CN107730835B - A vehicle driver fatigue recognition method based on stress response ability - Google Patents

Abstract

A vehicle driver fatigue identification method based on stress response ability belongs to the technical field of automobile safety. function, the driver's reaction time and execution time in the awake state obey the normal distribution, and the driver's reaction time and execution time in the fatigue state obey the log-normal distribution; the data transformation of the samples of two different distributions makes the measurement The standard is consistent; the binary joint distribution of the transformed reaction time and execution time is established, the reaction time and execution time and the prior probability of the two fatigue states are obtained, and then the posterior probability of the reaction time and execution time under the known fatigue state is obtained. Test probability; use Naive Bayes algorithm and Logistic algorithm to identify driver fatigue. The present invention can eliminate the influence of individual differences and driver's distraction on the accuracy of fatigue judgment.

Description

A vehicle driver fatigue recognition method based on stress response ability

technical field

The invention belongs to the technical field of automobile safety, and particularly relates to a method for identifying a driver's fatigue state.

Background technique

Driving fatigue is mainly manifested in that the driver's reaction time becomes longer and the alertness becomes lower after driving for a continuous period of time. Continuing to drive the vehicle after the driver enters a state of fatigue is extremely prone to road traffic accidents. At present, there are two main problems in driving fatigue recognition:

1. It is difficult to determine the threshold of the driving fatigue identification index. Each driver behaves differently when fatigued, and the fatigue identification threshold of one driver cannot be used to determine the fatigue status of other drivers.

2. During the long-term driving process, the driver will appear inattention and distraction in the awake state. In the fatigued state, the driver will resist fatigue and appear in a short-term awake state, which will lead to the recognition of the driver's fatigue state. mistake.

In this method, the reaction time and the execution time are selected to characterize the driver's stress response ability, and a new technical method is urgently needed in the field to solve the problem in view of the deficiencies in the prior art.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a method for identifying the fatigue of automobile drivers based on the ability of stress response. The indicators are classified to identify the driver's fatigue state. The invention effectively avoids the problem that the driver's fatigue state is identified incorrectly due to individual differences and the driver's distraction.

The present invention adopts the following technical scheme:

A vehicle driver fatigue identification method based on stress response ability, characterized in that it comprises the following steps:

Step 1. Design of the test plan: Using the car driving simulator, the driver drives continuously in the virtual road traffic environment until the driver refuses to drive due to subjective fatigue. A group of LED lights are set in the driver's front vision, every 1 minute. ～3 minutes of random flashing, the flashing duration is 1 second, it is used as a stimulus to the driver, and the driver takes the brakes when he finds external stimuli; the driver's subjective fatigue state Y is obtained manually, and the driver's subjective fatigue state Y is obtained. Fatigue state Y is divided into two states: fatigued and awake;

Step 2. Data collection and preprocessing: Collect and record the data of the reaction time R and execution time E of the driver taking braking when they find external stimuli during the entire simulated driving process, and record the driver's subjective fatigue. State Y is divided into two states: fatigue and awake. The fatigue state is recorded as 1, and the awake state is recorded as 0. According to the analysis of the characteristics of the driving state, the reaction time R and execution time E in the fatigue state obey the log-normal distribution, and the reaction time in the awake state is obtained. Time R and execution time E follow a normal distribution, that is:

Among them, N ₀ means that the reaction time R in the awake state obeys the normal distribution, N ₂ means that the execution time E obeys the normal distribution in the awake state, N ₁ means that the ln R in the fatigue state obeys the normal distribution, and N ₃ means that the ln E in the fatigue state Obey the normal distribution, μ ₀ is the mean value of the reaction time R in the awake state, μ ₂ is the mean value of the execution time E in the awake state, μ ₁ is the mean value of ln R in the fatigue state, μ ₃ is the mean value of ln E in the fatigue state , σ ₀ is the standard deviation of the reaction time R in the awake state, σ ₂ is the standard deviation of the execution time E in the awake state, σ ₁ is the standard deviation of ln R in the fatigue state, σ ₃ is the standard deviation of ln E in the fatigue state ;

Step 3. Data transformation: For the reaction time R and execution time E in any state, use the transformation f to map the reaction time R and execution time E to the neighborhood of 1, and get

In the neighborhood of 1, there is ln R ⁰ ≈R ⁰ -1, and ln E ⁰ ≈E ⁰ -1 is obtained,

Among them, f represents a transformation method to transform the reaction time R and execution time E in any state into R ⁰ and E ⁰ , N' ₀ means that R ⁰ obeys a normal distribution in the awake state, and N' ₂ means that E in the awake state ⁰ obeys the normal distribution, N' ₁ means that ln R ⁰ obeys the normal distribution in the fatigue state, N' ₃ means that the ln E ⁰ obeys the normal distribution under the fatigue state, and N″ ₀ means that the R ⁰ -1 obeys the normal distribution in the awake state distribution, N″ ₂ means E ⁰ -1 obeys normal distribution in awake state, N″ ₁ means R ⁰ -1 obeys normal distribution in fatigue state, N″ ₃ means E ⁰ -1 obeys normal distribution in fatigue state, R ⁰ is the data after the reaction time R is mapped to the neighborhood of 1 through the transformation f, E ⁰ is the data after the execution time E is mapped to the neighborhood of 1 through the transformation f, μ′ ₀ is the mean value of R ⁰ in the awake state, μ′ ₂ is the mean value of E ⁰ in the awake state, μ′ ₁ is the mean value of ln R ⁰ in the fatigue state, μ′ ₃ is the mean value of ln E ⁰ in the fatigue state, and σ′ ₀ is the standard deviation of R ⁰ in the awake state , σ′ ₂ is the standard deviation of E ⁰ in the awake state, σ′ ₁ is the standard deviation of ln R ⁰ in the fatigue state, and σ′ ₃ is the standard deviation of ln E ⁰ in the fatigue state;

In any state, R ⁰ -1, E ⁰ -1 obey the normal distribution, and the transformation is constructed:

which defines the transformation:

It is a function that transforms the reaction time R and execution time E into R', E', R', E' are respectively the reaction time R, the execution time E after the transformation The obtained data; x is the argument of the function, and the assignment of x is R or E;

f(R) represents a transformation method to transform the reaction time R in any state into R ⁰ ;

f(E) represents a transformation method for transforming the execution time E in any state into E ⁰ ;

Step 4. Establish a probability model based on Naive Bayes: set X=(R′,E′) ^T , X is a two-dimensional random vector, which obeys the bivariate normal distribution, that is, X～N(μ,Σ ), where N indicates that the two-dimensional random vector X obeys a bivariate normal distribution, μ is the mean vector of the two-dimensional random vector X, and Σ is the covariance matrix of the two-dimensional random vector X, Cov represents the covariance between random variables, T represents the transpose of the matrix, the driver's subjective fatigue state Y is a Bernoulli variable, and the probability density function of the driver's subjective fatigue state Y is:

P(Y)＝φ ^Y (1-φ) ^1-Y (1)

The conditional probability distributions of the two-dimensional random vector X in the two states of fatigue and wakefulness are:

According to the Naive Bayes formula and the full probability formula:

Bring equations (1), (2), (3) into equation (4), and simplify them to get:

in is a linear function of X, simplified to -θ ^T X-θ ₀ , φ is the probability that the driver state is fatigued, Y=0, 1, μ ₀ is the mean vector of the two-dimensional random vector X in the awake state, μ ₁ is the mean vector of the two-dimensional random vector X in the fatigue state, and Σ is the covariance matrix of the two-dimensional random vector X;

Step 5. Estimating model parameters: Using the maximum likelihood estimation method, the parameters in the formula (5) are the probability φ that the driver's state is fatigue, the mean vector μ ₀ of the two-dimensional random vector X in the awake state, and the two-dimension vector in the fatigue state. The mean vector μ ₁ of the random vector X and the covariance matrix Σ of the two-dimensional random vector X are estimated, and there is a likelihood equation:

Take the logarithm of the likelihood equation:

get:

Where {Y ⁽ⁱ⁾ = 1}, {Y ⁽ⁱ⁾ = 0} are implicit functions, the formula in parentheses is true, the function takes the value 1; otherwise the function takes the value 0, m is the number of samples in the training set ;

Step 6. Establish a logistic-based probability model: Obtain the logistic function from step 4:

Among them, θ ₀ , θ ₁ and θ ₂ are represented by θ _k , and θ _k is the probability φ that the driver state is fatigued, the mean vector μ ₀ of the two-dimensional random vector X in the awake state, and the two-dimensional random vector X in the fatigue state The mean vector μ ₁ of , a suitable function of the covariance matrix Σ of the two-dimensional random vector X, k=0,1,2;

Step 7. Fatigue identification: The input reaction time R and execution time E are transformed from the logistic function in step 6. The obtained R', E', the output probability is greater than 0.5, the driver is in the fatigue state; the output probability is less than 0.5, the driver is in the awake state.

Compared with the prior art, the beneficial effects brought by the present invention are:

1. For the first time, the present invention proposes a driver fatigue identification method using the naive Bayes algorithm and the Logistic algorithm with the driver's reaction time and execution time as indicators.

2. The logistic classification algorithm applied in the present invention establishes a fatigue recognition classifier to recognize the driver's fatigue state by training data samples of a certain driver, which effectively avoids recognition errors caused by individual differences in the process of driving fatigue recognition. The problem.

3. The naive Bayes algorithm applied in the present invention uses the known prior probability to infer the posterior probability that will occur to calculate the posterior probability and the error rate of each sample, and uses the maximum posterior probability to divide the classification of the samples . Using the known state information of the driver in the previous time period to infer the state probability of the subsequent driver, and then determine the exact time when the driver reaches fatigue, which can effectively eliminate the abnormal situation caused by the driver's own reasons.

4. The present invention provides a new means for improving the accuracy of driver fatigue identification, and provides a new idea for the fatigue early warning and assisted driving technology.

Detailed ways

In order to illustrate the present invention more clearly, the present invention will be further described below with reference to the preferred embodiments. Those skilled in the art will understand. The content specifically described below is illustrative rather than restrictive, and should not limit the protection scope of the present invention.

The present invention provides a method for identifying fatigue of automobile drivers based on the ability of stress response, which is characterized by comprising the following steps:

Step 1. Design of the test plan: Using the car driving simulator, the driver drives continuously in the virtual road traffic environment until the driver refuses to drive due to subjective fatigue; ~3 minutes of random flashing, the flashing duration is 1 second, it is used as a stimulus for the driver, and the driver takes the brakes when he finds external stimuli; the investigator sits on the co-pilot, inquires and records the driver's Subjective fatigue state Y;

Step 2. Data collection and preprocessing: Collect and record the data of the reaction time R and execution time E of the driver taking braking when they find external stimuli during the entire simulated driving process, and record the driver's subjective fatigue. The state Y is divided into fatigue and sobriety, fatigue is recorded as 1, and sobriety is recorded as 0; according to the analysis of the driving state characteristics, the reaction time R and execution time E in the fatigue state obey the log-normal distribution, and the reaction time R and the awake state are obtained. The execution time E follows a normal distribution, that is:

Among them, N ₀ means that the reaction time R in the awake state obeys the normal distribution, N ₂ means that the execution time E obeys the normal distribution in the awake state, N ₁ means that the ln R in the fatigue state obeys the normal distribution, and N ₃ means that the ln E in the fatigue state Obey the normal distribution, μ ₀ is the mean value of the reaction time R in the awake state, μ ₂ is the mean value of the execution time E in the awake state, μ ₁ is the mean value of ln R in the fatigue state, μ ₃ is the mean value of ln E in the fatigue state , σ ₀ is the standard deviation of the reaction time R in the awake state, σ ₂ is the standard deviation of the execution time E in the awake state, σ ₁ is the standard deviation of ln R in the fatigue state, σ ₃ is the standard deviation of ln E in the fatigue state ;

Step 3. Data transformation: For the reaction time R and execution time E in any state, use the transformation f to map the reaction time R and execution time E to the neighborhood of 1, and get

In the neighborhood of 1, there is ln R ⁰ ≈R ⁰ -1, and ln E ⁰ ≈E ⁰ -1 is obtained,

where f represents a transformation method for transforming reaction time R and execution time E into R ⁰ and E ⁰ , N' ₀ means that R ⁰ obeys a normal distribution in the awake state, and N' ₂ means that E ⁰ obeys a normal distribution in the awake state Distribution, N' ₁ means ln R ⁰ obeys normal distribution in fatigue state, N' ₃ means ln E ⁰ obeys normal distribution in fatigue state, N″ ₀ means R ⁰ -1 obeys normal distribution in awake state, N″ ₂ means E ⁰ -1 obeys normal distribution in awake state, N″ ₁ means R ⁰ -1 obeys normal distribution in fatigue state, N″ ₃ means E ⁰ -1 obeys normal distribution in fatigue state, R ⁰ is response Time R is the data after the transformation f is mapped to the neighborhood of 1, E ⁰ is the data after the execution time E is mapped to the neighborhood of 1 through the transformation f, μ′ ₀ is the mean value of R ⁰ in the awake state, μ′ ₂ is The mean value of E ⁰ in the awake state, μ′ ₁ is the mean value of ln R ⁰ in the fatigue state, μ′ ₃ is the mean value of ln E ⁰ in the fatigue state, σ′ ₀ is the standard deviation of R ⁰ in the awake state, σ′ ₂ is the standard deviation of E ⁰ in the awake state, σ′ ₁ is the standard deviation of ln R ⁰ in the fatigue state, and σ′ ₃ is the standard deviation of ln E ⁰ in the fatigue state;

In any state, R ⁰ -1, E ⁰ -1 obey the normal distribution, and the transformation is constructed:

which defines the transformation:

It is a function that transforms the reaction time R and execution time E into R', E', R', E' are respectively the reaction time R, the execution time E after the transformation The obtained data; x is the argument of the function, and the assignment of x is R or E;

f(R) represents a transformation method to transform the reaction time R in any state into R ⁰ ;

f(E) represents a transformation method for transforming the execution time E in any state into E ⁰ ;

Step 4. Establish a probability model based on Naive Bayes: set X=(R′,E′) ^T , X is a two-dimensional random vector, which obeys the bivariate normal distribution, that is, X～N(μ,Σ ), where N indicates that the two-dimensional random vector X obeys a bivariate normal distribution, μ is the mean vector of the two-dimensional random vector X, and Σ is the covariance matrix of the two-dimensional random vector X, Cov represents the covariance between random variables, T represents the transpose of the matrix, the driver's subjective fatigue state Y is a Bernoulli variable, and the probability density function of the driver's subjective fatigue state Y is:

P(Y)＝φ ^Y (1-φ) ^1-Y (1)

The conditional probability distributions of X under the two states of fatigue and wakefulness are:

According to the Naive Bayes formula and the full probability formula:

Bring equations (1), (2), (3) into equation (4), and simplify them to get:

in is a linear function of X, simplified to -θ ^T X-θ ₀ , φ is the probability that the driver state is fatigued, Y=0, 1, μ ₀ is the mean vector of the two-dimensional random vector X in the awake state, μ ₁ is the mean vector of the two-dimensional random vector X in the fatigue state, and Σ is the covariance matrix of the two-dimensional random vector X;

Step 5. Estimating model parameters: Using the maximum likelihood estimation method, the parameters in the formula (5) are the probability φ that the driver's state is fatigue, the mean vector μ ₀ of the two-dimensional random vector X in the awake state, and the two-dimension vector in the fatigue state. The mean vector μ ₁ of the random vector X and the covariance matrix Σ of the two-dimensional random vector X are estimated, and there is a likelihood equation:

Take the logarithm of the likelihood equation:

get:

Where {Y ⁽ⁱ⁾ = 1}, {Y ⁽ⁱ⁾ = 0} are implicit functions, the formula in parentheses is true, the function takes the value 1; otherwise the function takes the value 0, m is the number of samples in the training set ;

Step 6. Establish a logistic-based probability model: Obtain the logistic function from step 4:

Among them, θ ₀ , θ ₁ and θ ₂ are represented by θ _k , and θ _k is the probability φ that the driver state is fatigued, the mean vector μ ₀ of the two-dimensional random vector X in the awake state, and the two-dimensional random vector X in the fatigue state The mean vector μ ₁ of , a suitable function of the covariance matrix Σ of the two-dimensional random vector X, k=0,1,2;

Step 7. Fatigue identification: The input reaction time R and execution time E are transformed from the logistic function in step 6. The obtained R', E', the output probability is greater than 0.5, the driver is in the fatigue state; the output probability is less than 0.5, the driver is in the awake state.

The present invention provides a method for identifying fatigue of automobile drivers based on the ability of stress response. The method is as follows: collecting data of the driver's reaction time R and execution time E during the entire simulated driving process; obtaining probability densities in different states Function, the driver's reaction time R and execution time E in the awake state obey the normal distribution, and the driver's reaction time R and execution time E in the fatigue state obey the log-normal distribution; The data transformation makes the measurement standard consistent; the binary joint distribution of the transformed reaction time R and execution time E is established, and its distribution is a multivariate normal distribution; the reaction time R and execution time E and the prior probability of the two fatigue states are obtained , and then obtain the posterior probability of reaction time R and execution time E under known two fatigue states; obtain the distribution of two fatigue states Y under known reaction time R and execution time E according to the Naive Bayes formula; The logistic regression model classifies the two fatigue states Y, the probability is greater than 0.5, it is the fatigue state, and the probability is less than 0.5, the awake state. Compared with the existing fatigue identification algorithm, the invention has a great improvement, and can eliminate the influence of individual differences and driver distraction on the accuracy of fatigue judgment. A new method for judging the driver's fatigue state is proposed.

Claims (1)

1. a vehicle driver fatigue identification method based on stress response ability, is characterized in that, comprises the following steps: Step 1. Design of the test plan: Using the car driving simulator, the driver drives continuously in the virtual road traffic environment until the driver refuses to drive due to subjective fatigue. A group of LED lights are set in the driver's front vision, every 1 minute. ～3 minutes of random flashing, the flashing duration is 1 second, it is used as a stimulus to the driver, and the driver takes the brakes when he finds external stimuli; the driver's subjective fatigue state Y is obtained manually, and the driver's subjective fatigue state Y is obtained. Fatigue state Y is divided into two states: fatigued and awake; Step 2. Data collection and preprocessing: Collect and record the data of the reaction time R and execution time E of the driver taking braking when they find external stimuli during the entire simulated driving process, and record the driver's subjective fatigue. State Y is divided into two states: fatigue and awake. The fatigue state is recorded as 1, and the awake state is recorded as 0. According to the analysis of the characteristics of the driving state, the reaction time R and execution time E in the fatigue state obey the log-normal distribution, and the reaction time in the awake state is obtained. Time R and execution time E follow a normal distribution, that is: Among them, N ₀ means that the reaction time R in the awake state obeys the normal distribution, N ₂ means that the execution time E obeys the normal distribution in the awake state, N ₁ means that the ln R in the fatigue state obeys the normal distribution, and N ₃ means that the ln E in the fatigue state Obey the normal distribution, μ ₀ is the mean value of the reaction time R in the awake state, μ ₂ is the mean value of the execution time E in the awake state, μ ₁ is the mean value of ln R in the fatigue state, μ ₃ is the mean value of ln E in the fatigue state , σ ₀ is the standard deviation of the reaction time R in the awake state, σ ₂ is the standard deviation of the execution time E in the awake state, σ ₁ is the standard deviation of ln R in the fatigue state, σ ₃ is the standard deviation of ln E in the fatigue state ; Step 3. Data transformation: For the reaction time R and execution time E in any state, use the transformation f to map the reaction time R and execution time E to the neighborhood of 1, and get In the neighborhood of 1, there is ln R ⁰ ≈R ⁰ -1, and ln E ⁰ ≈E ⁰ -1 is obtained, Among them, f represents a transformation method to transform the reaction time R and execution time E in any state into R ⁰ and E ⁰ , N' ₀ means that R ⁰ obeys a normal distribution in the awake state, and N' ₂ means that E in the awake state ⁰ obeys the normal distribution, N' ₁ means that ln R ⁰ obeys the normal distribution in the fatigue state, N' ₃ means that the ln E ⁰ obeys the normal distribution under the fatigue state, and N″ ₀ means that the R ⁰ -1 obeys the normal distribution in the awake state distribution, N″ ₂ means E ⁰ -1 obeys normal distribution in awake state, N″ ₁ means R ⁰ -1 obeys normal distribution in fatigue state, N″ ₃ means E ⁰ -1 obeys normal distribution in fatigue state, R ⁰ is the data after the reaction time R is mapped to the neighborhood of 1 through the transformation f, E ⁰ is the data after the execution time E is mapped to the neighborhood of 1 through the transformation f, μ′ ₀ is the mean value of R ⁰ in the awake state, μ′ ₂ is the mean value of E ⁰ in the awake state, μ′ ₁ is the mean value of ln R ⁰ in the fatigue state, μ′ ₃ is the mean value of ln E ⁰ in the fatigue state, and σ′ ₀ is the standard deviation of R ⁰ in the awake state , σ′ ₂ is the standard deviation of E ⁰ in the awake state, σ′ ₁ is the standard deviation of ln R ⁰ in the fatigue state, and σ′ ₃ is the standard deviation of ln E ⁰ in the fatigue state; In any state, R ⁰ -1, E ⁰ -1 obey the normal distribution, and the transformation is constructed: which defines the transformation: It is a function that transforms the reaction time R and execution time E into R', E', R', E' are respectively the reaction time R, the execution time E after the transformation The obtained data; x is the argument of the function, and the assignment of x is R or E; f(R) represents a transformation method to transform the reaction time R in any state into R ⁰ ; f(E) represents a transformation method for transforming the execution time E in any state into E ⁰ ; Step 4. Establish a probability model based on Naive Bayes: set X=(R′,E′) ^T , X is a two-dimensional random vector, which obeys the bivariate normal distribution, that is, X～N(μ,Σ ), where N indicates that the two-dimensional random vector X obeys a bivariate normal distribution, μ is the mean vector of the two-dimensional random vector X, and Σ is the covariance matrix of the two-dimensional random vector X, Cov represents the covariance between random variables, T represents the transpose of the matrix, the driver's subjective fatigue state Y is a Bernoulli variable, and the probability density function of the driver's subjective fatigue state Y is: P(Y)＝φ ^Y (1-φ) ^1-Y (1) The conditional probability distributions of the two-dimensional random vector X in the two states of fatigue and wakefulness are: According to the Naive Bayes formula and the full probability formula: Bring equations (1), (2), (3) into equation (4), and simplify them to get: in is a linear function of X, simplified to -θ ^T X-θ ₀ , φ is the probability that the driver state is fatigued, Y=0, 1, μ ₀ is the mean vector of the two-dimensional random vector X in the awake state, μ ₁ is the mean vector of the two-dimensional random vector X in the fatigue state, and Σ is the covariance matrix of the two-dimensional random vector X; Step 5. Estimating model parameters: Using the maximum likelihood estimation method, the parameters in the formula (5) are the probability φ that the driver's state is fatigue, the mean vector μ ₀ of the two-dimensional random vector X in the awake state, and the two-dimension vector in the fatigue state. The mean vector μ ₁ of the random vector X and the covariance matrix Σ of the two-dimensional random vector X are estimated, and there is a likelihood equation: Take the logarithm of the likelihood equation: get: where {Y ⁽ⁱ⁾ = 1} and {Y ⁽ⁱ⁾ = 0} are implicit functions, the formula in parentheses is true, and the function takes the value 1; otherwise, the function takes the value 0, and m is the number of samples in the training set ; Step 6. Establish a logistic-based probability model: Obtain the logistic function from step 4: Among them, θ ₀ , θ ₁ and θ ₂ are represented by θ _k , and θ _k is the probability φ that the driver state is fatigued, the mean vector μ ₀ of the two-dimensional random vector X in the awake state, and the two-dimensional random vector X in the fatigue state The mean vector μ ₁ of , a suitable function of the covariance matrix Σ of the two-dimensional random vector X, k=0,1,2; Step 7. Fatigue identification: The input reaction time R and execution time E are transformed from the logistic function in step 6. The obtained R', E', the output probability is greater than 0.5, the driver is in the fatigue state; the output probability is less than 0.5, the driver is in the awake state.