Method for creating a trustworthiness evaluation network and device therefor

Abstract

The application discloses a method for creating a trust degree evaluation network and a device thereof, and the method comprises the following steps: obtaining a plurality of driving data corresponding to a plurality of drivers, performing the following preprocessing on each driving data, creating a trust degree evaluation network, and then training the trust degree evaluation network based on all the driving data. The trust degree evaluation network can evaluate the trust degree of the drivers.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a method and apparatus for creating a trust evaluation network. Background Technology

[0002] With the rapid development of autonomous driving technology, self-driving vehicles are becoming increasingly common worldwide. This technology aims to improve road safety and transportation efficiency by reducing human error. When the system requests the driver to take over control, the driver's level of trust in the autonomous driving system becomes a crucial issue.

[0003] Driver trust refers to the degree to which a driver trusts an autonomous driving system, specifically their confidence in the system's capabilities, reliability, and safety. Trust directly impacts a driver's behavior and decisions in autonomous driving mode, and also relates to the overall safety and user experience of the system. Drivers with high trust levels are more likely to rely on the autonomous driving system, while those with low trust levels may more frequently take over vehicle control, thus affecting the effectiveness of the technology. Insufficient trust can also lead to excessive monitoring of the system during autonomous driving, increasing cognitive load and fatigue, and reducing the overall driving experience. Therefore, accurately identifying and assessing driver trust in autonomous driving systems is crucial.

[0004] Existing methods for assessing driver trust primarily rely on subjective scales, which collect drivers' subjective evaluations of their trust in autonomous driving systems through standardized trust scales or customized experience questionnaires. However, existing subjective scale methods struggle to assess driver trust levels in real time, and the data analysis, employing machine learning and other methods, requires manual extraction of driver subjective characteristics, resulting in a significant workload. Furthermore, the trust assessment results are heavily influenced by individual drivers. Summary of the Invention

[0005] In view of this, the main objective of the present invention is to provide a method and apparatus for creating a trust evaluation network.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows: a method for creating a trust evaluation network, comprising the following steps: acquiring multiple driving data corresponding to multiple drivers, each driving data including: A EEG data points E1, E2, ..., E... corresponding to the driver in active driving mode. A And B ECG data points C1, C2, ..., C B Acquire A EEG data points E′1, E′2, ..., E′ corresponding to the driver in autonomous driving mode. A And B ECG data points C′1, C′2, ..., C′ BWhere A and B are natural numbers, A≥2, B≥2; then, the EEG data E1, E2, ..., E A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And ECG data C′1, C′2, ..., C′ B All data were preprocessed; for each driving data set, the following processing was performed: Num1 EEG attributes were obtained, and data were generated from A preprocessed EEG data sets E1, E2, ..., E... A Obtain Num1 first EEG attribute values ​​corresponding to Num1 EEG attributes, from A preprocessed EEG data E′1, E′2, ..., E′ A Obtain Num1 corresponding second EEG attribute values ​​for each of the Num1 EEG attributes. For any EEG attribute, obtain the EEG attribute difference value between the corresponding first and second EEG attribute values, thus obtaining Num1 EEG attribute difference values. Obtain Num2 ECG attributes from B preprocessed ECG data C1, C2, ..., C... B We obtain Num2 corresponding first ECG attribute values ​​for each of the Num2 ECG attributes, and then use B preprocessed ECG data C′1, C′2, ..., C′ B We obtain Num2 corresponding second ECG attribute values ​​for each of the Num2 ECG attributes. For any ECG attribute, we obtain the ECG attribute difference value between the corresponding first and second ECG attribute values, thus obtaining Num2 ECG attribute difference values. We also obtain the differences x1, x2, ..., x1 between the first and second EEG attribute values ​​corresponding to Num1 EEG attributes. Num1 Get the difference x between the first and second ECG attribute values ​​corresponding to Num2 ECG attributes. Num1+1 x Num1+2 ... x Num1+Num2 When x1, x2, ..., x Num1+Num2 All are greater than the preset value, sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 1, when x1, x2, ..., x Num1+Num2 All values ​​are less than or equal to the preset value, and the sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 0; a trust evaluation network is created, which first feeds the initial features into the first convolutional layer to extract local temporal and frequency domain features. in, The extracted features are represented by `filter`, the number of convolutional kernels is represented by `kernel_size`, the kernel size is represented by `3*3`, and the activation function is represented by `ReLU`. Then, the initially extracted features are fed into a pooling layer for downsampling to reduce computational complexity.

[0007] MaxPooling1D(pool_size=2), where pool_size represents the size of the pooling operation; then, a second convolutional layer is used to further extract higher-order features: Conv1D(32, filter=64, kernel_size=3, activation=ReLU); subsequently, downsampling is repeated to reduce computational complexity before being fed into two consecutive LSTM layers to continuously extract temporal features.

[0008] LSTM(64, return_sequences=True), LSTM(64), where return_sequences represents whether to return the complete output sequence; the loss function of the trust evaluation network is set as follows: Where θ is a parameter. It predicts labels; the trust rating network updates model parameters using gradient descent. Where, θ t Let be the parameter for the t-th iteration, and η be the learning rate. Let θ be the gradient of the parameter θ; the output of the trust rating network is changed from binary classification to a single value, and this value is transformed into a trust score between 0 and 1 using the Sigmoid function: The trust evaluation network is trained based on all driving data.

[0009] As an improvement to an embodiment of the present invention, the "electroencephalogram data E1, E2, ..., E A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And ECG data C′1, C′2, ..., C′ B "Preprocessing" specifically includes: processing E1, E2, ..., E A Filtering process is performed to obtain For ECG data C1, C2, ..., C B Filtering process is performed to obtain EEG data E′1, E′2, ..., E′ A Filtering process is performed to obtain For ECG data C′1, C′2, ..., C′ BAll were obtained by filtering.

[0010] As an improvement to this embodiment of the invention, the step of "acquiring Num1 EEG attributes from A preprocessed EEG data E1, E2, ..., E..." is described. A Obtain Num1 first EEG attribute values ​​corresponding to Num1 EEG attributes, from A preprocessed EEG data E′1, E′2, ..., E′ A Obtaining Num1 corresponding Num1 secondary EEG attribute values ​​for each of the Num1 EEG attributes specifically includes: Num1 EEG attributes including: mean, variance, and power spectral density at frequency f; obtaining EEG data. corresponding mean and the corresponding variance Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. The power spectral density represents the frequency f, and T represents the acquisition duration; thus, EEG data are obtained. corresponding mean and the corresponding variance Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. This represents the power spectral density at frequency f.

[0011] As an improvement to this embodiment of the invention, the step of "obtaining Num2 ECG attributes and processing B preprocessed ECG data C1, C2, ..., C..." is described. B Obtaining Num2 ECG attributes corresponding to Num2 primary ECG attribute values ​​specifically includes: the Num2 ECG attributes are: the mean of heart rate variability, the standard deviation of intervals, and the power spectral density at frequency f; obtaining ECG data. The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval and the corresponding power spectral density at frequency f Among them, RR 0,j The j-th element of the RR interval sequence representing the time-domain characteristic of heart rate variability is... Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function.

[0012] As an improvement to this embodiment of the invention, the "from B preprocessed electrocardiogram data C′1, C′2, ..., C′ B "Obtaining Num2 corresponding Num2 second ECG attribute values ​​for each of the Num2 ECG attributes" specifically includes: obtaining ECG data. The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval and the corresponding power spectral density at frequency f Among them, RR′ 0,j The j-th element of the mean RR interval sequence representing the time-domain characteristic of heart rate variability. Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T.

[0013] This invention also provides a device for creating a trust evaluation network, comprising the following modules: a data acquisition module, used to acquire multiple driving data corresponding to multiple drivers, each driving data including: A EEG data points E1, E2, ..., E1 corresponding to the driver in active driving mode. A And B ECG data points C1, C2, ..., C B Acquire A EEG data points E′1, E′2, ..., E′ corresponding to the driver in autonomous driving mode. A And B ECG data points C′1, C′2, ..., C′ B Where A and B are natural numbers, A≥2, B≥2; then, the EEG data E1, E2, ..., E A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And ECG data C′1, C′2, ..., C′ B All data undergoes preprocessing; the data processing module performs the following processing on each driving data set: acquires Num1 EEG attributes from A preprocessed EEG data sets E1, E2, ..., E... A Obtain Num1 first EEG attribute values ​​corresponding to Num1 EEG attributes, from A preprocessed EEG data E′1, E′2, ..., E′ AObtain Num1 corresponding second EEG attribute values ​​for each of the Num1 EEG attributes. For any EEG attribute, obtain the EEG attribute difference value between the corresponding first and second EEG attribute values, thus obtaining Num1 EEG attribute difference values. Obtain Num2 ECG attributes from B preprocessed ECG data C1, C2, ..., C... B We obtain Num2 corresponding first ECG attribute values ​​for each of the Num2 ECG attributes, and then use B preprocessed ECG data C′1, C′2, ..., C′ B We obtain Num2 corresponding second ECG attribute values ​​for each of the Num2 ECG attributes. For any ECG attribute, we obtain the ECG attribute difference value between the corresponding first and second ECG attribute values, thus obtaining Num2 ECG attribute difference values. We also obtain the differences x1, x2, ..., x1 between the first and second EEG attribute values ​​corresponding to Num1 EEG attributes. Num1 Get the difference x between the first and second ECG attribute values ​​corresponding to Num2 ECG attributes. Num1+1 x Num1+2 ... x Num1+Num2 When x1, x2, ..., x Num1+Num2 All are greater than the preset value, sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 1, when x1, x2, ..., x Num1+Num2 All values ​​are less than or equal to the preset value, and the sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 0; the network creation module is used to create a trust evaluation network, which first feeds the initial features into the first convolutional layer to extract local temporal and frequency domain features. in, The `filter` parameter represents the extracted features, `kernel_size` represents the kernel size (3x3), and `activation` represents the ReLU activation function. The initially extracted features are then fed into a pooling layer for downsampling to reduce computational complexity: `MaxPooling1D(pool_size=2)`, where `pool_size` represents the pooling operation size. Next, a second convolutional layer further extracts higher-order features: `Conv1D(32, filter=64, kernel_size=3, activation=ReLU)`. Afterward, downsampling is repeated to further reduce computational complexity before being fed into two consecutive LSTM layers to continuously extract temporal features.

[0014] LSTM(64, return_sequences=True), LSTM(64), where return_sequences represents whether to return the complete output sequence; the loss function of the trust evaluation network is set as follows: Where θ is a parameter. It predicts labels; the trust rating network updates model parameters using gradient descent. Where, θ t Let be the parameter for the t-th iteration, and η be the learning rate. Let θ be the gradient of the parameter θ; the output of the trust rating network is changed from binary classification to a single value, and this value is transformed into a trust score between 0 and 1 using the Sigmoid function: The training module is used to train the trust evaluation network based on all driving data.

[0015] As an improvement to this embodiment of the invention, the data processing module is further configured to: process the EEG data E1, E2, ..., E A Filtering process is performed to obtain For ECG data C1, C2, ..., C B Filtering process is performed to obtain EEG data E′1, E′2, ..., E′ A Filtering process is performed to obtain For ECG data C′1, C′2, ..., C′ B All were obtained by filtering.

[0016] As an improvement to this embodiment of the invention, the data processing module is further configured to: obtain EEG data by considering Num1 EEG attributes, including mean, variance, and power spectral density at frequency f. corresponding mean and the corresponding variance Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. The power spectral density represents the frequency f, and T represents the acquisition duration; thus, EEG data are obtained. corresponding mean and the corresponding variance Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. This represents the power spectral density at frequency f.

[0017] As an improvement to this embodiment of the invention, the data processing module is further configured to: obtain ECG data from Num2 ECG attributes, including the mean of heart rate variability, the standard deviation of intervals, and a power spectral density at frequency f; The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval and the corresponding power spectral density at frequency f Among them, RR 0,j The j-th element of the RR interval sequence representing the time-domain characteristic of heart rate variability is... Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function.

[0018] As an improvement to this embodiment of the invention, the data processing module is further configured to: obtain electrocardiogram data. The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval

[0019] and the corresponding power spectral density at frequency f Among them, RR′ 0,j The j-th element of the mean RR interval sequence representing the time-domain characteristic of heart rate variability. Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T.

[0020] The method and apparatus for creating a trust evaluation network provided in this invention have the following advantages: This invention discloses a method and apparatus for creating a trust evaluation network. The method includes the following steps: acquiring multiple driving data corresponding to multiple drivers; performing the following preprocessing on each driving data; creating a trust evaluation network; and then training the trust evaluation network based on all driving data. This trust evaluation network is capable of evaluating the trust level of drivers. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the method for creating the trust evaluation network in this embodiment.

[0022] Figure 2 This is a schematic diagram of the trust evaluation network in the embodiment. Detailed Implementation

[0023] The present invention will now be described in detail with reference to the embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of the present invention.

[0024] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some embodiments may include or substitute parts and features of other embodiments. The scope of the embodiments herein encompasses the entire scope of the claims and all available equivalents thereof. Throughout this document, the terms “first,” “second,” etc., are used only to distinguish one element from another without requiring or implying any actual relationship or order between the elements. Indeed, a first element can also be referred to as a second element, and vice versa. Furthermore, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a structure, apparatus, or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a structure, apparatus, or device. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the structure, apparatus, or device that includes said element. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0025] The terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer" used in this document to indicate orientations or positional relationships are based on the orientations or positional relationships shown in the accompanying drawings. They are used solely for the convenience of describing the document and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In the description herein, unless otherwise specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two elements; they can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0026] Embodiment 1 of the present invention provides a method for creating a trust evaluation network, such as... Figure 1 As shown, it includes the following steps:

[0027] Step 101: Obtain multiple driving data points corresponding to multiple drivers. Each driving data point includes: A EEG data points E1, E2, ..., E1 corresponding to the driver in active driving mode. A And B ECG data points C1, C2, ..., C B Acquire A EEG data points E′1, E′2, ..., E′ corresponding to the driver in autonomous driving mode. A And B ECG data points C′1, C′2, ..., C′ B Where A and B are natural numbers, A≥2, B≥2; then, the EEG data E1, E2, ..., E A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And ECG data C′1, C′2, ..., C′ B All were preprocessed;

[0028] To obtain this data, tests can be conducted under active driving conditions, collecting electroencephalogram (EEG) and electrocardiogram (ECG) data of the driver in this state. EEG is a method of recording brain activity using electrophysiological indicators. It is formed by summing the postsynaptic potentials that occur synchronously among a large number of neurons during brain activity. It records the changes in electrical waves during brain activity and is a comprehensive reflection of the electrophysiological activity of brain nerve cells on the surface of the cerebral cortex or scalp. ECG is a technique that uses an electrocardiograph to record the changes in electrical activity of the heart during each cardiac cycle from the body surface.

[0029] Physiological data in active driving mode can be represented as: D i ={E i C i}, Among them, E i This indicates EEG data in active driving mode, C i This represents ECG data in active driving mode, where N represents the number of electrodes on the EEG device, T represents the acquisition length, and f EEG f represents the sampling frequency of EEG data. ECG This indicates the sampling frequency of the electrocardiogram (ECG) data.

[0030] Physiological data in autonomous driving mode can be represented as: D′ j ={E′ j ,C′ j}, Among them, E′ j Represents EEG data in autonomous driving mode, C′ jThis represents ECG data in autonomous driving mode, where N represents the number of electrodes on the EEG device, T represents the data acquisition duration, and f EEG f represents the sampling frequency of EEG data. ECG This indicates the sampling frequency of the electrocardiogram (ECG) data.

[0031] Step 102: Perform the following processing on each driving data point: Obtain Num1 EEG attributes from A preprocessed EEG data points E1, E2, ..., E A Obtain Num1 first EEG attribute values ​​corresponding to Num1 EEG attributes, from A preprocessed EEG data E′1, E′2, ..., E′ A Obtain Num1 corresponding second EEG attribute values ​​for each of the Num1 EEG attributes. For any EEG attribute, obtain the EEG attribute difference value between the corresponding first and second EEG attribute values, thus obtaining Num1 EEG attribute difference values. Obtain Num2 ECG attributes from B preprocessed ECG data C1, C2, ..., C... B We obtain Num2 corresponding first ECG attribute values ​​for each of the Num2 ECG attributes, and then use B preprocessed ECG data C′1, C′2, ..., C′ B We obtain Num2 corresponding second ECG attribute values ​​for each of the Num2 ECG attributes. For any ECG attribute, we obtain the ECG attribute difference value between the corresponding first and second ECG attribute values, thus obtaining Num2 ECG attribute difference values. We also obtain the differences x1, x2, ..., x1 between the first and second EEG attribute values ​​corresponding to Num1 EEG attributes. Num1 Get the difference x between the first and second ECG attribute values ​​corresponding to Num2 ECG attributes. Num1+1 x Num1+2 ... x Num1+Num2 When x1, x2, ..., x Num1+Num2 All are greater than the preset value, sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 1, when x1, x2, ..., x Num1+Num2 All values ​​are less than or equal to the preset value, and the sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 0;

[0032] Step 103: Create a trust evaluation network. The trust evaluation network first feeds the initial features into the first convolutional layer to extract local temporal and frequency domain features. in, The extracted features are represented by `filter`, the number of convolutional kernels, `kernel_size` (3*3 kernel size), and `activation` (ReLU activation function). The initially extracted features are then fed into a pooling layer for downsampling to reduce computational complexity: `MaxPooling1D(pool_size=2)`, where `pool_size` represents the size of the pooling operation. Next, a second convolutional layer is used to further extract higher-order features: `Conv1D(32,filter=64,kernel_size=3,activation=ReLU)`. Afterward, downsampling is repeated to reduce computational complexity, and then the features are fed into two consecutive LSTM layers to continuously extract temporal features: `LSTM(64,return_sequences=True)`, `LSTM(64)`, where `return_sequences` represents whether a complete output sequence is returned. The loss function of the trust evaluation network is set as follows: Where θ is a parameter. It predicts labels; the trust rating network updates model parameters using gradient descent. Where, θ t Let be the parameter for the t-th iteration, and η be the learning rate. Let θ be the gradient of the parameter θ; the output of the trust rating network is changed from binary classification to a single value, and this value is transformed into a trust score between 0 and 1 using the Sigmoid function:

[0033] Step 104: Train the trust evaluation network based on all driving data.

[0034] Here, the trained trust evaluation network is used to detect the driver's trust in autonomous driving in real time, and the trust level is divided into three levels according to the output trust score: when the output trust score is 0-0.33, the driver is determined to be in a low trust state. At this time, the autonomous driving system should take into account the driver's low trust state when taking over or performing other operations to avoid or reduce the driver's discomfort; when the output trust score is 0.34-0.67, the driver is determined to be in a medium trust state. At this time, the autonomous driving system needs to intervene appropriately in the driver's state to prevent the driver's trust from decreasing further; when the output trust score is 0.68-1, the driver is determined to be in a high trust state. The autonomous driving system can continuously monitor the driver's trust state and road conditions and respond in real time.

[0035] This invention proposes a method for evaluating driver trust in autonomous vehicles based on physiological signals. It collects electrocardiogram (ECG) and electroencephalogram (EEG) data from drivers during active and autonomous driving using physiological devices. The data is then preprocessed, analyzed, and labeled. A deep learning network model is then trained, and finally, the driver trust level is output. This invention can provide a relatively objective trust evaluation of physiological data through deep learning methods, and the real-time trust assessment can help with subsequent implementation decisions for autonomous driving systems.

[0036] This invention proposes a method for evaluating driver trust in autonomous vehicles based on physiological signals. It collects physiological data from drivers during both manual and autonomous driving using multi-channel physiological devices, preprocesses and annotates the data, constructs a deep learning network model for training, and finally outputs the driver trust level. Its advantages are as follows:

[0037] 1. Utilizing physiological signals for analysis can provide a more objective and real-time assessment of trustworthiness;

[0038] 2. It can monitor and assess the driver's trust level in real time during driving, providing immediate feedback, which helps to adjust the behavior of the autonomous driving system in a timely manner and improve safety.

[0039] 3. Traditional machine learning methods rely on manual feature extraction, which is labor-intensive. Deep learning models, on the other hand, can automatically extract features from data through multi-layered neural networks, eliminating the need for manual intervention. This not only reduces the workload of feature extraction but also helps discover potential, complex features that are difficult to extract manually.

[0040] 4. A complex nonlinear relationship exists between physiological signals and driver trust levels. Deep learning models possess powerful nonlinear modeling capabilities, enabling them to capture and learn these complex relationships, thereby improving the model's prediction accuracy. Traditional machine learning methods, on the other hand, typically perform poorly when handling nonlinear relationships.

[0041] In summary, the embodiments of the present invention construct a trust evaluation network that can analyze the physiological data of drivers in autonomous driving mode, output relatively objective trust results in real time, and is not affected by subjective factors of the driver. It has high accuracy and real-time performance, realizes accurate assessment of driver trust, and provides important support for subsequent decision-making and adjustment of autonomous driving system.

[0042] In this embodiment, the phrase "for EEG data E1, E2, ..., E..." is used. A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And ECG data C′1, C′2, ..., C′B "Preprocessing" specifically includes: processing E1, E2, ..., E A Filtering process is performed to obtain For ECG data C1, C2, ..., C B Filtering process is performed to obtain EEG data E′1, E′2, ..., E′ A Filtering process is performed to obtain For ECG data C′1, C′2, ..., C′ B All were obtained by filtering.

[0043] Here, a bandpass filter can be selected for EEG data. and All data were filtered for ECG data. and All signals are filtered to retain the valid portion, as follows: in, and This represents the filtered EEG data. and denoted as the filtered ECG data, h(t) represents the impulse response of the bandpass filter, 0≤t≤T.

[0044] In this embodiment, the step of "acquiring Num1 EEG attributes from A preprocessed EEG data E1, E2, ..., E..." is described. A Obtain Num1 first EEG attribute values ​​corresponding to Num1 EEG attributes, from A preprocessed EEG data E′1, E′2, ..., E′ A "Obtaining Num1 secondary EEG attribute values ​​corresponding to Num1 EEG attributes" specifically includes:

[0045] Num1 EEG attributes include: mean, variance, and power spectral density at frequency f;

[0046] Obtaining EEG data corresponding mean and the corresponding variance

[0047]

[0048] Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. The power spectral density represents the frequency f, and T represents the acquisition duration;

[0049] Obtaining EEG data corresponding mean and the corresponding variance

[0050] Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. This represents the power spectral density at frequency f.

[0051] The mean and method are both time-domain features of EEG data, reflecting the overall voltage level and stability of the data. Power spectral density is a frequency-domain feature of EEG data, reflecting the energy distribution of the data at corresponding frequencies.

[0052] In this embodiment, the step of "obtaining Num2 ECG attributes and extracting B preprocessed ECG data C1, C2, ..., C..." is described. B The specific details of obtaining the Num2 first ECG attribute values ​​corresponding to the Num2 ECG attributes include: the Num2 ECG attributes include: the mean of heart rate variability, the standard deviation of interval, and the power spectral density at a frequency of f;

[0053] Obtain ECG data The mean of the corresponding heart rate variability

[0054] and the corresponding standard deviation of the interval

[0055] and the corresponding power spectral density at frequency f Among them, RR 0,j The j-th element of the RR interval sequence representing the time-domain characteristic of heart rate variability is... Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function.

[0056] In this embodiment, the phrase "from B preprocessed electrocardiogram data C′1, C′2, ..., C′" refers to... B "Obtaining Num2 corresponding Num2 second ECG attribute values ​​for each of the Num2 ECG attributes" specifically includes: obtaining ECG data. The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval and the corresponding power spectral density at frequency f Among them, RR′ 0,j The j-th element of the mean RR interval sequence representing the time-domain characteristic of heart rate variability. Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T.

[0057] For ECG data and ECG data To reflect the activity of the autonomic nervous system, and to assess the driver's level of trust by measuring the difference in heart rate intervals, we selected the mean RR interval and standard deviation of the time domain characteristics of heart rate variability (HRV) and the power spectral density of the frequency domain characteristics for evaluation.

[0058] ECG data The corresponding HRV data can be represented as:

[0059] ECG data The corresponding HRV data can be represented as:

[0060]

[0061] Obtain the feature matrices for active driving mode and autonomous driving mode. Obtain the feature matrices for active driving mode and autonomous driving mode.

[0062]

[0063] Since the autonomous driving system is not active during active driving, there is no trust level between the driver and the autonomous driving system. Trust is only relevant in autonomous driving mode; therefore, the trust threshold is set to 0, and the changes in different physiological data are calculated. The method is as follows: Δx = F1 - F0, where Δx represents the change in each corresponding variable in the feature matrix, i.e.: Δx = (ΔMeanΔVΔPΔHRV), where...

[0064]

[0065] Based on the calculated changes in different characteristics, due to the physiological characteristics of EEG, a smaller time-domain characteristic indicates a more relaxed driver and a tendency towards a state of trust; for the power spectral density of the corresponding waveform in the frequency domain, a higher value represents a higher level of trust; the higher the time- and frequency-domain characteristics of HRV, the higher the driver's level of trust. Therefore, the magnitude of the changes and thresholds are calculated, and the data is labeled accordingly.

[0066]

[0067] Where y represents the corresponding label, which is 0 or 1, where 0 represents distrust and 1 represents trust; threshold represents the threshold value, which is set to 0 based on the aforementioned physiological characteristics. Finally, all features are integrated, judged, and labeled.

[0068]

[0069] Based on each set of physiological data and its corresponding label, the final dataset D is constructed: D = {(x c ,y c )|c=1,2,...,C}, where x c The value represents the change in the feature vector, and C represents the number of data points. The training set and validation set are then divided according to the corresponding ratio for use in the next module.

[0070] Embodiment 2 of the present invention provides an apparatus for creating a trust evaluation network, comprising the following modules:

[0071] The data acquisition module is used to acquire multiple driving data points corresponding to multiple drivers. Each driving data point includes: A EEG data points E1, E2, ..., E1 corresponding to the driver in active driving mode. A And B ECG data points C1, C2, ..., C B Acquire A EEG data points E′1, E′2, ..., E′ corresponding to the driver in autonomous driving mode. A And B ECG data points C′1, C′2, ..., C′ B Where A and B are natural numbers, A≥2, B≥2; then, the EEG data E1, E2, ..., E A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And ECG data C′1, C′2, ..., C′ B All were preprocessed;

[0072] The data processing module performs the following processing on each driving data set: acquire Num1 EEG attributes from A preprocessed EEG data sets E1, E2, ..., E... A Obtain Num1 first EEG attribute values ​​corresponding to Num1 EEG attributes, from A preprocessed EEG data E′1, E′2, ..., E′ A Obtain Num1 corresponding second EEG attribute values ​​for each of the Num1 EEG attributes. For any EEG attribute, obtain the EEG attribute difference value between the corresponding first and second EEG attribute values, thus obtaining Num1 EEG attribute difference values. Obtain Num2 ECG attributes from B preprocessed ECG data C1, C2, ..., C... B We obtain Num2 corresponding first ECG attribute values ​​for each of the Num2 ECG attributes, and then use B preprocessed ECG data C′1, C′2, ..., C′ B We obtain Num2 corresponding second ECG attribute values ​​for each of the Num2 ECG attributes. For any ECG attribute, we obtain the ECG attribute difference value between the corresponding first and second ECG attribute values, thus obtaining Num2 ECG attribute difference values. We also obtain the differences x1, x2, ..., x1 between the first and second EEG attribute values ​​corresponding to Num1 EEG attributes. Num1 Get the difference x between the first and second ECG attribute values ​​corresponding to Num2 ECG attributes. Num1+1 x Num1+2 ... x Num1+Num2 When x1, x2, ..., x Num1+Num2 All are greater than the preset value, sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 1, when x1, x2, ..., x Nun1+Nun2 All values ​​are less than or equal to the preset value, and the sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 0;

[0073] The network creation module is used to create a trust evaluation network. This network first feeds initial features into the first convolutional layer to extract local temporal and frequency domain features. in, The extracted features are represented by `filter`, the number of convolutional kernels, `kernel_size` (3*3 kernel size), and `activation` (ReLU activation function). The initially extracted features are then fed into a pooling layer for downsampling to reduce computational complexity: `MaxPooling1D(pool_size=2)`, where `pool_size` represents the size of the pooling operation. Next, a second convolutional layer is used to further extract higher-order features: `Conv1D(32,filter=64,kernel_size=3,activation=ReLU)`. Afterward, downsampling is repeated to reduce computational complexity, and then the features are fed into two consecutive LSTM layers to continuously extract temporal features: `LSTM(64,return_sequences=True)`, `LSTM(64)`, where `return_sequences` represents whether a complete output sequence is returned. The loss function of the trust evaluation network is set as follows: Where θ is a parameter. It predicts labels; the trust rating network updates model parameters using gradient descent. Where, θ t Let be the parameter for the t-th iteration, and η be the learning rate. Let θ be the gradient of the parameter θ; the output of the trust rating network is changed from binary classification to a single value, and this value is transformed into a trust score between 0 and 1 using the Sigmoid function:

[0074] The training module is used to train the trust evaluation network based on all driving data.

[0075] In this embodiment, the data processing module is further configured to: process the EEG data E1, E2, ..., E A Filtering process is performed to obtain For ECG data C1, C2, ..., C B Filtering process is performed to obtain EEG data E′1, E′2, ..., E′ A Filtering process is performed to obtain For ECG data C′1, C′2, ..., C′ B All were obtained by filtering.

[0076]

[0077] In this embodiment, the data processing module is further configured to:

[0078] Num1 EEG attributes include: mean, variance, and power spectral density at frequency f;

[0079] Obtaining EEG data corresponding mean and the corresponding variance

[0080]

[0081] Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. The power spectral density represents the frequency f, and T represents the acquisition duration;

[0082] Obtaining EEG data corresponding mean and the corresponding variance

[0083] Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. This represents the power spectral density at frequency f.

[0084] In this embodiment, the data processing module is further configured to:

[0085] The Num2 ECG attributes include: the mean of heart rate variability, the standard deviation of intervals, and the power spectral density at frequency f; to obtain ECG data. The mean of the corresponding heart rate variability

[0086] and the corresponding standard deviation of the interval

[0087] and the corresponding power spectral density at frequency f Among them, RR 0,j The j-th element of the RR interval sequence representing the time-domain characteristic of heart rate variability is... Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function.

[0088] In this embodiment, the data processing module is further configured to:

[0089] Obtain ECG data The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval

[0090] and the corresponding power spectral density at frequency f Among them, RR′ 0,j The j-th element of the mean RR interval sequence representing the time-domain characteristic of heart rate variability. Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T.

[0091] It should be noted that although the steps are described in a specific order above, it does not mean that the steps must be executed in the above specific order. In fact, some of these steps can be executed concurrently, or even in a different order, as long as the required function can be achieved.

[0092] This invention can be a system, method, and / or computer program product. A computer program product may include a readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0093] A readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. Readable storage media can include, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof.

[0094] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for creating a trust rating network, characterized in that, Includes the following steps: Acquire multiple driving data points corresponding to multiple drivers, each driving data point including: A EEG data points E1, E2, ..., E1 corresponding to the driver in active driving mode. A And B ECG data C1, C2, ..., C B Acquire A EEG data points E′1, E′2, ..., E′ corresponding to the driver in autonomous driving mode. A And B ECG data C′1, C′2, ..., C′ B Where A and B are natural numbers, A≥2, B≥2; then, the EEG data E1, E2, ..., E A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And electrocardiogram data C′1, C′2, ..., C′ B All were preprocessed; For each driving data point, the following processing is performed: Num1 EEG attributes are obtained; from A preprocessed EEG data points E1, E2, ..., EA, Num1 corresponding first EEG attribute values ​​are obtained; and from A preprocessed EEG data points E′1, E′2, ..., E′... A Obtain Num1 corresponding second EEG attribute values ​​for each of the Num1 EEG attributes. For any EEG attribute, obtain the EEG attribute difference value between the corresponding first and second EEG attribute values, thus obtaining Num1 EEG attribute difference values. Obtain Num2 ECG attributes from B preprocessed ECG data C1, C2, ..., C... B We obtain Num2 corresponding first ECG attribute values ​​for each of the Num2 ECG attributes, and then use B preprocessed ECG data C′1, C′2, ..., C′ B We obtain Num2 corresponding second ECG attribute values ​​for each of the Num2 ECG attributes. For any ECG attribute, we obtain the ECG attribute difference value between the corresponding first and second ECG attribute values, thus obtaining Num2 ECG attribute difference values. We also obtain the differences x1, x2, ..., x1 between the first and second EEG attribute values ​​corresponding to Num1 EEG attributes. Num1 Get the difference x between the first and second ECG attribute values ​​corresponding to Num2 ECG attributes. Num1+1 x Num1+2 ... x Num 1 +Num2 When x1, x2, ..., x Num1+Num2 All values ​​are greater than the preset value, and the sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 1, when x1, x2, ..., x Num1+Num2 All values ​​are less than or equal to the preset value, and the sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 0: A trust evaluation network is created, which first feeds initial features into the first convolutional layer to extract local temporal and frequency domain features. in, The `filter` parameter represents the extracted features, `kernel_size` represents the kernel size (3x3), and `activation` represents the ReLU activation function. The initially extracted features are then fed into a pooling layer for downsampling to reduce computational complexity: `MaxPooling1D(pool_size=2)`, where `pool_size` represents the pooling operation size. Next, a second convolutional layer further extracts higher-order features: `Conv1D(32, filter=64, kernel_size=3, activation=ReLU)`. Afterward, downsampling is repeated to further reduce computational complexity before being fed into two consecutive LSTM layers to continuously extract temporal features. LSTM(64, return_sequences = True), LSTM(64), where return_sequences represents whether to return the complete output sequence; the loss function of the trust evaluation network is set as follows: Where θ is a parameter. It predicts labels; the trust rating network updates model parameters using gradient descent. Where, θ t Let be the parameter for the t-th iteration, and η be the learning rate. Let θ be the gradient of the parameter θ; the output of the trust rating network is changed from binary classification to a single value, and this value is transformed into a trust score between 0 and 1 using the Sigmoid function: The trust evaluation network is trained based on all driving data.

2. The creation method according to claim 1, characterized in that, The "electroencephalogram data E1, E2, ..., E A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And electrocardiogram data C′1, C′2, ..., C′ B "Preprocessing" specifically includes: EEG data E1, E2, ..., E A Filtering process is performed to obtain For ECG data C1, C2, ..., C B Filtering process is performed to obtain EEG data E′1, E′2, ..., E′ A Filtering process is performed to obtain For ECG data C′1, C′2, ..., C′ B All were obtained by filtering.

3. The creation method according to claim 2, characterized in that, The process involves obtaining Num1 EEG attributes, extracting Num1 first EEG attribute values ​​corresponding to each of the A preprocessed EEG data sets E1, E2, ..., EA, and then extracting the first EEG attribute values ​​corresponding to each of the Num1 EEG attributes from the A preprocessed EEG data sets E′1, E′2, ..., E′. A "Obtaining Num1 corresponding Num1 secondary EEG attribute values ​​for each of the Num1 EEG attributes" specifically includes: Num1 EEG attributes include: mean, variance, and power spectral density at frequency f; Obtaining EEG data corresponding mean and the corresponding variance Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. The power spectral density represents the frequency f, and T represents the acquisition duration; Obtaining EEG data corresponding mean and the corresponding variance Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. This represents the power spectral density at frequency f.

4. The creation method according to claim 3, characterized in that, The phrase "obtain Num2 ECG attributes, from B preprocessed ECG data C1, C2, ..., C..." is used to... B "Obtaining Num2 corresponding Num2 first ECG attribute values ​​for each of the Num2 ECG attributes" specifically includes: The Num2 ECG attributes include: mean heart rate variability, standard deviation of intervals, and power spectral density at frequency f; Obtain ECG data The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval and the corresponding power spectral density at frequency f Among them, RR 0，j The j-th element of the mean RR interval sequence representing the time-domain characteristic of heart rate variability′ is... Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function.

5. The creation method according to claim 4, characterized in that, The phrase "from B preprocessed electrocardiogram data C′1, C′2, ..., C′" refers to... B "Obtaining Num2 corresponding Num2 second ECG attribute values ​​for each of the Num2 ECG attributes" specifically includes: Obtain ECG data The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval and the corresponding power spectral density at frequency f Among them, RR′ 0，j The j-th element of the mean RR interval sequence representing the time-domain characteristic of heart rate variability. Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T.

6. An apparatus for creating a trust evaluation network, characterized in that, Includes the following modules: The data acquisition module is used to acquire multiple driving data points corresponding to multiple drivers. Each driving data point includes: A sets of EEG data points E1, E2, ..., EA corresponding to the driver in active driving mode, and B sets of ECG data points C1, C2, ..., C... B Acquire A EEG data points E′1, E′2, ..., E′ corresponding to the driver in autonomous driving mode. A and B ECG data C′1 、 C′2、…、C′ B Where A and B are natural numbers, A≥2, B≥2; then, the EEG data E1, E2, ..., E A ECG data C1, C2, ..., C B EEG data E′1, E′2, ..., E′ A And ECG data C′1, C′2, ..., C′ B All were preprocessed; The data processing module performs the following processing on each driving data set: acquire Num1 EEG attributes from A preprocessed EEG data sets E1, E2, ..., E... A Obtain Num1 first EEG attribute values ​​corresponding to Num1 EEG attributes, from A preprocessed EEG data E′1, E′2, ..., E′ A Obtain Num1 corresponding second EEG attribute values ​​for each of the Num1 EEG attributes. For any EEG attribute, obtain the EEG attribute difference value between the corresponding first and second EEG attribute values, thus obtaining Num1 EEG attribute difference values. Obtain Num2 ECG attributes from B preprocessed ECG data C1, C2, ..., C... B We obtain Num2 corresponding first ECG attribute values ​​for each of the Num2 ECG attributes, and then use B preprocessed ECG data C′1, C′2, ..., C′ B Obtain Num2 corresponding second ECG attribute values ​​for each of the Num2 ECG attributes. For any ECG attribute, obtain the ECG attribute difference value between the corresponding first and second ECG attribute values, thus obtaining Num2 ECG attribute difference values. Obtain the differences x1, x2, ..., xNum1 between the first and second EEG attribute values ​​corresponding to Num1 EEG attributes, and obtain the differences x between the first and second ECG attribute values ​​corresponding to the Num2 ECG attributes. Num1+1 x Num1+2 ... x Num1+Num2 When x1, x2, ..., x Num1+Num2 All are greater than the preset value, sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 1, when x1, x2, ..., x Num1+Num2 All values ​​are less than or equal to the preset value, and the sample D = {x1, x2, ..., x...} Num1+Num2 The corresponding trust level is 0; The network creation module is used to create a trust evaluation network. This network first feeds initial features into the first convolutional layer to extract local temporal and frequency domain features. in, The extracted features are represented by `filter`, the number of convolutional kernels, `kernel_size` (3*3 kernel size), and `activation` (ReLU activation function). The initially extracted features are then fed into a pooling layer for downsampling to reduce computational complexity: `MaxPooling1D(pool_size=2)`, where `pool_size` represents the size of the pooling operation. Next, a second convolutional layer is used to further extract higher-order features: `Conv1D(32, filter=64, kernel_size=3, activation=ReLU)`. Afterward, downsampling is repeated to reduce computational complexity, and then the features are fed into two consecutive LSTM layers to continuously extract temporal features: `LSTM(64, return_sequences=True)`, `LSTM(64)`, where `return_sequences` represents whether a complete output sequence is returned. The loss function of the trust evaluation network is set as follows: Where θ is a parameter. It predicts labels; the trust rating network updates model parameters using gradient descent. Where, θ t Let be the parameter for the t-th iteration, and η be the learning rate. Let θ be the gradient of the parameter θ; the output of the trust rating network is changed from binary classification to a single value, and this value is transformed into a trust score between 0 and 1 using the Sigmoid function: The training module is used to train the trust evaluation network based on all driving data.

7. The creation apparatus according to claim 6, characterized in that, The data processing module is also used for: EEG data E1, E2, ..., E A Filtering process is performed to obtain For ECG data C1, C2, ..., C B Filtering process is performed to obtain EEG data E′1, E′2, ..., E′ A Filtering process is performed to obtain For ECG data C′1, C′2, ..., C′ B All were obtained by filtering.

8. The creation apparatus according to claim 7, characterized in that, The data processing module is also used for: Num1 EEG attributes include: mean, variance, and power spectral density at frequency f; Obtaining EEG data corresponding mean and the corresponding variance Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. The power spectral density represents the frequency f, and T represents the acquisition duration; Obtaining EEG data corresponding mean and the corresponding variance Obtaining EEG data Power spectral density at the corresponding frequency f in, Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function. This represents the power spectral density at frequency f.

9. The creation apparatus according to claim 8, characterized in that, The data processing module is also used for: The Num2 ECG attributes include: mean heart rate variability, standard deviation of intervals, and power spectral density at frequency f; Obtain ECG data The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval and the corresponding power spectral density at frequency f Among them, RR 0，j The j-th element of the mean RR interval sequence representing the time-domain characteristic of heart rate variability′ is... Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T, where FFT() represents the Fast Fourier Transform function.

10. The creation apparatus according to claim 9, characterized in that, The data processing module is also used for: Obtain ECG data The mean of the corresponding heart rate variability and the corresponding standard deviation of the interval and the corresponding power spectral density at frequency f Among them, RR′ 0，j The j-th element of the mean RR interval sequence representing the time-domain characteristic of heart rate variability. Represents the standard deviation of the interval. Indicates by The fitted function, 0≤t≤T.

Images