Method and system for real-time measurement of human-machine trust based on heart rate variability

Abstract

The present application belongs to the field of human-computer interaction, and relates to a human-computer trust real-time measurement method and system based on heart rate variability. The present application performs principal component analysis on multiple time domain, frequency domain and nonlinear heart rate variability characteristics, and constructs a human-computer trust real-time measurement model on this basis; in application, user electrocardio signals are collected, heart rate variability characteristics are extracted from the user electrocardio signals; the heart rate variability characteristics are input into a load matrix, and a principal component matrix is calculated according to the load matrix; the principal component matrix is input into the human-computer trust real-time measurement model, and human-computer trust real-time measurement results are obtained. The present application is based on user heart rate variability characteristics, can realize real-time measurement of human-computer trust, and overcomes defects in measurement cost, accuracy, stability and anti-interference.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction, and more particularly to a method and system for real-time measurement of human-computer trust based on heart rate variability. Background Technology

[0002] Conditional automated driving (CED) has two key characteristics: autonomy within a feasible domain, meaning that CED vehicles can operate autonomously and reliably within that domain without requiring constant driver monitoring; and limited capabilities, meaning that CED vehicles cannot handle all driving scenarios, thus necessitating driver takeover when necessary. It's important to note that the feasible domain is dynamic, not fixed. Furthermore, takeover can be categorized into active takeover (initiated by the driver, typically due to low trust or dangerous situations) and passive takeover (initiated by the system, requiring effective human-machine interaction). Both excessively high and low trust can lead to untimely and unsafe takeovers, negatively impacting the safety of CED vehicles. Trust is a complex and multidimensional concept, and its effective measurement is fundamental to improving the safety of CED vehicles. However, currently, real-time measurement methods for driver trust are lacking.

[0003] Trust is the attitude of an agent in helping to achieve personal goals in situations of uncertainty and vulnerability. (Reference: Manchon JB, Bueno M, Navarro J. How the initial level of trust in automated driving impacts drivers' behavior and early trust construction[J]. Transportation research part F: traffic psychology and behaviour,2022,86:281-295.)

[0004] Existing methods for measuring driver trust can be mainly divided into three categories:

[0005] 1. Measurement method based on subjective scales. After a driver uses or experiences a specific model of autonomous driving system under one or more operating conditions, the driver completes a single-dimensional or multi-dimensional subjective scale. The scores for each item on the scale are then statistically calculated to obtain the driver's level of trust in the current autonomous driving system. Examples include the Autonomy Trust Scale (TiA) and the Autonomy Scenario Trust Scale (STS-AD). For the Autonomy Trust Scale (TiA), see the literature "...". M.Theoretical considerations and development of a questionnaire to measure trust inautomation[C] / / Proceedings of the 20th Congress of the InternationalErgonomics Association(IEA 2018)Volume VI:Transport Ergonomics and HumanFactors(TEHF),Aerospace Human Factors and Ergonomics 20.SpringerInternational Publishing, 2019:13-30.". For the automatic driving situational trust scale STS-AD, please refer to the literature "Holthausen BE, Wintersberger P, Walker BN, et al. Situational trust scale for automated driving (STS-AD): Development and initial validation [C] / / 12thInternational Conference on Automotive User Interfaces and InteractiveVehicular Applications.2020:40-47.".

[0006] 2. Behavioral performance-based measurement method. Driver trust is indirectly measured through driver behavior (e.g., the frequency of monitoring and takeover by the system). Driver facial video or operation input logs (typically including steering wheel, accelerator pedal, and brake pedal operations) within a specific time window are selected to obtain the driver's level of trust in the current autonomous driving system.

[0007] 3. Measurement based on physiological signals. Driver trust is measured using physiological signals such as eye movements or electroencephalograms (EEGs). By selecting temporal physiological signals from the driver over a specific period, preprocessing the signals, and extracting time-domain and frequency-domain features, the driver's level of trust in the current autonomous driving system can be obtained.

[0008] The existing technology has the following main drawbacks:

[0009] 1. Subjective scale-based measurement methods: These are highly invasive and difficult to conduct in real-time during daily driving, as filling out questionnaires not only interferes with the driver's ongoing tasks and activities but also interrupts their original state. Furthermore, subjective measurements cannot achieve continuous measurement and have low temporal resolution.

[0010] 2. Performance-based measurement methods: These methods have low validity and are not suitable for conditional automated driving. This is because conditional automated driving does not require drivers to continuously perform driving tasks, and the method is indirect. In actual driving, various other factors can interfere with the accuracy of the measurement (e.g., driver fatigue, distraction, and changes in the complexity of the driving environment can also lead to changes in performance), resulting in low validity.

[0011] 3. Measurement methods based on physiological signals: Eye movement-based and electroencephalogram (EEG)-based driver trust measurement methods are costly, have low accuracy, and are weak in stability and interference resistance. Specifically, eye movement measurement based on glasses-type eye trackers is costly and requires wearing wired equipment, which interferes with driving and monitoring; while eye movement measurement based on desktop eye trackers is easily affected by light interference (especially when the light intensity changes drastically), resulting in poor stability; EEG-based measurement methods require wearing an EEG cap and connecting to the EEG acquisition device via wires, which leads to high cost and poor practicality, and the EEG signal quality is easily degraded due to factors such as scalp sweating, resulting in low stability and interference resistance. Summary of the Invention

[0012] To address the issues of existing technologies being highly invasive, having low measurement validity, and being unsuitable for conditionally automated driving, this invention provides a real-time human-machine trust measurement method and system based on heart rate variability for scenarios such as conditionally automated driving. Based on the user's heart rate variability characteristics, it can achieve real-time measurement of the user's trust in the automated system, overcoming the shortcomings in measurement cost, accuracy, stability, and anti-interference.

[0013] The technical solution adopted in this invention is as follows:

[0014] A real-time human-computer trust measurement method based on heart rate variability includes the following steps:

[0015] Collect user electrocardiogram (ECG) signals in human-computer interaction scenarios and extract heart rate variability features from the user ECG signals;

[0016] Input the heart rate variability features into the loading matrix, and calculate the principal component matrix based on the loading matrix;

[0017] Input the principal component matrix into the real-time human-machine trust measurement model to obtain the real-time human-machine trust measurement results.

[0018] Furthermore, the real-time human-machine trust measurement model is constructed using the following steps:

[0019] Collect users' electrocardiogram signals in human-computer interaction scenarios and use a human-computer trust subjective scale to obtain users' trust in the automated system;

[0020] Heart rate variability features are extracted from user electrocardiogram signals, and principal component analysis is used to reduce the dimensionality of the heart rate variability features to obtain the loading matrix and principal component matrix.

[0021] Using multiple principal components in the principal component matrix as input and the trust level obtained using the human-machine trust subjective scale as output, a real-time user trust measurement model is constructed.

[0022] Furthermore, the heart rate variability features include frequency domain features, time domain features, and nonlinear features; the frequency domain features include LF / HF, LF Peak, and HF Peak; the time domain features include RMSSD, SDHR, and pNN50; and the nonlinear features include SampEn and SD1 / SD2.

[0023] Furthermore, the human-computer interaction scenario is a driving scenario; the process of collecting user electrocardiogram signals in the human-computer interaction scenario and using a human-computer trust subjective scale to obtain the user's trust in the automated system includes:

[0024] Multiple driving scenarios are built based on the driving simulator. Each scenario is divided into multiple events, and each event includes multiple road users. The driving scenario library includes straight roads and curves with different curvatures.

[0025] Recruit no fewer than 30 test drivers. Test drivers must have more than three years of driving experience and highway driving experience, have normal or corrected normal vision, and be of a gender balance and with the widest possible age distribution.

[0026] The experiment employed a within-subjects experimental design, using multiple driving scenarios as independent variables and selecting an appropriate subjective driver trust scale as the method for measuring driver trust during the experiment.

[0027] A driving simulation experiment was conducted, in which the driver's electrocardiogram signal was collected at a frequency of no less than 250Hz, and after each event, the driver's trust in the autonomous driving system was measured using an electronic questionnaire.

[0028] Furthermore, the step of using principal component analysis to reduce the dimensionality of heart rate variability features to obtain the loading matrix and principal component matrix includes:

[0029] For the distribution of 8 heart rate variability features of N valid data points, an index matrix F is established. Using principal component analysis, the principal component matrix P and loading matrix W are obtained based on the following formula:

[0030] P = FW

[0031] Among them, based on the eigenvalue being greater than or equal to 1, the number of principal components k to be retained, which is the number of columns of the principal component matrix P, is selected based on the existing N valid data.

[0032] Furthermore, the step of constructing a real-time human-machine trust measurement model by using multiple principal components in the principal component matrix as input and the trust level obtained using a human-machine trust subjective scale as output includes:

[0033] Using a stepwise regression method, with the principal component matrix P as input and the trust matrix T of N valid data points as output, a real-time human-machine trust measurement model α is established:

[0034] T = α(P) = α(FW)

[0035] The specific form of the real-time human-machine trust measurement model α is as follows:

[0036] α(P)=α1(P1)+α2(P2)+…α k (P k )+c

[0037] Among them, P1, P2, ... P k These are the first to kth columns of the principal component matrix, representing the first to kth principal components of the N valid data points; α1, α2, ... α k are the model coefficients corresponding to the first principal component to the kth principal component, and c is the constant term.

[0038] A real-time human-machine trust measurement system based on heart rate variability, comprising:

[0039] The user ECG acquisition module is used to acquire the user's ECG signal.

[0040] The heart rate variability analysis module is connected to the user's electrocardiogram acquisition module and is used to extract heart rate variability features from the user's electrocardiogram signal.

[0041] The trust measurement model storage module is used to store the real-time human-machine trust measurement model and load matrix;

[0042] The human-machine trust calculation module is connected to the heart rate variability analysis module and the trust measurement model storage module. It is used to input heart rate variability features into the load matrix, calculate the principal component matrix based on the load matrix, input the principal component matrix into the real-time human-machine trust measurement model, and obtain the real-time human-machine trust measurement result.

[0043] The key points and corresponding beneficial effects of this invention are as follows:

[0044] Key Point 1: The data acquisition experiment (i.e., the human-machine trust fluctuation induction experiment) of the trust measurement model proposed in this invention needs to simultaneously include multiple scenarios. For example, for autonomous driving, it needs to simultaneously include scenarios such as the autonomous driving system operating normally, the vehicle reaching the boundary of the autonomous driving system, and the autonomous driving system failing. Furthermore, it needs to cover multiple scenarios, rather than considering only a single situation or a fixed scenario. Beneficial technical effects: It can cover a wider range of scenarios, improve the external validity of the dataset, thereby improving the accuracy of subsequent real-time recognition models and their stability across scenarios and users.

[0045] Key Point 2: The trust measurement model proposed in this invention employs a multi-indicator method combining subjective and objective approaches, using a data-driven weighted calculation method. This is significantly different from existing methods that use only a single indicator or rely on expert scoring. Beneficial technical effects: It can utilize more comprehensive and abundant feature information while reducing the impact of multicollinearity, thereby reducing feature redundancy, simplifying the model, and mitigating the impact of abnormal fluctuations in a particular feature (e.g., a malfunction in a part of the feature extraction module), ultimately improving recognition accuracy and stability. Furthermore, compared to measurement methods based on EEG or other physiological signals, the heart rate variability signal of this invention is acquired through wearable devices (e.g., wristbands), making the acquisition method simple, easy to implement, highly stable, highly resistant to interference, and low-cost.

[0046] Key Point 3: Regarding the application (deployment) of the trust measurement model proposed in this invention, a two-step calculation method is employed. Using a sliding time window (60 seconds), features such as LF / HF, LFPeak, HF Peak, RMSSD, SDHR, pNN50, SampEn, and SD1 / SD2 are extracted from the user's ECG signal and input into a stored load matrix according to the required format. The principal component matrix is ​​then calculated and input into the real-time human-machine trust measurement model to calculate the real-time human-machine trust measurement result. Beneficial technical effects: On the one hand, it reduces the instability of results caused by large fluctuations in a certain indicator; on the other hand, it improves the transparency and interpretability of the output results, facilitating the optimization of real-time system performance indicators. Furthermore, compared to measurement methods based on subjective scales, this invention can achieve real-time measurement of human-machine trust, such as measuring driver trust in real-time during daily driving, with minimal intrusion into the user's tasks. Compared to measurement methods based on behavioral performance, the method proposed in this invention can directly measure trust, avoiding interference from other factors during the operation of automated systems, thus resulting in higher measurement accuracy and validity. Attached Figure Description

[0047] Figure 1 This is a flowchart for obtaining a real-time measurement model of human-machine trust.

[0048] Figure 2 It is a module architecture for a real-time human-machine trust measurement system.

[0049] Figure 3 This is the practical application process of the real-time measurement model of human-machine trust. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0051] The purpose of this invention is to achieve real-time measurement of human-machine trust in scenarios such as conditional autonomous driving, based on user heart rate variability characteristics that are easy to collect, highly stable, and have strong anti-interference capabilities, so as to provide technical support for research on human-machine interaction and collaborative behavior and user safety related to conditional autonomous driving.

[0052] The present invention provides a real-time human-machine trust measurement method based on heart rate variability, the main contents of which include:

[0053] 1. The data acquisition experiment of the trust measurement model (i.e., the human-machine trust fluctuation inducement experiment) needs to include multiple situations at the same time. For example, for autonomous driving, it needs to include the situations of the autonomous driving system operating normally, the vehicle reaching the boundary of the autonomous driving system, and the autonomous driving system failing at the same time, and cover multiple scenarios, rather than only considering a single situation and a fixed scenario.

[0054] 2. For the trust measurement model acquisition, the time-domain features (including LF / HF, LF Peak, and HF Peak), frequency-domain features (including RMSSD, SDHR, and pNN50), and nonlinear features (including SampEn and SD1 / SD2) of heart rate variability are first calculated. Then, principal component analysis is used to reduce the dimensionality of these eight heart rate variability features, obtaining the loading matrix and principal component matrix. Next, using multiple principal components from the principal component matrix (all calculated from objective heart rate variability indices) as input and the human-computer trust level obtained from a multidimensional subjective scale as output, a trust measurement model is constructed to reflect the specific parameter values ​​reflecting the time-dependent mapping relationship between the two (see details). Figure 1 Therefore, the acquisition part of the trust measurement model proposed in this invention is a multi-indicator method that combines subjective and objective methods, which is significantly different from existing methods that only use a single indicator or use methods based on expert scoring.

[0055] 3. In the application part of the trust measurement model, a sliding time window (60 seconds) is used to extract LF / HF, LF Peak, HF Peak, RMSSD, SDHR, pNN50, SampEn, and SD1 / SD2 features from the user's ECG signal. These features are then input into a stored load matrix according to the required format to calculate the principal component matrix, which is then input into the trust measurement model to calculate the real-time human-machine trust measurement result. Therefore, this invention adopts a two-step calculation method, which can reduce the instability of results caused by large fluctuations in a certain indicator, and improve the transparency and interpretability of the output results, facilitating the optimization of performance indicators.

[0056] The following uses a driving scenario as an example to specifically illustrate this invention. The real-time human-machine trust measurement method based on heart rate variability proposed in this invention comprises two main parts: the acquisition and application of the trust measurement model. It should be noted that this invention includes details regarding the relevant experimental design and procedures.

[0057] I. Acquisition of Trust Measurement Model

[0058] This phase primarily utilizes various heart rate variability characteristics of drivers across multiple driving scenarios, along with driver trust in the autonomous driving system obtained through scales, to establish a driver trust measurement model. This model takes various heart rate variability characteristics as input and the driver's trust in the autonomous driving system as output.

[0059] Figure 1 The flowchart for obtaining the real-time measurement model of driver trust is implemented as follows:

[0060] First, based on a driving simulator, several typical scenarios for conditional autonomous driving are constructed, including situations where the autonomous driving system is operating normally, the vehicle reaches the boundary of the autonomous driving system, and the autonomous driving system fails. Reaching the boundary of the autonomous driving system refers to the situation where the autonomous driving system sends a takeover request to the driver before reaching a condition it cannot handle. System failure refers to the situation where the system cannot handle the current condition and fails to return driving control to the driver. Each scenario includes multiple events, encompassing various road users such as pedestrians, cyclists, and commercial vehicles. Furthermore, the driving scenario library includes straight roads and curves with varying curvatures.

[0061] Then, recruit no fewer than 30 test drivers, who must have more than three years of driving experience, highway driving experience, and normal or corrected normal vision. Furthermore, the test drivers should be of balanced gender and have the widest possible age distribution.

[0062] Then, a within-subjects experimental design was adopted, using three scenarios (i.e., the autonomous driving system is operating normally, the vehicle reaches the boundary of the autonomous driving system, and the autonomous driving system fails) and multiple scenarios (e.g., highway construction zones, off-ramp, and vehicle malfunctions in front) as independent variables. A suitable subjective driver trust scale (e.g., TiA, STS-AD, or a single-dimensional scale) was selected as the method for measuring driver trust during the experiment.

[0063] Then, a driving simulation experiment was conducted based on the above experimental design. During the experiment, the driver's electrocardiogram signal was collected at a frequency of no less than 250Hz, and after each event, the driver's trust in the autonomous driving system during the experiment was measured using an electronic questionnaire (trust score denoted as T).

[0064] The electrocardiogram (ECG) signal was then preprocessed, and heart rate variability (HRV) signals were extracted. Frequency domain features (including LF / HF, LF Peak, and HF Peak), time domain features (including RMSSD, SDHR, and pNN50), and nonlinear features (including SampEn and SD1 / SD2) were then extracted from the HRV signals for subsequent analysis.

[0065] The definitions of the above-mentioned heart rate variability characteristics are as follows:

[0066] LF / HF: The ratio of the energy of the low-frequency band (0.04-0.15Hz) and the high-frequency band (0.15-0.4Hz) of a heart rate variability signal.

[0067] LF Peak: The highest frequency in a low-frequency band (0.04-0.15Hz) of a heart rate variability signal.

[0068] HF Peak: The highest frequency in a high-frequency band (0.15-0.4Hz) of a heart rate variability signal.

[0069] RMSSD: The root mean square value of the difference between two adjacent RR intervals in a heart rate variability signal.

[0070] SDHR: The standard deviation of the RR interval of a heart rate variability signal.

[0071] pNN50: The percentage of consecutive RR intervals in a heart rate variability signal that differ by more than 50 milliseconds.

[0072] SampEn: The sample entropy of a heart rate variability signal.

[0073] SD1 / SD2: The ratio of the standard deviation of the minor axis to the standard deviation of the major axis of an ellipse in a Poincaré plot of a heart rate variability signal.

[0074] Then, principal component analysis (PCA) was used to reduce the dimensionality of the heart rate variability features. Specifically, for the distribution of the above eight features of N valid data points, an index matrix F was established. Using PCA, the principal component matrix P and the loading matrix W can be obtained based on the following formula:

[0075] P = FW

[0076] Among them, based on the eigenvalue being greater than or equal to 1, the number of principal components k to be retained, which is the number of columns of the P matrix, is determined by the existing N valid data.

[0077] Then, a real-time driver trust measurement model is constructed. Using a generalized additive model and a stepwise regression method, with the principal component matrix P as input and the trust matrix T of N valid data points as output, model α is established:

[0078] T = α(P) = α(FW)

[0079] The specific form of α is:

[0080] α(P)=α1(P1)+α2(P2)+…α k (P k )+c

[0081] Among them, P1, P2, ... P k These are the first to kth columns of the principal component matrix, representing the first to kth principal components of the N valid data points; α1, α2, ... α k are the model coefficients corresponding to the first principal component to the kth principal component, and c is the constant term.

[0082] It should be noted that for the corresponding system, the information that needs to be recorded includes the driver trust real-time measurement model α (including model form and parameters) and the load matrix W.

[0083] II. Application of Trust Measurement Model

[0084] Figure 2 The architecture of the real-time driver trust measurement system requires the vehicle to have the following system modules: driver electrocardiogram acquisition module, heart rate variability analysis module, trust measurement model storage module, and driver trust calculation module.

[0085] Figure 3 The practical application process for establishing driver trust in real-time measurement models includes the following steps:

[0086] The driver's electrocardiogram (ECG) acquisition module needs to stably and reliably acquire the driver's ECG signal at a frequency of no less than 250Hz, and transmit the ECG signal to the heart rate variability analysis module in real time at this frequency. This can be achieved using a dedicated patch sensor or a wearable device (such as a smartwatch) connected to the system to measure the driver's ECG signal.

[0087] The heart rate variability analysis module needs to receive the ECG signals sent by the driver's ECG acquisition module in real time, automatically preprocess the signals and extract features, and send the extracted heart rate variability features to the driver trust calculation module in real time. Specifically, this module uses a sliding time window (60 seconds in length) to extract LF / HF, LF Peak, HF Peak, RMSSD, SDHR, pNN50, SampEn, and SD1 / SD2 from the driver's ECG signal.

[0088] The trust measurement model storage module stores the real-time driver trust measurement model α (including model form and parameters) and the load matrix W. The function of this module is to store and transmit this two pieces of information to the driver trust calculation module.

[0089] The driver trust calculation module receives heart rate variability features (i.e., LF / HF, LF Peak, HF Peak, RMSSD, SDHR, pNN50, SampEn, and SD1 / SD2) from the heart rate variability analysis module and algorithm information (i.e., the real-time driver trust measurement model α and the load matrix W) stored from the trust measurement model storage module. It then inputs the heart rate variability index matrix F* at that moment into the load matrix W to obtain the principal component matrix P* at that moment.

[0090] P*=F*W

[0091] Then, the principal component matrix P* at that moment is input into the real-time driver trust measurement model α to obtain the driver's trust level T* at the current moment:

[0092] T* = α(P*)

[0093] Note that T* here is a scalar, not a vector matrix.

[0094] At this point, the constructed trust measurement model can be used to measure the driver's level of trust in the conditional automated driving system in real time.

[0095] Experimental data:

[0096] 1) Test Scenario: Three scenarios were selected for testing autonomous driving systems on highways, focusing on situations where a vehicle cuts in front. These included: the autonomous driving system promptly and safely decelerating upon the vehicle cutting in; the autonomous driving system issuing a takeover request to the driver; and the autonomous driving system failing to decelerate in time and not issuing a takeover request, resulting in a collision (i.e., a silent failure). Each scenario was conducted on a clear, daytime with normal visibility on a four-lane highway. The vehicle maintained a speed of 110 km / h before being cut in by the vehicle in front. Each scenario lasted for 2 minutes.

[0097] 2) Experimental Design: A high-fidelity driving simulator with visual, auditory, and motion simulation capabilities was used as the experimental setup. A BIOPAC MP150 was used as the ECG acquisition device (sampling frequency set to 1000Hz). Drivers' subjective trust levels (score range 1-10) were obtained using an electronic questionnaire on the Wenjuanxing platform. A total of 38 drivers were recruited, and each driver was measured twice for each scenario, resulting in 3*2=6 data points per driver.

[0098] 3) Data Processing: Data from one driver was discarded due to poor ECG quality. A total of 216 valid data points were collected from 36 drivers. 180 data points from 30 drivers were used as the training set, and 36 data points from 6 drivers were used as the test set. Python was used to extract heart rate variability features: LF / HF, LF Peak, HF Peak, RMSSD, SDHR, pNN50, SampEn, and SD1 / SD2. The training set data was used to calculate the loading matrix and the principal component matrix for model training; the test set data was used to verify the effectiveness of the method.

[0099] 4) Experimental Results and Analysis: During principal component extraction, using an eigenvalue greater than or equal to 1 as the criterion, and based on 180 valid data points in the training set, 4 principal components were selected for retention. Therefore, the principal component matrix P is 180 rows and 4 columns; the loading matrix W is 8 rows and 4 columns. The loading matrix is ​​as follows:

[0100]

[0101] Using a generalized additive model and a stepwise regression method, the constructed model is as follows:

[0102] α(P) = 0.812·P1 - 0.283·P1 2 +0.114·P1 3 +0.553·P² - 0.264·P² 2 +1.297·P3-0.631·P4 2 +4.395

[0103] Among them, P1, P2, P3, and P4 are the four principal components extracted from the training set.

[0104] For the test set data, the payload matrix W and feature matrix F are first obtained based on the training set. test Calculate the principal component matrix P of the test set. test :

[0105] P test =F test W

[0106] Then, the calculation model outputs the driver trust matrix T. pred :

[0107] T pred =α(P test )

[0108] The model outputs a driver trust matrix T. pred Actual trust score T during driver experiment act By comparing the results, the effectiveness of the method proposed in this invention can be verified.

[0109] Choose the most commonly used coefficient of determination (R²) for regression tasks. 2 The above model's performance on the training set is R0, which is used as the evaluation metric. 2 train =0.856, with an R value of 0.856 on the test set. 2 test =0.728. In summary, the method proposed in this invention, verified by experimental data, can effectively achieve accurate measurement of driver trust and has high interpretability.

[0110] It should be understood that the methods and systems disclosed in the above embodiments of the present invention can be implemented in other ways. For example, the above module division can be implemented in other ways, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Each module in the present invention can be implemented as a software functional unit and can be stored in a computer-readable storage medium, including several instructions to cause a computer device to execute some or all of the steps of the method described in the present invention. For example, one embodiment of the present invention provides a computer device (computer, server, smartphone, etc.) including a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for executing each step of the method of the present invention. For example, another embodiment of the present invention provides a computer-readable storage medium (such as ROM / RAM, disk, optical disk, etc.) storing a computer program, which, when executed by a computer, implements each step of the method of the present invention.

[0111] The specific embodiments of the present invention disclosed above are intended to help understand the content of the present invention and to implement it accordingly. Those skilled in the art will understand that various substitutions, changes, and modifications are possible without departing from the spirit and scope of the present invention. The present invention should not be limited to the content disclosed in the embodiments of this specification; the scope of protection of the present invention is defined by the claims.

Claims

1. A real-time human-computer trust measurement method based on heart rate variability, characterized in that, Includes the following steps: Collect user electrocardiogram (ECG) signals in human-computer interaction scenarios, and extract heart rate variability features from user ECG signals using a sliding time window method; The heart rate variability features include frequency domain features, time domain features, and nonlinear features; the frequency domain features include LF / HF, LF Peak, and HF Peak; the time domain features include RMSSD, SDHR, and pNN50; the nonlinear features include SampEn and SD1 / SD2. Input the heart rate variability features into the loading matrix, and calculate the principal component matrix based on the loading matrix; Input the principal component matrix into the real-time human-machine trust measurement model to obtain the real-time human-machine trust measurement results; The real-time human-machine trust measurement model is constructed using the following steps: Collect users' electrocardiogram signals in human-computer interaction scenarios and use a human-computer trust subjective scale to obtain users' trust in the automated system; Heart rate variability features are extracted from user electrocardiogram signals, and principal component analysis is used to reduce the dimensionality of the heart rate variability features to obtain the loading matrix and principal component matrix. Using multiple principal components in the principal component matrix as input and the trust level obtained by using the human-machine trust subjective scale as output, a real-time measurement model for user trust is constructed. The principal component analysis is used to reduce the dimensionality of heart rate variability features, obtaining the loading matrix and principal component matrix, including: For the distribution of 8 heart rate variability features of N valid data points, an index matrix is ​​established. F Principal component analysis is used to obtain the principal component matrix based on the following formula. P and load matrix W : P = FW Among them, based on the eigenvalue being greater than or equal to 1, the number of principal components to be retained is selected from the existing N valid data. k That is, the principal component matrix P The number of columns; The method involves using multiple principal components from the principal component matrix as input and the trust level obtained using a subjective human-machine trust scale as output to construct a real-time human-machine trust measurement model, including: Using the stepwise regression method, with the principal component matrix P As input, a trust matrix containing N valid data points. T To output, establish a real-time human-machine trust measurement model. α : T = α ( P )= α ( FW ) Among them, the real-time measurement model of human-machine trust α The specific form is as follows: α ( P )= α 1( P 1 )+ α 2( P 2 )+… α k ( P k )+ c in, P 1 , P 2 … P k It is the first column to the first column of the principal component matrix. k The column, i.e., the first principal component to the last principal component of N valid data points. k principal component; α 1. α 2. ... α k It is the first principal component to the... k The model coefficients corresponding to the principal components, c This is a constant term.

2. The method according to claim 1, characterized in that, The human-computer interaction scenario is a driving scenario; the collection of user electrocardiogram signals in the human-computer interaction scenario and the acquisition of user trust in the automated system using a human-computer trust subjective scale include: Multiple driving scenarios are built based on the driving simulator. Each scenario is divided into multiple events, and each event includes multiple road users. The driving scenario library includes straight roads and curves with different curvatures. The experiment employed a within-subjects experimental design, using multiple driving scenarios as independent variables and selecting the Driver Trust Subjective Scale as the method for measuring driver trust during the experiment. A driving simulation experiment was conducted, in which the driver's electrocardiogram signal was collected at a frequency of no less than 250 Hz, and after each event, the driver's trust in the autonomous driving system was measured using an electronic questionnaire.

3. A real-time human-machine trust measurement system based on heart rate variability, characterized in that, The system employing the human-machine trust real-time measurement method based on heart rate variability as described in claim 1 or 2 includes: The user ECG acquisition module is used to acquire user ECG signals in human-computer interaction scenarios. The heart rate variability analysis module is connected to the user's electrocardiogram acquisition module and is used to extract heart rate variability features from the user's electrocardiogram signal. The trust measurement model storage module is used to store the real-time human-machine trust measurement model and load matrix; The human-machine trust calculation module is connected to the heart rate variability analysis module and the trust measurement model storage module. It is used to input heart rate variability features into the load matrix, calculate the principal component matrix based on the load matrix, input the principal component matrix into the real-time human-machine trust measurement model, and obtain the real-time human-machine trust measurement result.

4. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for performing the method of claim 1 or 2.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a computer, implements the method of claim 1 or 2.

Images