TW201625176A - Method of heartbeat detecting for drowsiness detection - Google Patents

Abstract

A method of heartbeat detecting for drowsiness detection is disclosed. The method comprises detecting the heartbeat signal by body senor. Calculating the heartbeat signal for R-wave to transfer the signal to RRI value. And then proceed the signal calculating process. The process includes to remove the anomaly value range of RRI and to retrieve a renew RRI value. Using the renew RRI value to proceed the sampling again to output a second renew RRI value. Enlarging the second renew RRI value of equality. And then proceeding to transfer the second renew RRI value to frequency spectrum signal. And calculating the he second renew RRI value for fast Fourier transform and power spectral density function. Using the calculating result to compare the numerical value of human body status for judging the driver status. It will help to warm the driver to avoid of car accident.

Description

Heartbeat detection method for detecting sleepy state

The invention relates to a sleep detection method, in particular to a heartbeat detection method using heart rate variability analysis.

At present, in the technical field of intelligent transportation systems, all countries have plans to increase the safe driving of driving, and whether driving safety comes from the driver's driving behavior, so how to detect whether the driver's driving behavior changes It has become the research goal of most current research projects.

In the current technology, the main method for detecting the driving behavior of the driver is to directly analyze the change of the driver's external behavior or the internal physiological signal change. The identification of the external behavior includes head movement and steering wheel. Movement trajectory or eye movement, etc., while the internal physiological signal change is to measure the change of heartbeat; however, there are quite a few technical obstacles to the identification of external behavior, because there is no certain change in the driver's external behavior. Judging the standard, it is still impossible to accurately judge whether the driver's driving behavior is dangerous. The system based on the human eye line of sight does not work after the driver wears the sunglasses, so the change of the external behavior is recognized to judge the driving. Action cannot be easily realized in real life.

The internal physiological signal changes are more directly responsive to the driver's driving state, especially the driver's fatigue state; when the driver has a sign of drowsiness, his heartbeat will The change is mainly based on the R-Wave Amplitude (RWA) as the source of the data. The RWA has obvious changes in the heart rate (RRI) during the transition period between waking, sleep and the two mental states. It is judged whether the driver has entered the initial threshold of the dozing state, so this method can be used as a detection method, and the warning design is used to remind the driver of his own safety state.

Although the aforementioned prior art is easier to alert the driver to the safety situation than the external behavior recognition, the RWA value may become wider as the physiological internal environment changes, causing the RWA to exceed the range of the originally set judgment state, and at the same time It is also affected by the measurement method and the RWA numerical anomaly is generated, so that the RWA value can not truly reflect the driver's actual situation, which becomes a drawback of this method.

In view of the above-mentioned shortcomings, the main object of the present invention is to provide a heartbeat detection method for detecting a doze state, which uses a non-intrusive method to measure an electrocardiogram signal, and performs signal pre-processing before the calculation program to exclude an abnormal signal. After that, the calculation results are used to determine whether the driver enters a doze state by using the standard deviation and the average value, which is helpful for early warning to prevent traffic accidents.

In order to achieve the above object, the present invention mainly provides a heartbeat detecting method for detecting a doze state, the steps of which include an external sensor to measure a driver's ECG signal, and the measured ECG signal is R. The wave value is calculated to generate the RRI value, and then the signal pre-processing procedure is performed, including removing the abnormal range of the calculated RRI value and generating a new RRI value, and then re-sampling the new RRI value, and then re-sampling the RRI value. Equal-rate amplification, the RRI value is converted into a spectrum signal, and finally, the RRI value is sequentially subjected to Fast Fourier Transform (FFT) and power spectral density (PSD). Calculated to produce the final judgment numerical result, the statistical result of the comparison of the human condition is compared with the designed conditional cross-match, and the driver is judged to judge whether the driver has a doze state.

The above and other objects, features and advantages of the present invention will become more <RTIgt;

The first figure is a measurement flow chart of the present invention.

The second figure is an electrocardiogram of the present invention.

The third figure is the RRI value correspondence table of the present invention.

The fourth figure is the RRI linear diagram of the present invention.

The fifth figure is a schematic diagram of the judgment condition 2 of the present invention.

Please refer to the first figure, which is a measurement flow chart of the present invention. As shown in the first figure, the heartbeat detection method of the present invention includes measuring an electrocardiography (ECG) (S1), and the method for measuring the ECG signal can pass through a non-invasive device, such as an external sensor. Directly measure the human body ECG signal; then calculate the R wave value of the measured ECG signal to generate the RRI value (S2); the R wave appearing in the ECG represents a heartbeat, and the RRI is the previous time. The time between the heartbeat and the last heartbeat, the Ratio data usually needs to accumulate the RRI data amount of five minutes to perform the FFT and PSD calculation; in this embodiment, the R wave numerical value uses the R wave numerical algorithm (So and Chan Method) performs the calculation and calculates the distance between each R wave, which is the RR interval value (RRI), as shown in the electrocardiogram of the second figure; the calculated RRI value is then subjected to the signal pre-processing procedure.

The pre-processing of the signal further includes removing the abnormal RRI value (S3). Since the heartbeat value of the normal person is between 60 and 100/min, in the case of actual measurement, the sensor in the external body is not attached. If the correct or measured person is in the active period, and the waveform generated by the ECG is abnormal or the ECG waveform signal is not measured, these factors will cause great error in the subsequent PSD calculation, so it must be measured. After the RRI value is removed, the abnormal RRI value is removed. The detailed method of removing the abnormal RRI value is to set five RRI average intervals first. As shown in the RRI value correspondence table of the third figure, in the embodiment, the electrocardiogram is The RRI average value is calculated as , where the value 255 refers to the detection point in the R wave. After calculating the original RRI average value, which range the RRI value is to be obtained belongs to, and then the abnormal RRI value is removed and the RRI value is complemented. Recalculate and generate new RRI values. For example, if the average RRI is calculated to be 164, then the segment selection shown in the third graph belongs to interval 3, and the RRI range obtained is between 219 and 139.

Then, the new RRI value is resampled (S4), and the obtained R wave is taken as the distance between adjacent R waves, and the sequence of each RRI value is the heart rate variability signal, and the original sampling is not equally spaced sampling. To perform the conversion, the original heart rate variability signal must be converted to the signal of the equidistant sampling. In this embodiment, the linear interpolation method (Linear Interpolation) is used, and the sampling frequency is 2 Hz. As shown in the RRI linear graph of the fourth graph, the calculation formula of the method is y=y1+(y2-y1)× , the procedure of re-sampling in this way, but not limited to this method; then the ratio of the RRI value after re-sampling is amplified (S5), in order to make the RRI value sequence have its correlation, to avoid the impact after the calculation The correctness; finally, the RRI value is converted into the spectrum signal (S6). Because the signal truncation is used for spectrum conversion, the Gibbs effect will be generated, causing severe distortion of the high and low frequency signals, which directly affects the subsequent measurement. As a result, in the present embodiment, the window function is used to perform spectrum signal conversion, such as the Hann Window function.

After the pre-signal processing procedure, the RRI values are sequentially subjected to Fast Fourier Transform (FFT) and power spectral density (PSD) calculation to generate the final LF (low frequency) / HF (high frequency). The power ratio value is a judgment result (S7), and after obtaining the above-described judgment value (S7) result, the conditional expression can be judged.

Judgment condition 1: Ratio=LF/HF_ratio _(Now)

Take the experimental data from the waking to the sleep of the experiment, and find out that under normal activities, the sympathetic/parasympathetic balance indicator (LF/HF-ratio(Average)) will be greater than 2; In the case of sleeping, most of the sympathetic/parasympathetic balance indicators (LF/HF-ratio (Average)) will be less than 0.6.

To this end, Condition 1 is a sympathetic/parasympathetic balance indicator (LF/HF- measured in the eight minutes of the test with the sympathetic/parasympathetic balance indicator (LF/HF-ratio (Average)). The average of ratio(Now) is compared.

In order to capture the trigger point of dozing in advance, the average calculation of the sympathetic/parasympathetic balance every ten minutes is called Mean. The calculation formula is as follows, updated every five minutes, and overlaps for five minutes, then use Mean. The condition 1 is divided into six levels, as shown in Table 1, according to which the mode level can be compared.

Where X is the sympathetic/parasympathetic balance, N is the number of LF/HF-ratio values, and X _i is the i-th group of data, with a total of N groups.

Judgment condition two: Diff=LF/HF_ratio(Now)-LF/HF_ratio(Average)

Condition 2 is to calculate the sympathetic/parasympathetic balance indicator (LF/HF-ratio(Now)) of the first three minutes of the test at the time of the test, and then subtracted from the sympathetic/parasympathetic balance indicator (LF/HF_ratio( Average)), as shown in the fifth picture.

When Diff is less than 0, it means that the balance indicator is decreasing; if Diff is greater than 0, it means that the balance indicator is rising. We use the standard deviation (SD) to divide the condition into five levels, as shown in Table 2.

SD( μ) is the average value of the data, N is the number of LF/HF-ratio values, X _i is the ith data of LF/HF_ratio(Now), and μ is LF/HF_ratio(Average).

The results analyzed by the conditions are called sleepiness indicators, as shown in Table 3.

Next, statistical analysis is performed for every 25 sleepiness indicators, and then updated every 90 seconds. The model index obtained by Condition 1 and Condition 2 is cross-compared in Table 3. The numerical judgment conditions are as follows:

a. Sleepiness indicator 1, judged to be awake.

b. Sleepiness indicator 3, judged to be severe sleepiness.

c. Others, judged to be slightly drowsy or drowsy.

According to this, the judged numerical result produced can be judged by the degree of sleepiness of the measured person, so as to promptly issue a warning to prevent the occurrence of a traffic accident.

However, the embodiments described above are preferred embodiments, and the scope of the invention is not limited thereto, and equivalent changes or modifications made in accordance with the scope of the invention and the contents of the specification should be The scope of patent coverage.

(S1)~(S7)

Claims (9)

A heartbeat detecting method for detecting a dozing state, the steps comprising: a. measuring an electrocardiogram signal; b. calculating the R wave value of the measured ECG signal to generate an RRI value; c. removing Abnormal RRI values and generate new RRI values; d. Resample new RRI values; e. Amplify RRI values after resampling; f. Perform fast Fourier transform and power spectral density calculations on RRI values to produce final LF (low frequency) / HF (high frequency) power ratio is the judgment value; g. measured by the sympathetic / parasympathetic balance indicator (LF / HF-ratio (Average)) and the testee's judgment value at the time of the test eight minutes The average value (LF/HF-ratio(Now)) is compared, and the judgment condition is a mode level; h. and the measured value calculated by the test subject in the first three minutes of the test (LF/HF- Ratio(Now)) is subtracted from the sympathetic/parasympathetic balance indicator (LF/HF-ratio(Average)), and the obtained values are divided into five grades, and the mode level of the judgment condition 2 is summarized accordingly; i. When the mode level of condition one is compared with the mode level of condition two, you can get 瞌Indicators. The heartbeat detecting method for detecting a doze state according to the first aspect of the patent application, wherein the electrocardiographic signal is measured by a non-invasive instrument in step a. The heartbeat detecting method for detecting a dozing state as described in claim 2, wherein the electrocardiographic signal is measured by an external sensor in step a. The heartbeat detecting method for detecting a doze state according to the first aspect of the patent application, wherein the R wave value in the step b is calculated by using a So and Chan Method. The heartbeat detecting method for detecting a doze state according to the first aspect of the patent application, wherein the d step is performed in an equidistant sampling manner. The heartbeat detecting method for detecting a doze state according to claim 5, wherein the equidistant frequency is 2 Hz. The heartbeat detecting method for detecting a dozing state as described in claim 5, wherein the linear interpolation method performs linear sampling. The heartbeat detecting method for detecting a doze state according to the first aspect of the patent application, wherein the f-step uses a window function to perform spectrum signal conversion. The heartbeat detection method for detecting a doze state as described in claim 8 of the patent application, wherein the window function is Hann Window function is optimal.