Using nine RX-M0404M pressure film sensors and a real road driving environment, we collected driving data. We then labeled the data, removed noise, and standardized it. We then trained an LSTM model. After 50 iterations, the results are shown in Table 1.

[0080] Table 1 Average optimal probability threshold Average test set accuracy 0.9372 0.875

[0081] For each sample, the data at each time point is summed, the difference between the sums of adjacent time points is calculated, and the average value is calculated to obtain the average rate of change. For samples of normal driving, the sample with the smallest average rate of change is selected as the typical sample; for samples of physically fatigued driving, the sample with the largest average rate of change is selected as the typical sample. The image is as follows: Figure 3 、 Figure 4 As shown;

[0082] It can be seen from Table 1 that the model established by the method provided in the embodiment of the present invention can effectively capture the long-distance dependency relationship in the driver's body pressure data and accurately judge the driver's physical fatigue state; Figure 3 、 Figure 4 It can be seen that when the driver is physically tired, the body pressure data shows obvious unstable fluctuations, while when the driver is driving normally, the body pressure data shows an obvious long-term stable state.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying a driver's fatigue state based on body pressure data, characterized in that: The following steps are involved: Obtain stress data during driving; Standardize the pressure data, perform window separation and label generation; Build an LSTM model based on the processed pressure data and train the model; Adjust the dropout rate of the Dropout layer in the LSTM model and the probability threshold of the sigmoid function of the fully connected layer in the LSTM model; Output driving status category.

2. The method for identifying a driver's fatigue state based on body pressure data according to claim 1, characterized in that: The step of obtaining pressure data during driving specifically includes: collecting data from pressure film sensors evenly arranged on the driver's seat based on a set collection frequency as the pressure data during driving.

3. The method for identifying a driver's fatigue state based on body pressure data according to claim 1, characterized in that: The steps of normalizing the pressure data, performing window separation and label generation specifically include: The pressure data is processed using mean-variance standardization to obtain the pressure data matrix; Divide the pressure data matrix into multiple matrices based on the set time nodes; A label is assigned to each matrix. When the pressure data changes from a steady state to a fluctuating state over time, it is considered to be in a fatigue state and the label is 1, otherwise it is 0.

4. The method for identifying fatigue status based on driver's body pressure data according to claim 1, characterized in that: The steps of constructing an LSTM model based on the processed pressure data and training the model specifically include: Build a two-layer LSTM model, add a Dropout layer between the LSTM layers, and add a fully connected layer after the LSTM layer for output; Use the training set to train the LSTM model, use the validation set to validate the LSTM model, and adjust the model's hyperparameters.

5. The method for identifying fatigue status based on driver's body pressure data according to claim 1, characterized in that: The step of adjusting the dropout rate of the Dropout layer in the LSTM model specifically includes: setting 0.2 as the initial value of the dropout rate, increasing the dropout rate if the model is overfitting, and reducing the dropout rate if it is overfitting, adjusting the dropout rate in steps of 0.1, and selecting the optimal dropout rate parameter by cross-validation.

6. The method for identifying fatigue status based on driver's body pressure data according to claim 1, characterized in that: The step of adjusting the probability threshold of the sigmoid function of the fully connected layer in the LSTM model specifically includes: using a precision-recall curve to find the optimal probability threshold.

7. A fatigue state recognition system based on driver body pressure data, used to implement the fatigue state recognition method based on driver body pressure data according to any one of claims 1 to 6, characterized in that: include: A data acquisition module, used to obtain pressure data during driving; The data processing module standardizes the pressure data, performs window separation and label generation; Model building module, used to build LSTM model based on processed pressure data and train the model; The model adjustment module is used to adjust the dropout rate of the Dropout layer in the LSTM model and the probability threshold of the sigmoid function of the fully connected layer in the LSTM model; The output module is used to output the driving status category.

Images