A PPG signal acquisition method and system based on control disturbance response fingerprint

Abstract

The present application relates to the technical field of biomedical signal processing, and particularly relates to a PPG signal acquisition method and system based on control disturbance response fingerprint, wherein the method comprises the following steps: collecting image sequences and constructing multiple space-time observation units; applying control disturbance to illumination intensity, exposure time or gain to obtain a control disturbance sequence; identifying the response fingerprint of each observation branch to the control disturbance based on the control disturbance sequence and the observation sequence to form a response matrix; performing control disturbance component suppression, physiological frequency band filtering, non-uniform time reconstruction of exposure integral forward consistency and virtual differential signal recombination based on the response matrix to obtain a target pulse wave signal; generating control prediction residual based on the response matrix to perform hierarchical quality judgment, and outputting the target pulse wave signal or re-measuring after adjusting parameters according to the judgment result. The present application can effectively suppress the influence of camera control disturbance and non-uniform exposure on the PPG signal, and improve the waveform reliability and time credibility.

Description

Technical Field

[0001] This invention relates to the field of biomedical signal processing technology, specifically to a method and system for acquiring PPG signals based on control perturbation response fingerprints. Background Technology

[0002] Photoplethysmography (PPG) obtains pulse waveforms by detecting changes in optical reflection or transmission in skin tissue during the cardiac cycle. With the development of mobile phone cameras and mobile computing capabilities, it is possible to use mobile phone cameras to capture image sequences of fingers or other skin areas and extract PPG signals from changes in color or brightness over time.

[0003] The illumination intensity, exposure time, analog or digital gain, rolling shutter readout, and image signal processing of a mobile phone camera can all alter the observed brightness. Finger pressure, contact area, local motion, and tissue optical pathways can also produce changes that overlap with the frequency bands of the pulse. Different spatial locations, line bands, color channels, pixel groups, or exposure branches respond differently to these changes. Using only frame averages, passive image quality indicators, or fixed filter parameters makes it difficult to determine whether the observed changes are caused by physiological pulse or by a combination of camera control and tissue optical pathway responses.

[0004] Existing non-uniform sampling processing typically approximates discrete observations as instantaneous sampling points, while camera pixels integrate the incident light within the exposure window. When there are long gaps, rolling shutter row-level sampling, or different exposure branches coexisting, if interpolation or resampling is performed directly without considering the exposure integration relationship, peaks and valleys that are not present in the original observation may be introduced, and the peak position, pulse interval, or other temporal characteristics may be altered.

[0005] Using only a single spectral signal-to-noise ratio or constructing and evaluating templates using the same batch of heartbeats may misinterpret high similarity caused by control response residues, cross-branch common artifacts, or template self-containment as physiological evidence. Directly synthesizing multiple quality indicators into a total quality score may also mask hard failures such as unidentifiable response matrices, long gaps, or excessive time uncertainty by other high scores.

[0006] Therefore, there is an urgent need for a mobile phone camera PPG signal acquisition scheme that can transform the composite response of control input to multiple observation branches into an online identifiable object in the same measurement, and enable the online identification result to pass through control-related component suppression, signal reconstruction and quality judgment, while combining the actual sampling time and exposure integration window to reconstruct the physically consistent trajectory. Summary of the Invention

[0007] To address the technical challenges in existing technologies, such as the difficulty in distinguishing between physiological changes and complex observational changes caused by camera control disturbances, tissue optical pathways, and contact state changes in passive image quality, the difficulty in ensuring consistency between continuous trajectory and exposure integration physical processes through ordinary interpolation, and the difficulty in reliably gating acquisition, identification, reconstruction, waveform, and temporal features using a single quality score, this invention provides a PPG signal acquisition method and system based on control disturbance response fingerprints.

[0008] In a first aspect, the present invention provides a method for acquiring PPG signals based on control perturbation response fingerprints, specifically including the following steps: During the stable pre-acquisition phase, image sequences and camera control parameters are acquired, the region of interest in the image is divided into multiple spatiotemporal observation units, and the observation sequences of multiple observation branches are obtained. Based on frame time information, exposure information and the position of the spatiotemporal observation units in the image, the actual sampling time, exposure integration window and quality information of each observation sample are determined. During the identification phase of the same measurement, control perturbations are applied to the illumination intensity, exposure time, or gain to obtain a control perturbation sequence; Based on the control disturbance sequence and the observation sequence, the response fingerprint of each observation branch to the control disturbance is identified online, and the response fingerprints are arranged according to the observation branches to form a response matrix; Based on the response matrix, the suppression constraints of the control perturbation component are determined, and the observation sequence is subjected to perturbation suppression and physiological frequency band filtering to obtain candidate pulse observations; Based on the actual sampling time, exposure integration window and quality information, the candidate pulse observations are reconstructed using non-uniform time to satisfy forward consistency of exposure integration, thereby obtaining multiple continuous candidate pulse trajectories. Virtual differential weights are determined based on the response matrix. The virtual differential weights satisfy control response residual constraints, weight bounded constraints, and pulse evidence preservation constraints. Based on the virtual differential weights, the candidate pulse trajectory is reconstructed to obtain the target pulse wave signal. Based on the response matrix, a control prediction residual is generated, and based on the control prediction residual, a hierarchical quality determination is performed on the target pulse wave signal; When the quality determination is passed, the target pulse wave signal is output; when the quality determination is failed, the acquisition parameters, response identification parameters, time reconstruction parameters, and signal reconstruction parameters are adjusted based on the failed quality level. If the preset conditions are still not met after adjustment, a remeasurement command is output.

[0009] Furthermore, the spatiotemporal observation unit includes line bands, color channels, effective pixel groups, and exposure branches divided along the rolling shutter scanning direction; the frame time information includes frame timestamps and the rolling shutter scanning direction; the exposure information includes the exposure midpoint; and the position of the spatiotemporal observation unit in the image includes the effective line position. A unified observation record is established for each observation branch, and the unified observation record includes the observed value, actual sampling time, exposure integration window, sampling time uncertainty, sample quality, response fingerprint, and source identifier.

[0010] Furthermore, the control disturbance is a disturbance with limited amplitude and balanced average dose, and the duration of the identification phase is 1 to 2 seconds.

[0011] Furthermore, the response fingerprint of a single observation branch includes response gain, response delay, saturation margin, exposure response, and stability. The response fingerprints of multiple observation branches are arranged in rows to form a response matrix, and the columns of the response matrix correspond to the control dimension or the hysteresis term of the control dimension.

[0012] Furthermore, the disturbance suppression includes: constructing an approximate null space projection based on the response matrix, projecting the multi-branch observation vectors onto the approximate null space direction of the response matrix; and limiting the control response residue of the processed observations relative to the response matrix through an optimization method. The physiological frequency band filter adaptively adjusts the filter parameters based on the candidate pulse main frequency, effective harmonics, motion information, control prediction residual information, and the time characteristics to be output, and records or compensates for the filter group delay; when the group delay compensation reliability does not meet the preset conditions, the corresponding time characteristics are prohibited from being output.

[0013] Furthermore, the non-uniform temporal reconstruction includes processing of missing intervals: for missing intervals that meet the shortage condition, linear interpolation or conformal interpolation weighted by sample quality and sampling time uncertainty is used; for long gaps that do not meet the shortage condition, the missing mask is retained, and the output of temporal features is prohibited using the intervals corresponding to the long gaps; the non-uniform temporal reconstruction obtains continuous candidate pulse trajectories by solving an objective function including an exposure integral kernel observation consistency term, a quality weight term, and a non-physiological change regularization term, and the continuous candidate pulse trajectories are mapped with the valid observations after being mapped by the exposure integral forward operator.

[0014] Furthermore, the virtual differential weights are determined through regularized projection, quadratic programming, or constraint optimization; the pulse evidence preservation constraint is used to limit the loss of candidate pulse dominant frequency energy, effective harmonic energy, template subspace energy, or temporal pulse morphology before and after recombination; after forming multiple virtual differential channels, a second-level robust signal recombination is performed, and the weights of the second-level robust signal recombination are jointly determined by sample quality, time quality, signal-to-noise ratio quality, template quality, and sentinel quality, and the maximum influence of any observation branch on the target pulse wave signal is limited.

[0015] Furthermore, the layer quality determination includes: The spectral signal-to-noise ratio is determined based on the relationship between the candidate pulse dominant frequency and the effective harmonic neighborhood energy and the energy of the other evaluation frequency bands; the identification residual signal-to-noise ratio is determined based on the relationship between the retained pulse prediction energy and the control prediction residual energy; and the branch consistency signal-to-noise ratio is determined based on the relationship between the common component energy of multiple observation branches and the branch difference residual energy. A heartbeat template is constructed using the current user's measurement data, and template matching is performed using correlation coefficient, derivative similarity, feature temporal difference, or dynamic time warping score. The quality determination is performed in the order of acquisition gate, identification gate, reconstruction gate, waveform gate, and time gate. The failure of any quality gate cannot be compensated by the score of other quality gates. When the waveform gate passes but the time gate fails, only the pulse waveform or heart rate is output, and the output of time features is prohibited.

[0016] Furthermore, after the identification phase ends, a stable acquisition phase begins. During this phase, a sparse sentinel perturbation is applied, whereby the brightness change caused by the perturbation does not exceed a preset sensing threshold, or the average brightness difference before and after the perturbation is within a preset tolerance. Based on the response drift of the observed response of the sparse sentinel perturbation relative to the response fingerprint, the corresponding response fingerprint is updated, downweighted, or rendered invalid. When the device cannot meet the preset sensing threshold or preset tolerance, the visible sentinel perturbation is turned off, and drift detection is performed using an imperceptible control quantity or the device is downgraded to a sentinel-free mode.

[0017] Secondly, the present invention provides a PPG signal acquisition system based on control perturbation response fingerprinting, comprising: The acquisition and control module is used to acquire image sequences and camera control parameters during the stable pre-acquisition phase, and to apply control perturbations during the identification phase to obtain control perturbation sequences. The observation construction module is used to divide the region of interest in the image into multiple spatiotemporal observation units, obtain the observation sequence of multiple observation branches, and determine the actual sampling time, exposure integration window and quality information of each observation sample based on frame time information, exposure information and the position of the spatiotemporal observation units in the image. The response fingerprint identification module is used to identify the response fingerprint of each observation branch to the control disturbance online based on the control disturbance sequence and the observation sequence, and arrange the response fingerprints according to the observation branches to form a response matrix; The filtering and reconstruction module is used to determine the suppression constraints of the control disturbance component based on the response matrix, perform disturbance suppression and physiological frequency band filtering on the observation sequence to obtain candidate pulse observations, and perform non-uniform time reconstruction of the candidate pulse observations to satisfy the forward consistency of exposure integration based on the actual sampling time, exposure integration window and quality information to obtain multiple continuous candidate pulse trajectories. The virtual differential and signal reconstruction module is used to determine virtual differential weights based on the response matrix. The virtual differential weights satisfy control response residual constraints, weight bounded constraints, and pulse evidence preservation constraints. Based on the virtual differential weights, the module reconstructs the candidate pulse trajectory to obtain the target pulse wave signal. The quality assessment module is used to generate control prediction residuals based on the response matrix and to perform hierarchical quality assessment on the target pulse wave signal based on the control prediction residuals. A closed-loop output module is used to output the target pulse wave signal when the quality judgment is passed, adjust the parameters of the corresponding module based on the failed quality level when the quality judgment is failed, and output a re-measurement command when the preset conditions are still not met after adjustment.

[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) By applying a known control perturbation in the same measurement, the present invention transforms the composite response of the camera control transmitted to multiple observation branches through the tissue optical path and imaging chain into an online identifiable response matrix, and enables the same response matrix to be used simultaneously for control perturbation component suppression, virtual differential weight constraint and quality judgment, thereby enhancing the technical correlation between each processing link and avoiding inconsistencies caused by mutually independent empirical parameters.

[0019] (2) This invention utilizes the actual sampling time, exposure integration window and quality information to perform non-uniform time reconstruction, and constrains the continuous trajectory through exposure integration forward consistency. At the same time, it determines the virtual differential weight through control response residual constraint, weight bounded constraint and pulse evidence maintenance constraint. This avoids the pseudo peaks and valleys and time deviations introduced by direct interpolation or resampling, as well as the problem of amplifying noise or eliminating real pulse components in pursuit of interference cancellation, thus ensuring the reliability of signal reconstruction and the accuracy of time characteristics.

[0020] (3) The present invention generates control prediction residuals based on the same response matrix, and evaluates the acquisition, identification, reconstruction, waveform and time quality by combining multidimensional signal-to-noise ratio, leave-one heartbeat template and hierarchical quality gating respectively, so that hard failures are not masked by the total score, and implements classification recovery and hierarchical output according to the failure level. When some quality conditions are not met, the output range can be limited to avoid unreliable measurement results. After automatic recovery failure, a targeted remeasurement instruction is given, which improves the credibility of quality judgment and the robustness of the system. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0023] Figure 2 This is a schematic diagram of the stable pre-acquisition, identification, and stable acquisition timing stages of the present invention.

[0024] Figure 3 This is a schematic diagram of the response fingerprint identification and response matrix construction of the present invention.

[0025] Figure 4 This is a schematic diagram of the control disturbance component suppression and physiological frequency band filtering of the present invention.

[0026] Figure 5 This is a schematic diagram of the non-uniform time reconstruction of the present invention.

[0027] Figure 6 This is a schematic diagram of the virtual differential and signal reconstruction of the present invention.

[0028] Figure 7 This is a schematic diagram illustrating the construction of waveform quality evidence for this invention.

[0029] Figure 8 This is a schematic diagram of the layered quality determination and graded output of the present invention.

[0030] Figure 9 This is a schematic diagram of the system modules and closed-loop feedback of the present invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] This invention provides the following technical solutions: like Figure 1 The method for acquiring PPG signals based on control perturbation response fingerprints, as shown, mainly includes the following steps: S1. In the stable pre-acquisition stage, image sequences and camera control parameters are acquired, the region of interest in the image is divided into multiple spatiotemporal observation units, and the observation sequences of multiple observation branches are obtained. Based on frame time information, exposure information and the position of the spatiotemporal observation units in the image, the actual sampling time, exposure integration window and quality information of each observation sample are determined.

[0034] In a preferred embodiment of this application, the spatiotemporal observation unit includes a row band, a color channel, an effective pixel group, and an exposure branch divided along the rolling shutter scanning direction; the frame time information includes a frame timestamp and the rolling shutter scanning direction; the exposure information includes the exposure midpoint; and the position of the spatiotemporal observation unit in the image includes the effective row position. A unified observation record is established for each observation branch, and the unified observation record includes the observation value, the actual sampling time, the exposure integration window, the sampling time uncertainty, the sample quality, the response fingerprint, and the source identifier.

[0035] S2. In the identification phase of the same measurement, apply control perturbations to the illumination intensity, exposure time, or gain to obtain a control perturbation sequence.

[0036] In a preferred embodiment of this application, the control disturbance is a disturbance with limited amplitude and balanced average dose, and the duration of the identification phase is 1 to 2 seconds.

[0037] S3. Based on the control disturbance sequence and the observation sequence, identify the response fingerprint of each observation branch to the control disturbance online, and arrange the response fingerprints according to the observation branches to form a response matrix.

[0038] In a preferred embodiment of this application, the response fingerprint of a single observation branch includes response gain, response delay, saturation margin, exposure response, and stability. The response fingerprints of multiple observation branches are arranged in rows to form a response matrix, and the columns of the response matrix correspond to the control dimension or the hysteresis term of the control dimension.

[0039] S4. Based on the response matrix, determine the suppression constraints of the control perturbation component, perform perturbation suppression and physiological frequency band filtering on the observation sequence, and obtain candidate pulse observations.

[0040] In a preferred embodiment of this application, the disturbance suppression includes: constructing an approximate null space projection based on the response matrix, projecting the multi-branch observation vectors onto the approximate null space direction of the response matrix; and limiting the control response residue of the processed observations relative to the response matrix through an optimization method. The physiological frequency band filter adaptively adjusts the filter parameters based on the candidate pulse main frequency, effective harmonics, motion information, control prediction residual information, and the time characteristics to be output, and records or compensates for the filter group delay; when the group delay compensation reliability does not meet the preset conditions, the corresponding time characteristics are prohibited from being output.

[0041] S5. Based on the actual sampling time, exposure integration window and quality information, perform non-uniform time reconstruction on the candidate pulse observation to satisfy the forward consistency of exposure integration, and obtain multiple continuous candidate pulse trajectories.

[0042] As a preferred embodiment of this application, the non-uniform temporal reconstruction includes processing of missing intervals: for missing intervals that meet the shortage condition, linear interpolation or conformal interpolation weighted by sample quality and sampling time uncertainty is used; for long gaps that do not meet the shortage condition, the missing mask is retained, and the output of temporal features is prohibited using the intervals corresponding to the long gaps; the non-uniform temporal reconstruction obtains continuous candidate pulse trajectories by solving an objective function including an exposure integral kernel observation consistency term, a quality weight term, and a non-physiological change regularization term, and the continuous candidate pulse trajectories are mapped with the valid observations after being mapped by the exposure integral forward operator.

[0043] S6. Based on the response matrix, determine the virtual differential weights, which satisfy the control response residual constraint, the weight bounded constraint, and the pulse evidence preservation constraint. Based on the virtual differential weights, reconstruct the candidate pulse trajectory to obtain the target pulse wave signal.

[0044] In a preferred embodiment of this application, the virtual differential weights are determined through regularized projection, quadratic programming, or constraint optimization; the pulse evidence preservation constraint is used to limit the loss of candidate pulse dominant frequency energy, effective harmonic energy, template subspace energy, or temporal pulse morphology before and after reconstruction; after forming multiple virtual differential channels, a second-level robust signal reconstruction is performed, the weights of which are jointly determined by sample quality, time quality, signal-to-noise ratio quality, template quality, and sentinel quality, and the maximum influence of any observation branch on the target pulse wave signal is limited.

[0045] S7. Generate control prediction residuals based on the response matrix, and perform hierarchical quality determination on the target pulse wave signal based on the control prediction residuals.

[0046] In a preferred embodiment of this application, the layer quality determination is carried out in the following manner: The spectral signal-to-noise ratio is determined based on the relationship between the candidate pulse dominant frequency and the effective harmonic neighborhood energy and the energy of the other evaluation frequency bands; the identification residual signal-to-noise ratio is determined based on the relationship between the retained pulse prediction energy and the control prediction residual energy; and the branch consistency signal-to-noise ratio is determined based on the relationship between the common component energy of multiple observation branches and the branch difference residual energy. A heartbeat template is constructed using the current user's measurement data, and template matching is performed using correlation coefficient, derivative similarity, feature temporal difference, or dynamic time warping score. The quality determination is performed in the order of acquisition gate, identification gate, reconstruction gate, waveform gate, and time gate. The failure of any quality gate cannot be compensated by the score of other quality gates. When the waveform gate passes but the time gate fails, only the pulse waveform or heart rate is output, and the output of time features is prohibited.

[0047] S8. When the quality determination is passed, the target pulse wave signal is output; when the quality determination is failed, the acquisition parameters, response identification parameters, time reconstruction parameters and signal reconstruction parameters are adjusted based on the failed quality level. If the preset conditions are still not met after adjustment, a remeasurement command is output.

[0048] In a preferred embodiment of this application, after the identification phase ends, a stable acquisition phase begins. During the stable acquisition phase, a sparse sentinel perturbation is applied. The brightness change caused by the sparse sentinel perturbation does not exceed a preset sensing threshold, or the average brightness difference before and after the perturbation is within a preset tolerance. Based on the response drift of the observed response of the sparse sentinel perturbation relative to the response fingerprint, the corresponding response fingerprint is updated, downweighted, or invalidated. When the device cannot meet the preset sensing threshold or preset tolerance, the visible sentinel perturbation is turned off, and drift detection is performed using an imperceptible control quantity or the device is downgraded to a no-sentinel mode.

[0049] This invention discloses a PPG signal acquisition system based on control perturbation response fingerprinting, comprising: The acquisition and control module is used to acquire image sequences and camera control parameters during the stable pre-acquisition phase, and to apply control perturbations during the identification phase to obtain control perturbation sequences. The observation construction module is used to divide the region of interest in the image into multiple spatiotemporal observation units, obtain the observation sequence of multiple observation branches, and determine the actual sampling time, exposure integration window and quality information of each observation sample based on frame time information, exposure information and the position of the spatiotemporal observation units in the image. The response fingerprint identification module is used to identify the response fingerprint of each observation branch to the control disturbance online based on the control disturbance sequence and the observation sequence, and arrange the response fingerprints according to the observation branches to form a response matrix; The filtering and reconstruction module is used to determine the suppression constraints of the control disturbance component based on the response matrix, perform disturbance suppression and physiological frequency band filtering on the observation sequence to obtain candidate pulse observations, and perform non-uniform time reconstruction of the candidate pulse observations to satisfy the forward consistency of exposure integration based on the actual sampling time, exposure integration window and quality information to obtain multiple continuous candidate pulse trajectories. The virtual differential and signal reconstruction module is used to determine virtual differential weights based on the response matrix. The virtual differential weights satisfy control response residual constraints, weight bounded constraints, and pulse evidence preservation constraints. Based on the virtual differential weights, the module reconstructs the candidate pulse trajectory to obtain the target pulse wave signal. The quality assessment module is used to generate control prediction residuals based on the response matrix and to perform hierarchical quality assessment on the target pulse wave signal based on the control prediction residuals. A closed-loop output module is used to output the target pulse wave signal when the quality judgment is passed, adjust the parameters of the corresponding module based on the failed quality level when the quality judgment is failed, and output a re-measurement command when the preset conditions are still not met after adjustment.

[0050] Example The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. This embodiment uses the example of a user covering the rear camera of a mobile phone with their fingertip to collect PPG signals. Figure 1 As shown, the overall process of this invention includes stable pre-acquisition, identification stage, response matrix identification, disturbance suppression and filtering, non-uniform time reconstruction, signal recombination, quality judgment, and closed-loop output. Figure 2 As shown, the acquisition sequence is as follows: stable pre-acquisition, identification phase, and stable acquisition, with sparse sentinel perturbation applied during the stable acquisition phase. The user covers the mobile terminal's camera and adjacent illumination units with their fingertip. The system continuously acquires image sequences, frame time information, exposure time, gain, and illumination control values ​​through the public camera interface or the manufacturer's camera interface. The acquisition window is 15 to 30 seconds, with stable pre-acquisition performed first, followed by 1 to 2 seconds for the identification phase, and the remaining time for the stable acquisition phase.

[0051] S1, Stable Pre-acquisition Phase During the stabilization pre-acquisition phase, the system extracts the region of interest from the image, excluding saturated, overly dark, blurred, obviously moving, and non-target skin pixels. The remaining valid pixels are then organized into multiple observation branches based on color channels, valid pixel groups, and line bands or exposure branches along the rolling shutter scanning direction. Each observation branch retains the valid pixel position, frame number, color channel, and source identifier.

[0052] The system determines the representative sampling time and exposure integration window for each observation sample based on the frame timestamp, rolling shutter scanning direction, effective line position, effective image height, frame-by-frame exposure time, and exposure midpoint. Frame timestamp error, scan duration error, line position error, and exposure semantic error are combined into the sampling time uncertainty.

[0053] For the Establish unified observation records for each spatiotemporal observation unit:

[0054] in, For the observed values, To represent the sampling time, and These represent the start and end times of the exposure integration window, respectively. For quality information, For sampling time uncertainty, To control the perturbation response fingerprint, This is a source identifier.

[0055] S2, Identification Stage During the identification phase, the system applies control perturbations to the illumination intensity, exposure time, or gain, acquiring a control perturbation sequence. These control perturbations are amplitude-limited and average dose-balanced, and the identification phase lasts for 1 to 2 seconds. The control sequence employs positive and negative variations, complementary pulses, pseudo-random binary sequences, or orthogonal sequences. By adjusting the duration, amplitude, or duty cycle of the positive and negative perturbations, the average brightness and average illuminance dose are kept balanced. The perturbation amplitude, duration, illumination dose, equipment temperature rise, and brightness change rate are all limited within the equipment safety and user comfort boundaries.

[0056] S3, Response Fingerprint Recognition like Figure 3 As shown, the system identifies the response fingerprint of each observation branch to the control disturbance online based on the control disturbance sequence and the observation sequence. The d-dimensional control disturbance sequence is represented as... The observation sequence of m observation branches is represented as To describe the response delay, the current control variable and one or more delayed control variables are concatenated into an extended control matrix U. e The relative U of each observation branch e The response coefficients form the response matrix. Each row corresponds to an observation branch, and each column corresponds to a control dimension or hysteresis term.

[0057] Before identification, saturation sample removal, slow baseline removal, outlier weighting, and temporary suppression of candidate physiological bands were performed on the observed sequence X to reduce the impact of uncontrolled related changes on identification.

[0058] Target identification using weighted regularization:

[0059] in, This is an estimate of the response matrix. The observation matrix after removing slow baselines and quality processing, To expand the control matrix, The weight matrix is ​​composed of sample quality. The parameter is a non-negative regularization parameter. It is the Frobenius norm.

[0060] The system is based on the extended control matrix U e The identifiability is determined by the rank, condition number, degree of independent excitation of each control dimension, observation response energy, and identification residual. When the control disturbance lacks independent excitation, the observation response is below the detectable condition, or the response matrix is ​​ill-conditioned, the identification gate fails to trigger, and the control disturbance, exposure baseline, observation unit division, or regularization parameters are adjusted within the safe and comfortable boundary.

[0061] The response fingerprint of a single observation branch includes response gain, response delay, saturation margin, exposure response, and stability. Multiple response fingerprints are arranged by branch to form a response matrix, with the columns of the response matrix corresponding to the control dimension or the hysteresis term of the control dimension.

[0062] S4. Control disturbance component suppression and physiological frequency band filtering like Figure 4 As shown, the system determines the suppression constraints of the control disturbance component based on the response matrix, performs disturbance suppression and physiological frequency band filtering on the observation sequence, and obtains candidate pulse observations.

[0063] The integral observations during the exposure period are represented as follows:

[0064] in, For the first One observation value, For the first Exposure integration kernel for each observation, The continuous PPG trajectory to be reconstructed. In response to fingerprints, In order to be with the first The known control inputs corresponding to each observation For ambient light, slow baseline, or unmodeled bias, To observe noise.

[0065] Constructing a regularized approximate null space projection based on the response matrix R:

[0066] in, For an m×m projection matrix, for 3D identity matrix for 3D identity matrix This is a non-negative regularization parameter used to mitigate ill-conditioned responses. Multi-branch observation X is based on... Suppress control disturbance components. In addition to projection, the residual control response relative to R in the processed observation can be directly limited through constraint optimization methods.

[0067] The observations after disturbance suppression enter the physiological frequency band filter. The physiological frequency band filter adaptively adjusts the filtering parameters based on the candidate pulse dominant frequency, effective harmonics, motion information, control prediction residual information, and the time characteristics to be output. The system records and compensates for the filter group delay; when the group delay compensation reliability does not meet the preset conditions, the corresponding time characteristics are prohibited from being output.

[0068] S5, Non-uniform temporal reconstruction like Figure 5As shown, the system performs non-uniform time reconstruction of candidate pulse observations based on the actual sampling time, exposure integration window, and quality information to satisfy the forward consistency of exposure integration, thereby obtaining multiple continuous candidate pulse trajectories.

[0069] The system sorts uniform observation records according to actual sampling time and establishes non-uniform observations using exposure integration windows and sample quality. The shortage condition is determined by the number of consecutive missing samples, the duration of the missing sample, or the ratio of the duration of the missing sample to the candidate pulse cycle. When the shortage condition is met, the system uses a method based on q... i and σ ti Commonly weighted linear interpolation or conformal interpolation.

[0070] Intervals that do not meet the shortfall condition are marked as long gaps and their missing masks are retained. The use of long gaps to output time features is prohibited. If a long gap results in insufficient valid heartbeats, the acquisition window is extended; if the number is still insufficient after extension, a remeasurement command is output.

[0071] The objective of continuous trajectory inversion is:

[0072] in, These are estimates of continuous candidate pulse trajectories. It is a second-order difference or second-order derivative operator. To suppress regularization terms that suppress non-physiological mutations, excessive oscillations, or changes exceeding the physiological frequency band, and The weights are non-negative.

[0073] After obtaining the continuous candidate pulse trajectories, their exposure integral forward operator K is mapped to K( The forward residual is checked for consistency with the valid observations; if the forward residual exceeds the preset conditions, the reconstruction gate fails.

[0074] S6, Virtual Differential and Signal Reconstruction like Figure 6 As shown, virtual difference weights w are determined based on the response matrix R, which limits the residual control response and the magnitude of the weights after recombination, while preserving candidate pulse evidence. The basic constraints are:

[0075] in, To control the residual tolerance of the response, This represents the upper limit of the weight magnitude. The pulse evidence matrix is ​​constructed based on candidate pulse frequency bands, effective harmonics, or a left-one-beat template subspace. Maintain conditions for the lowest pulse evidence.

[0076] The objective of weight calculation is:

[0077] in, For virtual difference weight vectors, and This is the regularization parameter. Regularized projection, quadratic programming, convex constraint optimization, or other methods that satisfy the objectives of identical control residues, bounded weights, and pulse evidence preservation can also be used.

[0078] Multiple virtual differential channels are obtained from different candidate branches or different constraint parameters. The second-level robust signal reconstruction is weighted according to sample quality, time quality, signal-to-noise ratio quality, template quality, and sentinel quality, and the maximum influence of any observation branch on the target pulse wave signal is limited.

[0079] S7, Layer Quality Judgment like Figure 7 As shown, the system generates control prediction residuals based on the response matrix, and performs hierarchical quality determination on the target pulse wave signal based on the control prediction residuals.

[0080] The spectral signal-to-noise ratio is determined based on the relationship between the candidate pulse dominant frequency and the energy of the effective harmonic neighborhood and the energy of the remaining evaluation frequency bands:

[0081] in, For the spectral signal-to-noise ratio, The energy of the candidate pulse dominant frequency and the effective harmonic neighborhood. To evaluate the remaining energy within the frequency band, To prevent regularization parameters with a denominator of zero.

[0082] The residual signal-to-noise ratio is determined based on the relationship between the retained pulse prediction energy and the control prediction residual energy. The control prediction residual obtained from the response matrix R is expressed as:

[0083]

[0084] in, To control the prediction residual, To identify the residual signal-to-noise ratio, To expose the integral forward operator, For pulse prediction observations corresponding to multiple observation branches, Predict the response to a known control disturbance.

[0085] The branch consistency signal-to-noise ratio is determined based on the relationship between the common component energy of multiple observation branches and the residual energy of branch differences:

[0086] in, For branch consistency signal-to-noise ratio, Energy as a common candidate physiological component across branches The residual energy of each branch relative to the common component is the difference in energy.

[0087] The system constructs a left-one-beat template using the current user's measurement data and performs template matching using correlation coefficient, derivative similarity, feature time difference, or dynamic time warping score.

[0088] like Figure 8 As shown, quality assessment is performed in the following order: acquisition gate, identification gate, reconstruction gate, waveform gate, and time gate. The acquisition gate checks saturation, underexposure, effective pixel ratio, motion and contact stability; the identification gate checks control excitation, response detectability, matrix rank, and ill-conditioned degree; the reconstruction gate checks the effective observation ratio, forward consistency, and long gaps; the waveform gate checks the multidimensional signal-to-noise ratio, left-one-beat template, and branch influence; and the time gate checks sampling time uncertainty, filter group delay compensation, and effective time interval.

[0089] The failure of any quality gate cannot be compensated by the score of other quality gates.

[0090] S8, Closed-loop output like Figure 8 As shown, when the quality judgment is passed, the system outputs the target pulse wave signal.

[0091] When a quality assessment fails, the system adjusts the acquisition parameters, response identification parameters, time reconstruction parameters, and signal reconstruction parameters based on the failed quality level. Specifically: it feeds back illumination, exposure, or gain parameters to the acquisition and control module; it feeds back control disturbance, observation unit, or regularization parameters to the response fingerprint identification module; it feeds back gap processing or trajectory reconstruction parameters to the filtering and reconstruction module; and it feeds back virtual differential or robust fusion parameters to the virtual differential and signal reconstruction module.

[0092] When the waveform gate passes but the time gate does not, only the allowed level of pulse waveform or heart rate is output, and the output of time characteristics is prohibited; when both the waveform gate and the time gate pass, the target pulse wave signal, heart rate and allowed time characteristics are output.

[0093] like Figure 2As shown, during the stable acquisition phase, the system applies sparse sentinel perturbations. The brightness change caused by the sparse sentinel perturbation does not exceed a preset sensing threshold, or the average brightness difference before and after the perturbation is within a preset tolerance. Based on the response drift of the observed response to the sparse sentinel perturbation relative to the response fingerprint, the system updates, downweights, or invalidates the corresponding response fingerprint. The system compares the sentinel response with the saved response fingerprint: slight and consistent drift triggers recursive updates; insufficient evidence or local anomalies trigger downweighting of the corresponding observation branch; persistent inconsistency, reversed response direction, or worsening matrix ill-conditioning triggers response fingerprint invalidation and re-identification.

[0094] When the device fails to meet the preset sensing threshold or preset tolerance, the visible sentinel disturbance is turned off, and drift detection is performed using an imperceptible control quantity or the device is downgraded to a no-sentinel mode.

[0095] If the preset conditions are still not met after adjustment, the system outputs a remeasurement command.

[0096] Compared with existing technologies, this invention enhances the technical correlation between various stages by applying control disturbances to identify the response matrix online and using this response matrix simultaneously for disturbance suppression, signal reconstruction, and quality judgment; it ensures the accuracy of waveforms and time by using actual sampling time and exposure integration window for non-uniform time reconstruction; and it improves the robustness of the system and the reliability of measurement results by combining multidimensional signal-to-noise ratio and hierarchical quality gating for quality evaluation and implementing classification recovery and graded output according to failure level.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for acquiring PPG signals based on control perturbation response fingerprints, characterized in that, Includes the following steps: During the stable pre-acquisition phase, image sequences and camera control parameters are acquired, the region of interest in the image is divided into multiple spatiotemporal observation units, and the observation sequences of multiple observation branches are obtained. Based on frame time information, exposure information and the position of the spatiotemporal observation units in the image, the actual sampling time, exposure integration window and quality information of each observation sample are determined. During the identification phase of the same measurement, control perturbations are applied to the illumination intensity, exposure time, or gain to obtain a control perturbation sequence; Based on the control disturbance sequence and the observation sequence, the response fingerprint of each observation branch to the control disturbance is identified online, and the response fingerprints are arranged according to the observation branches to form a response matrix; Based on the response matrix, the suppression constraints of the control perturbation component are determined, and the observation sequence is subjected to perturbation suppression and physiological frequency band filtering to obtain candidate pulse observations; Based on the actual sampling time, exposure integration window and quality information, the candidate pulse observations are reconstructed using non-uniform time to satisfy forward consistency of exposure integration, thereby obtaining multiple continuous candidate pulse trajectories. Virtual differential weights are determined based on the response matrix. The virtual differential weights satisfy control response residual constraints, weight bounded constraints, and pulse evidence preservation constraints. Based on the virtual differential weights, the candidate pulse trajectory is reconstructed to obtain the target pulse wave signal. Based on the response matrix, a control prediction residual is generated, and based on the control prediction residual, a hierarchical quality determination is performed on the target pulse wave signal; When the quality determination is passed, the target pulse wave signal is output; when the quality determination is failed, the acquisition parameters, response identification parameters, time reconstruction parameters, and signal reconstruction parameters are adjusted based on the failed quality level. If the preset conditions are still not met after adjustment, a remeasurement command is output.

2. The PPG signal acquisition method based on control perturbation response fingerprinting according to claim 1, characterized in that, The spatiotemporal observation unit includes a row band, color channel, effective pixel group, and exposure branch divided along the rolling shutter scanning direction; the frame time information includes the frame timestamp and rolling shutter scanning direction, the exposure information includes the exposure midpoint, and the position of the spatiotemporal observation unit in the image includes the effective row position; a unified observation record is established for each observation branch, and the unified observation record includes the observation value, actual sampling time, exposure integration window, sampling time uncertainty, sample quality, response fingerprint, and source identifier.

3. The PPG signal acquisition method based on control perturbation response fingerprinting according to claim 1, characterized in that, The control disturbance is a disturbance with limited amplitude and balanced average dose, and the identification phase lasts for 1 to 2 seconds.

4. The PPG signal acquisition method based on control perturbation response fingerprinting according to claim 1, characterized in that, The response fingerprint of a single observation branch includes response gain, response delay, saturation margin, exposure response, and stability. The response fingerprints of multiple observation branches are arranged in rows to form a response matrix, and the columns of the response matrix correspond to the control dimension or the hysteresis term of the control dimension.

5. The PPG signal acquisition method based on control perturbation response fingerprinting according to claim 1, characterized in that, The disturbance suppression includes: constructing an approximate null space projection based on the response matrix, projecting the multi-branch observation vectors onto the approximate null space direction of the response matrix; and limiting the control response residue of the processed observations relative to the response matrix through an optimization method. The physiological frequency band filter adaptively adjusts the filter parameters based on the candidate pulse main frequency, effective harmonics, motion information, control prediction residual information, and the time characteristics to be output, and records or compensates for the filter group delay; when the group delay compensation reliability does not meet the preset conditions, the corresponding time characteristics are prohibited from being output.

6. The PPG signal acquisition method based on control perturbation response fingerprinting according to claim 1, characterized in that, The non-uniform time reconstruction includes the processing of missing intervals: for missing intervals that meet the missing gap condition, linear interpolation or conformal interpolation weighted by sample quality and sampling time uncertainty is used; for long gaps that do not meet the missing gap condition, the missing mask is retained, and the output of time features is prohibited using the intervals corresponding to the long gaps. The non-uniform time reconstruction obtains continuous candidate pulse trajectories by solving an objective function that includes an exposure integral kernel observation consistency term, a quality weight term, and a non-physiological change regularization term. The continuous candidate pulse trajectories are then mapped with the valid observations using the exposure integral forward operator for consistency verification.

7. The PPG signal acquisition method based on control perturbation response fingerprinting according to claim 1, characterized in that, The virtual differential weights are determined through regularized projection, quadratic programming, or constraint optimization; the pulse evidence preservation constraint is used to limit the loss of candidate pulse dominant frequency energy, effective harmonic energy, template subspace energy, or temporal pulse morphology before and after reconstruction; after forming multiple virtual differential channels, a second-level robust signal reconstruction is performed, and the weights of the second-level robust signal reconstruction are jointly determined by sample quality, time quality, signal-to-noise ratio quality, template quality, and sentinel quality, and the maximum influence of any observation branch on the target pulse wave signal is limited.

8. The PPG signal acquisition method based on control perturbation response fingerprinting according to claim 1, characterized in that, The layer quality determination is performed using the following method: The spectral signal-to-noise ratio is determined based on the relationship between the candidate pulse dominant frequency and the effective harmonic neighborhood energy and the energy of the other evaluation frequency bands; the identification residual signal-to-noise ratio is determined based on the relationship between the retained pulse prediction energy and the control prediction residual energy; and the branch consistency signal-to-noise ratio is determined based on the relationship between the common component energy of multiple observation branches and the branch difference residual energy. A heartbeat template is constructed using the current user's measurement data, and template matching is performed using correlation coefficient, derivative similarity, feature temporal difference, or dynamic time warping score. The quality determination is performed in the order of acquisition gate, identification gate, reconstruction gate, waveform gate, and time gate. The failure of any quality gate cannot be compensated by the score of other quality gates. When the waveform gate passes but the time gate fails, only the pulse waveform or heart rate is output, and the output of time features is prohibited.

9. The PPG signal acquisition method based on control perturbation response fingerprinting according to claim 1, characterized in that, After the identification phase ends, the stable acquisition phase begins. During the stable acquisition phase, a sparse sentinel perturbation is applied. The brightness change caused by the sparse sentinel perturbation does not exceed a preset perception threshold, or the average brightness difference before and after the perturbation is within a preset tolerance. Based on the response drift of the observed response to the sparse sentinel disturbance relative to the response fingerprint, the corresponding response fingerprint is updated, downweighted, or invalidated; when the device cannot meet the preset sensing threshold or preset tolerance, visible sentinel disturbance is turned off, and drift detection is performed using an imperceptible control quantity or the device is downgraded to a no-sentinel mode.

10. A PPG signal acquisition system based on control disturbance response fingerprinting, used to implement the method according to any one of claims 1-9, characterized in that, include: The acquisition and control module is used to acquire image sequences and camera control parameters during the stable pre-acquisition phase, and to apply control perturbations during the identification phase to obtain control perturbation sequences. The observation construction module is used to divide the region of interest in the image into multiple spatiotemporal observation units, obtain the observation sequence of multiple observation branches, and determine the actual sampling time, exposure integration window and quality information of each observation sample based on frame time information, exposure information and the position of the spatiotemporal observation units in the image. The response fingerprint identification module is used to identify the response fingerprint of each observation branch to the control disturbance online based on the control disturbance sequence and the observation sequence, and arrange the response fingerprints according to the observation branches to form a response matrix; The filtering and reconstruction module is used to determine the suppression constraints of the control disturbance component based on the response matrix, perform disturbance suppression and physiological frequency band filtering on the observation sequence to obtain candidate pulse observations, and perform non-uniform time reconstruction of the candidate pulse observations to satisfy the forward consistency of exposure integration based on the actual sampling time, exposure integration window and quality information to obtain multiple continuous candidate pulse trajectories. The virtual differential and signal reconstruction module is used to determine virtual differential weights based on the response matrix. The virtual differential weights satisfy control response residual constraints, weight bounded constraints, and pulse evidence preservation constraints. Based on the virtual differential weights, the module reconstructs the candidate pulse trajectory to obtain the target pulse wave signal. The quality assessment module is used to generate control prediction residuals based on the response matrix and to perform hierarchical quality assessment on the target pulse wave signal based on the control prediction residuals. A closed-loop output module is used to output the target pulse wave signal when the quality judgment is passed, adjust the parameters of the corresponding module based on the failed quality level when the quality judgment is failed, and output a re-measurement command when the preset conditions are still not met after adjustment.

Images