JPH0257433A - Device for detecting doze - Google Patents

Abstract

PURPOSE:To accurately detect a doze by detecting the movement of the eyelid of a man and distinguishing a doze level according to a variable threshold level based on this detected result and giving an alarm. CONSTITUTION:A light emitting element 10a is driven by a driving circuit 11 and the emitted light is projected toward the eyeball of a man and, then received by a light receiving element 10b, to detect the movement of the eyelid of the man by means of a sensor 10. At this time, the output signal of the light receiving element 10b is inputted into a CPU 14 via a light receiving quantity/voltage converting circuit 12 and an A/D converter 13. The CPU 14 extracts a doze detecting evaluation parameter while operating a doze level. The data of the doze level is inputted into a comparator 17 via a D/A converting circuit 16. When the voltage of the doze level is higher than the reference voltage of a circuit 18, an alarm device 19 is operated.

Description

DETAILED DESCRIPTION OF THE INVENTION BACKGROUND OF THE INVENTION This invention relates to a device for detecting dozing off based on eyelid movement detection.

There are devices that detect drowsiness by detecting eyelid movement, measuring eye-opening time, and comparing this eye-opening time with a predetermined threshold time, but the sensor is based on mere discrimination of eye-opening time. The installation is based on position, eye movement,
It was easily affected by facial movements and often caused malfunctions. Also, the threshold time varies considerably depending on the person, situation, and purpose of use.

It has been difficult to always perform appropriate dozing detection.

SUMMARY OF THE INVENTION An object of the present invention is to provide a dozing detection device that is less likely to malfunction due to external influences.

Another object of the present invention is to provide a dozing detection device that can be manually or automatically adjusted depending on individual differences, purpose of use, etc.

The dozing detection device according to the present invention includes means for detecting eyelid movements, means for extracting evaluation parameters for dozing detection from the detected eyelid movements, a predetermined membership function regarding the extracted evaluation parameters, and a predetermined membership function regarding the dozing level. It is equipped with an inference calculation means for calculating a dozing level by applying a predetermined rule using a membership function, and a means for discriminating the calculated dozing level at a variable threshold level and generating an alarm output. It is characterized by

According to this invention, since dozing is detected using fuzzy inference, sensor installation is not affected by position, sensor misalignment, eye movements, facial movements, etc., depending on the rule settings. A device that can accurately detect dozing is realized.

In addition, the threshold level for discriminating the dozing level to finally obtain the alarm output is variable.
The dozing level at which the alarm is output can be changed depending on the purpose of use, increasing the degree of freedom of use. In other words, it is possible to meet the demands of detecting dozing at an early stage and reliably detecting a state that is visible to the human eye.

Preferably, the apparatus includes means for collecting data related to evaluation parameters during a predetermined observation period, and means for setting the threshold level using the collected data.

As a result, the personal characteristics of the person to whom the sensor is attached are first observed, and the level of drowsiness at which the alarm is output can be changed depending on the observation results.

Appropriate dozing detection can be achieved depending on individual differences.

The dozing detection device according to this invention also includes: Using a means for detecting eyelid movement, a means for extracting an evaluation parameter for dozing detection from the detected eyelid movement, a predetermined membership function regarding the extracted evaluation parameter, and a predetermined membership function regarding the dozing level, An inference calculating means for calculating a drowsiness level by applying a predetermined rule, a means for discriminating the calculated drowsiness level by a threshold level and generating an alarm output, and a means for generating an alarm output by discriminating the calculated drowsiness level by a threshold level, data related to evaluation parameters in a predetermined observation period. The method is characterized by comprising means for collecting data, and means for changing at least one of the membership function and the rules based on the collected data.

With this invention, individual differences related to dozing, such as short or long eye opening times when awake, have been observed.
Accordingly, fuzzy inference is executed to accurately detect dozing off without the influence of individual differences.

Description of examples FIG. 1 shows a schematic configuration of a dozing detection device.

The sensor lO that detects the movement of the eyelids has two light emitting elements 10a.
and a light receiving element lOb. The sensor IO is attached to glasses, for example, and a light emitting element 10a driven by a drive circuit 11 projects light toward the position of the eyeball, and the diffusely reflected light is received by a light receiving element 10b. Although it is sufficient to detect the movement of the eyelid of one eye with the sensor 10, it is preferable to detect the movement of the eyelids of both eyes and calculate the average value or MAX value.

The output signal of the light receiving element 10b is transmitted to the received light amount/voltage conversion circuit 12.
The signal is converted into a voltage signal by the A/D conversion circuit 13, and then converted into a digital signal by the A/D conversion circuit 13 and taken into the CPU 14. C
The PU 14 extracts feature quantities, ie, evaluation parameters, which will be described later, from the input data representing eyelid movements, performs fuzzy inference, and outputs data representing the dozing level.

In the memory 15, membership functions related to evaluation parameters and membership functions related to dozing levels are preset, as will be described later. C.P.U.I.
4 performs fuzzy Hfi theory using these membership functions and predetermined rules (so-called If, tl+er+rules).

Fuzzy inference can be applied not only to binary-type computers or processors such as CPUI4, but also to specialized devices (regardless of analog or digital type) for fuzzy inference (for example, "Nikkei Electronics J, July 27, 1987, 148th to 15th
2, p. 1, Nikkei McGraw-Hill).

Data representing the dozing level, which is the fuzzy inference result outputted from the CPU 14, is converted into an analog voltage signal by the D/A conversion circuit I6, and is applied to the positive input terminal of the comparator 17. On the other hand, the reference voltage outputted from the reference voltage generation circuit 18 is applied to a variable voltage dividing resistor R, and a voltage V 1 that determines a threshold level is outputted from this resistor R. This voltage V 1 is r iIT t and is input to the negative input terminal of the comparator 17 .

The comparator 17 generates an output when the voltage of the D/A-converted dozing level signal exceeds the voltage V 1 and activates a buzzer or other alarm device 19 .

The output signal of the light receiving element lOb, that is, the diffusely reflected light, exhibits a high level when the eyelids are closed. An example of the received light signal waveform (referred to as the "blink waveform") obtained when the eyelids are closed from the open state and then opened again (in other words, when you "blink") is shown in Figure 3. has been done.

The eye-closing time T as shown in Figure 1 is used as a feature of the blink waveform.
(for example, the half width of the waveform), the peak value H, and the rising angle A. In addition to these feature amounts, evaluation parameters for detecting dozing include the number of blinks N per unit time, the blink interval, and the like. In this example, an awake state, a tilted eyes state (nearly falling asleep), and a hypnotic state (
The eye-closing time T and wave height value H, which show a relatively clear difference between the two states (mostly sleeping), are used as evaluation parameters. The dozing level indicates a level corresponding to the degree of dozing, and is lowest in the awake state and highest in the 1-fall asleep state.

The eye-closing time T tends to increase as the dozing level increases. Furthermore, the peak value H tends to decrease as the dozing level increases. This can be interpreted to mean that when the drowsiness level increases, the eyelids become half-closed, and as a result, the average level of the light reception signal increases, so the peak value decreases relatively.

The blink signal has large fluctuations, and it is difficult to obtain the correct eye-closing time T and peak value H from a single blink waveform. Therefore, the average value of these parameters over a fixed period of 1.7 seconds (for example, 30 seconds) is determined. Each of these
The measured average eye-opening time TV is referred to as nl-determined average wave height value H. Further, as evaluation parameters used in the fuzzy trE theory, VOshi0 data T , H obtained by dividing these values T , H by the average v v eye opening time T during awakening and the average wave height value H during awakening, respectively, are used. That is, T −p
p pT /T , H
-H/H. sensor.

v vo p v
lol.

It is considered that the user is awake when the user attaches the doze detector or starts using the doze detector. Therefore, the average eye-opening time and average wave height value during a certain period of time (for example, 1 minute) immediately after the power is turned on are T.

It can be H.

VO Figure 4 shows an example of the membership function of the binary T, H and dozing level p. This membership function is a self-effective form for detecting dozing off by statistically analyzing the collected data. In the membership function of T and H, S is Sa+all (small, low) 2M is M
ediu+g (medium), B stands for 131g (large, high). Also, in the membership function of the dozing level, A is Avaak (awake) and S is So
mnolence (somnolence), D stands for Dozing (falling asleep).

Figure 5 (A) summarizes the rules used in fuzzy inference in the form of a table. For example, he expresses the following shillel:

Or Ir) T or thick (T-B) and H.

If p is small (H-3), then falling asleep (D).

This rule further includes the following: ``If T -3 and H-
There are also rules such as ``If it is 3, it is an awakening (A)'' or ``If it is T - B and HB, it is an awakening (A).'' These rules take into consideration the case where there is an influence of disturbance such as displacement of the sensor O, and the existence of such rules enables fuzzy inference that is resistant to the influence of disturbance.

FIG. 5(B) shows an example of another rule. According to this rule, it becomes difficult to determine whether a person has fallen asleep.

Although nine rules are shown in each of FIGS. 5A and 5B, the number of rules can be set arbitrarily.

Further, the membership function may also be changed as appropriate.

FIG. 6 shows an example of a process in which fuzzy inference is performed, and this is an example of fuzzy inference that follows the M I N/MAX operation rule.

In a certain rule, for the membership function of T. The goodness of fit (function value) of the calculated T value T,l is determined and similarly for the membership function of H. The appropriateness (function value) of the calculated H value H11 is determined,
The one with smaller fitness is selected (MIN calculation)
. The membership function of the dozing level that follows the same rule is cut by the result of the M I N calculation (equivalent to truncation, MINei calculation).

The above-mentioned processing is performed for the rules of the schedule, and the cutting results of the membership functions of the dozing level are negotiated (MAX operation). This MAX operation result is defuzzified by, for example, finding its center of gravity, and a doze relevel representing the inference result is found.

Such fuzzy inference is repeated at appropriate time intervals.

FIG. 7(A) shows the level change of the light reception signal of the sensor O and the blink waveform during the transition from the awake state to the sleep state via the somnolence state.

FIG. 7 (13) shows feature extraction of such a received light signal.

It shows the change in dozing level obtained by performing fuzzy inference. This dozing level signal is the threshold level V set by the adjustable low voltage dividing resistor R.
The level is discriminated by

Alert when the drowsiness level signal exceeds level V. information is output.

In addition, in FIGS. 7(A) and 7(B), in order to express changes in the dozing level over a short period of time, a sudden transition from an awake state to a falling asleep state is shown.

Since the threshold level (1) is variable, this level (V) can be changed depending on the purpose of use. For example, for a car driver, the threshold level V is set so that a warning is output as soon as the drowsiness level increases by 1.
Set the threshold level V to a low value.Also, if it is not necessary to detect dozing off too early and the purpose is to reliably detect falling asleep, it is sufficient to set the threshold level V to a high value. .

Set the threshold level V by CPU instead of manually.
14 may be automatically set. This is the measured value T in the awake state when the power is turned on, etc.
, H may be set or changed based on νOvO.

Since vO −T /T becomes lower, the threshold v
It is preferable to set the vO level ■ to a low value.

In FIG. 2, a voltage dividing circuit is provided in which resistors R and R2 are connected in series, and the output of this voltage dividing circuit is provided to a ratio q converter 17. A plasma voltage is applied to the resistor R1. Further, the resistor R2 is connected to the port A of the CPU14. This port p A is normally in a floating state. I wanted to,
Comparator I7 has a base '$ output base of lightning positive pressure generation circuit 8! fI
The voltage can be detected. Referring to FIG. 8, if the mean value of the eye open time for one minute immediately after the power supply official (awake state) is, for example, 0.4 seconds or more, port A of the CPU 14 is set to the L level. As a result, the basic voltage becomes the resistance R
and R2 and applied to the comparator 17.

That is, the threshold level becomes lower.

It is also possible to automatically set and change other factors based on measurement (learning) of various values during wakefulness. For example, FIG. 9 shows an example of the process of changing the rule based on the T/7) measurement result. If the ratio between the maximum value and the minimum value of the eye opening time per minute in the awake state immediately after the power is turned on is 2 or more, the rule shown in Figure 5 (13) is selected; otherwise, the rule shown in Figure 5 (13) is selected. (A
) is selected. As a result, a rule is adopted in which somnolence detection is delayed in order to avoid false detections for people whose eye opening times vary widely even when awake.

Rules can also be set and changed based on measured values of other parameters. It is also possible to change the form of the membership function based on the measurement results of the evaluation parameters. For example, shift the membership function on the horizontal axis. Since the membership function is stored in memory, shifting the membership function is easy.

The memory arrangement may be changed, or the storage state of the memory may be left unchanged and the address may be shifted.

As described above, it becomes possible to accurately detect dozing off by IT, corresponding to individual differences in habits.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing the configuration of the dozing detection device;
The figure is a block diagram showing a modified example, Figure 3 is a waveform diagram showing an example of a blinking waveform, and Figures 4 (Δ) (B) and (C) are examples of membership functions regarding T, H and dozing level p. The graphs shown in FIGS. 5(A) and 5(B) respectively show examples of different rules in the form of tables, and FIG. 6 is a diagram for explaining the fuzzy inference process, and FIG. 7(A) 7(B) is a chart showing a received light signal waveform, and FIG. 7(B) is a chart showing a dozing level which is an inference result. FIG. 8 is a flow chart showing an example of a process for automatically setting a threshold level, and FIG. 9 is a flow chart 1 showing an example of a process for changing a fuzzy rule. Figure 3 Figure 4 +A+ 10...Sensor, 10a...Light emitting element. 10b... Light receiving element, 14... CPU. 15...Memory, 17...Comparator. 18...Reference voltage generation circuit. 19...Alarm device. R...Variable voltage dividing resistor. R, R,... Voltage dividing resistance. that's all

Claims (4)

[Claims] (1) means for detecting eyelid movements; means for extracting evaluation parameters for dozing detection from the detected eyelid movements; and determining a predetermined membership function regarding the extracted evaluation parameters and a predetermined membership function regarding the dozing level. a dozing detection device comprising: inference calculating means for calculating a dozing level by applying a predetermined rule; and means for discriminating the calculated dozing level using a variable threshold level and generating an alarm output. (2) means for collecting data related to the evaluation parameter during a predetermined observation period; and means for setting the threshold level using the collected data. The dozing detection device described. (3) means for detecting eyelid movements; means for extracting evaluation parameters for dozing detection from the detected eyelid movements; and determining a predetermined membership function regarding the extracted evaluation parameters and a predetermined membership function regarding the dozing level. an inference calculation means for calculating a dozing level by applying a predetermined rule using the method; a means for discriminating the calculated dozing level by a threshold level and generating an alarm output; A dozing detection device comprising: means for collecting data; and means for changing at least one of the membership function and the rules based on the collected data. (4) Any one of claims (1) to (3), wherein the evaluation parameters are a parameter related to the average eye-closing time and a parameter related to the average peak value of a waveform representing eyelid movement. The dozing detection device described in .