CN104952210A - Fatigue driving state detecting system and method based on decision-making level data integration - Google Patents

Abstract

The invention discloses a fatigue driving state detection system and method based on decision-making level data fusion. The system first uses an acceleration sensor to collect the motion acceleration of the steering wheel, judges the motion state of the steering wheel based on the dynamic threshold of the acceleration, and preliminarily judges the driving state based on the 4s non-moving theory of the steering wheel. The fault tolerance of the detection is enhanced by setting the error value of the left and right swing of the steering wheel; the pulse sensor is used to collect the pulse time domain value of the driver, and the physiological state of the driver is detected based on the pulse rate dynamic threshold; through the two kinds of detection The results are fused at the decision level, and the fused detection results can be obtained. A fatigue driving state detection prototype system is designed and constructed based on the fatigue driving state detection method based on decision-level data fusion. Compared with existing methods, the method for detecting driver fatigue driving state based on decision-level data fusion disclosed by the present invention has certain advantages in detection accuracy, and is superior in response time, time complexity and memory consumption of the algorithm. It has ideal performance.

Description

A fatigue driving state detection system and method based on decision-level data fusion

technical field

The invention relates to a driving state detection method, in particular to a fatigue driving state detection method, which belongs to the cross-technical application field of mobile computing, sensing technology and data fusion.

Background technique

Fatigue driving generally refers to the situation in which the driver's control ability is abnormal due to fatigue changes in the body mechanism during driving, which seriously endangers road traffic safety and has become a serious problem facing the world. The report of the National Highway Traffic Safety Administration of the United States shows that traffic accidents caused by driver fatigue driving account for 20%-30% of the total number of traffic accidents. For example, at about 2:31 on August 26, 2012, the driver Chen X drove a sleeper bus and collided with a heavy tank semi-trailer train, causing 36 people in the bus to die on the spot; System, GPS) records, the driver of the bus, Mr. Chen, drove continuously for 4 hours and 22 minutes, did not stop to rest halfway, fatigue driving caused inattention when driving, and decreased reaction and judgment ability, which led to the accident.

Efficient detection of the driver's fatigue driving state and timely feedback can effectively prevent similar traffic accidents. By collecting the driver's blood, the blood sugar, blood urea and creatinine of the driver can be analyzed in the state of fatigue driving, and a comprehensive analysis of these types of information can obtain an estimate of whether the driver is fatigued. This method has a high detection accuracy. rate, the experimental results can be used as a reference for other methods, but the real-time performance is not good, and professional medical equipment is required. At present, researchers and technicians have researched and developed a series of research results and products, which are mainly divided into three categories: the first category is the detection technology based on physiological signals, mainly based on changes in brain waves, heart rate, pulse and skin voltage, etc. ; The second type is based on the physical state of the driver's body, mainly based on the tilt of the head, changes in the eyes, changes in the mouth, and the strength of the steering wheel; the third type is based on the running state of the vehicle, mainly based on the movement of the steering wheel The law, the speed of the vehicle, the acceleration of the vehicle, and the trajectory of the vehicle, etc.

At present, some researchers have designed a wearable electroencephalogram (Electroencephalograph, EEG) detection system, which can detect the driver's vigilance of his own driving behavior in real time, thus reflecting the driver's fatigue; In the case of sleep deprivation, the driver’s electrocardiogram (ECG) data is collected, and the two indicators of heart rate and blinking frequency are integrated to analyze the driver’s fatigue driving state; there are also researchers who analyze the driver’s EEG data. The change information of each frequency energy of the driver in the fatigue stage is used to obtain the fatigue variability of the driver in the fatigue stage. There are also researchers who propose a mechanism for real-time detection of the driver's fatigue driving state, using the driver's EEG data, ECG data and myoelectric signals to comprehensively judge the driver's fatigue driving state; there are also researchers based on kernel-based principal component analysis Using the method to analyze the experimental samples, selecting the appropriate kernel function and related parameters can effectively separate the normal samples and fatigue samples, and linearly analyze the driver's ECG data to obtain the experimental samples of the driver's EEG data. The analysis of the experimental samples is normal It still belongs to the fatigue sample, and then detects whether the driver is in a fatigue driving state. The fatigue state detection method based on EEG and ECG has good real-time performance and high detection accuracy, but the hardware cost is high and it is not easy to wear.

At present, some researchers analyze and judge the driver's fatigue driving state based on the driver's facial behavior characteristics. For example: comprehensively use the driver's eye closure time percentage, the degree of mouth opening and the tilt of the head to comprehensively judge the driver's fatigue driving state; use the frame difference method, template matching method and Kalman method to locate the position of the human eye , and then locate the open and closed state of the human eye, and detect the driver's fatigue driving state based on the Percentage of Eyelid Closure (PERCLOS) feature value. The detection method of driver fatigue driving state based on eye behavior has good real-time performance and high detection accuracy, but this kind of method is limited to the situation where the cab is well lit, and cannot be applied to driving at night, which has great limitations. .

The steering wheel detection device SAM developed and produced by Electronic Safety Products is a device that can detect the abnormal movement of the steering wheel of the car. In the case of the situation, the sensor device will send out an alarm to remind the driver until the steering wheel is controlled by the driver to return to the normal motion state. There is also a method for detecting the driver's fatigue driving state based on the change of the steering wheel angle proposed by the researchers. The angular displacement sensor and the GPS module are embedded in the steering wheel, and the pattern recognition theory is applied to the collected angle change to judge whether the driver is fatigued. Driving status, while using the GPS module to judge the driving status of the vehicle. There are also researchers who have designed a method to detect the driver's fatigue driving state based on the steering wheel angle signal. Regression model, through which it can be analyzed whether the driver is in a state of fatigue driving. These methods have good real-time performance and low overhead, but the detection accuracy is relatively low.

In short, the current research results generally have defects such as low detection accuracy, high hardware cost, difficult to wear equipment, and greater influence by environmental factors. And the present invention can well solve the above problems.

Contents of the invention

The purpose of the present invention is to propose a fatigue driving state detection system based on decision-making level data fusion, which can detect driver fatigue state efficiently and conveniently. The pulse data of the driver; the two kinds of data are preprocessed respectively, and the dynamic thresholds are calculated respectively for the two kinds of data after preprocessing, and the preliminary detection result about whether the driver is in a fatigue state is obtained; fusion to obtain more accurate fusion detection results.

The technical solution adopted by the present invention to solve the technical problem is: a fatigue driving state detection system based on decision-level data fusion, the system includes an acceleration data acquisition module, an acceleration data transmission and preprocessing module, and an acceleration data dynamic threshold training module , Algorithm application module for detecting driver fatigue driving state based on acceleration data, pulse data acquisition module, pulse data storage and preprocessing module, pulse data dynamic threshold training module, algorithm application module for detecting driver fatigue driving state based on pulse data, data fusion module, etc.; comprehensively utilize the indirect information of the motion acceleration of the steering wheel and the direct pulse information of the driver; the system uses the acceleration data acquisition module to collect the motion acceleration data of the steering wheel based on the acceleration sensor, and uses the acceleration data transmission and preprocessing module to process these raw data. The data is smoothed using the moving average method; the algorithm application module for detecting the driver's fatigue driving state based on acceleration data adopts the theory of steering wheel not moving for 4 seconds, and initially judges the detection result of the fatigue driving state; at the same time, the pulse data acquisition module is used to collect the The pulse data during the driver’s driving process; use the pulse data storage and preprocessing module to store the collected pulse data first, then remove the pulse signal noise based on the threshold method of wavelet transform, and then use the weighted moving average method to smooth the data; use The pulse data dynamic threshold training module analyzes and calculates the driver's pulse rate changes, and establishes the threshold of the corresponding normal driving state for different individuals; the algorithm application module for detecting the driver's fatigue driving state based on the pulse data compares with this normal threshold. Judging whether the currently obtained driver’s pulse rate is normal, and then judging whether the driver is in a fatigue driving state; the data fusion module applies evidence theory to the decision-level fusion of these two recognition results, and detects the fatigue driving state based on decision-level data fusion The algorithm obtains the detection result of whether the driver is in a fatigue driving state.

The function of the acceleration data acquisition module is to use the acceleration sensor to collect the steering wheel motion acceleration data.

The function of the acceleration data transmission and preprocessing module is to apply the weighted moving average method to smooth the original data of the steering wheel motion acceleration collected by the acceleration sensor.

The function of the acceleration data dynamic threshold training module is to calculate the average value of the data processed by the acceleration data transmission and preprocessing module to obtain the dynamic threshold of the driver on the current road section, and then obtain the fluctuation range of the steering wheel by comparing the acceleration data.

The function of the algorithm application module for detecting the driver's fatigue driving state based on the acceleration data is: to use the theory of the steering wheel not moving for 4 seconds to preliminarily judge the detection result of the fatigue driving state.

The function of the pulse data acquisition module is to collect the pulse data of the driver during driving based on the pulse sensor fixed at the wrist radial artery of the human body.

The function of the pulse data storage and preprocessing module is: store the collected pulse data, then remove the pulse signal noise based on the wavelet transform threshold method, and then use the weighted moving average method to smooth the data.

The function of the pulse data dynamic threshold training module is to analyze and calculate the change of the driver's pulse rate, and establish the corresponding normal driving state threshold for different individuals.

The function of the algorithm application module for detecting the driver’s fatigue driving state based on the pulse data is to judge whether the currently obtained driver’s pulse rate is normal by comparing it with the threshold of the normal driving state, and then determine whether the driver is in a fatigue driving state.

The function of the data fusion module is to apply evidence theory to carry out decision-level fusion on the recognition results of the algorithm application module for detecting driver fatigue driving state based on acceleration data and the algorithm application module for detecting driver fatigue driving state based on pulse data. The fatigue driving state detection algorithm of data fusion obtains the detection result of whether the driver is in a fatigue driving state.

In the system of the present invention, when the driver is driving the vehicle, the acceleration and angle change information of the vehicle steering wheel motion is important information for analyzing the driver's driving state, and can indirectly reflect the driver's driving state to a certain extent. On a normal driving road, if the steering wheel does not move for more than 4 seconds, the driver is likely to be in a state of fatigue driving, but if the driver is driving on a straight road with few vehicles around, then the steering wheel does not move for 4 seconds When driving, it is impossible to judge whether the driver is in a fatigue driving state, so the motion information of the steering wheel can only indirectly reflect the driving state of the driver to a certain extent. Changes in the pulse of a person can show a person's physiological state such as fatigue. The change in the pulse of the driver during driving, especially the change in the pulse rate, is a direct reflection of the driver's driving state.

The system of the present invention is a fatigue driving state detection method based on decision-making level data fusion, and comprehensively utilizes the indirect information of the motion acceleration of the steering wheel and the direct pulse information of the driver to effectively improve the detection accuracy of the fatigue driving state: the steering wheel is collected by the acceleration sensor The motion acceleration data is smoothed by applying the weighted moving average method to these original data, and based on the theory of the steering wheel not moving for 4 seconds, the detection result of the fatigue driving state is preliminarily judged; And calculate the driver's pulse rate change, establish the threshold of the corresponding normal driving state for different individuals, and judge whether the current driver's pulse rate is normal by comparing with this normal threshold, and then can judge whether the driver is in fatigue Driving state: apply evidence theory to the two recognition results for decision-level fusion, and obtain more accurate detection results about whether the driver is in a fatigue driving state.

The fatigue driving state detection model of the present invention includes a pulse data acquisition module, a pulse data storage and preprocessing module, a pulse data dynamic threshold training module, an algorithm application module for detecting driver fatigue driving state based on pulse data, an acceleration data acquisition module, and an acceleration data acquisition module. Transmission and preprocessing module, acceleration data dynamic threshold training module, algorithm application module for detecting driver fatigue driving state based on acceleration data, data fusion module, etc.

1. Data collection and preprocessing

(1) Data collection

The advantage of the steering wheel motion acceleration is that the characteristic changes are relatively obvious, which can reflect the changes of the steering wheel motion in real time, and is easy to capture. The present invention first collects the motion acceleration of the steering wheel through an acceleration sensor as one of the basis for judging the driving state of the driver.

For the collection of pulse data, a pulse sensor is used. The pulse sensor is fixed at the wrist radial artery of the human body, because the radial artery is the strongest place in the human body.

(2) Data preprocessing

Acceleration data and pulse data are both time-series data, and there is noise in the data collected by the sensor. There are common errors in the data collected by sensors, so it is necessary to perform basic data smoothing processing on the data collected by sensors. The present invention uses the weighted moving average method to smooth the data. The pulse signal has the characteristics of weak signal, low frequency and strong noise. Noise interference may cause the pulse signal to be distorted and cause a large detection error. It is necessary to denoise the pulse signal before feature extraction. The pulse signal of the human body is mostly distributed in the low frequency region, while the noise signal is generally evenly distributed in the high frequency region, and the wavelet component with larger amplitude usually appears in the signal mutation region. The present invention removes noise based on the threshold value method of wavelet transform.

2. Dynamic threshold training

(1) Acceleration dynamic threshold

The method for detecting the fatigue state of the driver in the present invention is based on the theory that the steering wheel does not move for 4 seconds, and is improved on this basis. The method of dynamic threshold value is added to remove the influence of the change of the angle between the steering wheel and the car body on the detection results, and more accurate Determine the state of the driver.

(2) Pulse dynamic threshold

The present invention proposes a method for detecting the driver's fatigue driving state based on a dynamic threshold, and judges the driver's fatigue degree according to the degree of the driver's heart rate cycle drop: the driver is generally awake for a period of time when he just starts driving, and during this period Detect the driver's pulse data and calculate the corresponding heart rate cycle value, take this heart rate cycle value as the driver's normal heart rate cycle value, and continuously detect the driver's pulse after this awake time and analyze the heart rate cycle value , compared with the normal heart rate cycle, if the degree of decrease exceeds 10%, the system determines that the driver is in a state of mild fatigue, and if the degree of decrease exceeds 20%, the system determines that the driver is in a state of fatigue.

3. Fatigue driving state detection algorithm based on decision-level data fusion

The present invention fuses the driver fatigue detection result based on the steering wheel motion acceleration and the driver fatigue detection result based on the pulse at the decision-making level.

The present invention also provides a fatigue state recognition method based on decision-making level data fusion, the method includes the following steps:

Step 1 constructs the basic probability distribution function. For the two evidence bodies of acceleration and pulse, calculate the probability distribution function f of the two, and at the same time, ensure that the two evidence bodies do not affect each other and are independent of each other.

Step 2 Apply the combination rule of evidence theory to obtain a new evidence body, which is composed of two evidence bodies, acceleration and pulse. The closer the basic probability distribution of the new evidence body is to 1, it means that The higher the accuracy of propositional judgment is.

Step 3 applies the decision rules to obtain the judgment and decision results about the fatigue state and output them.

Step 3 of the present invention adopts the decision rule based on probability distribution, expresses as for the decision rule based on probability distribution: for any set M, set ${\&ForAll;S}_1,S_2\&Subset;m,$ satisfy: $f\left(S_1\right)=\max \{f\left(S_i\right),S_i\&Subset;m\},$ If f(S ₁ )-f(S ₂ )>θ ₁ , f(M)<θ ₂ , f(S ₁ )>f(M), then S ₁ is the decision result of the event, where θ ₁ and θ ₂ represents the set threshold value.

Beneficial effect:

1. The fatigue driving state detection accuracy rate of the present invention is relatively high; experiments show that the fatigue driving state detection accuracy rate of the present invention reaches 91.67%.

2. The response time of the system of the present invention is short; the driver's angry driving state detection system has a delay of about 1 second, which is mainly spent on pulse data collection and transmission.

3. The space-time complexity of the present invention is low; the time complexity of the present invention is O(n); the system memory consumption is between 50-70M.

Description of drawings

Fig. 1 is a fatigue driving state detection model diagram of the present invention.

Figure 2 is a measured pulse waveform diagram of one cycle.

FIG. 3 is a diagram of a decision-level data fusion model of the present invention.

Fig. 4 is a flow chart of the method of the present invention.

Detailed ways

The invention will be described in further detail below in conjunction with the accompanying drawings.

The fatigue driving state detection method based on the decision-making level data fusion of the present invention comprehensively utilizes the indirect information of the motion acceleration of the steering wheel and the direct pulse information of the driver, which can effectively improve the detection accuracy of the fatigue driving state: the acceleration sensor is used to collect the motion acceleration data of the steering wheel, The weighted moving average method is applied to these raw data for smoothing, and based on the theory of 4s non-movement of the steering wheel, the detection result of the fatigue driving state is preliminarily judged; at the same time, the pulse sensor is used to collect the pulse data of the driver during driving, and the driving force is analyzed and calculated. According to the change of pulse rate of the driver, the threshold of the corresponding normal driving state is established for different individuals. By comparing with this normal threshold, it can be judged whether the current pulse rate of the driver is normal, and then it can be judged whether the driver is in a fatigue driving state; These two recognition results apply evidence theory for decision-level fusion to obtain more accurate detection results about whether the driver is in a fatigue driving state.

As shown in Figure 1, the system of the present invention includes a pulse data acquisition module, a pulse data storage and preprocessing module, a pulse data dynamic threshold training module, an algorithm application module for detecting driver fatigue driving state based on pulse data, an acceleration data acquisition module, Acceleration data transmission and preprocessing module, acceleration data dynamic threshold training module, algorithm application module for detecting driver fatigue driving state based on acceleration data, data fusion module, etc.

1. Data collection and preprocessing

(1) Data collection

The advantage of the steering wheel motion acceleration is that the characteristic changes are relatively obvious, which can reflect the changes of the steering wheel motion in real time, and is easy to capture. The present invention first collects the motion acceleration of the steering wheel through an acceleration sensor as one of the basis for judging the driving state of the driver.

Pulse wave (Pulse Wave) is the pressure wave formed when the blood flow starts from the aorta and propagates along the arterial system, and the propagation of blood flow in the arterial system is due to the contraction and relaxation of the aorta caused by the periodic contraction and relaxation of the ventricle of the heart. Each time the left ventricle contracts, blood is ejected into the aorta, causing the aortic wall to expand, and when the left ventricle relaxes, the aortic wall elastically recoils. The pulse originates from the beating of the aortic root and spreads along the arterial wall to the arteries of the whole body in turn. The pulse reflects the periodic rise and fall of pressure in the aorta. Arteries open and contract as the heart contracts and relaxes. Under normal circumstances, the pulse is consistent with the beating of the heart, the pulse is strong, the rhythm is uniform, the strength is consistent, and the interval is equal. The blood pumped by the heart flows into the aorta, causing the aorta to contract and relax. The blood flow starts from the root of the aorta in the form of pressure waves and propagates along the arterial system, forming a pulse. Every time the heart contracts and relaxes, a cycle can be generated. pulse wave. The waveform diagram shown in Figure 2 is a periodic pulse waveform intercepted in the periodic pulse waveform diagram collected by the system.

In Fig. 2, the abscissa T represents time, and the ordinate P represents the pressure value. The waveform within the time [0, t] is a typical pulse cycle waveform, and t represents a cycle of the pulse waveform. h1 represents the amplitude of the main wave, which is the height from the peak of the main wave to the baseline of the pulse wave diagram. The height from the bottom of the canyon to the baseline of the pulse wave diagram, h4 represents the amplitude of the dicrotic wave, and is the height between the parallel lines of the baseline drawn from the peak of the dicrotic wave to the bottom of the middle canyon, and t1 represents the time from the starting point of the pulse wave diagram to the main peak point value, t1 corresponds to the rapid ejection period of the left ventricle, t2 represents the time value between the starting point of the pulse wave diagram and the descending gorge, t2 corresponds to the systolic period of the left ventricle, and the period t2-t represents the period from the descending gorge to the pulse wave The time value between the end points of the graph corresponds to the diastolic period of the left ventricle, 0-t represents the time value between the starting point and the end point of the pulse wave graph, t corresponds to a cardiac cycle of the left ventricle, corresponding to the pulse, and A pulse cycle.

The physiological significance corresponding to each feature point in Figure 2 is as follows:

Point p1: represents the opening point of the aorta, that is, the starting point of injection. It is the lowest point of the entire pulse waveform diagram, marking the beginning of the rapid ejection period of the heart, and mainly reflecting the pressure and volume in the blood vessel at the end of systole.

Point p2: the highest point of aortic pressure. Here is the main wave, which is an ascending curve from the baseline of the waveform diagram to the peak of the main wave. The peak reflects the maximum pressure and volume in the artery, which constitutes the ascending branch of the main wave, reflecting the rapid ejection of blood from the ventricle and the rapid rise of arterial pressure. Sudden expansion of the tube wall. Its rising speed is mainly related to cardiac output, ventricular ejection speed, arterial resistance and wall elasticity, which can be expressed by the slope of the ascending branch. If the cardiac output is large, the ejection speed is faster, and the aortic elasticity decreases, the slope is larger and the amplitude is higher; if the cardiac output is lower, the ejection speed is slower, and the aortic elasticity is greater, the slope decreases. Small, low volatility.

Point p4: left heart ejection stop point, here is tidal wave, also called dicrotic wave front wave. It is located in the descending branch of the waveform diagram, generally delayed after the main wave, lower than the main wave and higher than the dicrotic wave. It is a reflected wave formed by the ventricle stopping blood ejection, arterial expansion, blood pressure drop, and arterial blood reverse flow in the late period of slowing ejection, mainly related to the change speed of peripheral resistance, vascular elasticity, and descending artery descending speed.

Point p5: the trough of the dicrotic wave, which is a downward notch wave formed by the descending branch of the main wave and the rising branch of the dicrotic wave. It mainly reflects the emptying time of aortic static pressure, which is the dividing point between systole and diastole, and is easily affected by peripheral resistance and descending branch velocity.

Point p6: aortic elastic recoil wave, that is, dicrotic wave. It is a prominent wavelet located after the trough of the dicrotic wave. It is formed when the ventricle begins to relax after the slow ejection period of the ventricle, and the intraventricular pressure drops rapidly to significantly lower than the aortic pressure. directional regurgitation. Due to the impact of the regurgitated blood, the aortic valve suddenly closes, and the regurgitated blood hits the suddenly closed aortic valve and bounces back, causing the aortic pressure to rise slightly again and the arterial wall to expand slightly. Therefore, an upward wavelet is formed in the middle of the descending branch, that is, the descending medium wave. It can reflect the functional status of the aortic valve, blood vessel elasticity and blood flow status.

For the collection of pulse data, a pulse sensor is used. The pulse sensor is fixed at the radial artery of the wrist of the human body, because the radial artery is the strongest place of the human pulse, it is convenient and accurate to obtain the pulse data of the human body, and it is easy to wear.

(2) Data preprocessing

Acceleration data and pulse data are both time-series data, and there is noise in the data collected by the sensor. There are common errors in the data collected by sensors, so it is necessary to perform basic data smoothing processing on the data collected by sensors. The system of the present invention at first utilizes the acceleration sensor to collect the motion acceleration of the steering wheel by the acceleration data acquisition module, as one of the basis for judging the driving state of the driver; the pulse data acquisition module utilizes the pulse sensor for the acquisition of the pulse data, and the pulse sensor is fixed on At the wrist radial artery of the human body; the acceleration data transmission and preprocessing module and the pulse data storage and preprocessing module all use the weighted moving average method to smooth the data. Points are gradually added to the average, gradually removing their influence on the overall smoothness:

${\mathrm{the\; s}}_i=\underset{j=-k}{\overset{k}{\mathrm{\&Sigma;}}}{w}_j{x}_{i+j}$ in $\underset{j=-k}{\overset{k}{\mathrm{\&Sigma;}}}{w}_j=1---\left(1\right)$

In formula (1), s _i represents the smoothing value of the i-th point; x _i+j represents the data point; w _j represents the weight, the weight near the middle is larger, the weight near the edge is smaller, and the sum of the weight is 1. In order to obtain a better data smoothing effect succinctly, the present invention selects the weighted moving average method as the method for acceleration data smoothing, such as using (1/4, 1/2, 1/4) as a weight, for the collected Steering wheel motion acceleration data, three adjacent points are used as a processing item, and the result after smoothing is updated to the corresponding data item in the middle, and the processing is cyclic until the data is processed.

The pulse signal has the characteristics of weak signal, low frequency and strong noise. The interference of human body state and external environment will have a relatively large impact on the collection of human physiological signals. These noise interference may lead to pulse signal distortion, which will cause a large detection error. It is necessary to denoise the pulse signal before feature extraction. The pulse signal of the human body is mostly distributed in the low frequency region, while the noise signal is generally evenly distributed in the high frequency region, and the wavelet component with larger amplitude usually appears in the signal mutation region. The present invention removes noise based on the threshold value method of wavelet transform.

First, the pulse signal needs to be transformed by discrete wavelet. A typical basic wavelet is:

suppose yes The Fourier transform of called the wavelet mother function. by wavelet mother function Translating and stretching yields discrete wavelet families:

In formula (3), a is the scaling factor, and b is the translation factor.

Suppose the pulse signal is signal(t), signal(t)=start(t)+noise(t), start(t) represents the original pulse signal, and noise(t) represents noise. Perform discrete sampling on the pulse signal: signal(t),t=0,1,2,...,N-1.

The coefficients of the wavelet transform are:

${\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; signal</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{\frac{\mathrm{<span\; class="notranslate">\; a</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}}\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; a</span>}}\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

After the wavelet coefficient W _signal (a, b) is obtained by formula (4), it is processed with a threshold value to determine the estimated value of the wavelet coefficient to ensure The value of is the smallest, and the threshold adopts the general threshold selection method:

$\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; med</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; 0.6475</span>}\sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; ln</span>}\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

In formula (5), med represents the median value of high-frequency orthogonal wavelet coefficients.

The estimated values of the wavelet coefficients are obtained Reconstruct the wavelet with wavelet inverse transform to get the estimated signal It is the pulse signal after denoising.

2. Dynamic threshold training

(1) Acceleration dynamic threshold

The motion acceleration data of the steering wheel is a time-series data flow, and the present invention studies the motion acceleration of the steering wheel based on a sliding window model. The sliding window model is a common model for dealing with time series data streams. The characteristic of the sliding window model is that the size of the window where the data it processes is fixed, and the end point of the sliding window is always the current time. This limitation also ensures that the valid data processed by the sliding window model is always the data in the latest arrival window in the data stream. .

The method for detecting the fatigue state of the driver in the present invention is based on the theory that the steering wheel does not move for 4 seconds, and is improved on this basis by adding a dynamic threshold method to remove the influence of the change in the angle between the steering wheel and the car body on the detection results, and more accurately Determine the state of the driver.

Due to the fact that the angle of the steering wheel and the condition of the road on which the vehicle is located are different, a fixed threshold cannot be used as the basis for detecting the state of the steering wheel. The present invention proposes a method for dynamically training the acceleration threshold of the steering wheel: first, through the acceleration data analysis of the steering wheel Whether the vehicle is in a relatively stable driving state, monitor the continuous time during which the left and right fluctuations of the steering wheel are less than 15 degrees during this period, divide this continuous time into multiple consecutive time periods, and obtain the weighted average of the acceleration data of the steering wheel in these time periods Then take the average value of these weighted averages to get the dynamic threshold of the driver on the current road section. After getting the threshold, the fluctuation range of the steering wheel is obtained by comparing the acceleration data. Apply the theory of the steering wheel not moving for 4s, and judge when the steering wheel does not move for 4s continuously The driver is in a state of fatigue driving.

(2) Pulse dynamic threshold

The pulse of a normal person is consistent with the heartbeat, which is 60-100 beats per minute, usually 70-80 beats per minute, with an average of about 72 beats per minute. The elderly are slower, at 55 to 60 beats/min. The pulse rate of normal people is regular, and there will be no phenomenon of different pulse intervals. The pulse strength of normal people is equal, and there will be no phenomenon of alternating strong and weak pulses. The frequency of the pulse is affected by age and gender. In addition, exercise and emotional agitation can make the pulse faster, while rest and sleep make the pulse slow down.

In order to obtain the relationship between pulse rate and fatigue, the present invention has carried out comparative experiment test. There were 12 experimenters in total, all healthy adults, aged between 23 and 26 years old. When the experimenter is awake, measure 5 sets of pulse values; when the experimenter is a little tired, measure 5 sets of pulse values; when the experimenter is relatively tired, continue to measure 5 sets of pulse values. According to the measured pulse value, the corresponding pulse rate was calculated, and the pulse rate changes of the experimenters in different physical states were counted, as shown in Table 1.

Table 1: The pulse rate changes of the experimenters in different states

It can be seen from Table 1 that there is a close relationship between pulse rate and human fatigue. When the experimenter is in a state of fatigue, the pulse rate decreases between 8.57% and 12.50%, and most of them are above 10%; when the experimenter is in a fatigue state, the pulse rate decreases between 19.44% and 24.66% , and most of them are above 20%. According to the conclusions drawn from the experiment, the present invention proposes a method for detecting the driver's fatigue driving state based on a dynamic threshold, and judges the driver's fatigue degree according to the degree of the driver's heart rate cycle decline: the driver's period of time when he just started driving is generally relatively Awake, detect the driver's pulse data during this period and calculate the corresponding heart rate cycle value, take this heart rate cycle value as the driver's normal heart rate cycle value, and continuously detect the driver after this awake time Compared with the normal heart rate cycle, if the decrease exceeds 10%, the system determines that the driver is in a state of mild fatigue, and if the decrease exceeds 20%, the system determines that the driver is in a fatigue state.

3. Fatigue driving state detection algorithm based on decision-level data fusion

The present invention fuses the driver fatigue detection result based on the steering wheel motion acceleration and the driver fatigue detection result based on the pulse at the decision-making level.

The present invention is based on the theory of finite set Θ, what Θ represents is an identification frame, includes all objects to be detected by the system, and the relationship between the objects is mutually exclusive; in the present invention Θ represents the set of two objects, fatigue and non-fatigue . That is, Θ={fatigue, not fatigue}.

Let the subset of θ be 2 ^θ , f is the mapping function from 2 ^θ to [0,1], and satisfy f(Φ)=0, for any S∈2 ^Θ , f(s)≥0 and ∑f (s)=1. f(s) represents the basic probability value of a recognition frame, reflecting the reliability of s. f ₁ () and f ₂ () represent the basic probability distribution functions derived from two independent evidence sources, which in the present invention correspond to the driver's fatigue driving state detected based on acceleration and the driver's fatigue driving detected based on pulse State these two bodies of evidence. A basic probability distribution function is calculated to reflect the fusion information of the joint action of the two evidence bodies:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&CirclePlus;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{\underset{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&cap;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; A</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\underset{{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&cap;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&Phi;</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

make $k=\underset{B_i\&cap;C_j=\mathrm{\&Phi;}}{\mathrm{\&Sigma;}}{f}_1\left(B_i\right)f_2\left(C_j\right)$

but $1-\underset{B_i\&cap;C_j=\mathrm{\&Phi;}}{\mathrm{\&Sigma;}}{f}_1\left(B_i\right)f_2\left(C_j\right)=\frac{1}{1-k}\underset{B_i\&cap;C_j=A}{\mathrm{\&Sigma;}}{f}_1\left(B_i\right)f_2\left(C_j\right)$

The present invention adopts the decision rule based on probability distribution, and can be expressed as follows for the decision rule based on probability distribution:

For any set M, let ${\&ForAll;S}_1,S_2\&Subset;m,$ satisfy: $f\left(S_1\right)=\max \{f\left(S_i\right),S_i\&Subset;m\},$ If f(S ₁ )-f(S ₂ )>θ ₁ , f(M)<θ ₂ , f(S ₁ )>f(M), then S ₁ is the decision result of the event, where θ ₁ and θ ₂ represents the set threshold value.

As shown in Figure 4, the present invention provides a kind of implementation method of the fatigue driving state detection system based on decision-making level data fusion, and this method comprises the following steps:

Step 1 constructs the basic probability distribution function. For the two evidence bodies of acceleration and pulse, calculate the probability distribution function f of the two, and at the same time, ensure that the two evidence bodies do not affect each other and are independent of each other.

Step 2 Apply the combination rule of evidence theory to obtain a new evidence body, which is composed of two evidence bodies, acceleration and pulse. The closer the basic probability distribution of the new evidence body is to 1, it means that The higher the accuracy of propositional judgment is.

Step 3 applies the decision rules to obtain the judgment and decision results about the fatigue state and output them.

Claims (8)

1. the fatigue driving state detection system based on decision making level data fusion, it is characterized in that, described system comprises acceleration information acquisition module, acceleration information transmission stores with pretreatment module, acceleration information dynamic threshold training module, the algorithm application module detecting driver tired driving state based on acceleration information, pulse data acquisition module, pulse data and pretreatment module, pulse data dynamic threshold training module, detect algorithm application module, the data fusion module of driver tired driving state based on pulse data;

The function of acceleration information acquisition module is: utilize acceleration transducer to gather recent movement acceleration information;

Acceleration information transmission with the function of pretreatment module is: to gathering the smoothing process of the recent movement acceleration raw data application method of weighted moving average with acceleration transducer;

The function of acceleration information dynamic threshold training module is: to the data calculating mean value again through acceleration information transmission and pretreatment module process, obtaining the dynamic threshold of driver at current road segment, then obtaining the waving interval of bearing circle by comparing acceleration information;

The function detecting the algorithm application module of driver tired driving state based on acceleration information is: adopt the motionless theory of bearing circle 4s, tentatively judges the testing result of fatigue driving state;

The function of pulse data acquisition module is: the pulse transducer based on the wrist radial artery place being fixed on human body gathers the pulse data in driver process;

Pulse data stores and with the function of pretreatment module is: the pulse data that storage of collected arrives, and the threshold method then based on wavelet transformation removes pulse signal noise, re-uses the method for weighted moving average to the smoothing process of data;

The function of pulse data dynamic threshold training module is: analyze and calculate driver pulse frequency change, set up the threshold value of corresponding abnormal driving state for different individualities;

The function detecting the algorithm application module of driver tired driving state based on pulse data is: judge that whether the current driver's pulse frequency obtained is normal by comparing with the threshold value of abnormal driving state, and then judge whether driver is in fatigue driving state;

The function of data fusion module is: to detecting the algorithm application module of driver tired driving state based on acceleration information and carrying out decision level fusion, by obtaining based on the fatigue driving state detection algorithm of decision making level data fusion the testing result whether driver is in fatigue driving state based on the recognition result application evidence theory of the algorithm application module of pulse data detection driver tired driving state.

2. a kind of fatigue driving state detection system based on decision making level data fusion according to claim 1, it is characterized in that: first described system utilizes acceleration transducer to gather the acceleration of motion of bearing circle by acceleration information acquisition module, as judge driver driving condition according to one of; Utilize pulse transducer for the collection of pulse data by pulse data acquisition module, pulse transducer is fixed on the wrist radial artery place of human body; Acceleration information transmission stores with pretreatment module and pulse data and all uses the method for weighted moving average to the smoothing process of data with pretreatment module, point weights the closer to smooth window edge are less, the point entering smooth window is all little by little count in mean value, eliminates the impact on overall smoothness gradually:

$s_i=\underset{j=-k}{\overset{k}{\mathrm{\&Sigma;}}}{w}_j{x}_{i+j}$ Wherein $\underset{j=-k}{\overset{k}{\mathrm{\&Sigma;}}}{w}_i=1---\left(1\right)$

In formula (1), s _irepresent the smooth value of i-th point; x _i+jrepresentative data point; w _jrepresent weights, comparatively large near middle weights, submarginal weights are less, and weights summation is 1; Select the method for weighted moving average as the method for acceleration information smoothing processing, for the recent movement acceleration information collected, within adjacent three o'clock, as a processing item, the result after smoothing processing is updated to middle corresponding data item, circular treatment, until process data.

3. a kind of fatigue driving state detection system based on decision making level data fusion according to claim 1, is characterized in that: described system removes pulse signal noise based on the threshold method of wavelet transformation, comprising:

First, need to carry out wavelet transform to pulse signal, wavelet is:

Suppose be fourier transform, be called wavelet mother function, by wavelet mother function translation can obtain discrete wavelet race with flexible:

In formula (3), a is contraction-expansion factor, and b is shift factor;

Suppose that pulse signal is signal (t), signal (t)=start (t)+noise (t), start (t) represents original pulse signal, noise (t) represents noise, carries out discrete sampling: signal (t), t=0 to pulse signal, 1,2 ..., N-1;

The coefficient of described wavelet transformation is:

$W_{\mathrm{signal}}(a,b)=2^{\frac{a}{2}}\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}f\left(n\right)(2^{a}n-b)---\left(4\right)$

Through type (4) obtains wavelet coefficient W _signalafter (a, b), process by threshold value, determine the estimated value of wavelet coefficient guarantee value minimum, threshold value adopts generic threshold value choosing method, comprising:

$T=\mathrm{med}/0.6475\sqrt{2\ln N}---\left(5\right)$

In formula (5), med represents the intermediate value of high frequency orthogonal wavelet coefficient;

Obtain the estimated value of wavelet coefficient with wavelet inverse transformation to wavelet reconstruction, obtain estimated signal it is exactly the pulse signal after denoising.

4. a kind of fatigue driving state detection system based on decision making level data fusion according to claim 1, it is characterized in that: first analyze vehicle by the acceleration information of the bearing circle collected by acceleration information acquisition module and whether be in metastable transport condition, during this period of time monitoring direction is faced left the continuous time that right fluctuation is less than 15 degree, decile this continuous time is multiple continuous print time periods, is transmitted the weighted mean value obtaining the acceleration information of bearing circle in these time periods with pretreatment module by acceleration information; Then the mean value of these weighted mean values is got by acceleration information dynamic threshold training module, obtain the dynamic threshold of driver at current road segment, obtain the waving interval of bearing circle by comparing acceleration information after obtaining threshold value, detect the algorithm application module of driver tired driving state according to the motionless theory of bearing circle 4s by based on acceleration information again, judge that when the continuous 4s of bearing circle is motionless driver is in fatigue driving state.

5. based on an implementation method for the fatigue driving state detection system of decision making level data fusion, it is characterized in that, described method comprises the steps:

Step 1: build Basic probability assignment function, for acceleration and pulse two evidence bodies, calculate both probability distribution function f respectively, guarantee to be independent of each other between these two evidence bodies simultaneously, separate;

Step 2: the rule of combination of application evidence theory obtains a new evidence body, this new evidence body is combined out by acceleration and these two evidence bodies of pulse, the basic probability assignment that new evidence body shows, more close to 1, shows the accuracy of proposition judgement higher;

Step 3: application decision rule, obtains the judgement result of decision about fatigue state and export.

6. the implementation method of a kind of fatigue driving state detection system based on decision making level data fusion according to claim 5, is characterized in that: described method carries out decision level fusion by the driver fatigue testing result based on recent movement acceleration and the driver fatigue testing result based on pulse.

7. the implementation method of a kind of fatigue driving state detection system based on decision making level data fusion according to claim 5, it is characterized in that: described method is based on the theory of finite set Θ, what Θ represented is a framework of identification, comprising the entire objects wanting system to detect, is the relation of mutual exclusion between object; Θ represents set that is tired and not tired two objects, i.e. Θ={ tired, not tired };

If the subset of θ is 2 ^θ, f is 2 ^θto the mapping function of [0,1], and meet f (Φ)=0, to arbitrary S ∈ 2 ^Θ, have f (s)>=0 and Σ f (s)=1, f (s) represents the elementary probability value of an identification framework, reflects the size to s reliability; f ₁() and f ₂() represents two independently evidence source Basic probability assignment function of deriving, the driver tired driving state that corresponding is goes out based on acceleration detection and these two the evidence bodies of driver tired driving state gone out based on pulse detection; Calculate a Basic probability assignment function, to reflect two coefficient fuse informations of evidence body, comprising:

$f\left(A\right)=f_1\&CirclePlus;f_2=\frac{\underset{B_i\&cap;C_j=A}{\mathrm{\&Sigma;}}{f}_1\left(B_i\right)f_2\left(C_j\right)}{1-\underset{B_i\&cap;C_j=\mathrm{\&Phi;}}{\mathrm{\&Sigma;}}{f}_1\left(B_i\right)f_2\left(C_j\right)}k---\left(6\right)$

Order $k=\underset{B_i\&cap;C_j=\mathrm{\&Phi;}}{\mathrm{\&Sigma;}}{f}_1\left(B_i\right)f_2\left(C_j\right)$

Then $1-\underset{B_i\&cap;C_j=\mathrm{\&Phi;}}{\mathrm{\&Sigma;}}{f}_1\left(B_i\right)f_2\left(C_j\right)=\frac{1}{1-k}\underset{B_i\&cap;C_j=A}{\mathrm{\&Sigma;}}{f}_1\left(B_i\right)f_2\left(C_j\right)$

8. the implementation method of a kind of fatigue driving state detection system based on decision making level data fusion according to claim 5, it is characterized in that: described step 3 adopts the decision rule based on probability assignments, decision rule based on probability assignments is expressed as: for arbitrary set M, if meet: and S _i≠ S ₁; If there is f (S ₁)-f (S ₂) > θ ₁, f (M) < θ ₂, f (S ₁) > f (M), then S ₁the result of decision to event, wherein θ ₁and θ ₂the threshold value of representative setting.