CN111209816B - Non-contact fatigue driving detection method based on regular extreme learning machine - Google Patents
Non-contact fatigue driving detection method based on regular extreme learning machine Download PDFInfo
- Publication number
- CN111209816B CN111209816B CN201911382493.9A CN201911382493A CN111209816B CN 111209816 B CN111209816 B CN 111209816B CN 201911382493 A CN201911382493 A CN 201911382493A CN 111209816 B CN111209816 B CN 111209816B
- Authority
- CN
- China
- Prior art keywords
- learning machine
- fatigue
- extreme learning
- follows
- respiratory
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 30
- 230000009466 transformation Effects 0.000 claims abstract description 14
- 238000001228 spectrum Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 31
- 239000011159 matrix material Substances 0.000 claims description 30
- 230000000241 respiratory effect Effects 0.000 claims description 25
- 238000000034 method Methods 0.000 claims description 21
- 230000029058 respiratory gaseous exchange Effects 0.000 claims description 18
- 238000012549 training Methods 0.000 claims description 17
- 239000000523 sample Substances 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 11
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 238000006467 substitution reaction Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 230000036387 respiratory rate Effects 0.000 description 8
- 206010039203 Road traffic accident Diseases 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000002829 reductive effect Effects 0.000 description 5
- 210000002569 neuron Anatomy 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000035565 breathing frequency Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000001815 facial effect Effects 0.000 description 2
- 230000002427 irreversible effect Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000035790 physiological processes and functions Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000004962 physiological condition Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a non-contact fatigue driving detection method based on a regular extreme learning machine, which comprises the following steps: s10, acquiring physiological signals of a driver through a Doppler radar module; s20, classifying physiological signals; s30, performing discrete Fourier transform on the physiological signal to obtain frequency spectrum characteristics; s40, performing characteristic transformation on the frequency spectrum characteristics; s50, designing a regular extreme learning machine model to train the data set, so as to obtain an algorithm model for detecting the fatigue state of the driver, and detecting the fatigue state through the model. The invention can efficiently and accurately detect the fatigue state of the driver while avoiding influencing the normal driving of the driver.
Description
Technical Field
The invention belongs to the field of modeling detection, and particularly relates to a non-contact fatigue driving detection method based on a regular extreme learning machine.
Background
Fatigue driving is one of the most common causes of traffic accidents worldwide. According to WHO (world health organization) reports, more than 130 tens of thousands of people die each year from traffic accidents, and 2 to 5 tens of thousands of people suffer non-fatal injuries due to traffic accidents, of which about 20% of fatal traffic accidents are caused by fatigue driving. Therefore, if a system for automatically detecting the fatigue driving can be developed and a driver can be warned in advance that the system is in the fatigue driving state, a large number of traffic accidents can be effectively avoided, and the occurrence rate of the traffic accidents is reduced.
The current methods for detecting fatigue states are mainly divided into two main categories: 1. detecting contact fatigue state; 2 non-contact fatigue state detection.
The contact fatigue detection method mainly detects the physiological state of a driver. Although the data obtained by the method is reliable, the error is small and the external interference is small, the method needs to install a device for detecting the physiological signal on the driver, and the interference to the driver is too large. For this reason, researchers have matured by using radio to measure physiological signals and by ZigBee, bluetooth, etc. to acquire signals, but the accuracy has been greatly reduced and the artificial interference has caused detection artefacts and errors.
The non-contact fatigue detection method mainly monitors facial features of a driver and vehicle parameter detection. The analysis individual difference of the facial features of the driver is large, and the change of brightness or the wearing of sunglasses, masks and other articles shielding the face by the driver can cause great interference to detection, so that the cost required by the whole device can be increased; for detecting the state and the running track of the vehicle, the required hardware support is high and the cost is high. And the requirements on the external conditions are severe (such as road signs, climates, illumination conditions and the like). One of the great limitations of this approach is that it is a vehicle detection, not a direct detection of the driver, which is greatly reduced in reliability and accuracy.
In summary, although there are various methods for measuring the fatigue state of the driver in real time at present, most of the methods are limited to theoretical research level, and the existing monitoring devices have various limitations and have various problems to be solved. Each method for detecting fatigue driving has advantages and limitations, so that the method for detecting fatigue driving of a driver should not be used in a single way. Many studies have shown that the reliability and accuracy of the hybrid detection method is higher than that of the single detection method. Therefore, to develop an effective fatigue driving detection system, various detection methods should be combined in a hybrid system to detect, and the reliability of data obtained by detecting physiological conditions is high, but the interference to the driver is large.
Disclosure of Invention
In view of the technical problems, the invention realizes non-contact detection of the physiological state of the driver, avoids physical and driving interference to the driver and improves the accuracy; the difference between different individuals can be eliminated; the data can be processed rapidly and efficiently; the non-contact fatigue driving detection method based on the regular extreme learning machine is simple in algorithm model, fast in learning efficiency, few in iteration times and high in accuracy.
The method comprises the following steps:
s10, acquiring physiological signals of a driver through a Doppler radar module;
s20, classifying physiological signals;
s30, performing discrete Fourier transform on the physiological signal to obtain frequency spectrum characteristics;
s40, performing characteristic transformation on the frequency spectrum characteristics;
s50, designing a regular extreme learning machine model to train the data set, so as to obtain an algorithm model for detecting the fatigue state of the driver, and detecting the fatigue state through the model.
Preferably, the physiological signal includes at least a respiratory signal and a heartbeat signal of the driver.
Preferably, the discrete fourier transform is performed on the physiological signal to obtain spectral characteristics, thereby obtaining the amplitude B of the respiratory signal A And period B T Period H of heartbeat signal T 。
Preferably, the feature transformation is:
wherein R is T Representing respiratory cycle B T And heart cycle H T Ratio of (h) θ (x) For the hypothesized functions obtained by gradient descent algorithm, B is respectively T And H is T Substitution into h θ (x) And h is set θ (B T ) And h θ (H T ) R is used for the ratio of A And (3) representing.
Preferably, the regular extreme learning machine trains according to the training set data and the randomly set input layer weight matrix omega to obtain an output layer weight matrix
Preferably, the weight calculation equation of the output layer of the regular extreme learning machine is as follows:
wherein H is a hidden layer activation term matrix, C is a regularization coefficient, I is a unit matrix, and L is a desired output matrix, namely an output label.
Preferably, the doppler radar module employs a microwave doppler radar detector probe sensor HB100 module.
Compared with the prior art, the Doppler radar module adopted by the invention can realize non-contact accurate detection of the physiological signals of the driver, and the physiological signals can accurately reflect the fatigue state of the driver, so that the difference of the physiological signals of different individuals in different fatigue states can be effectively solved. The used regular extreme learning machine model has the advantages of high learning efficiency, less iteration times, high accuracy and greatly reduced calculated amount.
Drawings
FIG. 1 is a step flow diagram of a non-contact fatigue driving detection method based on a regular extreme learning machine according to an embodiment of the present invention;
FIG. 2 is a standard chart of an expert evaluation method of a non-contact fatigue driving detection method based on a regular extreme learning machine according to an embodiment of the invention;
FIG. 3 is a graph of physiological signals collected by a Doppler radar module in a non-contact fatigue driving detection method based on a regular extreme learning machine according to an embodiment of the present invention;
FIG. 4 is a graph of physiological signal amplitude-frequency characteristics of a non-contact fatigue driving detection method based on a regular extreme learning machine according to an embodiment of the present invention;
FIG. 5 is a scatter plot of respiration cycle and amplitude and a linear fitting graph thereof of a non-contact fatigue driving detection method based on a regular extreme learning machine according to an embodiment of the present invention;
FIG. 6 is a graph showing the effect of time domain change of physiological signals after feature transformation of a non-contact fatigue driving detection method based on a regular extreme learning machine according to an embodiment of the present invention;
FIG. 7 is a regular extreme learning machine network model diagram of a non-contact fatigue driving detection method based on a regular extreme learning machine according to an embodiment of the present invention;
fig. 8 is a regular extreme learning machine network training result diagram of a non-contact fatigue driving detection method based on a regular extreme learning machine according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The method comprises data acquisition, data processing and data training. The data acquisition part mainly comprises a Doppler radar module as a core and a simulated driving software system matched with a simulated driver and is used for acquiring respiration and heartbeat signals of the driver. The data processing part is mainly used for classifying the acquired respiratory and heartbeat signals by the class of the signals through an expert judging method, then carrying out filtering processing on each group of signals, and then carrying out discrete Fourier transformation on the signals to obtain a spectrogram, so as to extract characteristic values such as respiratory cycle, amplitude, heart rate and the like. The data training part mainly designs a regular extreme learning machine network model, trains the acquired data and obtains an algorithm model for detecting the fatigue state of the driver.
Example 1
Referring to fig. 1, a step flow chart of a non-contact fatigue driving detection method based on a regular extreme learning machine, S10, collecting physiological signals of a driver through a doppler radar module;
s20, classifying physiological signals by adopting an expert evaluation method;
s30, performing discrete Fourier transform on the physiological signal to obtain frequency spectrum characteristics;
s40, performing characteristic transformation on the frequency spectrum characteristics;
s50, designing a regular extreme learning machine model to train the data set, so as to obtain an algorithm model for detecting the fatigue state of the driver, and detecting the fatigue state through the model.
The physiological signals include at least a respiration signal and a heartbeat signal of the driver.
Performing discrete Fourier transform on the physiological signal to obtain frequency spectrum characteristics, and further obtaining amplitude B of the respiratory signal A And period B T Period H of heartbeat signal T 。
The characteristic transformation is as follows:
wherein R is T Representing respiratory cycle B T And heart cycle H T Ratio of (h) θ (x) For the hypothesized functions obtained by gradient descent algorithm, B is respectively T And H is T Substitution into h θ (x) And h is set θ (B T ) And h θ (H T ) R is used for the ratio of A And (3) representing.
The regular extreme learning machine trains according to the training set data and randomly sets the input layer weight matrix omega to obtain an output layer weight matrix
The weight calculation equation of the output layer of the regular extreme learning machine is as follows:
wherein H is a hidden layer activation term matrix, C is a regularization coefficient, I is a unit matrix, and L is a desired output matrix, namely an output label.
The Doppler radar module adopts a microwave Doppler radar detector probe sensor HB100 module.
See fig. 2 for expert criteria for fatigue grade. Driver fatigue status grade classification 4 grades: an awake state, a stage I fatigue state, a stage II fatigue state, and a stage III fatigue state. Each fatigue level has corresponding characteristic expression, such as blink frequency, and the number of times of breathing, and the like, and the fatigue level of the driver in the video information is judged by expert analysis on the video information, so that the fatigue level of the physiological signal corresponding to the video information can be obtained. All physiological signals acquired are classified by this method.
Referring to fig. 3 and fig. 4, there are respectively a physiological signal acquired by the doppler radar module and a spectrum characteristic diagram thereof. It can be seen that the physiological signals of the human body can be successfully acquired through the radar module. The physiological signal comprises a respiratory signal, a heartbeat signal and noise, the noise can be effectively filtered by filtering the signal and performing Discrete Fourier Transform (DFT), a frequency spectrum characteristic diagram is obtained, and then the frequency and the amplitude of the signal are extracted.
See fig. 5 for a scatter plot of respiratory cycles and amplitudes for different individuals and a linear fit thereof. Respiratory and heartbeat signals vary from person to person. Too large individual differences are detrimental to neural networksClassification of data and greatly reduces classification accuracy. In a normal state, the respiratory rate is generally 1/5-1/4 of the heart rate, and as the fatigue degree increases, the life activity becomes slower, the energy consumption decreases, and the ratio of the respiratory rate to the heart rate decreases. The ratio of respiratory rate to heart rate is defined as R T . On the basis, the relation between the breathing frequency and the breathing amplitude is found by modeling the breathing frequency and the breathing amplitude of each individual, and then the relation between the breathing amplitude and the heart rate can be found. Shown in the left hand graph of fig. 5 is a scatter plot of respiratory rate and respiratory amplitude for two different testers in different fatigue states, and it can be seen that the linear relationship between respiratory rate and respiratory amplitude is different for different testers. The data for each tester was fitted using a gradient descent algorithm. The fitted curve can be seen in the right hand graph in fig. 5. The specific training method for the regular extreme learning machine model comprises the following steps:
according to the gradient descent algorithm principle, firstly, a hypothesis function h is defined θ (X) is:
wherein θ is a weight, x (i) Is the ith input data.
Next, a cost function J (θ) is defined. Cost functions, also known as square error functions, are the most common method of solving the regression problem. The definition is as follows:
wherein m is the total number of samples, x (i) For the ith input data, y (i) Is the i-th target output.
According to the gradient algorithm principle, a group of theta values need to be found to enable the cost function to be converged, and according to the formula (3), the theta values enabling the cost function to be converged can be obtained.
θ j Defined as a hypothetical function h θ (x) Alpha is the learning rate. Substitution of formula (2) into formula (3) can result in:
finally, assuming that the relationship between the respiratory amplitude and the respiratory cycle is linear, then the assumed function is h θ (X)：
h θ (x)＝θ 0 +θ 1 x (5)
Substituting equation (5) into equation (4) and decreasing the gradient can yield the θ value. Thus, the relationship between the breathing amplitude and the breathing cycle is found. Note that: the values of θ are different for each set of data.
According to the above method, by appropriately discarding the measurement accuracy, the relationship between the breathing cycle and the amplitude of each person can be described by a linear curve. Thus, the respiratory signal may be converted as follows:
f(t)＝Asin(B T t+b)→f(t)＝h θ (B T )sin(B T t+b) (6)
where A is amplitude and b is phase. The ratio R of respiratory rate to heart rate has been defined above T The heart rate of a normal person is consistent with the respiratory rate, i.e. the ratio of heart rate to respiratory rate varies within a certain range.
Defining a new variable R A ，R A The definition is as follows:
wherein h is θ (x) Namely h in formula (5) θ (x)。
As the fatigue level deepens, R T To a certain extent, then it is only necessary to prove that R is increasing with increasing fatigue A Regular variations occur over a range. For equation (8), the following is developed:
for the function V (x), V (x) is defined as follows:
differentiating V (x):
thus, from formulas (9) and (11), the following can be concluded:
1) Function R A Decreasing progressively;
2) For each tester, i.e. θ 1 The higher the heart rate, the unchanged, R A The smaller;
3)R A ＜＝1。
in combination with the above conclusion, a new variable R is obtained through eigenvalue transformation, which is called a ratio point:
R＝(R A ,R T ) (12)
r is unique to each tester. In this way, the respiration signal can be mapped to a new two-dimensional space.
f(t)＝Asin(Tt+b)→f(t)＝R A sin(R T t+b) (13)
Referring to fig. 6, the effect of the time domain change of the physiological signal after the feature transformation is shown. In fig. 6 (a), these two curves represent the respiration signal of the tester a in the stage I fatigue state and the respiration signal of the tester B in the stage II fatigue state, respectively. It can be seen that the period and amplitude of the two curves differ little. At this time, the result of converting the respiratory signal according to equation (13) is shown in fig. 6 (b). It can be seen that the two signals at this time differ greatly in period and amplitude. Thus, the difference in breathing amplitude between individuals can be reduced by appropriately discarding the measurement accuracy.
The following characteristic values are finally determined as training sample data by classifying, filtering, discrete Fourier transformation and characteristic value conversion of the data: r is R A ，R T ，H T . Wherein H is T Representing the heart cycle.
The sample library that was ultimately used for neural network training was as follows:
X (i) ＝(R A R T H T ) (i) (14)
L (i) ＝(h(s) h(s-1) h(s-2) h(s-3)) (i) (15)
S＝{(X (i) T (i) )},i∈[1,m] (16)
wherein X is input data, L is an output tag corresponding to X, S is a sample data set, h (X) is an impact function, and the definition is as follows:
s is a fatigue state, and s can be 0, 1, 2 or 3, which respectively represent an awake state, a I-stage fatigue state, a II-stage fatigue state and a III-stage fatigue state, I is a sample index, and m is a total sample size.
See fig. 7 for a regular extreme learning machine network model. Extreme Learning Machines (ELMs) are a proposed algorithm for solving single hidden layer neural networks by the professor yellow light. ELM has the advantages of high learning efficiency and strong generalization capability, and is widely applied to the problems of classification, regression, clustering, feature learning and the like. For the single hidden layer neural network in fig. 7, the total number of input samples is m, the input data is X, the network output is O, the hidden layer activation term matrix is H, the hidden layer input weight matrix is ω, the hidden layer output weight matrix is β, b is a threshold, the activation function is selected as a sigmoid function, and the mathematical expression of the sigmoid function is as follows:
by using the activation function, nonlinear characteristics can be added, so that the learning speed is faster and the learning efficiency is higher, and then the calculation equation of the activation term H is as follows:
the output O can be expressed as:
βH＝O (20)
the goal of single hidden layer neural network learning is to minimize the output error, namely:
||O-L|| n×m ＝0 (21)
where n represents the number of eigenvalues and m represents the total number of samples. L is the desired output matrix, i.e. the output label. Then β, ω, and b are present such that:
βH＝L (22)
according to the principle of the extreme learning machine: as long as the activation function g (·) meets the infinite microability in any interval, the input weight matrix ω and the bias b can be randomly generated, that is, the single hidden layer feedforward neural network does not need to adjust ω and b any more; also because of the ω and b randomly generated by the continuous probability distribution and assuming the number of hidden layer neurons is k, then there is k.ltoreq.N, so that L- βH L n×m ε must be true, and the bias of the output layer is found to be unnecessary. Therefore, after the hidden layer input weight matrix omega and the bias b are determined, the hidden layer activation term H can be determined, and only the hidden layer output matrix beta is required after the activation term H is determined. According to the theory proposed by the Huang An teachings: when the number k of hidden layer neurons is equal to the number m of training samples, the hidden layer activation term matrix H is a reversible matrix (the possibility that H is irreversible occurs in a very small degree), and then the hidden layer output weight β that enables the neural network to output with 0 error fit can be obtained by the equation (22). However, the process is not limited to the above-described process,in most cases, the number k of hidden layer neurons is much smaller than the number m of training set samples, where the matrix H is irreversible. At this point, a solution to minimize the loss function is required, namely:
according to the minimum norm criterion, a solution is obtained by least squares:
wherein the method comprises the steps ofIs the moore-coarse generalized inverse of the hidden layer output matrix H, called pseudo-inverse. Solving->The method comprises the following steps:
in order to obtain a more practical and reliable regression coefficient, a biased estimated regression method, called ridge regression, is required. When X is not a full order matrix or the linear correlation between certain columns is large, its determinant is close to 0, which will lead to large errors in computation, thus requiring the addition of regularization terms in the loss function.
Where I is the identity matrix and C is the regularization coefficient. Studies have shown that relatively small weighting coefficients can improve the stability and generalization capability of single hidden layer feedforward neural networks (SLFNs), which means that regularization of ELMs is necessary under complex problems.
In summary, when training data is input and the input weight matrix is randomly initialized, the output weight matrix can be obtained by equation (36).
Through multiple training, the number of hidden layer neurons is finally determined to be 500, and the regularization coefficient C=1e5. See fig. 8 for regular extreme learning machine network training results. In FIG. 8, the entire canvas is divided into four sections, each section displaying the output of each fatigue level. The abscissa represents the fatigue level and the ordinate represents the corresponding probability. In practice, for each input, the corresponding output is 4 points, for example, the output of the input signal is represented as follows:
Test＝((0,0.2),(1,0.1),(2,0.4),(3,0.3)) (37)
the meaning of Test is: input signal X (i) The probability corresponding to fatigue level 0 is 0.2, the probability of level I is 0.1, the probability of level II is 0.4, and the probability of level III is 0.3, so this signal corresponds to fatigue level II. The predicted and total predicted results for each fatigue level test set sample are as follows:
1) Awake state: 0.983
2) I fatigue state: 0.867
3) II fatigue state: 0.883
4) III fatigue status: 0.967
Total probability: 0.925.
it should be understood that the exemplary embodiments described herein are illustrative and not limiting. Although one or more embodiments of the present invention have been described with reference to the accompanying drawings, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present invention as defined by the following claims.
Claims (1)
1. The non-contact fatigue driving detection method based on the regular extreme learning machine is characterized by comprising the following steps of:
s10, acquiring physiological signals of a driver through a Doppler radar module;
s20, classifying physiological signals;
s30, performing discrete Fourier transform on the physiological signal to obtain frequency spectrum characteristics;
s40, performing characteristic transformation on the frequency spectrum characteristics;
s50, designing a regular extreme learning machine model to train the data set so as to obtain an algorithm model for detecting the fatigue state of the driver, and detecting the fatigue state through the model;
the physiological signals at least comprise a respiration signal and a heartbeat signal of the driver;
the physiological signal is subjected to discrete Fourier transform to obtain frequency spectrum characteristics, and then the amplitude B of the respiratory signal is obtained A And period B T Period H of heartbeat signal T ；
The feature transformation is as follows:
wherein R is T Representing respiratory cycle B T And heart cycle H T Ratio of (h) θ (x) For the hypothesized functions obtained by gradient descent algorithm, B is respectively T And H is T Substitution into h θ (x) And h is set θ (B T ) And h θ (H T ) R is used for the ratio of A A representation;
the regular extreme learning machine trains according to training set data and randomly sets an input layer weight matrix omega to obtain an output layer weight matrix
The weight calculation equation of the output layer of the regular extreme learning machine is as follows:
wherein H is a hidden layer activation term matrix, C is a regularization coefficient, I is a unit matrix, and L is a desired output matrix, namely an output label;
the Doppler radar module adopts a microwave Doppler radar detector probe sensor HB100 module;
the specific training method for the regular extreme learning machine model comprises the following steps:
according to the gradient descent algorithm principle, firstly, a hypothesis function h is defined θ (X) is:
wherein θ is a weight, x (i) Is the ith input data;
next, a cost function J (θ), also called a square error function, is defined as follows:
wherein m is the total number of samples, x (i) For the ith input data, y (i) Output for the ith target;
finding a group of theta values according to the gradient algorithm principle, converging the cost function, and obtaining the theta values converging the cost function according to the formula (3);
θ j defined as a hypothetical function h θ (x) Alpha is the learning rate; substitution of formula (2) into formula (3) can result in:
finally, assuming that the relationship between the respiratory amplitude and the respiratory cycle is linear, then the assumed function is h θ (x)：
h θ (x)＝θ 0 +θ 1 x (5)
Substituting the formula (5) into the formula (4), and obtaining a theta value through gradient descent, so that the relation between the respiratory amplitude and the respiratory period is found, wherein the theta value obtained by each group of data is different;
according to the above method, the relationship between the breathing cycle and the amplitude of each person can be described by a linear curve, and thus the breathing signal can be converted as follows:
f(t)＝Asin(B T t+b)→f(t)＝h θ (B T )sin(B T t+b) (6)
wherein A is amplitude, b is phase, and the ratio R of respiratory cycle to heartbeat cycle is defined T The respiratory cycle of normal people is consistent with the heartbeat cycle, namely the ratio of the respiratory cycle to the heartbeat cycle is changed in a certain range,
defining a new variable R A ，R A The definition is as follows:
wherein h is θ (x) Namely h in formula (5) θ (x)，
As the fatigue level deepens, R T To a certain extent, then it is only necessary to prove that R is increasing with increasing fatigue A Regular changes occur within a certain range, and for equation (8), the development is as follows:
for the function V (x), V (x) is defined as follows:
differentiating V (x):
thus, from formulas (9) and (11), the following can be concluded:
1) Function R A Decreasing progressively;
2) For each tester, i.e. θ 1 The higher the heart rate, the unchanged, R A The smaller;
3)R A ＜＝1；
in combination with the above conclusion, a new variable R is obtained through eigenvalue transformation, which is called a ratio point:
R＝(R A ,R T ) (12)
r is unique to each tester; the respiration signal can be mapped to a new two-dimensional space;
f(t)＝Asin(Tt+b)→f(t)＝R A sin(R T t+b) (13)
the following characteristic values are finally determined as training sample data by classifying, filtering, discrete Fourier transformation and characteristic value conversion of the data: r is R A ，R T ，H T Wherein H is T Representing a heartbeat cycle;
the sample library that was ultimately used for neural network training was as follows:
X (i) ＝(R A R T H T ) (i) (14)
L (i) ＝(h(s) h(s-1) h(s-2) h(s-3)) (i) (15)
S＝{(X (i) T (i) )},i∈[1,m] (16)
wherein X is input data, L is an output tag corresponding to X, S is a sample data set, h (X) is an impact function, and the definition is as follows:
s is a fatigue state, and s can be 0, 1, 2 or 3, which respectively represent an awake state, a I-stage fatigue state, a II-stage fatigue state and a III-stage fatigue state, I is a sample index, and m is the total number of samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911382493.9A CN111209816B (en) | 2019-12-28 | 2019-12-28 | Non-contact fatigue driving detection method based on regular extreme learning machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911382493.9A CN111209816B (en) | 2019-12-28 | 2019-12-28 | Non-contact fatigue driving detection method based on regular extreme learning machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111209816A CN111209816A (en) | 2020-05-29 |
| CN111209816B true CN111209816B (en) | 2023-10-24 |
Family
ID=70785943
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911382493.9A Active CN111209816B (en) | 2019-12-28 | 2019-12-28 | Non-contact fatigue driving detection method based on regular extreme learning machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111209816B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3960078A1 (en) * | 2020-08-31 | 2022-03-02 | Srett (Sas) | Respiration monitoring |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104724123A (en) * | 2013-12-19 | 2015-06-24 | 罗伯特·博世有限公司 | Method for ascertaining the stress level of the driver |
| CN105354988A (en) * | 2015-12-11 | 2016-02-24 | 东北大学 | Driver fatigue driving detection system based on machine vision and detection method |
| CN106264501A (en) * | 2016-10-13 | 2017-01-04 | 杭州电子科技大学 | A kind of bio-signal acquisition system based on continuous wave Doppler radar |
| CN106491156A (en) * | 2016-10-13 | 2017-03-15 | 杭州电子科技大学 | A kind of fatigue drive of car detection method based on Multi-source Information Fusion |
| CN107595242A (en) * | 2017-07-26 | 2018-01-19 | 来邦科技股份公司 | A kind of sleep physiology signal monitoring method, device, electronic equipment and storage medium |
| CN108335457A (en) * | 2018-02-08 | 2018-07-27 | 吉林大学 | A kind of driving safety monitoring method based on driver's bracelet |
| EP3479759A1 (en) * | 2017-11-04 | 2019-05-08 | Tata Consultancy Services Limited | System and method to classify cardiopulmonary fatigue |
| CN110420019A (en) * | 2019-07-29 | 2019-11-08 | 西安电子科技大学 | A kind of depth recurrence heart rate estimation method of ballistocardiography signal |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10618522B2 (en) * | 2018-03-27 | 2020-04-14 | Hong Kong Productivity Council (HKPC) | Drowsiness detection and intervention system and method |
-
2019
- 2019-12-28 CN CN201911382493.9A patent/CN111209816B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104724123A (en) * | 2013-12-19 | 2015-06-24 | 罗伯特·博世有限公司 | Method for ascertaining the stress level of the driver |
| CN105354988A (en) * | 2015-12-11 | 2016-02-24 | 东北大学 | Driver fatigue driving detection system based on machine vision and detection method |
| CN106264501A (en) * | 2016-10-13 | 2017-01-04 | 杭州电子科技大学 | A kind of bio-signal acquisition system based on continuous wave Doppler radar |
| CN106491156A (en) * | 2016-10-13 | 2017-03-15 | 杭州电子科技大学 | A kind of fatigue drive of car detection method based on Multi-source Information Fusion |
| CN107595242A (en) * | 2017-07-26 | 2018-01-19 | 来邦科技股份公司 | A kind of sleep physiology signal monitoring method, device, electronic equipment and storage medium |
| EP3479759A1 (en) * | 2017-11-04 | 2019-05-08 | Tata Consultancy Services Limited | System and method to classify cardiopulmonary fatigue |
| CN108335457A (en) * | 2018-02-08 | 2018-07-27 | 吉林大学 | A kind of driving safety monitoring method based on driver's bracelet |
| CN110420019A (en) * | 2019-07-29 | 2019-11-08 | 西安电子科技大学 | A kind of depth recurrence heart rate estimation method of ballistocardiography signal |
Non-Patent Citations (3)
| Title |
|---|
| Abdu Gumaei 等.A Hybrid Feature Extraction Method With Regularized Extreme Learning Machine for Brain Tumor Classification.《IEEE Access》.2019,第7卷全文. * |
| Long Chen 等.Non-contact physiological signal monitoring system based on Doppler radar.《IEICE Electronics Express》.2017,第14卷(第14期),全文. * |
| 杨希宁.基于多源信息融合的非接触式疲劳驾驶检测***研究.《中国优秀硕士学位论文全文数据库信息科技辑》.2018,(第undefined期),第1-69页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111209816A (en) | 2020-05-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113627518B (en) | Method for realizing neural network brain electricity emotion recognition model by utilizing transfer learning | |
| CN109124625B (en) | Driver fatigue state level grading method | |
| CN109994203A (en) | A kind of epilepsy detection method based on EEG signal depth multi-angle of view feature learning | |
| CN105877766A (en) | Mental state detection system and method based on multiple physiological signal fusion | |
| TWI500411B (en) | Assessment system and method for pulse and constitution health risks | |
| CN109620244A (en) | The Infants With Abnormal behavioral value method of confrontation network and SVM is generated based on condition | |
| Singh et al. | Classification of short unsegmented heart sound based on deep learning | |
| CN113743471B (en) | Driving evaluation method and system | |
| CN112617835B (en) | Multi-feature fusion fatigue detection method based on transfer learning | |
| CN107205652A (en) | The sleep analysis system with automatic mapping is generated with feature | |
| CN110110809A (en) | The construction method of fuzzy automata based on Machine Fault Diagnosis | |
| CN113188807A (en) | Abs result automatic judging algorithm | |
| CN105303771A (en) | Fatigue judging system and method | |
| CN111209816B (en) | Non-contact fatigue driving detection method based on regular extreme learning machine | |
| CN111209815B (en) | Non-contact fatigue driving detection method based on BP neural network with momentum optimization | |
| CN117598700B (en) | Intelligent blood oxygen saturation detection system and method | |
| Herath et al. | Autism spectrum disorder diagnosis support model using Inception V3 | |
| Lopes et al. | Ensemble deep neural network for automatic classification of EEG independent components | |
| CN112215044A (en) | Driving tendency identification method based on probabilistic neural network | |
| CN106580324A (en) | Method and device for extracting respiratory signal | |
| CN110801228B (en) | Brain effect connection measurement method based on neural network prediction | |
| CN114464319B (en) | AMS susceptibility assessment system based on slow feature analysis and deep neural network | |
| Quanyu | Pulse signal analysis based on deep learning network | |
| CN112274154B (en) | Cross-subject fatigue driving classification method based on electroencephalogram sample weight adjustment | |
| CN107967943A (en) | A kind of children ADHD analysis system based on Functional imagnetic resonance imaging |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |