CN113040734A - Non-contact blood pressure estimation method based on signal screening - Google Patents
Non-contact blood pressure estimation method based on signal screening Download PDFInfo
- Publication number
- CN113040734A CN113040734A CN202110242252.5A CN202110242252A CN113040734A CN 113040734 A CN113040734 A CN 113040734A CN 202110242252 A CN202110242252 A CN 202110242252A CN 113040734 A CN113040734 A CN 113040734A
- Authority
- CN
- China
- Prior art keywords
- signal
- screening
- blood pressure
- rppg
- difference
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000012216 screening Methods 0.000 title claims abstract description 82
- 230000036772 blood pressure Effects 0.000 title claims abstract description 62
- 238000000034 method Methods 0.000 title claims abstract description 61
- 108010076504 Protein Sorting Signals Proteins 0.000 claims abstract description 33
- 238000004364 calculation method Methods 0.000 claims description 25
- 230000008569 process Effects 0.000 claims description 20
- 210000001061 forehead Anatomy 0.000 claims description 19
- 238000001228 spectrum Methods 0.000 claims description 18
- 238000005070 sampling Methods 0.000 claims description 11
- 238000001914 filtration Methods 0.000 claims description 4
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 claims description 4
- 230000009191 jumping Effects 0.000 claims description 4
- 239000010931 gold Substances 0.000 claims description 3
- 229910052737 gold Inorganic materials 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 2
- 238000001514 detection method Methods 0.000 abstract description 11
- 238000000605 extraction Methods 0.000 abstract description 9
- 238000005259 measurement Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000035487 diastolic blood pressure Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000033001 locomotion Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000000926 separation method Methods 0.000 description 3
- 230000035488 systolic blood pressure Effects 0.000 description 3
- 206010020772 Hypertension Diseases 0.000 description 2
- 210000004204 blood vessel Anatomy 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 239000011475 Accrington brick Substances 0.000 description 1
- 208000024172 Cardiovascular disease Diseases 0.000 description 1
- 208000012661 Dyskinesia Diseases 0.000 description 1
- 208000015592 Involuntary movements Diseases 0.000 description 1
- 206010028347 Muscle twitching Diseases 0.000 description 1
- 208000003443 Unconsciousness Diseases 0.000 description 1
- 238000009530 blood pressure measurement Methods 0.000 description 1
- 208000026106 cerebrovascular disease Diseases 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000002526 effect on cardiovascular system Effects 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 210000004373 mandible Anatomy 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000017311 musculoskeletal movement, spinal reflex action Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 230000008733 trauma Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/021—Measuring pressure in heart or blood vessels
Abstract
The invention relates to a non-contact blood pressure estimation method based on signal screening, and belongs to the field of blood pressure detection. The method comprises the steps of synchronously shooting the face and palm interest areas of a user through a camera, and carrying out signal processing on video streams in the interest areas to obtain a two-channel rPPG signal sequence from the face and the palm. And synchronously carrying out window-by-window processing on the signal sequence by using a sliding window so as to extract a signal segment for screening. And constructing a screening layer comprising five dimensions of frequency comparison, confidence ratio, signal-to-noise ratio, peak height difference and pulse width difference to screen the candidate signal sections so as to screen out the signal sections with reliable signal quality. And extracting time difference combinations into time difference sequences aiming at all reliable signal segments, and extracting final time lags from the time difference sequences. And finally, constructing a regression model of time lag and calibration blood pressure for non-contact estimation of the blood pressure. The method improves the accuracy and robustness of time lag extraction based on the dual-channel rPPG signal sequence, and has important significance for building a high-performance and high-reliability blood pressure estimation model in the future.
Description
Technical Field
The invention belongs to the technical field of blood pressure detection, and relates to a blood pressure non-contact estimation method based on signal screening.
Background
The blood pressure is closely related to the health of human body, is an important index for evaluating the functional status of heart and blood vessel of human body, and is an important basis for evaluating the health condition of human body. Studies have shown that hypertension is becoming a major cause of cardiovascular and cerebrovascular disease, known as a silent killer of death, with approximately 13% of people dying from hypertension each year. The traditional blood pressure detection mode needs a user to wear a cuff before detection, the comfort of the user can be reduced by inflation and pressurization operation in the use process, meanwhile, the blood pressure is prone to fluctuation caused by various stimuli such as physiological cycles, outside and inside, and the change condition of the blood pressure of a person in a period of time cannot be accurately reflected by the result of single measurement, so that the non-contact continuous detection of the blood pressure is gradually a hot point of research at present.
The current blood pressure measuring methods can be classified into invasive and non-invasive. The invasive measurement mode is that a catheter needs to be punctured into an arterial blood vessel of a human body during measurement, a sensor connected with the catheter reads a current blood pressure value, the measurement mode can continuously monitor the blood pressure change of the human body, the blood pressure change is regarded as a gold standard of blood pressure detection by the medical field, and the invasive measurement mode is usually used in Intensive Care Unit (ICU) and other serious diseases monitoring places. However, this measurement method is not suitable for daily blood pressure measurement because it causes trauma to the skin tissue of the human body, and requires a complicated operation and professional medical staff for measurement. The non-invasive blood pressure detection method mainly comprises the following steps: auscultatory method, oscillometric method, arterial tension method, volume compensation method, and plethysmography. The detection methods are contact type, devices such as cuffs or sensors need to be arranged on corresponding parts of a human body when the method is used, and the method has poor flexibility and high load when a user uses the method.
At present, non-contact blood pressure detection technology is becoming a hot point of research. The comparison is representative of a blood pressure estimation method based on rPPG, which is a pulse wave volume change signal. Unlike plethysmography, rPPG is captured in a non-contact manner, specifically, a two-channel rPPG signal at the proximal end (usually the face) and the distal end (usually the palm) is acquired by a camera, and a regression model between the time lag (the time required for pulse waves to be transmitted from the proximal end to the distal end) of the two-channel rPPG signal and the calibrated blood pressure is further established for blood pressure estimation. In this method, the computational accuracy of the time lag affects the final model performance, and the computational accuracy of the time lag is closely related to the signal quality of rPPG.
However, in an actual detection scenario, despite various signal processing methods, the rPPG extracted in the existing rPPG-based blood pressure estimation method is still susceptible to noise influence to cause signal overall or local distortion, which affects time-lag extraction accuracy and further affects final blood pressure estimation performance.
Disclosure of Invention
Technical problem to be solved
The invention provides a non-contact blood pressure estimation method based on signal screening, aiming at the defects that in an actual scene, the extracted rPPG in the existing non-contact blood pressure estimation method based on the rPPG is still easily influenced by noise to cause signal integral or local distortion, influence time-lag extraction accuracy and further influence final blood pressure estimation performance. A non-contact time-lag extraction method is improved by adopting a segmented signal screening mechanism, the association degree of time lag and calibration blood pressure and extraction robustness are improved, a high-performance blood pressure estimation model is conveniently constructed, and the blood pressure estimation effect is improved.
Technical scheme
A non-contact blood pressure estimation method based on signal screening is characterized by comprising the following steps:
step 1: synchronously shooting the face and the palm of a human body corresponding to an interest region by using a camera to obtain a video stream;
step 2: separating a video stream into image frames frame by frame, respectively calculating a gray average value of three channels of the image frames, further removing nonlinear trend, normalizing, blind source separating and band-pass filtering, extracting a component with the maximum correlation degree with the heart beat as rPPG, and combining the rPPG signals respectively derived from a forehead interest region and a palm interest region acquired by the method into a two-channel signal sequence;
and step 3: dividing the two-channel signal sequence into signal segments by using sliding window sub-windows, and constructing a five-dimensional screening layer containing frequency comparison, a confidence ratio, a signal-to-noise ratio, a peak height difference and a pulse width difference so as to screen a high-reliability signal segment;
and 4, step 4: calculating the phase difference of the two-channel signals in the high-reliability signal segment, further converting the phase difference into a time difference, and combining all the time differences obtained by the high-reliability signal segment into a time difference sequence;
and 5: and calculating the mode of the time difference sequence as an actual time lag, and establishing a blood pressure estimation model for estimating the blood pressure by using the time lag and the calibration blood pressure.
The technical scheme of the invention is further that: in step 1, a video camera with a sampling rate of 30 frames per second and a resolution of 720P is selected to respectively extract video streams of two interest areas, namely a forehead area and a palm area, wherein the forehead interest area is set to be a rectangle with the length of 180 pixels and the width of 90 pixels, and the palm interest area is set to be a rectangle with the length of 90 pixels and the width of 90 pixels.
The technical scheme of the invention is further that: in step 3, a sliding window with a window size of 120 sampling points is synchronously used for the two-channel rPPG signal sequence, the sliding window with a sliding step length of 60 sampling points divides the signal sequence into signal segments, and each signal segment comprises a forehead interest region rPPG and a palm interest region rPPG; constructing a screening layer containing five dimensions of frequency comparison, confidence ratio, signal-to-noise ratio, peak height difference and pulse width difference, screening the signal sections to screen out high-reliability signal sections, wherein the screening sequence in the screening layer sequentially comprises the frequency comparison, the confidence ratio, the signal-to-noise ratio, the peak height difference and the pulse width difference, two-channel signals in the signal sections all need to sequentially meet the five screening dimensions, and otherwise, skipping out of the current screening process to perform screening of the next signal section;
the frequency comparison refers to that Fourier transformation is respectively carried out on signals in the signal section to obtain the frequency corresponding to the maximum value in the power spectrum, if the two-channel rPPG frequencies are the same, confidence ratio screening is carried out, otherwise, the current screening process is skipped to execute next signal section screening;
the calculation formula of the confidence ratio is as follows:
wherein P isfaExpressed as a power value, P, corresponding to a dominant frequency in the power spectrumfbExpressed as the power value corresponding to the secondary dominant frequency in the power spectrum, if QCR>10, carrying out signal-to-noise ratio screening, otherwise jumping out of the current screening process and executing next signal segment screening;
the calculation formula of the signal-to-noise ratio is as follows:
wherein P isfFor the corresponding spectral power at frequency f, Ω is the frequency range, corresponding to the human heart rate set in this example to [0.7Hz,3Hz]；faD is a parameter for determining a frequency range, wherein d is set to be 0.1Hz and corresponds to 6 heartbeats per minute; if QSNR>10, screening the peak height difference, otherwise, skipping the current screening process and executing the next signal segment screening;
the calculation formula of the peak height difference is as follows:
wherein h isiIs the peak value of the ith pulse wave crest in the signal sequence in the signal segment,
the average value of all pulse wave peak values in the signal sequence in the signal section is obtained; if PHV<0.1, evaluating the pulse width difference, otherwise jumping out of the current screening process and executing next signal segment screening;
the calculation formula of the pulse width difference is as follows:
wherein wiIs the ith pulse width in the signal sequence, the pulse width is the number of sampling points spaced between adjacent wave troughs of the signal sequence,
is the mean value of the pulse width; if PWV<And 2, performing the step 4, otherwise, skipping the current screening process to perform the next signal segment screening.
The technical scheme of the invention is further that: in step 4, for all reliable signal segments meeting the screening condition, phase spectra are used for respectively countingCalculating the phase difference between the initial phase of the frontal region rPPG and the initial phase of the palmar region rPPG in the segment, and using
Converting the phase difference into a time difference, wherein
The phase difference between rPPG of the forehead interest region and rPPG of the palm interest region in the signal section is f, the frequency of the rPPG is the frequency of the rPPG, and the frequency corresponds to the heart rate of the human body, namely the dominant frequency corresponding to the maximum power value in the power spectrum; all time differences extracted from the reliable signal segments are then combined into a time difference sequence.
The technical scheme of the invention is further that: step 5, calculating a mode of the time difference sequence, and taking the mode as an actual time lag of the pulse wave in the video stream, wherein the pulse wave is transmitted from the face to the palm; in order to facilitate the calculation of the mode, the time difference sequences are firstly sorted from big to small, and then the mode of the time difference sequences is calculated by using a gold interpolation method, wherein the calculation formula is as follows:
wherein M isoIs a mode, L is a lower bound of a time difference sequence, faFrequency number adjacent to the lower limit of the time difference sequence, fbAnd i is the frequency number adjacent to the upper limit of the time difference sequence, i is the group distance, and the group distance is calculated by the difference between the upper limit and the lower limit of the time difference sequence.
Advantageous effects
The invention provides a non-contact blood pressure estimation method based on signal screening, which divides a dual-channel rPPG signal sequence into signal segments, screens a high-reliability signal segment by using a five-dimensional signal screening layer containing frequency comparison, a confidence ratio, a signal-to-noise ratio, a peak height difference and a pulse width difference, and further calculates a mode of a time difference sequence extracted from the signal segment as a final time lag. The method improves the accuracy and robustness of the time lag extracted in a non-contact manner based on the dual-channel rPPG under the noise interference condition, improves the correlation degree between the time lag and the calibrated blood pressure under the noise interference condition, provides a solution for robustly and accurately extracting the time lag in the non-contact blood pressure estimation method based on the rPPG, and has important significance for constructing a high-performance and high-reliability blood pressure estimation model in the future.
Drawings
FIG. 1 is a schematic flow chart of a non-contact blood pressure estimation method based on signal screening according to the present invention.
Fig. 2 is a schematic diagram of acquiring a video stream by synchronously shooting interest areas corresponding to the face and the palm of a human body by using a camera in the embodiment of the invention.
Fig. 3 is a schematic diagram of a selected location of a region of interest according to an embodiment of the present invention.
1 is the selected forehead region of interest, which is a rectangle 180 pixels long and 90 pixels wide. 2 is the selected palm region of interest, which is a rectangle 90 pixels long and 90 pixels wide.
Fig. 4 is a schematic diagram of the combination of rPPG derived from the forehead region of interest and the palm region of interest into a two-channel signal sequence according to an embodiment of the present invention.
Detailed Description
The invention will now be further described with reference to the following examples and drawings:
the invention provides a non-contact blood pressure estimation method based on signal screening, which improves a non-contact time-lag extraction method by adopting a segmented signal screening mechanism, improves the association degree of time lag and calibration blood pressure and extraction robustness, is convenient to construct a high-performance blood pressure estimation model, and improves the blood pressure estimation effect.
As shown in fig. 1, a non-contact blood pressure estimation method based on signal screening includes the following steps:
(1) and synchronously shooting the face and the palm of the human body corresponding to the interest region by using a camera to obtain a video stream.
In the embodiment of the invention, a signal acquisition schematic diagram is shown in fig. 1, a camera with a sampling rate of 30 frames per second and a resolution of 720P is placed on a 30cm support and placed on a table, a participant sits on an automatically adjustable chair, and the sight line is kept parallel to the camera through the adjustable chair. Keep the camera apart from participant face 1m, the participant lifts up the right hand and keeps right hand palm and human face mandible level simultaneously, and in the acquisition process, eyes need watch the camera and keep palm and face to be in same vertical plane as far as possible, guarantee face and palm simultaneous occurrence in the video picture. The calculation amount of extracting the rPPG directly from the human face and the hand is large, and the interference of the human blink, the facial twitching and other involuntary movements is easy to occur. Therefore, in this example, two regions of interest are selected, one is a rectangular region located at the forehead and having a length and width set to 180 pixels and 90 pixels, respectively, and the other is a square region located at the palm of the hand and having a length and width set to 90 pixels and 90 pixels, respectively. In addition, as the human face and the palm can not keep static in an absolute sense, namely, unconscious movement exists, in order to reduce the influence of motion artifacts on the signal quality, the invention marks the interest region on the first frame image in the video stream, and tracks the two interest regions by using a KLT tracking algorithm. In this example, 40s of video streams are continuously acquired simultaneously for two regions of interest, respectively.
(2) The method comprises the steps of separating a video stream into image frames frame by frame, then respectively calculating a gray average value of three channels of the image frames, further removing nonlinear trend, normalizing, blind source separation and band-pass filtering, extracting a component with the maximum correlation degree with the heart beat as rPPG, and combining the rPPG which is obtained by the method and respectively comes from a forehead interest region and a palm interest region into a two-channel signal sequence.
The video stream is separated into successive image frames, and the variation of the mean gray value of three channels in the successive image frames R, G, B in each interest region over time contains rPPG related to heart beat and noise information, which is usually regarded as a mixed signal and is calculated by the following formula:
wherein W represents the width of the region of interest, L represents the length of the region of interest, IijIndicating the gray value at the abscissa i and the ordinate j in the region of interest. To avoid interference with subsequent calculations and to facilitate calculations, use is made ofCubic spline interpolation removes the trend term in the signal, and then normalizes the data using Z-Score, which can be expressed as:
wherein XnorIs a normalized value, XiIs the ith value in the signal sequence, μ is the mean of the signal sequence, and σ is the standard deviation of the signal sequence. To this end, the R, G, B three-channel mixed signal still contains a lot of noise, so this example uses ICA to perform blind source separation on the signal, and inputs the signal combination matrix obtained by normalizing R, G, B three channels in each interest region into ICA to solve the source signal component reversely. Blind source separation cannot determine which component is the desired rPPG, which is a heart beat related signal, so embodiments of the present invention use butterworth bandpass filters to bandpass filter the three channel components separately to remove frequency components that are not related to heart beat. Researches show that the heart rate range of a normal person is 50-100 times/minute, and in order to filter noise components in input signals to the maximum extent, a Butterworth filter with the order of 10 and the passband frequency of 0.89-3 Hz is used for filtering three-channel components respectively. And performing fast Fourier transform on the filtered components to obtain power spectrums of the components, and selecting the component with the highest energy spectrum peak value to be characterized as the rPPG of the corresponding interest region. According to the foregoing method, the present example can acquire rPPG derived from the forehead region of interest and the palm region of interest, respectively, and combine them into a two-channel signal sequence.
(3) And dividing the two-channel signal sequence into signal segments by using sliding window division windows, and constructing a five-dimensional screening layer containing frequency comparison, a confidence ratio, a signal-to-noise ratio, a peak height difference and a pulse width difference so as to screen the high-reliability signal segments.
Ambient noise and motion disturbances can distort the whole or local area of rPPG causing difficulty in time lag extraction and thus affect blood pressure detection. In the embodiment, a sliding window is used for dividing signal segments of the dual-channel rPPG, and reliable signal segments which can be used for follow-up time lag extraction in a signal sequence are screened. In this example, the sliding window size is set to 120 samples, the sampling rate of the camera is 30 frames per second, so that at least 4 seconds of rPPG signal are contained in the window, and the sliding step is set to 60 samples. This example proposes that the screening layer based on five dimensions carries out the reliability screening to the signal section, and the screening dimension specifically includes: frequency comparison, confidence ratio, signal-to-noise ratio, peak height difference and pulse width difference, wherein the two-channel signals in the signal section all need to sequentially satisfy the five screening dimensions, otherwise, the current screening process is skipped to execute the screening of the next signal section.
And the frequency comparison refers to that the signals in the signal section are respectively subjected to Fourier transform to obtain the frequency corresponding to the maximum value in the power spectrum, if the two-channel rPPG frequencies are the same, the confidence ratio screening is carried out, otherwise, the current screening process is skipped to execute the next signal section screening.
The calculation formula of the confidence ratio is as follows:
wherein P isfaExpressed as a power value, P, corresponding to a dominant frequency in the power spectrumfbExpressed as the power value corresponding to the secondary dominant frequency in the power spectrum, if QCR>And 10, carrying out signal-to-noise ratio screening, otherwise, skipping the current screening process and carrying out next signal segment screening.
The calculation formula of the signal-to-noise ratio is as follows:
wherein P isfFor the corresponding spectral power at frequency f, Ω is the frequency range, corresponding to the human heart rate set in this example to [0.7Hz,3Hz]。faD is a parameter determining the frequency range for the frequency corresponding to the maximum power value in the power spectrum, where d is set to 0.1Hz corresponding to 6 beats per minute. If QSNR>And 10, screening the peak height difference, otherwise, skipping the current screening process and executing the next signal segment screening.
The calculation formula of the peak height difference is as follows:
wherein h isiIs the peak value of the ith pulse wave crest in the signal sequence in the signal segment,
the average value of all pulse wave peak values in the signal sequence in the signal segment is obtained. If PHV<And 0.1, evaluating the pulse width difference, and otherwise, skipping the current screening process to carry out next signal segment screening.
The calculation formula of the pulse width difference is as follows:
wherein wiIs the ith pulse width in the signal sequence, the pulse width is the number of sampling points spaced between adjacent wave troughs of the signal sequence,
is the average of the pulse widths. If PWV<And 2, performing the step (4), otherwise, skipping the current screening process to perform next signal segment screening.
(4) And calculating the phase difference of the two-channel signals in the high-reliability signal segment, then further converting the phase difference into a time difference, and combining all the time differences obtained by the high-reliability signal segment into a time difference sequence.
In the existing method, the time difference between two-channel signal sequences is calculated by comparing key characteristic points of two-channel signals on a time domain level, however, when noise interference exists, the characteristic points are easy to shift, so that the accuracy of time difference calculation is influenced. In this example, the phase difference between the two-channel signals in the high-reliability signal segment is calculated by solving the phase spectrum and converted into the time difference, so as to reduce the influence of noise interference on the time difference calculation. The calculation formula for converting the phase difference into the time difference is as follows:
wherein the content of the first and second substances,
is the phase difference between rPPG of the forehead interest region and rPPG of the palm interest region in the signal segment, and f is the frequency of rPPG, which in this example corresponds to the human heart rate, i.e. the dominant frequency corresponding to the maximum power value in the power spectrum. And calculating the time difference between the signals in each high-reliability signal segment, and combining the time differences into a time difference sequence.
(5) And calculating the mode of the time difference sequence as an actual time lag, and establishing a blood pressure estimation model for estimating the blood pressure by using the time lag and the calibration blood pressure.
The small time difference obtained in step (4) may still be interfered by noise to cause deviation from the actual time lag. To eliminate the interference factor, the present example finds the mode of the time difference series as the actual skew. In order to facilitate the calculation of the mode, the time difference sequences are firstly sorted from big to small, and then the mode of the time difference sequences is calculated by using a gold interpolation method, wherein the calculation formula is as follows:
wherein M isoIs a mode, L is a lower bound of a time difference sequence, faFrequency number adjacent to the lower limit of the time difference sequence, fbAnd i is the frequency number adjacent to the upper limit of the time difference sequence, i is the group distance, and the group distance is calculated by the difference between the upper limit and the lower limit of the time difference sequence.
30 groups of video streams of the forehead interest region and the palm interest region are acquired for an individual, the time lag of the forehead interest region and the palm interest region is extracted, each group is 40 seconds, and the diastolic pressure and the systolic pressure are synchronously acquired by using an ohm dragon sphygmomanometer to serve as calibrated blood pressures. A blood pressure estimation model was constructed as follows:
a, B are specificity parameters related to individuals respectively, PTT is the final time lag of a single group of video stream, BP is the calibrated blood pressure and is divided into two diastolic pressure and systolic pressure together, the diastolic pressure and the systolic pressure are substituted into the formula simultaneously with the time lag respectively, and a least square method is used for fitting individual parameters A, B to obtain a final blood pressure estimation model so as to realize blood pressure estimation.
Claims (5)
1. A non-contact blood pressure estimation method based on signal screening is characterized by comprising the following steps:
step 1: synchronously shooting the face and the palm of a human body corresponding to an interest region by using a camera to obtain a video stream;
step 2: separating a video stream into image frames frame by frame, respectively calculating a gray average value of three channels of the image frames, further removing nonlinear trend, normalizing, blind source separating and band-pass filtering, extracting a component with the maximum correlation degree with the heart beat as rPPG, and combining the rPPG signals respectively derived from a forehead interest region and a palm interest region acquired by the method into a two-channel signal sequence;
and step 3: dividing the two-channel signal sequence into signal segments by using sliding window sub-windows, and constructing a five-dimensional screening layer containing frequency comparison, a confidence ratio, a signal-to-noise ratio, a peak height difference and a pulse width difference so as to screen a high-reliability signal segment;
and 4, step 4: calculating the phase difference of the two-channel signals in the high-reliability signal segment, further converting the phase difference into a time difference, and combining all the time differences obtained by the high-reliability signal segment into a time difference sequence;
and 5: and calculating the mode of the time difference sequence as an actual time lag, and establishing a blood pressure estimation model for estimating the blood pressure by using the time lag and the calibration blood pressure.
2. The signal-screening-based non-contact blood pressure estimation method according to claim 1, wherein in step 1, a video camera with a sampling rate of 30 frames per second and a resolution of 720P is selected to extract video streams of two interest areas, namely a forehead area and a palm area, wherein the forehead interest area is set to be a rectangle with a length of 180 pixels and a width of 90 pixels, and the palm interest area is set to be a rectangle with a length of 90 pixels and a width of 90 pixels.
3. The signal-screening-based non-contact blood pressure estimation method according to claim 1, wherein in step 3, a sliding window with a window size of 120 sampling points and a sliding step size of 60 sampling points is used synchronously for the two-channel rPPG signal sequence to divide the signal sequence into signal segments, and each signal segment includes rPPG of a forehead interest region and rPPG of a palm interest region; constructing a screening layer containing five dimensions of frequency comparison, confidence ratio, signal-to-noise ratio, peak height difference and pulse width difference, screening the signal sections to screen out high-reliability signal sections, wherein the screening sequence in the screening layer sequentially comprises the frequency comparison, the confidence ratio, the signal-to-noise ratio, the peak height difference and the pulse width difference, two-channel signals in the signal sections all need to sequentially meet the five screening dimensions, and otherwise, skipping out of the current screening process to perform screening of the next signal section;
the frequency comparison refers to that Fourier transformation is respectively carried out on signals in the signal section to obtain the frequency corresponding to the maximum value in the power spectrum, if the two-channel rPPG frequencies are the same, confidence ratio screening is carried out, otherwise, the current screening process is skipped to execute next signal section screening;
the calculation formula of the confidence ratio is as follows:
wherein P isfaExpressed as a power value, P, corresponding to a dominant frequency in the power spectrumfbExpressed as the power value corresponding to the secondary dominant frequency in the power spectrum, if QCR>10, carrying out signal-to-noise ratio screening, otherwise jumping out of the current screening process and executing next signal segment screening;
the calculation formula of the signal-to-noise ratio is as follows:
wherein P isfFor the corresponding spectral power at frequency f, Ω is the frequency range, corresponding to the human heart rate set in this example to [0.7Hz,3Hz]；faD is a parameter for determining a frequency range, wherein d is set to be 0.1Hz and corresponds to 6 heartbeats per minute; if QSNR>10, screening the peak height difference, otherwise, skipping the current screening process and executing the next signal segment screening;
the calculation formula of the peak height difference is as follows:
wherein h isiThe peak value of the ith pulse wave peak in the signal sequence in the signal segment is hh, and the average value of all the pulse wave peak values in the signal sequence in the signal segment is hh; if PHV<0.1, evaluating the pulse width difference, otherwise jumping out of the current screening process and executing next signal segment screening;
the calculation formula of the pulse width difference is as follows:
wherein wiIs the ith pulse width in the signal sequence, the pulse width is the number of sampling points spaced between adjacent wave troughs of the signal sequence,
is the mean value of the pulse width; if PWV<And 2, performing the step 4, otherwise, skipping the current screening process to perform the next signal segment screening.
4. The signal-screening-based non-contact blood pressure estimation method according to claim 1, wherein in step 4, all the results are satisfiedScreening reliable signal segments of conditions, calculating phase difference between initial phase of forehead region rPPG and initial phase of palm region rPPG respectively by using phase spectrum
Converting the phase difference into a time difference, wherein
The phase difference between rPPG of the forehead interest region and rPPG of the palm interest region in the signal section is f, the frequency of the rPPG is the frequency of the rPPG, and the frequency corresponds to the heart rate of the human body, namely the dominant frequency corresponding to the maximum power value in the power spectrum; all time differences extracted from the reliable signal segments are then combined into a time difference sequence.
5. The method of claim 1, wherein in step 5, the time difference sequence is subjected to mode calculation to obtain the actual time lag of the pulse wave in the video stream from the face to the palm; in order to facilitate the calculation of the mode, the time difference sequences are firstly sorted from big to small, and then the mode of the time difference sequences is calculated by using a gold interpolation method, wherein the calculation formula is as follows:
wherein M isoIs a mode, L is a lower bound of a time difference sequence, faFrequency number adjacent to the lower limit of the time difference sequence, fbAnd i is the frequency number adjacent to the upper limit of the time difference sequence, i is the group distance, and the group distance is calculated by the difference between the upper limit and the lower limit of the time difference sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110242252.5A CN113040734B (en) | 2021-03-04 | Non-contact blood pressure estimation method based on signal screening |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110242252.5A CN113040734B (en) | 2021-03-04 | Non-contact blood pressure estimation method based on signal screening |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113040734A true CN113040734A (en) | 2021-06-29 |
| CN113040734B CN113040734B (en) | 2024-05-03 |
Family
ID=
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023193711A1 (en) * | 2022-04-07 | 2023-10-12 | Faceheart Corporation | Contactless physiological measurement device and method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103908236A (en) * | 2013-05-13 | 2014-07-09 | 天津点康科技有限公司 | Automatic blood pressure measuring system |
| CN105748053A (en) * | 2016-04-21 | 2016-07-13 | 华为技术有限公司 | Terminal and blood pressure measuring method |
| CN106236049A (en) * | 2016-10-12 | 2016-12-21 | 南京工程学院 | Blood pressure measuring method based on video image |
| CN108272449A (en) * | 2018-01-30 | 2018-07-13 | 中国科学院电子学研究所 | Contactless monitoring of blood pressure method and system |
| CN110090010A (en) * | 2019-06-17 | 2019-08-06 | 北京心数矩阵科技有限公司 | A kind of contactless blood pressure measuring method and system |
| KR20200064744A (en) * | 2018-11-29 | 2020-06-08 | 한국전자통신연구원 | Non contact blood pressure measuring method and apparatus using video |
| CN111611888A (en) * | 2020-05-12 | 2020-09-01 | 西安爱特眼动信息科技有限公司 | Non-contact blood pressure estimation method and device |
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103908236A (en) * | 2013-05-13 | 2014-07-09 | 天津点康科技有限公司 | Automatic blood pressure measuring system |
| CN105748053A (en) * | 2016-04-21 | 2016-07-13 | 华为技术有限公司 | Terminal and blood pressure measuring method |
| CN106236049A (en) * | 2016-10-12 | 2016-12-21 | 南京工程学院 | Blood pressure measuring method based on video image |
| CN108272449A (en) * | 2018-01-30 | 2018-07-13 | 中国科学院电子学研究所 | Contactless monitoring of blood pressure method and system |
| KR20200064744A (en) * | 2018-11-29 | 2020-06-08 | 한국전자통신연구원 | Non contact blood pressure measuring method and apparatus using video |
| CN110090010A (en) * | 2019-06-17 | 2019-08-06 | 北京心数矩阵科技有限公司 | A kind of contactless blood pressure measuring method and system |
| CN111611888A (en) * | 2020-05-12 | 2020-09-01 | 西安爱特眼动信息科技有限公司 | Non-contact blood pressure estimation method and device |
Non-Patent Citations (1)
| Title |
|---|
| 冯军;汤文明;曹剑剑;余瑛;: "非接触式心率测量研究初步", 中国生物医学工程学报, no. 05, 20 October 2017 (2017-10-20) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023193711A1 (en) * | 2022-04-07 | 2023-10-12 | Faceheart Corporation | Contactless physiological measurement device and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| McDuff et al. | Improvements in remote cardiopulmonary measurement using a five band digital camera | |
| Bousefsaf et al. | Continuous wavelet filtering on webcam photoplethysmographic signals to remotely assess the instantaneous heart rate | |
| Villarroel et al. | Non-contact vital sign monitoring in the clinic | |
| Blöcher et al. | An online PPGI approach for camera based heart rate monitoring using beat-to-beat detection | |
| CN111345803B (en) | Heart rate variability measuring method based on mobile device camera | |
| CN111387959A (en) | Non-contact physiological parameter detection method based on IPPG | |
| US10694959B2 (en) | Image based blood pressure monitoring method | |
| CN110236511A (en) | A kind of noninvasive method for measuring heart rate based on video | |
| Huang et al. | A motion-robust contactless photoplethysmography using chrominance and adaptive filtering | |
| KR20200079691A (en) | Camera-based contactless bio-signal measurement apparatus and operating method thereof | |
| Guede-Fernández et al. | A photoplethysmography smartphone-based method for heart rate variability assessment: Device model and breathing influences | |
| CN112294282A (en) | Self-calibration method of emotion detection device based on RPPG | |
| US20230005295A1 (en) | Non-contact facial blood pressure measurement method based on 3d cnn | |
| Li et al. | An improvement for video-based heart rate variability measurement | |
| CN113040734B (en) | Non-contact blood pressure estimation method based on signal screening | |
| Wittenberg et al. | Evaluation of HRV estimation algorithms from PPG data using neural networks | |
| CN113040734A (en) | Non-contact blood pressure estimation method based on signal screening | |
| Sharma et al. | Non-contact measurement of human heart rate using low cost video camera | |
| CN114331998A (en) | Non-contact cardiopulmonary coupling evaluation method | |
| Motin et al. | Robust heart rate estimation during physical exercise using photoplethysmographic signals | |
| Maji et al. | Contactless heart rate monitoring using A standard RGB camera | |
| Molinaro et al. | Heart Rate Monitoring With Smartphone Built-In Frontal Digital Camera | |
| Ozawa et al. | Improving the accuracy of noncontact blood pressure sensing using near-infrared light | |
| Chen et al. | A new approach to HR monitoring using photoplethysmographic signals during intensive physical exercise | |
| KR20200140469A (en) | Method for measuringing skin and health indicators of an user using a video image and apparatus using the same |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |