A method for evaluating safety of vehicle-mounted voice interaction system considering interaction complexity

Abstract

The application discloses a kind of vehicle-mounted voice interaction system safety evaluation method considering interaction complexity, and belongs to the technical field of automobile man-machine engineering, specifically relates to vehicle-mounted voice interaction system safety evaluation method considering interaction complexity.The purpose of the present application is to solve the problem of high traffic collision risk and accident rate caused by the distraction of existing vehicle-mounted voice interaction system;At the same time, the technical pain points of weak practicality of evaluation results guiding design, complicated evaluation process and large amount of calculation are solved.Process:1, construct voice interaction complexity function;2, get the real label of driving risk level;3, obtain the optimal parameters of membership function, and the values of and under the optimal parameters;4, solve the interaction complexity risk threshold and;5, based on the new total number of voice interaction, total time of voice interaction, average interaction time, calculate the voice interaction complexity value;Obtain the driving risk level corresponding to the voice interaction complexity value.

Description

Technical Field

[0001] This invention belongs to the field of automotive human-machine engineering technology, specifically relating to a safety evaluation method for in-vehicle voice interaction systems that takes into account the complexity of interaction. Background Technology

[0002] With the continuous development of automotive intelligence and informatization, voice interaction has become an important component of human-machine interaction in intelligent cockpits. While voice interaction can minimize the time spent "eyes off the road" and "hands off the steering wheel" during human-vehicle interaction, it also poses a certain threat to driving safety. Especially when voice commands malfunction in certain tasks or environments, drivers may choose to use visual aids and manual interaction to complete the task, potentially leading to visual and operational distractions. Furthermore, the continuous and uninterrupted voice broadcasting of the voice control system requires the driver to process auditory feedback, which can easily cause auditory cognitive distraction, causing them to overlook potential hazards in the surrounding environment and reducing their reaction time to emergencies. Therefore, before mass production of voice control systems, their impact on driving safety must be assessed.

[0003] Currently, domestic and foreign scholars mainly divide the evaluation methods of vehicle voice control systems into two categories: (1) post-event evaluation, which is to compare and analyze the differences in visual behavior, driving performance and other indicators of drivers with voice interaction tasks, without voice interaction tasks and using other interaction methods in mass-produced intelligent vehicles that have integrated voice control systems, and evaluate the impact of voice control systems on driving safety; (2) pre-event evaluation, which is to evaluate the voice control system that will be applied to automotive products in the laboratory, mainly based on the voice technology perspective to evaluate the usability of the system and driver preferences of the relevant design parameters of the voice control system (such as wake-up ability, recognition ability and understanding ability, etc.). This type of research ignores the safety needs of drivers in the actual interaction process.

[0004] In summary, existing safety assessment methods for in-vehicle voice interaction systems mostly focus on comparing the impact of voice interaction with other human-computer interaction methods on characteristic parameters such as driver visual behavior, cognitive load, and driving performance. However, they neglect the non-equivalent relationship between the parameters of the in-vehicle voice interaction system itself and its negative impacts and driving risks. Furthermore, they lack universally applicable quantitative thresholds for assessing driving risks, resulting in limited guiding value of the assessment results for practical applications. Summary of the Invention

[0005] The purpose of this invention is to address the problems of existing evaluation methods neglecting the impact of the in-vehicle voice interaction system's own parameters on driving safety, lacking a universally applicable risk threshold for judging driving safety, and resulting in high collision risks and accident rates due to distraction caused by the in-vehicle voice interaction system. At the same time, it addresses the technical pain points of existing safety evaluation methods, which rely on human subject experiments, are costly and have poor repeatability, resulting in weak practicality of evaluation results in guiding design, cumbersome evaluation processes, and large computational loads. Therefore, this invention proposes a safety evaluation method for in-vehicle voice interaction systems that considers the complexity of the interaction.

[0006] The specific process of a safety evaluation method for an in-vehicle voice interaction system that considers interaction complexity is as follows:

[0007] Step 1: Based on the total number of voice interaction rounds, the total voice interaction time, and the average interaction time, construct a voice interaction complexity function that satisfies the assumptions.

[0008] The complexity function of voice interaction is expressed as:

[0009]

[0010] in, This represents the complexity value of voice interaction;

[0011] Indicates the total number of voice interaction rounds; Indicates the total time of voice interaction; Indicates the average interaction time;

[0012] A function representing the total number of voice interaction rounds, the total voice interaction time, and the average interaction time;

[0013] , , , , , Represents the complexity function of voice interaction;

[0014] Indicates the error term;

[0015] Step 2: Classify driving risk levels; obtain pseudo-labels for driving risk levels based on the driving risk levels, and obtain real labels for driving risk levels based on the pseudo-labels;

[0016] Step 3: Obtain the voice interaction complexity value based on the voice interaction complexity function and the driving risk level. The value belongs to the first Individual driving risk levels Optimal parameters of membership function and optimal parameters The following is the voice interaction complexity function , , , , , The possible values ​​of ;

[0017] Step 4: Based on optimal parameters The following is the voice interaction complexity function , , , , , Solve the interaction complexity risk threshold between the first and second driving risk levels. The interaction complexity risk threshold between the second and third driving risk levels. ;

[0018] Step 5: Based on the new total number of voice interaction rounds, total voice interaction time, average interaction time, and... , , , , , Calculate the complexity value of voice interaction ;

[0019] If the complexity value of voice interaction The value is less than or equal to the interaction complexity risk threshold. The corresponding driving risk level is... ;

[0020] If the complexity value of voice interaction The value is greater than the interaction complexity risk threshold Less than or equal to the interaction complexity risk threshold The corresponding driving risk level is... ;

[0021] If the complexity value of voice interaction The value is greater than the interaction complexity risk threshold The corresponding driving risk level is... .

[0022] The beneficial effects of this invention are as follows:

[0023] This invention proposes a safety evaluation method and system for in-vehicle voice interaction systems that considers interaction complexity. It possesses clear engineering application significance and practical value: On the one hand, by outputting quantitative safety evaluation thresholds, it provides a quantitative basis for parameter design and interaction logic optimization of in-vehicle voice interaction control systems, effectively reducing driver distraction caused by voice interaction operations, thereby reducing the risk of collisions and accident rates due to distraction. On the other hand, it significantly optimizes the safety evaluation process and reduces the computational load. It eliminates the need for repeated experiments with human subjects; by collecting two core indicators—interaction time and number of interaction rounds—through a single system, it can complete the quantitative calculation of interaction complexity and directly output the corresponding safety level by comparing it with preset safety thresholds. This allows for accurate evaluation of the safety and rationality of the voice interaction system design, significantly improving evaluation efficiency and engineering applicability.

[0024] This invention provides a quantitative model for measuring the complexity of voice interaction, comprehensively considering the difficulty of voice interaction task types, interaction time, and interaction rounds. It solves the problem of difficulty in quantifying the correlation of multiple interaction parameters in existing technologies. It improves the fuzzy comprehensive evaluation method and classifies driving risk levels. It uses Bayes' theorem to determine the probability of driving risk under different interaction complexity levels, constructs an optimization function with the objective of minimizing the mean square error between membership degree and posterior probability, calibrates the complexity function parameters, and uses decision trees to determine the risk threshold of interaction complexity. This invention considers the impact of voice interaction system design parameters on driving risk, providing a new methodological framework and quantifiable risk threshold reference for the safety evaluation of in-vehicle voice interaction systems, and providing a theoretical basis for the design optimization and safety evaluation of voice interaction systems. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention;

[0026] Figure 2 Membership function graph for the standard deviation of following distance;

[0027] Figure 3 This is a membership function plot showing the horizontal offset standard deviation.

[0028] Figure 4 Membership function graph for velocity standard deviation;

[0029] Figure 5 The membership function graph represents the total time the line of sight leaves the road ahead.

[0030] Figure 6 Membership function plot for the standard deviation of horizontal fixations;

[0031] Figure 7 Membership function plot for the standard deviation of the vertical gaze point;

[0032] Figure 8 This is a schematic diagram illustrating the mapping relationship between interaction complexity and driving risk. Detailed Implementation

[0033] Specific Implementation Method 1: The specific process of this implementation method for evaluating the safety of an in-vehicle voice interaction system considering interaction complexity is as follows:

[0034] Step 1: Based on the total number of voice interaction rounds, the total voice interaction time, and the average interaction time, construct a voice interaction complexity function that satisfies the assumptions.

[0035] The complexity function of voice interaction is expressed as:

[0036]

[0037] in, This represents the complexity value of voice interaction;

[0038] Indicates the total number of voice interaction rounds; Indicates the total time of voice interaction; Indicates the average interaction time;

[0039] A function representing the total number of voice interaction rounds, the total voice interaction time, and the average interaction time;

[0040] , , , , , Represents the complexity function of voice interaction;

[0041] This represents the error term, used to characterize the model's fitting error or other minor influencing factors that are not taken into consideration;

[0042] Step 2: Classify driving risk levels; obtain pseudo-labels for driving risk levels based on the driving risk levels, and obtain real labels for driving risk levels based on the pseudo-labels;

[0043] Step 3: Obtain the voice interaction complexity value based on the voice interaction complexity function and the driving risk level. The value belongs to the first Individual driving risk levels Optimal parameters of membership function and optimal parameters The following is the voice interaction complexity function , , , , , The possible values ​​of ;

[0044] Step 4: Based on optimal parameters The following is the voice interaction complexity function , , , , , Solve the interaction complexity risk threshold between the first and second driving risk levels. The interaction complexity risk threshold between the second and third driving risk levels. ;

[0045] Step 5: Based on the new total number of voice interaction rounds, total voice interaction time, average interaction time, and... , , , , , Calculate the complexity value of voice interaction ;

[0046] If the complexity value of voice interaction The value is less than or equal to the interaction complexity risk threshold. The corresponding driving risk level is... ;

[0047] If the complexity value of voice interaction The value is greater than the interaction complexity risk threshold Less than or equal to the interaction complexity risk threshold The corresponding driving risk level is... ;

[0048] If the complexity value of voice interaction The value is greater than the interaction complexity risk threshold The corresponding driving risk level is... .

[0049] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the assumptions made in step one are as follows:

[0050] (1) Monotonicity Assumption: Follow The extension, It increases monotonically with the increase of , that is , ;

[0051] (2) Acceleration Effect Hypothesis: When or hour, and right The impact is accelerating;

[0052] in, This is the critical value for each round. This is the time critical value;

[0053] (3) Information content hypothesis: The longer the interactive statement, the more information it conveys. The higher;

[0054] (4) Risk Assumptions: The higher the value, the greater the probability of high driving risk.

[0055] The other steps and parameters are the same as in Specific Implementation Method 1.

[0056] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that: in step two, driving risk levels are divided; pseudo-labels for driving risk levels are obtained based on the driving risk levels; and real labels for driving risk levels are obtained based on the pseudo-labels. The specific process is as follows:

[0057] Step Two: First, based on paired t-tests, two-way repeated measures ANOVA, and correlation analysis, feature screening is performed on all sample data to select a set of universally applicable driving risk characterization indicators capable of distinguishing driving risks. The set of driving risk characterization index observations corresponding to the elements of the set. ;

[0058] Step 22: Based on the selected set of universal driving risk characterization indicators The set of driving risk characterization index observations corresponding to the elements of the set. The Davies-Bouldin Index was used to analyze the set of observations representing driving risk indicators. Processing yields the driving risk level number. The K-means algorithm was used to classify driving risk levels into 10 categories. The risk level is denoted as a set of driving risk levels. ;

[0059] in, Represents a set of driving risk levels;

[0060] This indicates the first driving risk level; Indicates the first Each driving risk level; Indicates the first Each driving risk level; ; ;

[0061] A smaller Davies-Bouldin Index (DB Index) indicates better clustering results. In this embodiment of the invention, the DB Index is minimized when the number of clusters is 3; therefore, the driving risk level is divided into 3 levels, i.e. , ;

[0062] Level 1 Driving Risk The meaning is low risk;

[0063] Level 2 driving risk The meaning is medium risk;

[0064] Level 3 of driving risk The meaning is high risk;

[0065] Based on the K-means algorithm, each sample is assigned to a risk level cluster, thus obtaining a pseudo-label for the driving risk level of each sample. The pseudo-label of the driving risk level corresponding to each sample is denoted as follows: The pseudo-label is the initial risk level given by the clustering algorithm, while the true risk label will be ultimately determined by the membership function.

[0066] Both the pseudo-labels and real labels for driving risk levels are taken from the driving risk level set. , No. The pseudo-label of the driving risk level corresponding to each sample is denoted as follows: ;

[0067] hour, ;

[0068] Indicates the first The pseudo-label for the driving risk level of each sample is low risk;

[0069] Indicates the first The pseudo-label for the driving risk level of each sample is medium risk.

[0070] Indicates the first The pseudo-label for the driving risk level of each sample is high risk;

[0071] Steps 2 and 3: Based on the kernel density estimation method, determine the probability density distribution function of the driving risk characterization index under the sample set corresponding to each driving risk level pseudo-label. ;

[0072] Based on probability density distribution function Bayes' theorem is used to calculate the posterior probability of each driving risk level corresponding to the driving risk characterization index. ;

[0073] Step 24: Construct membership functions adapted to various driving risk characterization indicators Quantifying various driving risk characteristics for the first Individual driving risk levels The degree of membership;

[0074] Step 25: Construct the posterior probability based on the driving risk level. and membership function An optimization model with the objective of minimizing the mean square error is used to solve the optimization model and obtain the optimal parameters of the driving risk characterization index with respect to the membership function.

[0075] Substitute the optimal parameters of the membership function of the driving risk characterization index into the membership function adapted to each driving risk characterization index constructed in step two or four. The calibrated membership function is obtained.

[0076] The dimension is constructed based on the calibrated membership function. The membership matrix corresponding to the driving risk characterization indicators;

[0077] Step 26: Based on the set of driving risk characterization indicators obtained in Step 21, the objective weighting method is used to calculate the weights of the set of driving risk characterization indicators obtained in Step 21, and the indicator weight vector is obtained.

[0078] A single indicator corresponds to a component value in the indicator weight vector. The component values ​​of each indicator are arranged in columns to form... A weight vector of dimensions, where each index corresponds one-to-one with a component in the weight vector.

[0079] The indicators are: following distance standard deviation, lateral deviation standard deviation, speed standard deviation, total time the eyes are off the road ahead, horizontal fixation point standard deviation, and vertical fixation point standard deviation.

[0080] Multiply the index weight vector and the membership matrix corresponding to the driving risk characterization index to obtain the comprehensive membership vector.

[0081] Based on the principle of maximum membership degree, the driving risk level corresponding to the maximum value of the comprehensive membership degree vector is determined;

[0082] A true label is obtained based on the driving risk level.

[0083] Other steps and parameters are the same as in specific implementation method one or two.

[0084] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that: in step two, all sample data are sequentially screened using paired T-tests, two-way repeated measures ANOVA, and correlation analysis to select a set of universally applicable driving risk characterization indicators capable of distinguishing driving risks. The set of driving risk characterization index observations corresponding to the elements of the set. The specific process is as follows:

[0085] Each sample data includes driving performance data and visual behavior data of all drivers when performing various voice interaction tasks (voice interaction tasks include simple phone calls, complex phone calls, simple text messages, complex text messages, simple navigation tasks, and complex navigation tasks);

[0086] The driving performance data includes: lateral driving behavior (such as lane departure, steering wheel angle, etc.) and longitudinal driving behavior (such as mean and standard deviation of speed, mean and standard deviation of acceleration, mean and standard deviation of following distance, etc.).

[0087] The visual behavior data includes: the driver's gaze and saccade behavior, such as the duration of the gaze leaving the road ahead, the standard deviation of the horizontal gaze point position, and the standard deviation of the vertical gaze point position.

[0088] Feature screening was performed on all sample data using paired t-tests, two-way repeated measures ANOVA, and correlation analysis to select a set of universally applicable driving risk characterization indicators capable of distinguishing driving risks. The set of driving risk characterization index observations corresponding to the elements of the set. ;

[0089] Thirty drivers were selected to conduct voice interaction tests in a driving simulator. The interaction tasks included multiple rounds of continuous interaction tasks such as "setting navigation destination," "sending text message," and "making phone calls." Drivers' visual behavior was collected using an eye tracker, and their driving performance was collected using driving simulation software. A total of 180 data samples were collected (30 drivers, each performing 6 tasks, for a total of 180 samples).

[0090] Feature screening of all driver data was performed using paired t-tests, two-way repeated measures ANOVA, and correlation analysis. The process was as follows:

[0091] Paired T-test screening:

[0092] Paired t-tests were performed on the baseline and experimental group sample data respectively, and significance levels were set. When the paired t-test results At that time, the judgment The risk characterization indices corresponding to the paired t-test results showed significant differences between the baseline and experimental groups, preserving... The risk characterization index corresponding to the T-test results and the sample data of the experimental group under that index; delete. The risk characterization index corresponding to the T-test results and the sample data of the experimental group under the index;

[0093] The baseline group sample data consists of driving performance and visual behavior data of all drivers performing only driving tasks;

[0094] The experimental group sample data consisted of driving performance and visual behavior data of all drivers simultaneously performing driving tasks and voice interaction tasks.

[0095] Voice interaction tasks include making phone calls, sending text messages, and navigation.

[0096] Task Description: Diverse voice interaction tasks can generate different interaction durations and rounds. To obtain data on different interaction rounds and durations, we selected typical driving assistance tasks (telephone, SMS, navigation) as the main voice interaction tasks, and designed two sub-task types, simple and complex, for each main task based on the differences in interaction rounds.

[0097] The interaction rounds and specific processes for each task are defined as follows: The driver initiates a voice command, and the voice assistant provides a response feedback to the command. One "command, feedback" dialogue loop is counted as one interaction round.

[0098]

[0099] (2) Screening using two-way repeated measures ANOVA:

[0100] For the candidate risk characterization indicators selected and retained by paired t-tests, and the sample data under these indicators, a two-way repeated measures ANOVA was conducted with voice interaction task type (making phone calls, sending text messages, navigation tasks) and task difficulty (simple, complex) as independent variables, and a significance level was set. ;

[0101] When the analysis results show that the main effect of task type, the main effect of difficulty level, and the interaction effect of the two are all significant (i.e., the performance differences of all voice interaction task types under different difficulties simultaneously meet the significance requirement), the results are considered significant. When a risk characterization indicator is determined to have significant differences in the sample data across different voice interaction tasks and difficulty combinations, the sample data under that risk characterization indicator is retained; if any effect does not reach a significant level, the sample data under the corresponding indicator is deleted.

[0102] (3) Correlation analysis screening:

[0103] The retained variance of two-way repeated measures ANOVA The Pearson correlation coefficients of any one of the candidate risk indicators from the two-way repeated measures ANOVA results with all other risk indicators on the experimental group sample data are used, and a significance level is set. When the analysis results When it is determined that the corresponding risk characterization indicator has no significant correlation with other risk characterization indicators, it is ultimately retained. The analysis results correspond to risk characterization indicators;

[0104] Set of driving risk characterization indicators Represented as:

[0105]

[0106] in,

[0107] This represents the first indicator of driving risk.

[0108] This represents the second indicator of driving risk.

[0109] Indicates the first Individual indicators characterizing driving risk; ;

[0110] Indicates the first One driving risk characterization indicator;

[0111] This represents the total number of indicators representing driving risk. ;

[0112] hour, These represent the standard deviations of following distance, lateral offset, speed, total time the line of sight is off the road, horizontal fixation point, and vertical fixation point, respectively.

[0113] Set of driving risk characterization indicators The corresponding set of driving risk characterization index observations ; ;

[0114] This represents the value of the first driving risk characteristic index for the first sample;

[0115] This indicates the first sample's... The values ​​of each driving risk characterization indicator; ;

[0116] This indicates the first sample's... The values ​​of each driving risk characterization indicator;

[0117] Indicates the first The value of the first driving risk characterization index for each sample; ;

[0118] Indicates the first The first sample The values ​​of each driving risk characterization indicator;

[0119] Indicates the first The first sample The values ​​of each driving risk characterization indicator;

[0120] Indicates the first The value of the first driving risk characterization index for each sample;

[0121] Indicates the first The first sample The values ​​of each driving risk characterization indicator;

[0122] Indicates the first The first sample The values ​​of each driving risk characterization indicator;

[0123] Represent real numbers, Represents the total sample size. ;

[0124] hour, They represent the first The observed values ​​of following distance standard deviation, lateral offset standard deviation, speed standard deviation, total time the line of sight is away from the road ahead, horizontal fixation point standard deviation, and vertical fixation point standard deviation for each sample.

[0125] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0126] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One through Four in that: in steps two and three, the probability density distribution function of the driving risk characterization index under the sample set corresponding to each driving risk level pseudo-label is determined based on the kernel density estimation method. ;

[0127] Based on probability density distribution function Bayes' theorem is used to calculate the posterior probability of each driving risk level corresponding to the driving risk characterization index. ;

[0128] The specific process is as follows:

[0129] Posterior probabilities of driving risk characterization indicators corresponding to each driving risk level The calculation formula is:

[0130]

[0131] in, Indicates the first Individual driving risk levels The prior probability, i.e., the pseudo-label of driving risk level, is The proportion of the total sample; Indicates the first Individual driving risk levels The prior probability, i.e., the pseudo-label of driving risk level, is The proportion of the total sample; ;

[0132] Indicates the first The first sample The values ​​of each driving risk characterization index Clustered into the th group by the Kmeans algorithm Individual driving risk levels The conditional probability; Indicates the first The first sample The values ​​of each driving risk characterization index Clustered into the th group by the Kmeans algorithm Individual driving risk levels The conditional probability;

[0133] Estimated using the Gaussian kernel function:

[0134]

[0135] In the formula, This indicates that the clustering is performed by the k-means algorithm as the th Under each driving risk level The number of samples is a known constant;

[0136] This represents bandwidth, which is a pre-defined constant.

[0137] Indicates the first The first sample The values ​​of the first driving risk characterization index are based on known current observation data (used to estimate the first...). A sample set of driving risk levels any sample The Individual driving risk characterization indicators (corresponding probability density value) ; ;

[0138] Indicates the first The first sample The values ​​of each driving risk indicator are known historical observation values; ; .

[0139] The other steps and parameters are the same as those in one of the specific implementation methods one to four.

[0140] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One through Five in that: in step two and four, the membership function adapted to each driving risk characterization index is constructed. Quantifying various driving risk characteristics for the first Individual driving risk levels The degree of membership; the specific process is as follows:

[0141] In step two and four, the membership functions of each driving risk indicator are calculated using sine functions to quantify the impact of each driving risk indicator on different driving risk levels. The degree of membership;

[0142] The formula for calculating the membership function is as follows:

[0143]

[0144] In the formula, Indicates the first The first sample The values ​​of each driving risk characterization index Belongs to the Individual driving risk levels The degree of membership;

[0145] Indicates the first The first sample The values ​​of each driving risk characterization index The value is in the first driving risk level. The parameter values ​​of the membership function;

[0146] Indicates the first The first sample The values ​​of each driving risk characterization index The value in the first Individual driving risk levels The parameter values ​​of the membership function;

[0147] Indicates the first The first sample The values ​​of each driving risk characterization index The value in the first Individual driving risk levels The parameter values ​​of the membership function;

[0148] Indicates the first The first sample The values ​​of each driving risk characterization index The value in the first Individual driving risk levels The parameter values ​​of the membership function;

[0149] Indicates the first The first sample The values ​​of each driving risk characterization index The value in the first Individual driving risk levels The parameter values ​​of the membership function.

[0150] The other steps and parameters are the same as those in any of the specific implementation methods one to five.

[0151] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One through Six in that: in step two-five, the posterior probability of the driving risk level is constructed. and membership function An optimization model with the objective of minimizing the mean square error is used to solve the optimization model and obtain the optimal parameters of the driving risk characterization index with respect to the membership function.

[0152] Substitute the optimal parameters of the membership function of the driving risk characterization index into the membership function adapted to each driving risk characterization index constructed in step two or four. The calibrated membership function is obtained.

[0153] Construct the membership matrix corresponding to the driving risk characterization index based on the calibrated membership function;

[0154] The specific process is as follows:

[0155] Step 2.51: Optimize the model expression as follows:

[0156]

[0157] in,

[0158] Indicates the first The first sample The values ​​of each driving risk characterization index In the Individual driving risk levels The parameter values ​​of the membership function;

[0159] Indicates the first The first sample The values ​​of each driving risk characterization index In the Individual driving risk levels The optimal parameter values ​​for the membership function;

[0160] Step 252, the first The first sample The values ​​of each driving risk characterization index In the Individual driving risk levels Optimal parameter values ​​of the membership function Substitute the membership function constructed in step two or four to fit each driving risk characterization index. The calibrated membership function is obtained.

[0161] , The membership functions of the six calibrated driving risk characterization indicators are as follows:

[0162]

[0163]

[0164]

[0165]

[0166]

[0167]

[0168] Steps 2, 5, and 3: Construct the membership matrix corresponding to the driving risk characterization index based on the calibrated membership function; represented as:

[0169]

[0170] in, Indicates the first Membership matrix corresponding to each driving risk characterization indicator;

[0171] This indicates the first sample's... Observations of driving risk characterization indicators It belongs to the first level of driving risk. The degree of membership;

[0172] This indicates the first sample's... Observations of driving risk characterization indicators It belongs to the second driving risk level. The degree of membership;

[0173] This indicates the first sample's... Observations of driving risk characterization indicators Belongs to the Individual driving risk levels The degree of membership;

[0174] Indicates the second sample's... Observations of driving risk characterization indicators It belongs to the first level of driving risk. The degree of membership;

[0175] Indicates the second sample's... Observations of driving risk characterization indicators It belongs to the second driving risk level. The degree of membership;

[0176] Indicates the second sample's... Observations of driving risk characterization indicators Belongs to the Individual driving risk levels The degree of membership;

[0177] Indicates the first The first sample Observations of driving risk characterization indicators It belongs to the first level of driving risk. The degree of membership;

[0178] Indicates the first The first sample Observations of driving risk characterization indicators It belongs to the second driving risk level. The degree of membership;

[0179] Indicates the first The first sample Observations of driving risk characterization indicators Belongs to the Individual driving risk levels The degree of subordination.

[0180] The other steps and parameters are the same as those in any of the specific implementation methods one to six.

[0181] Specific Implementation Method Eight: This implementation method differs from one of Specific Implementation Methods One to Seven in that: in step 26, based on the set of observation values ​​of driving risk characterization indicators obtained in step 21, the objective weighting method is used to perform weighting calculations on the set of driving risk characterization indicators selected in step 21 to obtain the indicator weight vector.

[0182] Multiply the index weight vector and the membership matrix corresponding to the driving risk characterization index by matrix multiplication to obtain the comprehensive membership vector.

[0183] Based on the principle of maximum membership degree, the driving risk level corresponding to the maximum value of the comprehensive membership degree vector is determined;

[0184] A true label is obtained based on the driving risk level;

[0185] The specific process is as follows:

[0186] Step 261: Based on the set of driving risk characterization index observations obtained in Step 21. The objective weighting method was used to analyze the set of driving risk characterization indicators selected in step two. Perform weighted calculations to obtain the dimension as follows: Indicator weight vector ;

[0187] Step 262, Each indicator Corresponding to a single-factor fuzzy membership matrix ;

[0188] For the Extract each indicator from a sample. The Line, i.e., the first The first sample Membership matrix corresponding to each driving risk characterization index ;

[0189] Using weight vector For the The first sample The weighted sum of the membership degree vectors of all indicators in the membership matrix corresponding to the i-th driving risk characterization indicator is obtained by taking the membership matrix of the i-th indicator. The combined membership vector of each sample:

[0190]

[0191] Right now ;

[0192] Indicates the first The probability that the comprehensive membership vector of a sample belongs to the first risk level; Indicates the first The probability that the comprehensive membership vector of a sample belongs to the second risk level; Indicates the first The comprehensive membership vector of the sample belongs to the first sample. The probability of each risk level; Indicates the first The comprehensive membership vector of the sample belongs to the first sample. The probability of each risk level;

[0193] After calculation, the first number is obtained. The comprehensive membership vector of each sample It is a 1-line Risk matrix of columns, Each column vector here Corresponding to the The probability of each risk level;

[0194] At that time, six driving risk indicators These represent the standard deviations of following distance, lateral deviation, speed, total time the gaze is off the road, horizontal fixation point, and vertical fixation point, respectively, with corresponding weights of... , , , , , Then the weight vector is ;

[0195] The risk levels are divided into three categories. The first driving risk level means low risk, the second driving risk level means medium risk, and the third driving risk level means high risk.

[0196] , Let represent the observed values ​​of the following distance standard deviation, lateral offset standard deviation, speed standard deviation, total time the line of sight leaves the road ahead, horizontal fixation point standard deviation, and vertical fixation point standard deviation for the first sample, respectively. , , , , , .

[0197] Substituting the membership function from step 2.52, we get The corresponding membership vector . This means that under indicator 1, the probability that the first sample belongs to low risk is 1, the probability that it belongs to medium risk is 0, and the probability that it belongs to high risk is 0. , , , , The meaning is the same as ;

[0198] but ;

[0199] Correspondingly, This indicates the overall membership degree of the first sample, which belongs to the low-risk category.

[0200] This indicates the overall membership degree of the first sample, which belongs to the medium-risk category.

[0201] This indicates the overall membership degree of the first sample, which belongs to the high-risk category.

[0202] The comprehensive membership vector of each sample is used as the rows of a matrix to form a comprehensive fuzzy membership matrix. ;

[0203] Step 263: Based on the principle of maximum membership degree, analyze the comprehensive membership matrix. The driving risk level corresponding to each sample is determined; the specific process is as follows:

[0204] Regarding the first In the comprehensive membership vector of each sample Through formula Finding the risk level index that maximizes membership allows us to determine the [missing information - likely a specific risk level]. The driving risk level corresponding to each sample;

[0205] Step 264: Obtain the real-world label based on the driving risk level; the real-world label includes low risk, medium risk, and high risk.

[0206] The actual label for driving risk level 1 is low risk;

[0207] The actual label for driving risk level 2 is medium risk;

[0208] The actual label for driving risk level 3 is high risk.

[0209] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0210] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One through Eight in that: in step three, step three, based on the voice interaction complexity function and driving risk level, obtains the voice interaction complexity value. The value belongs to the first Individual driving risk levels Optimal parameters of membership function and optimal parameters The voice interaction complexity function under the corresponding driving risk level , , , , , The value of is determined; the specific process is as follows:

[0211] Step 3.1: Construct a mapping relationship between the complexity of voice interaction and the driving risk obtained in Step 2; represented as:

[0212]

[0213] in,

[0214] Indicates the interaction round Interaction time The complexity value of voice interaction at that time;

[0215] This indicates the first driving risk level. The parameter values ​​of the membership function;

[0216] Indicates the first Individual driving risk levels The parameter values ​​of the membership function;

[0217] Indicates the first Individual driving risk levels The parameter values ​​of the membership function;

[0218] Indicates the first Individual driving risk levels The parameter values ​​of the membership function;

[0219] Indicates the first Individual driving risk levels The parameter values ​​of the membership function;

[0220] Represents the complexity value of voice interaction Belonging to the Individual driving risk levels The probability of;

[0221] Step 3.2: Calculate the known number of interaction rounds based on Bayes' theorem. Interaction time Under the conditions, the first Individual driving risk levels posterior probability ; indicates as:

[0222]

[0223] In the formula,

[0224] Indicates the first Prior probabilities of each level; Indicates the first Prior probabilities of each level; ;

[0225] Indicates the first Individual driving risk levels Next, the interaction rounds Interaction time The conditional probability of occurrence;

[0226] Indicates the first Individual driving risk levels Next, the interaction rounds Interaction time The conditional probability of occurrence;

[0227] Represented as:

[0228]

[0229] In the formula, Indicates the first Individual driving risk levels The number of interaction rounds is The number of samples at that time is a known constant.

[0230] Indicates in Individual driving risk levels The number of interaction rounds is sample set The value of the next interaction time is based on the known current observation data (used to estimate the driving risk level). The number of interaction rounds is The interaction time in the sample set is (corresponding probability density value)

[0231] Indicates the first Individual driving risk levels The following are the interaction rounds. , No. The interaction time observations corresponding to each sample are known historical observation data;

[0232] Step 33: Constructing and The optimization function that aims to minimize the mean squared error across all samples and all risk levels is:

[0233]

[0234] In the formula, The complexity value of voice interaction belongs to the first... Individual driving risk levels The optimal parameters of the membership function; For the first Each sample had a sample observation interaction time of [time value]. The number of interaction rounds is Under the conditions, the first Individual driving risk levels The posterior probability of occurrence; In the first Interaction time corresponding to each sample Interaction rounds Under the condition of, interaction complexity Belonging to the Individual driving risk levels The probability of.

[0235] The optimization objective takes all samples as the traversal subject and indexes... Traverse the entire sample set ( For each sample Extract the observation interaction time corresponding to the sample. Interaction rounds and calculate ( , The corresponding interaction complexity At the same time, for each sample ,index Traverse all driving risk levels ( For each risk level Calculate samples Corresponding observation interaction time Interaction rounds In risk level Posterior probability and the complexity of voice interaction at this level Risk membership probability The process involves squared differences; finally, the arithmetic mean of the squared errors between all samples and all risk levels is taken, and the optimal parameters for the voice interaction complexity belonging to each driving risk level are solved with the goal of minimizing this average error. ;

[0236] Steps 3 and 4: Solve for the complexity value of voice interaction Belonging to the Individual driving risk levels Optimal parameters of membership function and optimal parameters The voice interaction complexity function under the corresponding driving risk level , , , , , The value of is determined; the specific process is as follows:

[0237] Step 3-41: Let the number of iterations be... Initialize the membership function parameters , , , , , , Traverse all samples and iteratively execute steps 31 to 33;

[0238] Step 3-42: Interaction time based on each sample Interaction rounds Calculating the complexity of voice interaction Risk level probability mapped to voice interaction complexity ;

[0239] Step 3.43: Interaction time based on each sample Interaction rounds Calculate the posterior probability of driving risk level ;

[0240] Steps 3 and 4: Calculation and Mean squared error across all samples and all risk levels;

[0241] Steps 3, 4, and 5: Update the membership function parameters using the optimized function constructed in step 3.3. , , , , , , ;

[0242] Steps 3, 4, and 6: Let the number of iterations be... Repeat steps 342 to 345 until the mean square error change between two adjacent iterations is less than the preset convergence threshold. At that time To solve for the complexity value of voice interaction Belonging to the Individual driving risk levels Optimal parameters of membership function Based on optimal parameters Obtain the voice interaction complexity function under the corresponding driving risk level. , , , , , The value of .

[0243] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0244] Specific Implementation Method Ten: This implementation method differs from Specific Implementation Methods One through Nine in that: in step four, the optimal parameters are used... The following is the voice interaction complexity function , , , , , Solve the interaction complexity risk threshold between the first and second driving risk levels. The interaction complexity risk threshold between the second and third driving risk levels. The specific process is as follows:

[0245] Step 41: Based on optimal parameters The voice interaction complexity function under the corresponding driving risk level , , , , , Calculate the interaction complexity value; the specific process is as follows:

[0246] Optimal parameters The voice interaction complexity function under the corresponding driving risk level , , , , , Substitute into the voice interaction complexity function

[0247] To obtain the optimal parameters Voice interaction complexity value under the corresponding driving risk level

[0248] Step 4.2: Use a decision tree to determine the optimal parameters. The optimal parameters are obtained by processing the voice interaction complexity value under the corresponding driving risk level. The interaction complexity risk threshold under the corresponding driving risk level, and the interaction complexity risk threshold for the first and second driving risk levels. The interaction complexity risk threshold between the second and third driving risk levels. ; ;

[0249] Level 1 Driving Risk The pseudo-label is low risk;

[0250] Level 2 driving risk The pseudo-label is medium risk;

[0251] Level 3 of driving risk The false label is high risk;

[0252] Interaction complexity risk threshold between the first and second driving risk levels The threshold between low and medium risk is defined as the complexity value of voice interaction. The value is less than or equal to the interaction complexity risk threshold. If so, the corresponding driving risk level is low risk;

[0253] Interaction complexity risk threshold between the second and third driving risk levels The threshold between medium and high risk is defined as the complexity value of voice interaction. The value is greater than the interaction complexity risk threshold Less than or equal to the interaction complexity risk threshold If the voice interaction complexity value is [value missing], then the corresponding driving risk level is medium risk; The value is greater than the interaction complexity risk threshold If so, the corresponding driving risk level is high risk.

[0254] The other steps and parameters are the same as those in any of the specific implementation methods one to nine.

[0255] This invention may have other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A safety evaluation method for an in-vehicle voice interaction system considering interaction complexity, characterized in that: The specific process of the method is as follows: Step 1: Based on the total number of voice interaction rounds, the total voice interaction time, and the average interaction time, construct a voice interaction complexity function that satisfies the assumptions. The complexity function of voice interaction is expressed as: in, This represents the complexity value of voice interaction; Indicates the total number of voice interaction rounds; Indicates the total time of voice interaction; Indicates average time; A function representing the total number of voice interaction rounds, the total voice interaction time, and the average interaction time; , , , , , Represents the complexity function of voice interaction; Indicates the error term; Step 2: Classify driving risk levels; obtain pseudo-labels for driving risk levels based on the driving risk levels, and obtain real labels for driving risk levels based on the pseudo-labels; Step 3: Obtain the voice interaction complexity value based on the voice interaction complexity function and the driving risk level. The value belongs to the first Individual driving risk levels Optimal parameters of membership function and optimal parameters The following is the voice interaction complexity function , , , , , The possible values ​​of ; Step 4: Based on optimal parameters The following is the voice interaction complexity function , , , , , Solve the interaction complexity risk threshold between the first and second driving risk levels. The interaction complexity risk threshold between the second and third driving risk levels. ; Step 5: Based on the new total number of voice interaction rounds, total voice interaction time, average interaction time, and... , , , , , Calculate the complexity value of voice interaction ; If the complexity value of voice interaction The value is less than or equal to the interaction complexity risk threshold. The corresponding driving risk level is... ; If the complexity value of voice interaction The value is greater than the interaction complexity risk threshold Less than or equal to the interaction complexity risk threshold The corresponding driving risk level is... ; If the complexity value of voice interaction The value is greater than the interaction complexity risk threshold The corresponding driving risk level is... .

2. The safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 1, characterized in that: The assumptions made in step one are as follows: (1) Monotonicity assumption: Follow The extension, It increases monotonically with the increase of , that is , ; (2) Acceleration Effect Hypothesis: When or hour, and right The impact is accelerating; in, This is the critical value for each round. This is the time critical value; (3) Information content hypothesis: The longer, The higher; (4) Risk Assumptions: The higher the value, the greater the probability of driving risk.

3. The safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 2, characterized in that: In step two, driving risk levels are divided; pseudo-labels for driving risk levels are obtained based on the driving risk levels; and real labels for driving risk levels are obtained based on the pseudo-labels for driving risk levels. The specific process is as follows: Step Two: First, based on paired t-tests, two-way repeated measures ANOVA, and correlation analysis, feature screening is performed on all sample data to select a set of universally applicable driving risk characterization indicators capable of distinguishing driving risks. The set of driving risk characterization index observations corresponding to the elements of the set. ; Step 22: Based on the selected set of universal driving risk characterization indicators The set of driving risk characterization index observations corresponding to the elements of the set. The Davies-Bouldin Index was used to analyze the set of observations representing driving risk indicators. Processing yields the driving risk level number. The K-means algorithm was used to classify driving risk levels into 10 categories. The risk level is denoted as a set of driving risk levels. ; in, Represents a set of driving risk levels; This indicates the first driving risk level; Indicates the first Each driving risk level; Indicates the first Each driving risk level; ; ; Level 1 Driving Risk The meaning is low risk; Level 2 driving risk The meaning is medium risk; Level 3 of driving risk The meaning is high risk; Based on the K-means algorithm, each sample is assigned to a risk level cluster, thus obtaining a pseudo-label for the driving risk level of each sample. The pseudo-label of the driving risk level corresponding to each sample is denoted as follows: ; Both the pseudo-labels and real labels for driving risk levels are taken from the driving risk level set. , No. The pseudo-label of the driving risk level corresponding to each sample is denoted as follows: ; hour, ; Indicates the first The pseudo-label for the driving risk level of each sample is low risk; Indicates the first The pseudo-label for the driving risk level of each sample is medium risk. Indicates the first The pseudo-label for the driving risk level of each sample is high risk; Steps 2 and 3: Based on the kernel density estimation method, determine the probability density distribution function of the driving risk characterization index under the sample set corresponding to each driving risk level pseudo-label. ; Based on probability density distribution function Bayes' theorem is used to calculate the posterior probability of each driving risk level corresponding to the driving risk characterization index. ; Step 24: Construct membership functions adapted to various driving risk characterization indicators Quantifying various driving risk characteristics for the first Individual driving risk levels The degree of membership; Step 25: Construct the posterior probability based on the driving risk level. and membership function An optimization model with the objective of minimizing the mean square error is used to solve the optimization model and obtain the optimal parameters of the driving risk characterization index with respect to the membership function. Substitute the optimal parameters of the membership function of the driving risk characterization index into the membership function adapted to each driving risk characterization index constructed in step two or four. The calibrated membership function is obtained. The dimension is constructed based on the calibrated membership function. The membership matrix corresponding to the driving risk characterization indicators; Step 26: Based on the set of driving risk characterization indicators obtained in Step 21, the objective weighting method is used to calculate the weights of the set of driving risk characterization indicators obtained in Step 21, and the indicator weight vector is obtained. Multiply the index weight vector and the membership matrix corresponding to the driving risk characterization index to obtain the comprehensive membership vector. Based on the principle of maximum membership degree, the driving risk level corresponding to the maximum value of the comprehensive membership degree vector is determined; A true label is obtained based on the driving risk level.

4. The safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 3, characterized in that: In step two, all sample data are sequentially screened using paired t-tests, two-way repeated measures ANOVA, and correlation analysis to identify a set of universally applicable driving risk characterization indicators capable of distinguishing driving risks. The set of driving risk characterization index observations corresponding to the elements of the set. ; The specific process is as follows: Each sample data includes driving performance data and visual behavior data of all drivers when performing various voice interaction tasks; The driving performance data includes: lateral driving behavior and longitudinal driving behavior; The visual behavior data includes: the driver's gaze and saccade behavior; Feature screening was performed on all sample data using paired t-tests, two-way repeated measures ANOVA, and correlation analysis to select a set of universally applicable driving risk characterization indicators capable of distinguishing driving risks. The set of driving risk characterization index observations corresponding to the elements of the set. ; Set of driving risk characterization indicators Represented as: in, This represents the first indicator of driving risk. This represents the second indicator of driving risk. Indicates the first Individual indicators characterizing driving risk; ; Indicates the first One driving risk characterization indicator; This represents the total number of indicators representing driving risk. ; hour, These represent the standard deviations of following distance, lateral offset, speed, total time the line of sight is off the road, horizontal fixation point, and vertical fixation point, respectively. Set of driving risk characterization indicators The corresponding set of driving risk characterization index observations ; This represents the value of the first driving risk characteristic index for the first sample; This indicates the first sample's... The values ​​of each driving risk characterization indicator; ; This indicates the first sample's... The values ​​of each driving risk characterization indicator; Indicates the first The value of the first driving risk characterization index for each sample; ; Indicates the first The first sample The values ​​of each driving risk characterization indicator; Indicates the first The first sample The values ​​of each driving risk characterization indicator; Indicates the first The value of the first driving risk characterization index for each sample; Indicates the first The first sample The values ​​of each driving risk characterization indicator; Indicates the first The first sample The values ​​of each driving risk characterization indicator; Represent real numbers, Represents the total sample size. ; hour, They represent the first The observed values ​​of following distance standard deviation, lateral offset standard deviation, speed standard deviation, total time the line of sight is away from the road ahead, horizontal fixation point standard deviation, and vertical fixation point standard deviation for each sample.

5. The safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 4, characterized in that: In steps two and three, the probability density distribution function of the driving risk characterization index is determined based on the kernel density estimation method under the sample set corresponding to each driving risk level pseudo-label. ; Based on probability density distribution function Bayes' theorem is used to calculate the posterior probability of each driving risk level corresponding to the driving risk characterization index. ; The specific process is as follows: Posterior probabilities of driving risk characterization indicators corresponding to each driving risk level The calculation formula is: in, Indicates the first Individual driving risk levels The prior probability, i.e., the pseudo-label of driving risk level, is The proportion of the total sample; Indicates the first Individual driving risk levels The prior probability, i.e., the pseudo-label of driving risk level, is The proportion of the total sample; ; Indicates the first The first sample The values ​​of each driving risk characterization index Clustered into the th group by the Kmeans algorithm Individual driving risk levels The conditional probability; Indicates the first The first sample The values ​​of each driving risk characterization index Clustered into the th group by the Kmeans algorithm Individual driving risk levels The conditional probability; Estimated using the Gaussian kernel function: In the formula, This indicates that the clustering is performed by the k-means algorithm as the th Under each driving risk level The number of samples; Indicates bandwidth; Indicates the first The first sample The values ​​of each driving risk characterization indicator; ; ; Indicates the first The first sample The values ​​of each driving risk characterization indicator; ; .

6. The safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 5, characterized in that: In step two and four, a membership function is constructed to adapt to each driving risk characterization index. Quantifying various driving risk characteristics for the first Individual driving risk levels The degree of membership; the specific process is as follows: The formula for calculating the membership function is as follows: In the formula, Indicates the first The first sample The values ​​of each driving risk characterization index Belongs to the Individual driving risk levels The degree of membership; Indicates the first The first sample The values ​​of each driving risk characterization index The value is in the first driving risk level. The parameter values ​​of the membership function; Indicates the first The first sample The values ​​of each driving risk characterization index The value in the first Individual driving risk levels The parameter values ​​of the membership function; Indicates the first The first sample The values ​​of each driving risk characterization index The value in the first Individual driving risk levels The parameter values ​​of the membership function; Indicates the first The first sample The values ​​of each driving risk characterization index The value in the first Individual driving risk levels The parameter values ​​of the membership function; Indicates the first The first sample The values ​​of each driving risk characterization index The value in the first Individual driving risk levels The parameter values ​​of the membership function.

7. A safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 6, characterized in that: In step two-five, the posterior probability of the driving risk level is constructed. and membership function An optimization model with the objective of minimizing the mean square error is used to solve the optimization model and obtain the optimal parameters of the driving risk characterization index with respect to the membership function. Substitute the optimal parameters of the membership function of the driving risk characterization index into the membership function adapted to each driving risk characterization index constructed in step two or four. The calibrated membership function is obtained. Construct the membership matrix corresponding to the driving risk characterization index based on the calibrated membership function; The specific process is as follows: Step 2.51: Optimize the model expression as follows: in, Indicates the first The first sample The values ​​of each driving risk characterization index In the Individual driving risk levels The parameter values ​​of the membership function; Indicates the first The first sample The values ​​of each driving risk characterization index In the Individual driving risk levels The optimal parameter values ​​for the membership function; Step 252, the first The first sample The values ​​of each driving risk characterization index In the Individual driving risk levels Optimal parameter values ​​of the membership function Substitute the membership function constructed in step two or four to fit each driving risk characterization index. The calibrated membership function is obtained. Steps 2, 5, and 3: Construct the membership matrix corresponding to the driving risk characterization index based on the calibrated membership function; represented as: in, Indicates the first Membership matrix corresponding to each driving risk characterization indicator; This indicates the first sample's... Observations of driving risk characterization indicators It belongs to the first level of driving risk. The degree of membership; This indicates the first sample's... Observations of driving risk characterization indicators It belongs to the second driving risk level. The degree of membership; This indicates the first sample's... Observations of driving risk characterization indicators Belongs to the Individual driving risk levels The degree of membership; Indicates the second sample's... Observations of driving risk characterization indicators It belongs to the first level of driving risk. The degree of membership; Indicates the second sample's... Observations of driving risk characterization indicators It belongs to the second driving risk level. The degree of membership; Indicates the second sample's... Observations of driving risk characterization indicators Belongs to the Individual driving risk levels The degree of membership; Indicates the first The first sample Observations of driving risk characterization indicators It belongs to the first level of driving risk. The degree of membership; Indicates the first The first sample Observations of driving risk characterization indicators It belongs to the second driving risk level. The degree of membership; Indicates the first The first sample Observations of driving risk characterization indicators Belongs to the Individual driving risk levels The degree of subordination.

8. The safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 7, characterized in that: In step 26, based on the set of driving risk characterization index observations obtained in step 21, the objective weighting method is used to calculate the weights of the set of driving risk characterization indexes selected in step 21, and the index weight vector is obtained. Multiply the index weight vector and the membership matrix corresponding to the driving risk characterization index by matrix multiplication to obtain the comprehensive membership vector. Based on the principle of maximum membership degree, the driving risk level corresponding to the maximum value of the comprehensive membership degree vector is determined; A true label is obtained based on the driving risk level; The specific process is as follows: Step 261: Based on the set of driving risk characterization index observations obtained in Step 21. The objective weighting method was used to analyze the set of driving risk characterization indicators selected in step two. Perform weighted calculations to obtain the dimension as follows: Indicator weight vector ; Step 262, Each indicator Corresponding to a single-factor fuzzy membership matrix ; For the Extract each indicator from a sample. The Line, i.e., the first The first sample Membership matrix corresponding to each driving risk characterization index ; Using weight vector For the The first sample The weighted sum of the membership degree vectors of all indicators in the membership matrix corresponding to the i-th driving risk characterization indicator is obtained by taking the membership matrix of the i-th indicator. The combined membership vector of each sample: Right now ; Indicates the first The probability that the comprehensive membership vector of a sample belongs to the first risk level; Indicates the first The probability that the comprehensive membership vector of a sample belongs to the second risk level; Indicates the first The comprehensive membership vector of the sample belongs to the first sample. The probability of each risk level; Indicates the first The comprehensive membership vector of the sample belongs to the first sample. The probability of each risk level; Step 263: Based on the principle of maximum membership degree, analyze the comprehensive membership matrix. The driving risk level corresponding to each sample is determined in the process; The specific process is as follows: Regarding the first In the comprehensive membership vector of each sample Through formula Finding the risk level index that maximizes membership allows us to determine the [missing information - likely a specific risk level]. The driving risk level corresponding to each sample; Step 264: Obtain the true label based on the driving risk level; The actual risk level is categorized as low, medium, and high. The actual label for driving risk level 1 is low risk; The actual label for driving risk level 2 is medium risk; The actual label for driving risk level 3 is high risk.

9. A safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 8, characterized in that: In step three, based on the voice interaction complexity function and the driving risk level, the voice interaction complexity value is obtained. The value belongs to the first Individual driving risk levels Optimal parameters of membership function and optimal parameters The voice interaction complexity function under the corresponding driving risk level , , , , , The value of is determined; the specific process is as follows: Step 3.1: Construct a mapping relationship between the complexity of voice interaction and the driving risk obtained in Step 2; represented as: in, Indicates the interaction round Interaction time The complexity value of voice interaction at that time; This indicates the first driving risk level. The parameter values ​​of the membership function; Indicates the first Individual driving risk levels The parameter values ​​of the membership function; Indicates the first Individual driving risk levels The parameter values ​​of the membership function; Indicates the first Individual driving risk levels The parameter values ​​of the membership function; Indicates the first Individual driving risk levels The parameter values ​​of the membership function; Represents the complexity value of voice interaction Belonging to the Individual driving risk levels The probability of; Step 3.2: Calculate the known number of interaction rounds based on Bayes' theorem. Interaction time Under the conditions, the first Individual driving risk levels posterior probability ; indicates as: In the formula, Indicates the first Prior probabilities of each level; Indicates the first Prior probabilities of each level; ; Indicates the first Individual driving risk levels Next, the interaction rounds Interaction time The conditional probability of occurrence; Indicates the first Individual driving risk levels Next, the interaction rounds Interaction time The conditional probability of occurrence; Represented as: In the formula, Indicates the first Individual driving risk levels The number of interaction rounds is The number of samples at that time; Indicates in Individual driving risk levels The number of interaction rounds is sample set The value of the next interaction time; Indicates the first Individual driving risk levels The following are the interaction rounds. , No. The interaction time observations corresponding to each sample; Step 33: Constructing and The optimization function that aims to minimize the mean squared error across all samples and all risk levels is: In the formula, The complexity value of voice interaction belongs to the first... Individual driving risk levels The optimal parameters of the membership function; For the first Each sample had a sample observation interaction time of [time value]. The number of interaction rounds is Under the conditions, the first Individual driving risk levels The posterior probability of occurrence; In the first Interaction time corresponding to each sample Interaction rounds Under the condition of, interaction complexity Belonging to the Individual driving risk levels The probability of; Steps 3 and 4: Solve for the complexity value of voice interaction Belonging to the Individual driving risk levels Optimal parameters of membership function and optimal parameters The voice interaction complexity function under the corresponding driving risk level , , , , , The value of is determined; the specific process is as follows: Step 3-41: Let the number of iterations be... Initialize the membership function parameters , , , , , , Traverse all samples and iteratively execute steps 31 to 33; Step 3-42: Interaction time based on each sample Interaction rounds Calculating the complexity of voice interaction Risk level probability mapped to voice interaction complexity ; Step 3.43: Interaction time based on each sample Interaction rounds Calculate the posterior probability of driving risk level ; Steps 3 and 4: Calculation and Mean squared error across all samples and all risk levels; Steps 3, 4, and 5: Update the membership function parameters using the optimized function constructed in step 3.

3. , , , , , , ; Steps 3, 4, and 6: Let the number of iterations be... Repeat steps 342 to 345 until the mean square error change between two adjacent iterations is less than the preset convergence threshold. At that time To solve for the complexity value of voice interaction Belonging to the Individual driving risk levels Optimal parameters of membership function Based on optimal parameters Obtain the voice interaction complexity function under the corresponding driving risk level. , , , , , The value of .

10. A safety evaluation method for an in-vehicle voice interaction system considering interaction complexity according to claim 9, characterized in that: Step four is based on optimal parameters The following is the voice interaction complexity function , , , , , Solve the interaction complexity risk threshold between the first and second driving risk levels. The interaction complexity risk threshold between the second and third driving risk levels. ; The specific process is as follows: Step 41: Based on optimal parameters The voice interaction complexity function under the corresponding driving risk level , , , , , Calculate the interaction complexity value; The specific process is as follows: Optimal parameters The voice interaction complexity function under the corresponding driving risk level , , , , , Substitute into the voice interaction complexity function To obtain the optimal parameters The voice interaction complexity value under the corresponding driving risk level; Step 4.2: Use a decision tree to determine the optimal parameters. The optimal parameters are obtained by processing the voice interaction complexity value under the corresponding driving risk level. The interaction complexity risk threshold under the corresponding driving risk level, and the interaction complexity risk threshold for the first and second driving risk levels. The interaction complexity risk threshold between the second and third driving risk levels. ; Level 1 Driving Risk The pseudo-label is low risk; Level 2 driving risk The pseudo-label is medium risk; Level 3 of driving risk The false label is high risk; Interaction complexity risk threshold between the first and second driving risk levels The threshold between low and medium risk is defined as the complexity value of voice interaction. The value is less than or equal to the interaction complexity risk threshold. If so, the corresponding driving risk level is low risk; Interaction complexity risk threshold between the second and third driving risk levels The threshold between medium and high risk is defined as the complexity value of voice interaction. The value is greater than the interaction complexity risk threshold Less than or equal to the interaction complexity risk threshold If the voice interaction complexity value is [value missing], then the corresponding driving risk level is medium risk; The value is greater than the interaction complexity risk threshold If so, the corresponding driving risk level is high risk.

Images