TW202409905A - Method for detecting car driver drowsiness at night using Taguchi method capable of quickly designing new combinations to increase the accuracy of detection and shortening time for development - Google Patents

Abstract

A method for detecting car driver drowsiness at night using the Taguchi method is disclosed, which first detects the car driver's face, then locates the facial feature points to find out the locations of the left eye, right eye, mouth, chin, nose, left eyebrow and right eyebrow of the face from the image, and determines whether the car driver is dozing off by using the eye aspect ratio. Since the accuracy of identification is low at night, the disclosed uses the Taguchi method to conduct night experiments to achieve a robust design in an uncertain environment. In the Taguchi method experiment, the aspect ratio threshold, the infrared intensity of the infrared light source, the image color, and the frame size are set as control factors, the distance from the infrared night vision camera and the infrared lamp to the eyes is the input, wearing glasses or not wearing glasses are set as interference factors, and an accuracy rate is the output. After the obtained device design results are implemented by the embedded system, the accuracy rate can be greatly improved.

Description

Using Taguchi method to detect driver’s drowsiness at night

The present invention relates to a method for detecting drowsiness of car drivers at night using the Taguchi method. This method, especially involving a Taguchi orthogonal experiment, can quickly design new combinations to increase the accuracy of detection and shorten development time. In particular, it means that it can be applied in uncertain environments and can still maximize the signal to noise. The one that obtains the best combination of robust control factors (design parameters) based on the Signal-to-Noise Ratio (S/N) ratio and taking into account quality characteristics.

Nowadays, the proportion of vehicles used for transportation in society is increasing, especially every household has a small car. Cars are needed for daily work and holiday travel, so car driving safety is very important. If a car accident occurs during driving, it will cause injury or death to the driver or passengers. The top 10 traffic accidents in 2011 include failure to yield according to regulations, failure to pay attention to the situation in front of the car, and failure to turn left according to regulations. These are often caused by poor mental state of the driver. According to the National Highway Traffic Safety Administration (NHTSA), in 2019, statistics found that the number of deaths caused by highway driving accidents due to drowsiness was 697, accounting for 1.9% of the total number of deaths in traffic accidents. This death rate is quite high. Therefore, if the car has a drowsiness detection device, it will be able to effectively prevent the driver from falling asleep and causing a car accident.

The common solution is to use a trial-and-error approach to find Develop the design parameters of a drowsiness detection device. However, such an experimental method not only takes a lot of time, but also does not guarantee that the obtained design parameters can be used in different situations, such as: different distances between the driver’s face and the camera, the environment Use in uncertain environments such as different ambient illumination levels and whether the driver is wearing glasses. Since this conventional solution method does not systematically select the product control factors, it cannot achieve the optimal selection of the optimal product qualification (correctness) rate.

At present, drowsiness detection is widely used in the automotive industry because it requires A high degree of attention can prevent driving accidents, so drowsiness detection is very suitable for use in the automotive industry. For this reason, installing a drowsiness detection device in a car's assisted driving system can not only prevent driving accidents, but also effectively increase the annual sales of cars. Therefore, it is necessary to develop an invention that can solve the technical shortcomings of the previous case and use the Taguchi method for driver drowsiness detection and embedded system implementation when driving at night.

The main purpose of the present invention is to overcome the above-mentioned problems encountered in the conventional art and to provide Provides a Taguchi orthogonal experiment that can quickly design new combinations to increase detection accuracy and shorten development time. It can be applied in uncertain environments and can still maximize the S/N ratio while taking into account quality characteristics. Obtain the best combination of robust control factors (design parameters) and use the Taguchi method to detect driver drowsiness at night.

In order to achieve the above purpose, the present invention is a method for detecting drowsiness of car drivers at night using the Taguchi method. It is applied to a drowsiness detection device and includes a micro single-board computer and an infrared night vision camera in the drowsiness detection device. and an infrared lamp for nighttime driver drowsiness detection (Nodding-off Detection), which uses real-time and continuous automatic detection of human faces and facial-related features to calculate the eye aspect ratio to determine whether the driver is dozing. The method at least includes the following steps: a detection environment establishment step: using the infrared night vision camera and the infrared lamp to photograph the driver's face at night, where the infrared night vision camera and the infrared lamp are located on the driver's seat of the car In the front, the infrared night vision camera is connected to the micro single-board computer; the factor selection step that affects the quality characteristics: select several control factors that affect the quality characteristics. These control factors are parameters that can be controlled, including eye length. The threshold of the Eye Aspect Ratio (EAR), infrared light source (mW/cm ² ), image color, and frame size; the level determination steps for various control factor changes: each control factor has three levels. Among these control factors, the threshold of the eye aspect ratio has three levels of variation: 0.21, 0.22 and 0.23, and the infrared intensity of the infrared light source has 4, 5 and 6 mW/cm ² as its variation levels. Three levels, the image color uses gray scale, color and black and white as its three levels of change, the frame size uses 520, 620 and 720 pixel as its three levels of change; experimental orthogonal table selection steps: based on the various controls In the step of determining the level of factor changes, an orthogonal table is established for each control factor and its level. The orthogonal table selected is L ₉ (3 ⁴ ). The distance from the infrared night vision camera and the infrared lamp to the eyes is used as input. Wearing glasses and not wearing glasses as interference factors. According to these inputs and these interference factors, an accuracy rate output of the corresponding measurement value is obtained, and based on the orthogonal table experiment, the accuracy rate is recorded and identified; the experimental data and S/N Ratio calculation steps: Use the L ₉ (3 ⁴ ) Taguchi experiment orthogonal table analysis, conduct orthogonal experiments, and calculate the average and standard deviation of the measurement values (i.e., accuracy) obtained from each group of experiments in the orthogonal table. Finally, the S/N ratio is calculated based on the average value and the standard deviation of the measurement values (i.e., the accuracy) to obtain the degree of variation of the measurement values; the factor response analysis step: calculate the S/N ratio or The quality characteristics are put into the factor response analysis table and classified according to levels. Then the average value is obtained through level classification, and a new S/N ratio or factor response analysis table of quality characteristics is obtained again. From the new S/N ratio or quality characteristics The factor response analysis table classifies the eye aspect ratio threshold, the infrared light source, the image color, and the frame size into four categories of control factors to maximize the S/N ratio. The maximum value is obtained through mean analysis to obtain the new best A combination of control factors, in which the threshold of the eye aspect ratio is 0.22, the infrared intensity of the infrared light source is 4 mW/cm ² , the image color is grayscale, and the frame size is 620 pixel is the new best control factor combination; and the steps to confirm the design results: after re-conducting the new optimal control factor combination for nighttime driver drowsiness detection experiments, compared with the original design, a higher S/N ratio was obtained, confirming that the new The optimal control factor combination is the optimal and robust control factor combination.

In the above embodiment of the present invention, the micro single board computer further includes a touch screen display display module.

In the above embodiment of the present invention, the distance from the infrared night vision camera to the eye is equal to the distance from the infrared lamp to the eye.

In the above embodiment of the present invention, the distance between the infrared night vision camera and the infrared lamp to the eye is 55 to 65 cm.

In the above embodiment of the present invention, the micro single board computer is equipped with an application program library. It is used to automatically detect face and eye recognition in images continuously photographed on the face. Through facial feature point positioning, it can find out the position of the left eye, the position of the right eye, the position of the mouth, the position of the chin, the position of the nose, and the position of the left eye from the image. The eyebrow position and the right eyebrow position are used to calculate the eye aspect ratio.

In the above embodiment of the present invention, the micro single board computer uses the application program library The front face detection and facial feature point acquisition instructions provided by the Dlib-ml program module locate the eye coordinates and calculate the eye aspect ratio, and then use the convex command provided by the OpenCV program module to draw the Eye contour.

In the above embodiment of the present invention, the formula of the eye aspect ratio is: EAR= ; Among them, the It is the leftmost end of the white eyeball; the It is the upper left corner of the black eyeball; the It is the upper right corner of the black eyeball; the It is the right end of the white eyeball; the is the lower right end of the black eyeball; and the It is the lower left end of the black eyeball.

In the above embodiments of the present invention, the average value of the measured values (i.e., accuracy) is , and the standard deviation S of the measured values, and the formula for the S/N ratio is: = ; S= ; S/N= ; Among them, the is the i-th measurement value.

Please refer to "Figures 1 to 4", which are respectively the drowsiness detection device of the present invention. Schematic diagram of the structure, schematic diagram of the detection process of the drowsiness detection method of the present invention, photos of the open and closed eyes detection of the present invention, and photos of the eye aspect ratio of the present invention. As shown in the figure: the present invention is a method for detecting driver drowsiness at night using the Taguchi Method (Taguchi Method) for driver drowsiness detection (Nodding-off Detection), and uses Raspberry Pi 4 Model B is a micro single board computer 1, Raspberry Pi Noir Camera V2 8MP is an infrared night vision camera 2, and 48 external light bulbs 850 nm are infrared lamps 3 to implement a drowsiness detection device 100, and the drowsiness detection device 100 It is installed in vehicle 4. When used, the control factors and interference factors are first selected to perform orthogonal experiments to obtain the signal-to-noise ratio, and the optimal and robust control factor combination is finally obtained through mean analysis.

The control factor mentioned above is a controllable design parameter. The present invention uses the threshold of the eye aspect ratio (EAR), infrared light source (mW/cm ² ), image color, and frame size to conduct experiments and as control factors. Each control factor has three levels, so the Taguchi method orthogonal table uses L ₉ ( ) Orthogonal table analysis. The control factor level table designed by the present invention is shown in Table 1, where:

The threshold of the eye aspect ratio is the A factor, and three values of 0.21, 0.22, and 0.23 were selected for the experiment. If the threshold is too high, the eyes will be open but the eyes will be mistakenly judged as closed; if the threshold is too low, the eyes will be mistakenly judged as open.

The infrared intensity of the infrared light source is the B factor, and its unit is mW/ ^cm2 . If the infrared intensity is too weak, the Raspberry Pi Noir Camera V2 8MP infrared night vision camera will not be able to recognize the face and eyes; if the infrared intensity is too strong, the Raspberry Pi Noir Camera V2 8MP infrared night vision camera will have blurred recognition. Therefore, the infrared light source is set to 4, 5 and 6 mW/ ^cm2 .

The image color is the C factor. Since infrared recognition is poor at night, grayscale or black and white is added for experimentation to reduce the effect of light on face and eye recognition. Therefore, experiments are conducted on grayscale, color, and black and white.

The frame size is the D factor. The frame size will affect the face, eye interpretation, and program running speed, so the frame is set to 520, 620 and 720 pixels for experiments. Table I factor instruction Level 1 Level 2 Level 3 A Eye aspect ratio threshold 0.21 0.22 0.23 B Infrared light source (mW/ ) 4 5 6 C Image color Grayscale color black and white D Frame size (pixel) 520 620 720

The following embodiments are merely examples for understanding the details and connotations of the present invention, but are not intended to limit the scope of the patent application of the present invention.

In a preferred embodiment of the present invention, the present invention is targeted at the night time period, using Taguchi method orthogonal experiment to design a detection device based on an embedded system. Through real-time and continuous automatic detection of human faces and face-related features, it can calculate the eye aspect ratio to determine whether the car driver is dozing. Thereby improving the accuracy of nighttime drowsiness detection. The method proposed by the present invention is shown in Figure 2, which at least includes the following steps:

Detection environment establishment step s11: Use the infrared night vision camera 2 with the infrared The linear light 3 takes pictures of the driver's face at night, in which the infrared night vision camera 2 and the infrared light 3 are located in front of the driver's seat of the car, and the infrared night vision camera 2 is connected to the micro single board computer 1 .

Step s12 of selecting factors that affect quality characteristics: Select several control factors that affect quality characteristics. These control factors are parameters that can be controlled, including the threshold of eye aspect ratio (EAR), infrared light source (mW) /cm ² ), image color, and frame size.

Step s13 of determining the levels of changes of various control factors: each control factor is provided with three levels. Among these control factors, the threshold of the eye aspect ratio is 0.21, 0.22 and 0.23 as its three levels of change, the infrared intensity of the infrared light source is 4, 5 and 6 mW/ ^cm2 as its three levels of change, the image color is grayscale, color and black and white as its three levels of change, and the frame size is 520, 620 and 720 pixels as its three levels of change.

Experimental orthogonal table selection step s14: establish an orthogonal table based on each control factor and its level in the step of determining the level of changes in various control factors. The orthogonal table selected is L ₉ (3 ⁴ ), and the infrared night vision camera 2 and The distance from the infrared lamp 3 to the eyes is the input, and wearing glasses or not wearing glasses is the interference factor. According to the inputs and the interference factors, a correct rate output of the corresponding measurement value is obtained, and based on the orthogonal table experiment, Record the identification accuracy.

Experimental data and S/N ratio calculation step s15: Use the L ₉ (3 ⁴ ) Taguchi experimental orthogonal table analysis to conduct an orthogonal experiment, calculate the average value and standard deviation of the measurement values (i.e., accuracy) obtained from each group of experiments in the orthogonal table, and finally calculate the signal-to-noise ratio (Signal-to-Noise Ratio, S/N) based on the average value and the standard deviation of the measurement values to obtain the degree of variation of the measurement values.

Factor response analysis step s16: Put the S/N ratio or quality characteristics into the factor response analysis table and classify them according to levels, and then average the values through level classification to obtain a new factor response analysis table for S/N ratio or quality characteristics. , classify the eye aspect ratio threshold, the infrared light source, the image color, and the frame size from the new factor response analysis table of S/N ratio or quality characteristics into four categories of control factors to maximize the S/N ratio, The maximum value is obtained through mean analysis to obtain the new best control factor combination, in which the threshold of the eye aspect ratio is 0.22, the infrared intensity of the infrared light source is 4 mW/cm ² , the color of the image is gray scale, and the A frame size of 620 pixel is the new best combination of control factors.

Design result confirmation step s17: Re-run the new optimal control factor combination overnight After experimental comparison of driver drowsiness detection between cars, compared with the original design, a higher S/N ratio was obtained, confirming This new optimal control factor combination is an optimal and robust control factor combination. If so, through the above disclosure The process constitutes a brand-new method for detecting driver drowsiness at night using the Taguchi method.

In the above experimental step s11, the present invention uses a Raspberry Pi Noir Camera V2 8MP infrared night vision camera 2, and an external 48-bulb 850 nm infrared light 3 to take photos at night, and the micro single-board computer 1 further includes a touch screen display module 11, which has an application library 12. The image of the facial photography is automatically detected by the application library 12 of the Raspberry Pi 4 Model B micro single-board computer 1. The left eye position, right eye position, mouth position, chin position, nose position, left eyebrow position and right eyebrow position are found from the image through facial feature point positioning to calculate the eye aspect ratio. The experimental framework diagram is shown in Figure 1. The present invention uses the front face detection and facial feature point capture instructions provided by the Dlib-ml program module 121 in the application library 12 to locate the eye coordinates and calculate the eye aspect ratio, and then uses the convex instruction provided by the OpenCV program module 122 to draw the eye outline, and then burns the above program modules 121 and 122 into the Raspberry Pi 4 Model B micro single board computer 1 for real-time detection, as shown in Figures 3 and 4.

From Figure 3, we can see that the formula for the eye aspect ratio is: EAR= ; Among them, the It is the leftmost end of the white eyeball; the It is the upper left corner of the black eyeball; the It is the upper right corner of the black eyeball; the It is the right end of the white eyeball; the is the lower right end of the black eyeball; and the It is the lower left end of the black eyeball.

In the implementation of the experiment, we first need to select an orthogonal table. The selected orthogonal table is L ₉ ( ), and then the input is the distance between the eye and the infrared night vision camera and the infrared light (the two distances of the infrared night vision camera and the infrared light are equal), and the distance is 55, 60, and 65 cm. The interference factor is wearing glasses and not wearing glasses, and the output is the accuracy rate (percentage, symbol y). After the experiment is completed, Calculation, is the average value of n measured values (i.e., accuracy), calculated as in equation (1), where is the i-th accuracy rate: = Next, we calculate S, which is the standard deviation of the n measured values. The calculation formula is as follows: S = Finally, the S/N ratio is calculated. The S/N ratio is the degree of variation of the measured value. The calculation formula is as follows: S/N = After calculation, the average value, standard deviation, and S/N ratio of each group of experiments are shown in Table 2. Experiment A B C D Distance = 55cm Distance = 60cm Distance = 65cm S S/N Wear glasses Without glasses Wear glasses Without glasses Wear glasses Without glasses 1 1 1 1 1 0.82 0.93 0.91 0.99 0.96 0.92 0.922 0.058 24.06 2 1 2 2 2 0.96 0.97 0.96 0.98 0.93 0.91 0.952 0.026 31.14 3 1 3 3 3 0.81 0.81 0.84 0.99 0.88 0.92 0.875 0.071 21.86 4 2 1 2 3 0.91 0.92 0.92 0.92 0.87 0.94 0.913 0.023 31.84 5 2 2 3 1 0.91 0.89 0.95 0.95 0.85 0.92 0.912 0.038 27.56 6 2 3 1 2 0.92 0.95 0.98 0.94 0.92 0.9 0.935 0.028 30.44 7 3 1 3 2 0.89 0.88 0.91 0.91 0.92 0.86 0.895 0.023 31.96 8 3 2 1 3 0.93 0.92 0.84 0.89 0.93 0.9 0.902 0.034 28.39 9 3 3 2 1 0 1 0 0.98 0 0.95 0.488 0.535 -0.80 average value: 25.1617

The so-called factor response refers to the impact of changes in control factors on the S/N ratio or quality characteristics. For example, the A factor changes from Level 1 to Level 2, that is, the threshold is 0.21 to 0.22. When , the average variation of S/N ratio (or quality characteristics) is called the factor response of A, which can be expressed as ; When the B factor changes from the second level to the third level, the average change in the S/N ratio (or quality characteristics) can be expressed as . in Average the sum of the S/N ratios of all Level1 in the A factor, Calculate by averaging the sum of the S/N ratios of all Level2 in the A factor The method is as follows: = = 25.686 = = 29.946 = 29.946-25.686 = 4.260 The calculations from all combinations can be compiled into the factor response table of the S/N ratio shown in Table 3. Table 3 A B C D Level 1 25.69 29.28 27.63 16.94 Level 2 29.95 29.03 20.73 31.18 Level 3 19.85 17.17 27.13 27.36 E(1→2) 4.26 -0.25 -6.90 14.24 E(2→3) -10.09 -11.86 6.40 -3.82 Range 10.09 12.12 6.90 14.24 Classification(Rank) 3 2 4 1 Salience NO YES NO YES

From the above, we can see that when the change of a factor has a significant impact on the S/N ratio (or quality characteristics), this factor is called an important factor. The analysis of variance in statistics is used to determine the importance of factors. Currently, the present invention uses the half principle to determine important factors.

Next is the factor response table of quality characteristics, as shown in Table 4. Table 4 A B C D Level 1 0.916 0.910 0.919 0.774 Level 2 0.920 0.922 0.784 0.927 Level 3 0.762 0.766 0.894 0.897 E(1→2) 0.004 0.012 -0.135 0.153 E(2→3) -0.158 -0.156 0.109 -0.031 Range 0.158 0.156 0.135 0.153 Classification(Rank) 1 2 4 3 Salience YES YES NO NO

The first category is factors that have an impact on S/N, which can be used to maximize the S/N ratio, that is, to reduce variation. The second category is factors that have an impact on quality characteristics, which are used to adjust the average value of quality characteristics to the target value without changing the variation of quality characteristics. This type of factor is called an adjustment factor. The third category is factors that have no impact on S/N and quality characteristics. This factor is used to reduce costs, as shown in Table 5. The classification of control factors. Table 5 Factor category Does it affect S/N? Does it affect the quality characteristics? Control Factors use 1 YES YES/NO B.D Used to reduce variation 2 NO YES A Used to adjust quality characteristics to target values 3 NO NO C To reduce costs

Next for process optimization, first adjust the first type of control factors (B, D) to maximize Chemical S/N ratio: A? B1 C? D2 Then adjust the second type of control factor (A) to make the quality characteristics reach the target value: A2 B1 C? D2 Next, adjust the third type of control factor (C). Since the third type of control factor will not have an impact, it is used to reduce costs, so the selection is as follows: A2 B1 C1 D2

That is, the best robust design parameters are eye aspect ratio (A control factor) of 0.22 (Level 2), infrared intensity of infrared light source (B control factor) of 4 mW/ ^cm2 (Level 1), image color (C control factor) of grayscale (Level 1), and frame size (D control factor) of 620 pixels (Level 2). Experiment 10 in Table 6 is the final accuracy confirmation experiment of the best robust design. Table 6 EXP. A B C D Distance = 55cm Distance = 60cm Distance = 65cm S S/N Wear glasses Without glasses Wear glasses Without glasses Wear glasses Without glasses 10 2 1 1 2 0.94 0.93 0.94 0.92 0.94 0.91 0.93 0.012 37.32

It can be seen from Experiment 10 in Table 6 that the S/N ratio is higher than any S/N of the 9 experiments in Table 2. ratio, so Experiment 10 is the best and robust combination of control factors.

As can be seen from the above, the present invention first detects the face of the car driver, and then locates the left eye, right eye, mouth, chin, nose, left eyebrow and right eyebrow of the face from the image through facial feature point positioning, so as to use the aspect ratio of the eyes to determine whether the car driver is dozing off. Since the recognition accuracy is low at night, the present invention uses the Taguchi method to conduct night experiments to achieve a robust design in an uncertain environment. In the Taguchi method experiment, the aspect ratio threshold, the infrared intensity of the infrared light source, the image color, and the frame size are set as control factors, the infrared night vision camera and the distance from the infrared light to the eyes are inputs, wearing glasses and not wearing glasses are interference factors, and the accuracy is the output. After the device design results were implemented by the embedded system, the accuracy was greatly improved.

In this way, the present invention can be applied to the design of a drowsiness detection device for car drivers at night. Method, it is proposed that the Taguchi orthogonal experiment determines the control factors and possible levels of the drowsiness detection device, and the best combination of robust control factors (design parameters) obtained is the eye aspect ratio threshold, infrared intensity of the infrared light source, image color, and frame size, it can maintain the device's excellent recognition accuracy in uncertain environments. This best-in-class robust design will be used with a tiny single-board computer, an infrared night vision camera, an infrared lamp, and an application library that detects eye aspect ratio.

In summary, the present invention is a method for detecting drowsiness of car drivers at night by using the Taguchi method. It can effectively improve various shortcomings of the method and can be applied in uncertain environments. It can still obtain the best combination of robust control factors (design parameters) by considering the maximum S/N ratio and taking into account the quality characteristics. Because the Taguchi orthogonal experiment is used, the best new combination of control factors for detecting drowsiness of drivers at night can be quickly designed to increase the accuracy of detection, thus shortening the development time, making the invention more advanced, more practical, and more in line with the needs of users. It has indeed met the requirements for the invention patent application, and a patent application is filed in accordance with the law.

However, the above is only a preferred embodiment of the present invention and should not be used to limit the scope of implementation of the present invention; therefore, any simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the content of the invention specification should still fall within the scope of the present invention patent.

100: Sleepiness detection device 1: Micro single-board computer 11: Touch screen display module 12: Application library 121: Dlib-ml program module 122: OpenCV program module 2: Infrared night vision camera 3: Infrared light 4: Vehicle s11～s17: Steps

Figure 1 is a schematic diagram of the structure of the drowsiness detection device of the present invention. Figure 2 is a schematic diagram of the detection flow of the drowsiness detection method proposed by the present invention. Figure 3 is a photo of the invention with eyes open and eyes closed. Figure 4 is a photo of the aspect ratio of the eye of the present invention.

s11~s17: Steps

Claims (8)

A method for detecting nodding-off of a car driver at night using the Taguchi method is applied to a nodding-off detection device and a micro single-board computer, an infrared night vision camera and an infrared light in the nodding-off detection device are used to detect nodding-off of a car driver at night. The method determines whether the car driver is nodding off by automatically detecting the face and facial features in real time and continuously and calculating the aspect ratio of the eyes. The method comprises at least the following steps: The detection environment establishment step: using the infrared night vision camera and the infrared light to take facial photography of the car driver at night, wherein the infrared night vision camera and the infrared light are located in front of the car driver's seat, and the infrared night vision camera is connected to the micro single board computer; the factor selection step affecting the quality characteristics: selecting several control factors affecting the quality characteristics, which are controllable parameters, including the threshold of the eye aspect ratio (Eye Aspect Ratio, EAR), the infrared light source (mW/ ^cm2 ), the image color, and the frame size; The step of determining the level of change of various control factors: each control factor is set to three levels. Among the control factors, the threshold of the eye aspect ratio is 0.21, 0.22 and 0.23 as its three levels of change, the infrared intensity of the infrared light source is 4, 5 and 6 mW/ ^cm2 as its three levels of change, the image color is gray, color and black and white as its three levels of change, and the frame size is 520, 620 and 720 pixels as its three levels of change; the step of selecting an experimental orthogonal table: according to the step of determining the level of change of various control factors, an orthogonal table is established for each control factor and its level. The orthogonal table selected is L ₉ (3 ⁴ ), taking the distance between the infrared night vision camera and the infrared lamp and the eye as input, wearing glasses and not wearing glasses as interference factors, obtaining the accuracy of the corresponding measurement value as output according to the inputs and the interference factors, and recording the recognition accuracy according to the orthogonal table experiment; Experimental data and S/N ratio calculation step: using the L ₉ (3 ⁴ ) Taguchi experimental orthogonal table analysis, conducting an orthogonal experiment, calculating the average value and standard deviation of the measurement values (i.e., accuracy) obtained from each group of experiments in the orthogonal table, and finally calculating the signal-to-noise ratio (Signal-to-Noise Ratio, S/N) according to the average value and the standard deviation of the measurement values (i.e., accuracy) to obtain the degree of variation of the measurement values; Factor response analysis steps: put the S/N ratio or quality characteristics into a factor response analysis table and differentiate them according to the level, then take the average value through level classification to re-obtain a new factor response analysis table of the S/N ratio or quality characteristics, and classify the four control factors of the eye aspect ratio threshold, the infrared light source, the image color, and the frame size from the new factor response analysis table of the S/N ratio or quality characteristics to maximize the S/N ratio, and take the maximum value through mean analysis to obtain a new optimal control factor combination, wherein the eye aspect ratio threshold is 0.22, the infrared intensity of the infrared light source is 4 mW/ ^cm2 , the image color is grayscale, and the frame size is 620 pixel is the new optimal control factor combination; and a design result confirmation step: after the new optimal control factor combination is re-tested for nighttime car driver drowsiness detection, a higher S/N ratio is obtained compared with the original design, confirming that the new optimal control factor combination is the best and most robust control factor combination. According to the method for detecting a car driver's drowsiness at night using the Taguchi method as described in item 1 of the patent application, the micro single-board computer further includes a touch screen display module. According to the method described in item 1 of the patent application, the Taguchi method is used to detect driver's drowsiness at night, wherein the distance from the infrared night vision camera to the eyes is equal to the distance from the infrared lamp to the eyes. According to the method for detecting a car driver's drowsiness at night using the Taguchi method as described in item 1 or 3 of the patent application, the distance between the infrared night vision camera and the infrared light and the eyes is 55 to 65 centimeters. According to the method for detecting drowsiness of car drivers at night using the Taguchi method as described in item 1 of the patent application scope, the micro single-board computer is equipped with an application library for continuous automatic detection of images taken by the face. Face and eye recognition, through facial feature point positioning, find the left eye position, right eye position, mouth position, chin position, nose position, left eyebrow position and right eyebrow position from the image to calculate the length and width of the eyes Compare. According to the method for detecting driver drowsiness at night using the Taguchi method described in item 5 of the patent application, the micro single-board computer uses the front face detection provided by the Dlib-ml program module in the application library. Use the facial feature point acquisition command to locate the eye coordinates and calculate the eye aspect ratio, and then use the convex command provided by the OpenCV program module to draw the eye outline. A method for detecting a car driver's drowsiness at night using the Taguchi method as described in item 1, 5 or 6 of the patent application, wherein the formula for the eye aspect ratio is: EAR = ; Among them, the The leftmost end of the white of the eye; The upper left end of the black eyeball; The upper right corner of the black eyeball; The rightmost end of the white of the eye; The lower right corner of the black eyeball; and It is the lower left end of the black eyeball. According to the method for detecting driver drowsiness at night using the Taguchi method as described in item 1 of the patent application, the average value of these measurement values (i.e., the accuracy rate) , the standard deviation S of these measured values, and the formula of the S/N ratio is: = ; S= ; S/N= ; Among them, the is the i-th measurement value.