Remote physiological signal continuous and stable measurement method and system in complex dynamic scene

Abstract

The application discloses a remote physiological signal continuous stable measurement method and system under a complex dynamic scene, and the method comprises the following steps: acquiring video data of a to-be-detected object in a current time slice, wherein the video data comprises a target detection region; dividing the target detection region into a plurality of sub-regions, and extracting candidate physiological feature signals corresponding to each sub-region; calling historical physiological feature signals stored in a time sequence memory library to constrain and correct the candidate physiological feature signals, so as to obtain high-quality regional signals; inputting the high-quality regional signals into a backbone network; splicing the high-quality regional signals and the historical physiological feature signals according to a time dimension, and constructing a time sequence sequence fusing a time sequence context; performing multi-layer feature extraction and mapping conversion on the time sequence sequence, so as to obtain a reconstructed signal; and performing parameter prediction based on the reconstructed signal, so as to obtain physiological parameters of the to-be-detected object. The application can effectively improve the stability, accuracy and robustness of physiological signal detection.

Description

Technical Field

[0001] This application relates to the field of biometric detection technology, and in particular to a method and system for continuous and stable measurement of remote physiological signals in complex dynamic scenarios. Background Technology

[0002] Remote physiological signal detection refers to the use of cameras to collect video of human body areas, such as the face and neck, in a non-contact manner, and extracting physiological parameters such as heart rate and respiratory rate based on changes in skin pixel light and shadow. Due to its advantages of not requiring contact with the human body and being easy to deploy, it is widely used in scenarios such as smart terminals, vehicle monitoring, and health screening.

[0003] Currently, remote physiological detection methods are mainly divided into two categories: one is the traditional signal processing method, which uses channel mapping, independent component analysis and combined with region weighting and attention mechanisms to select high-quality signal regions; the other is the deep learning method, which relies on convolutional networks and temporal modeling to automatically extract latent physiological features in videos and complete physiological signal prediction.

[0004] However, existing technologies still have significant drawbacks. On the one hand, current methods evaluate signal quality based solely on the current frame or a short time window, failing to fully utilize historical temporal information. When there is head movement, facial expression changes, or lighting fluctuations, local artifacts can easily occur in the facial region. Existing technologies struggle to distinguish between genuine physiological fluctuations and environmental noise, which can easily lead to signal waveform distortion and parameter jumps.

[0005] On the other hand, conventional deep learning models rely on massive amounts of labeled data, resulting in high training costs; moreover, their feature learning capabilities are limited, and they exhibit weak generalization ability and insufficient anti-interference performance in complex lighting and motion interference scenarios. Therefore, how to stably output continuous and high-quality physiological signals in complex dynamic scenarios is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0006] Therefore, it is necessary to provide a method and system for continuous and stable measurement of remote physiological signals in complex dynamic scenarios to address the above-mentioned technical problems, so as to solve at least one of the problems existing in the prior art.

[0007] Firstly, a method for continuous and stable remote physiological signal measurement in complex dynamic scenarios is provided, including: Acquire video data of the object to be detected within the current time slice, wherein the video data includes the target detection area; The target detection area is divided into several sub-regions, and candidate physiological feature signals corresponding to each sub-region are extracted; By calling the historical physiological feature signals stored in the time-series memory, the candidate physiological feature signals are constrained and corrected to obtain high-quality regional signals; The high-quality regional signal is input into the backbone network; the high-quality regional signal and the historical physiological feature signal are concatenated according to the time dimension to construct a time sequence that integrates temporal context; multi-level feature extraction and mapping transformation are performed on the time sequence to obtain the reconstructed signal, and parameter prediction is performed based on the reconstructed signal to obtain the physiological parameters of the object to be detected.

[0008] In one possible implementation, the step of calling historical physiological feature signals stored in the time-series memory to constrain and correct the candidate physiological feature signals to obtain high-quality region signals includes: Based on the historical physiological feature signals, a time-series prediction model is constructed, which is used to characterize the continuous evolution relationship of physiological feature signals in the time dimension. Based on the time-series prediction model, the current physiological characteristic signal corresponding to the current time slice is predicted; The candidate physiological feature signals are matched with the current physiological feature signal to obtain a similarity score; Based on the similarity score, the high-quality region signal is determined.

[0009] In one possible implementation, the step of calling historical physiological feature signals stored in the time-series memory to constrain and correct the candidate physiological feature signals to obtain high-quality region signals includes: A baseline physiological feature signal is selected from the historical physiological feature signals stored in the time-series memory bank as a template signal; The similarity score is obtained by matching each candidate physiological feature signal with the template signal using a template matching algorithm. Based on the similarity score, the high-quality region signal is determined.

[0010] In one possible implementation, determining the high-quality region signal based on the similarity score includes: Based on the similarity score, the candidate physiological feature signal with the highest similarity score is selected as the high-quality region signal; or Based on the similarity score, the weights corresponding to each candidate physiological feature signal are determined; Based on the weights, the candidate physiological feature signals are weighted and fused to obtain the high-quality region signal.

[0011] In one possible implementation, the step of predicting parameters based on the reconstructed signal to obtain the physiological parameters of the object to be detected includes: The reconstructed signal is subjected to frequency domain transformation within a preset time window to obtain the signal spectrum distribution; Based on the signal spectrum distribution, the effective physiological frequency band is determined; Within the effective physiological frequency band, identify the dominant frequency component with the largest energy or amplitude; Based on the correspondence between the dominant frequency component and the physiological parameters, the physiological parameter values ​​corresponding to the object to be detected are calculated.

[0012] In one possible implementation, the target detection region is divided into several sub-regions, and candidate physiological feature signals corresponding to each sub-region are extracted, including: Based on human body geometric region division rules or equal grid division rules, the target detection region is divided into several sub-regions; Extract the corresponding initial candidate physiological feature signals from each sub-region; The mean of each initial candidate physiological feature signal is adjusted to zero through zero-mean processing; Bandpass filtering and normalization are performed on the zero-mean signal to obtain standard candidate physiological feature signals.

[0013] Secondly, a remote, continuous, and stable physiological signal measurement system for complex dynamic scenarios is provided, including: A video data acquisition unit is used to acquire video data of the object to be detected within the current time slice, wherein the video data includes the target detection area; The candidate physiological feature signal extraction unit is used to divide the target detection area into several sub-regions and extract the candidate physiological feature signals corresponding to each sub-region. A high-quality region signal filtering unit is used to call up historical physiological feature signals stored in the time-series memory bank, constrain and correct the candidate physiological feature signals, and obtain high-quality region signals. The physiological parameter generation unit is used to input the high-quality regional signal into the backbone network; to concatenate the high-quality regional signal with the historical physiological feature signal according to the time dimension to construct a time sequence that integrates the temporal context; to perform multi-layer feature extraction and mapping transformation on the time sequence to obtain the reconstructed signal; and to perform parameter prediction based on the reconstructed signal to obtain the physiological parameters of the object to be detected.

[0014] In one embodiment of this application, the physiological parameter generation unit is further configured to: The reconstructed signal is subjected to frequency domain transformation within a preset time window to obtain the signal spectrum distribution; Based on the signal spectrum distribution, the effective physiological frequency band is determined; Within the effective physiological frequency band, identify the dominant frequency component with the largest energy or amplitude; Based on the correspondence between the dominant frequency component and the physiological parameters, the physiological parameter values ​​corresponding to the object to be detected are calculated.

[0015] Thirdly, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and running on the processor, wherein the processor executes the computer-readable instructions to implement the method for continuous and stable measurement of remote physiological signals in complex dynamic scenarios as described above.

[0016] Fourthly, a readable storage medium is provided, on which computer-readable instructions are stored, which, when executed by a processor, implement the remote continuous and stable measurement method for physiological signals in complex dynamic scenarios as described above.

[0017] The method and system for continuous and stable measurement of remote physiological signals in the aforementioned complex dynamic scenarios include the following steps: acquiring video data of the target object within the current time slice, the video data including a target detection region; dividing the target detection region into several sub-regions and extracting candidate physiological feature signals corresponding to each sub-region; calling historical physiological feature signals stored in a temporal memory to constrain and correct the candidate physiological feature signals to obtain high-quality region signals; inputting the high-quality region signals into a backbone network; concatenating the high-quality region signals and the historical physiological feature signals according to the time dimension to construct a temporal sequence fused with temporal context; performing multi-layer feature extraction and mapping transformation on the temporal sequence to obtain a reconstructed signal; and performing parameter prediction based on the reconstructed signal to obtain the physiological parameters of the target object. In this embodiment, the target detection region is first cropped, aligned, and segmented to form multiple candidate physiological signals; historical physiological features are continuously stored and tracked using a temporal memory, and a temporal prediction model is constructed to provide temporal prior constraints. The model performs correlation matching between candidate signals and predicted signals in each sub-region, and performs weighted fusion of signals using similarity as the weight. This adaptively filters out low-quality signals and continuously tracks effective regions, significantly reducing the negative impacts of motion interference, illumination fluctuations, and local artifacts. The constrained and calibrated signals are then fed into the backbone network for reconstruction, outputting temporally continuous and rhythmically stable physiological signals. Frequency domain analysis is then used to accurately calculate parameters such as heart rate and respiratory rate. By fully utilizing historical temporal information, the model's dependence on labeled data is reduced, significantly improving the stability, accuracy, and robustness of physiological signal detection. It can maintain effective operation even with insufficient performance or failure, and can be stably applied to scenarios such as non-contact long-term health monitoring, ward care, and motion state vital sign detection. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of an application environment for a method for continuous and stable remote physiological signal measurement in a complex dynamic scenario according to an embodiment of this application; Figure 2 This is a flowchart illustrating a method for continuous and stable remote physiological signal measurement in a complex dynamic scenario according to one embodiment of this application. Figure 3 This is a schematic diagram of a remote physiological signal continuous and stable measurement system in a complex dynamic scenario according to one embodiment of this application; Figure 4 This is a schematic diagram of a computer device according to one embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The remote physiological signal continuous and stable measurement method for complex dynamic scenarios provided in this embodiment can be applied to, for example... Figure 1 In this application environment, the video acquisition device can communicate with the data processing unit and the storage unit, and the data processing unit and the storage unit can also communicate. Specifically, the video acquisition device is used to acquire human video sequences of the object to be detected, serving as the input source for physiological signal detection. The data processing unit is used to perform computational tasks such as feature tracking, signal analysis, and prediction based on the human video sequences acquired by the video acquisition device. This data processing unit can be a computer, an embedded device, or a server. If it is a server, it can be implemented using a standalone server or a server cluster composed of multiple servers. The storage unit is used to store the acquired video data, processed intermediate features, and historical data in the time-series memory.

[0022] In one embodiment, such as Figure 2 As shown, a method for continuous and stable remote physiological signal measurement in complex dynamic scenarios is provided, including the following steps: In step S110, video data of the object to be detected within the current time slice is acquired, and the video data includes the target detection area; Optionally, the system can preset time slices (fixed time segments, such as 1 second or 2 seconds) as the acquisition cycle to acquire raw video data of the object to be detected in real time. This video data can be continuously captured by a video acquisition device, such as a mobile phone camera, a vehicle-mounted camera, or a dedicated camera device. After acquiring the video image data corresponding to the current time slice, the system uses the corresponding detection algorithm to identify the target region (such as the face region, neck region, hand region, etc.) according to the specific detection scenario, locates and selects the target detection region containing the detected part of the object to be detected. For example, taking the face region as an example, this target detection region mainly covers the forehead, cheeks, and other areas sensitive to blood flow changes, while removing invalid areas such as hair, eyes, and background. For example, taking the face region as an example, firstly, the face key points (landmarks) in the video frame are obtained through a face detector, and then alignment changes (such as affine transformation) are calculated based on the key points. The aligned face image is then cropped to a uniform size to obtain the target detection region. The target detection region can serve as the basic region of interest for subsequent physiological signal extraction. All subsequent sub-region division and candidate physiological signal extraction operations are completed based on this target detection region.

[0023] It is important to note that after acquiring video data, the effectiveness of the target detection area in the video data can be evaluated and located to avoid subsequent detection anomalies caused by target area occlusion, posture shift, or region failure. For example, taking the face region as an example, the facial rPPG physiological signal mainly relies on the pixel temporal changes in the capillary-rich area of ​​the face for solution. When the subject undergoes significant facial posture changes such as turning their face to the side, looking down, looking up, or turning their head left or right, or when the face is obscured by hair, masks, or other objects, the effective blood sampling area of ​​the frontal face will be missing or completely ineffective, directly causing the extraction of physiological feature signals to be interrupted and waveform distortion to occur, which in turn causes drastic jumps in heart rate and respiratory rate parameters, seriously affecting the overall detection effect. To address this, the original video data can be split frame by frame to obtain multiple temporally continuous images. For each frame, facial key point detection and facial posture calculation can be performed simultaneously, and facial posture parameters such as face pitch angle, yaw angle, and roll angle can be calculated in real time to comprehensively determine the current orientation and region integrity of the face. Based on this, combined with preset posture thresholds, the system dynamically filters and updates high-quality target detection areas: when the signal quality of the frontal face area is attenuated due to posture deviation, the system automatically discards the invalid area and switches to alternative effective sub-regions with dense blood vessel distribution and stronger anti-posture interference capabilities, such as the nose, jaw, and both cheeks; at the same time, it identifies and removes invalid detection areas in the image that have occlusion, heavy shadow coverage, or low pixel signal-to-noise ratio in real time, and finally achieves dynamic optimal selection of target areas, ensuring the continuity and stability of physiological signal acquisition in complex posture scenarios.

[0024] In step S120, the target detection area is divided into several sub-regions, and candidate physiological feature signals corresponding to each sub-region are extracted; Optionally, the target detection region is divided into several sub-regions according to a preset partitioning rule. Then, a preset backbone model (such as MobileNet, ResNet, 3D-CNN, etc.) is called to analyze the image pixel information of each sub-region and extract the candidate physiological feature signals corresponding to each sub-region one by one. Specifically, all RGB pixel information in each sub-region can be extracted first. Then, multi-layer convolution operations are performed through the backbone model to statistically analyze the pixel brightness and color features of each frame and capture the subtle light intensity changes between adjacent frames, thereby obtaining the initial candidate physiological feature signals corresponding to each sub-region. Subsequently, preprocessing operations such as zero-mean, bandpass filtering, and normalization are performed based on each candidate signal to obtain standard candidate physiological feature signals, providing basic data for signal matching and temporal correction.

[0025] It should be noted that if there are multiple video frames within the current time slice, the target detection region in each video frame can be divided into regions. Then, all video frame images corresponding to the same sub-region in each video frame within the current time slice are collected. All RGB pixel information within each sub-region is extracted, and multi-layer convolution operations are performed through the backbone model to statistically analyze the pixel brightness and color features of each frame. The frame features are then concatenated and mapped along the time dimension to generate a one-dimensional temporal numerical sequence, which serves as the initial candidate physiological feature signal for the current sub-region. By traversing all sub-regions in the above manner, multiple independent initial candidate physiological feature signals can be obtained.

[0026] In step S130, historical physiological feature signals stored in the time-series memory are called to constrain and correct the candidate physiological feature signals to obtain high-quality regional signals; Optionally, the system can access historical physiological feature signals corresponding to several consecutive historical time slices of the object to be detected, continuously cached and updated in a time-series memory, to constrain and correct the current candidate signals and filter high-quality regional signals. These historical physiological feature signals can reflect the trends and change patterns of physiological signals. The time-series memory can be maintained in a queue format, storing historical physiological feature signals, typical template signals, and temporal change patterns corresponding to each sub-region under multiple past time slices. The system can use different tracking mechanisms based on the historical physiological feature signals in the memory to infer the signal characteristics of the next stage. For example, it can use an autoregressive model to predict future signal patterns or use template matching to find the optimal correspondence between historical and current signals. Then, by matching and filtering each candidate physiological feature signal with the predicted signal or template signal, high-quality regional signals are obtained. This tracking mechanism not only relies on the short-term information of the current frame but also utilizes historical features to constrain and correct the signal, thus providing a reliable basis for signal quality assessment and region selection in complex environments.

[0027] It should be noted that the maximum length of the signal sequence stored in the time-series memory can be L_{max}. When the current length of the memory is less than the preset length L, the system does not perform memory-based prediction or correction operations, but instead directly writes the candidate physiological feature signals extracted by the backbone model into the time-series memory in chronological order. When the length of the memory reaches or exceeds the preset length L, the system predicts and corrects the candidate physiological feature signals corresponding to each sub-region of the current time slice based on the historical physiological feature signals in the time-series memory, and replaces or appends the corrected physiological features into the time-series memory.

[0028] In step S140, the high-quality region signal is input into the backbone network; the high-quality region signal and the historical physiological feature signal are concatenated according to the time dimension to construct a time sequence with fused temporal context; multi-layer feature extraction and mapping transformation are performed on the time sequence to obtain the reconstructed signal, and parameter prediction is performed based on the reconstructed signal to obtain the physiological parameters of the object to be detected.

[0029] Optionally, after acquiring a high-quality regional signal, it can be input into a backbone network (such as a convolutional neural network) for signal reconstruction, thereby optimizing and repairing the high-quality regional signal. Specifically, the high-quality regional signal corresponding to the current time slice can be concatenated with the historical physiological feature signals of its adjacent preset time slices according to the time dimension, such as chronological order, to form a time-series sequence that integrates temporal context, fully utilizing historical physiological rhythm features to constrain the current signal. At this time, the backbone network performs multi-layer convolution operations on the concatenated time-series sequence, extracting shallow detailed features and deep abstract features step by step, comprehensively capturing the signal's variation patterns and physiological rhythm information. After completing the deep feature extraction, the backbone network achieves feature mapping transformation through nonlinear transformation and feature recombination, relying on contextual temporal information to perform overall correction and optimization of the signal, further filtering out noise and repairing waveform distortion. After multi-layer network processing, a standard physiological signal waveform with continuous temporal sequence, regular shape, and stronger anti-interference ability is finally obtained. Before the high-quality regional signal is input into the backbone network, motion artifacts and illumination interference have been suppressed through mechanisms such as temporal prediction constraints, regional consistency matching, and weighted fusion. This significantly weakens the expression of interference terms in the feature space. Therefore, the reconstructed signal output by the backbone network is continuous, smooth, and has stable physiological rhythm characteristics in the time dimension.

[0030] It should be noted that after obtaining physiological parameters such as heart rate and respiratory rate of the subject to be tested, the system will compare them with normal physiological index ranges to determine physiological abnormalities. It can effectively identify various abnormal physiological states such as tachycardia, bradycardia, respiratory rhythm disorders, and abnormal parameter drift. For different types and severity of abnormal physiological states, multi-level early warning mechanisms can be set and graded prompts can be executed, such as voice alarms, SMS reminders, notification of staff, nearby personnel, or related personnel for manual handling. At the same time, it can also link with the temporal memory database to retrieve the original video frames, preprocessed physiological waveforms, facial pose data, lighting environment parameters, and historical physiological parameters within the abnormal time period. It can complete the abnormality source tracing from multiple dimensions such as external environment, acquisition status, signal quality, and the physical condition of the subject to be tested, distinguishing between detection noise interference and real physiological abnormalities, and improving the accuracy and reliability of the system's early warning.

[0031] This application provides a method for continuous and stable remote physiological signal measurement in complex dynamic scenarios, comprising: acquiring video data of the target object within the current time slice, the video data including a target detection region; dividing the target detection region into several sub-regions and extracting candidate physiological feature signals corresponding to each sub-region; calling historical physiological feature signals stored in a temporal memory to constrain and correct the candidate physiological feature signals to obtain high-quality region signals; inputting the high-quality region signals into a backbone network; concatenating the high-quality region signals and the historical physiological feature signals according to the time dimension to construct a temporal sequence fused with temporal context; performing multi-layer feature extraction and mapping transformation on the temporal sequence to obtain reconstructed signals, and performing parameter prediction based on the reconstructed signals to obtain the physiological parameters of the target object. In this application embodiment, the target detection region is first cropped, aligned, and segmented to form multiple candidate physiological signals; historical physiological features are continuously stored and tracked using a temporal memory, and a temporal prediction model is constructed to provide temporal prior constraints. The model performs correlation matching between candidate signals and predicted signals in each sub-region, and performs weighted fusion of signals using similarity as the weight. This adaptively filters out low-quality signals and continuously tracks effective regions, significantly reducing the negative impacts of motion interference, illumination fluctuations, and local artifacts. The constrained and calibrated signals are then fed into the backbone network for reconstruction, outputting temporally continuous and rhythmically stable physiological signals. Frequency domain analysis is then used to accurately calculate parameters such as heart rate and respiratory rate. By fully utilizing historical temporal information, the model's dependence on labeled data is reduced, significantly improving the stability, accuracy, and robustness of physiological signal detection. It can maintain effective operation even with insufficient performance or failure, and can be stably applied to scenarios such as non-contact long-term health monitoring, ward care, and motion state vital sign detection.

[0032] In one embodiment of this application, the step of calling historical physiological feature signals stored in the time-series memory to constrain and correct the candidate physiological feature signals to obtain high-quality region signals includes: Based on the historical physiological feature signals, a time-series prediction model is constructed, which is used to characterize the continuous evolution relationship of physiological feature signals in the time dimension. Based on the time-series prediction model, the current physiological characteristic signal corresponding to the current time slice is predicted; The candidate physiological feature signals are matched with the current physiological feature signal to obtain a similarity score; Based on the similarity score, the high-quality region signal is determined.

[0033] Optionally, a temporal prediction model for physiological characteristics is constructed based on continuously stored historical physiological characteristic signals in a temporal memory bank. Specifically, it is assumed that in the current time slice... The previous recent Within each time slice, the stored historical physiological characteristic sequences are: The system then uses the historical features to determine the physiological characteristics of the current time slice. To make predictions and obtain current physiological characteristics This is used as a temporal prior constraint for the current physiological characteristics. After filtering and normalization, the predicted current physiological characteristics yield a predicted signal. Based on this, candidate physiological signals are extracted from each region in the current time slice. The current physiological characteristic signals obtained by prediction The system performs similarity matching and comparison to obtain a similarity score. Based on the similarity score, the system selects regions that meet the score and have excellent signal quality, which are then used as high-quality regions for subsequent processing.

[0034] The similarity score can be specifically expressed as: ; in, This represents the Pearson correlation coefficient. It is used to measure the consistency between the candidate signals in a region and the predicted current physiological characteristic signals in terms of temporal morphology and changing trends, thereby providing a basis for subsequent region selection and signal fusion.

[0035] In one embodiment of this application, the step of calling historical physiological feature signals stored in the time-series memory to constrain and correct the candidate physiological feature signals to obtain high-quality region signals includes: A baseline physiological feature signal is selected from the historical physiological feature signals as a template signal; The similarity score is obtained by matching each candidate physiological feature signal with the template signal using a template matching algorithm. Based on the similarity score, the high-quality region signal is determined.

[0036] Optionally, historical physiological feature signals stored over a long period can be retrieved from a time-series memory, and signals that reflect the normal physiological rhythms of the subject under test can be selected as benchmark template signals. Then, template matching algorithms (such as Pearson correlation coefficient, cosine similarity, mean square error, etc.) are used to compare each candidate physiological feature signal with the template signal point by point, calculating the degree of agreement in waveform and trend, and generating a corresponding similarity score. A higher score indicates a higher degree of consistency between the current sub-region signal and the standard physiological features, and a lower degree of interference. Based on the similarity score, the system selects regions with high scores and excellent signal quality, which are then used as high-quality regions for subsequent processing.

[0037] In one embodiment of this application, determining the high-quality region signal based on the similarity score includes: Based on the similarity score, the candidate physiological feature signal with the highest similarity score is selected as the high-quality region signal; or Based on the similarity score, the weights corresponding to each candidate physiological feature signal are determined; Based on the weights, the candidate physiological feature signals are weighted and fused to obtain the high-quality region signal.

[0038] Optionally, after calculating the similarity score, the similarity scores of all candidate physiological feature signals can be compared, and the signal with the highest score can be selected as the high-quality region signal. Alternatively, a fixed scoring threshold and selection number can be set to filter out K target candidate physiological feature signals with scores higher than the threshold; then, the weights corresponding to each target candidate physiological feature signal can be calculated based on the similarity scores of each signal. Specifically, it can be calculated using the following formula: ; in, It is a numerically stable term.

[0039] The weights corresponding to each candidate physiological feature are calculated sequentially using the above method. Then, the candidate physiological feature signals are weighted and fused based on these weights to obtain the correction signal. This refers to high-quality regional signals, as detailed below: ; in, Candidate physiological features extracted for each region.

[0040] Then, the correction signal can be The data is fed back to the backbone network for subsequent reconstruction and parameter prediction, outputting continuous, smooth, and physiologically realistic remote physiological signal waveforms and corresponding physiological parameters. The corrected physiological features output by the backbone network can be updated to the temporal memory bank in chronological order to achieve continuous tracking and dynamic correction.

[0041] In one embodiment of this application, the step of predicting parameters based on the reconstructed signal to obtain the physiological parameters of the object to be detected includes: The reconstructed signal is subjected to frequency domain transformation within a preset time window to obtain the signal spectrum distribution; Based on the signal spectrum distribution, the effective physiological frequency band is determined; Within the effective physiological frequency band, identify the dominant frequency component with the largest energy or amplitude; Based on the correspondence between the dominant frequency component and the physiological parameters, the physiological parameter values ​​corresponding to the object to be detected are calculated.

[0042] Optionally, after signal reconstruction, a stationary physiological signal within a fixed time window is extracted from the reconstructed signal. This extracted physiological signal is then converted into a frequency domain signal using a frequency domain transformation algorithm (such as Fourier transform) to obtain the complete signal spectrum distribution. Then, effective physiological frequency bands can be defined based on the normal fluctuation range of human physiological characteristics (such as heart rate and respiratory rate) to distinguish real physiological signals from environmental noise and clutter interference. Within the selected effective physiological frequency bands, all frequency points are traversed, and the energy and amplitude of each frequency component are compared one by one to identify the dominant frequency component with the highest peak value. This dominant frequency component corresponds to the true physiological rhythm frequency of the human body. Finally, calculations are performed based on the fixed conversion relationship between the dominant frequency and physiological parameters to determine the current values ​​of the subject's heart rate, respiratory rate, and other physiological parameters. For example, in heart rate prediction, the dominant frequency is mapped to the number of heartbeats per minute. If the dominant frequency is in Hz (beats / second), the heart rate (beats / minute) = dominant frequency × 60. If the dominant frequency = 1.2 Hz, the heart rate = 1.2 × 60 = 72 beats / minute. In respiratory rate prediction, the dominant frequency is mapped to the number of breaths per minute. If the respiratory rate (breaths / minute) = dominant frequency × 60, when the dominant frequency = 0.25 Hz, the respiratory rate = 0.25 × 60 = 15 breaths / minute. By performing frequency analysis on continuous and smooth physiological signals, the system can stably and accurately output the required physiological parameters.

[0043] It should be noted that the system can personalize the effective detection frequency bands and normal parameter ranges for heart rate and respiratory rate for different users based on their age, gender, skin color, body shape, and real-time movement status. For example, the baseline heart rate range for infants and young children is generally higher, while the heart rate and respiratory rate of the elderly are generally lower, and the ranges of human physiological indicators also shift significantly during exercise. The system can dynamically adjust the bandpass filter range and spectrum selection rules based on these characteristics. This avoids misjudgment of the main frequency and distortion of parameter detection caused by using a uniform standard, effectively improving the algorithm's adaptability to different populations and overall detection accuracy.

[0044] In one embodiment of this application, the target detection region is divided into several sub-regions, and candidate physiological feature signals corresponding to each sub-region are extracted, including: Based on human body geometric region division rules or equal grid division rules, the target detection region is divided into several sub-regions; Extract the corresponding initial candidate physiological feature signals from each sub-region; The mean of each initial candidate physiological feature signal is adjusted to zero through zero-mean processing; Bandpass filtering and normalization are performed on the zero-mean signal to obtain standard candidate physiological feature signals.

[0045] Optionally, the target detection area in the acquired video data can be sampled and preprocessed by region. Specifically, the corresponding region rules can be selected according to the actual application scenario. One is the human body geometric region division rule, which is based on the physiological structure characteristics of the human body. For example, taking the face as an example, the target detection area can be divided into independent key areas with rich blood circulation and high physiological signal quality, such as the forehead, left cheek, and right cheek, based on the physiological structure characteristics of the face. Another is the equal grid division rule, which uniformly cuts the complete target detection area into several grid sub-regions of the same size and non-overlapping. For example, the entire target detection area can be divided into 16 square sub-regions of the same size (4×4).

[0046] After region segmentation, the backbone model can be used to analyze the temporal variation information of pixels within each sub-region, simultaneously extracting multiple original candidate physiological feature signals. To eliminate the negative impacts of baseline offset, high and low amplitude differences, and environmental stray noise, each candidate physiological feature signal can be preprocessed using standardization. Specifically, zero-mean processing can be performed on the original candidate signals first, shifting the overall numerical distribution of the signal to zero, thus eliminating static baseline bias. Subsequently, bandpass filtering is applied to the zero-mean processed signal, retaining signals within a specified frequency range, such as 40–220 bpm, to filter out low-frequency motion noise and high-frequency image salt-and-pepper noise outside the frequency band. Finally, amplitude normalization is performed on the filtered signal, uniformly constraining all signal amplitudes to a fixed numerical range, weakening amplitude deviations caused by differences in illumination intensity and area between different sub-regions, ultimately obtaining standardized basic physiological feature signals with uniform format, low noise content, and significant rhythmic characteristics.

[0047] In this embodiment, the target detection region is first cropped, aligned, and segmented to form multiple candidate physiological signals. A temporal memory is used to continuously store and track historical physiological features, and a temporal prediction model is constructed to provide temporal prior constraints. The candidate signals and predicted signals of each sub-region are correlated and weighted by similarity to achieve signal weighted fusion. This adaptively filters out low-quality signals and continuously tracks effective regions, significantly reducing the negative impacts of motion interference, illumination fluctuations, and local artifacts. The constrained and corrected signals are then fed into the backbone network for reconstruction, outputting temporally continuous and rhythmically stable physiological signals. Frequency domain analysis is then used to accurately calculate parameters such as heart rate and respiratory rate. By fully utilizing historical temporal information, the model's dependence on labeled data is reduced, significantly improving the stability, accuracy, and robustness of physiological signal detection. It can maintain effective operation even with insufficient performance or failure, and can be stably applied to scenarios such as non-contact long-term health monitoring, ward care, and motion state vital sign detection.

[0048] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0049] In one embodiment, a remote physiological signal continuous and stable measurement system for complex dynamic scenarios is provided, which corresponds one-to-one with the remote physiological signal continuous and stable measurement method for complex dynamic scenarios described in the above embodiments. For example... Figure 3 As shown, the remote physiological signal continuous and stable measurement system in this complex dynamic scenario includes a video data acquisition unit 10, a candidate physiological feature signal extraction unit 20, a high-quality region signal screening unit 30, and a physiological parameter generation unit 40. Detailed descriptions of each functional module are as follows: Video data acquisition unit 10 is used to acquire video data of the object to be detected in the current time slice, wherein the video data includes the target detection area; The candidate physiological feature signal extraction unit 20 is used to divide the target detection area into several sub-regions and extract the candidate physiological feature signals corresponding to each sub-region. The high-quality region signal filtering unit 30 is used to call up historical physiological feature signals stored in the time-series memory bank, constrain and correct the candidate physiological feature signals, and obtain high-quality region signals. The physiological parameter generation unit 40 is used to input the high-quality region signal into the backbone network; to concatenate the high-quality region signal and the historical physiological feature signal according to the time dimension to construct a time sequence that integrates the temporal context; to perform multi-layer feature extraction and mapping transformation on the time sequence to obtain the reconstructed signal; and to perform parameter prediction based on the reconstructed signal to obtain the physiological parameters of the object to be detected.

[0050] In one embodiment of this application, the high-quality area signal filtering unit 30 is further configured to: Based on the historical physiological feature signals, a time-series prediction model is constructed, which is used to characterize the continuous evolution relationship of physiological feature signals in the time dimension. Based on the time-series prediction model, the current physiological characteristic signal corresponding to the current time slice is predicted; The candidate physiological feature signals are matched with the current physiological feature signal to obtain a similarity score; Based on the similarity score, the high-quality region signal is determined.

[0051] In one embodiment of this application, the high-quality area signal filtering unit 30 is further configured to: A baseline physiological feature signal is selected from the historical physiological feature signals as a template signal; The similarity score is obtained by matching each candidate physiological feature signal with the template signal using a template matching algorithm. Based on the similarity score, the high-quality region signal is determined.

[0052] In one embodiment of this application, the high-quality area signal filtering unit 30 is further configured to: Based on the similarity score, the candidate physiological feature signal with the highest similarity score is selected as the high-quality region signal; or Based on the similarity score, the weights corresponding to each candidate physiological feature signal are determined; Based on the weights, the candidate physiological feature signals are weighted and fused to obtain the high-quality region signal.

[0053] In one embodiment of this application, the physiological parameter generation unit 40 is further configured to: The reconstructed signal is subjected to frequency domain transformation within a preset time window to obtain the signal spectrum distribution; Based on the signal spectrum distribution, the effective physiological frequency band is determined; Within the effective physiological frequency band, identify the dominant frequency component with the largest energy or amplitude; Based on the correspondence between the dominant frequency component and the physiological parameters, the physiological parameter values ​​corresponding to the object to be detected are calculated.

[0054] In one embodiment of this application, the candidate physiological feature signal extraction unit 20 is further configured to: Based on human body geometric region division rules or equal grid division rules, the target detection region is divided into several sub-regions; Extract the corresponding initial candidate physiological feature signals from each sub-region; The mean of each initial candidate physiological feature signal is adjusted to zero by zero-mean processing; Bandpass filtering and normalization are performed on the zero-mean signal to obtain standard candidate physiological feature signals.

[0055] In this embodiment, the target detection region is first cropped, aligned, and segmented to form multiple candidate physiological signals. A temporal memory is used to continuously store and track historical physiological features, and a temporal prediction model is constructed to provide temporal prior constraints. The candidate signals and predicted signals of each sub-region are correlated and weighted by similarity to achieve signal weighted fusion. This adaptively filters out low-quality signals and continuously tracks effective regions, significantly reducing the negative impacts of motion interference, illumination fluctuations, and local artifacts. The constrained and corrected signals are then fed into the backbone network for reconstruction, outputting temporally continuous and rhythmically stable physiological signals. Frequency domain analysis is then used to accurately calculate parameters such as heart rate and respiratory rate. By fully utilizing historical temporal information, the model's dependence on labeled data is reduced, significantly improving the stability, accuracy, and robustness of physiological signal detection. It can maintain effective operation even with insufficient performance or failure, and can be stably applied to scenarios such as non-contact long-term health monitoring, ward care, and motion state vital sign detection.

[0056] Specific limitations regarding the remote physiological signal continuous and stable measurement system in complex dynamic scenarios can be found in the limitations of the remote physiological signal continuous and stable measurement method in complex dynamic scenarios described above, and will not be repeated here. Each module in the aforementioned remote physiological signal continuous and stable measurement system in complex dynamic scenarios can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0057] In one embodiment, a computer device is provided, which may be a terminal device, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes a readable storage medium storing computer-readable instructions. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer-readable instructions implement a method for continuous and stable remote measurement of physiological signals in complex dynamic scenarios. The readable storage medium provided in this embodiment includes both non-volatile and volatile readable storage media.

[0058] In this application embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor. When the processor executes the computer-readable instructions, it implements the steps of the remote physiological signal continuous and stable measurement method under complex dynamic scenarios as described above.

[0059] In this embodiment of the application, a readable storage medium is provided, which stores computer-readable instructions. When the computer-readable instructions are executed by a processor, they implement the steps of the remote physiological signal continuous and stable measurement method in the complex dynamic scenario described above.

[0060] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a non-volatile readable storage medium or a volatile readable storage medium. When executed, these computer-readable instructions can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0061] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0062] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for continuous and stable remote measurement of physiological signals in complex dynamic scenarios, characterized in that, The method includes: Acquire video data of the object to be detected within the current time slice, wherein the video data includes the target detection area; The target detection region is divided into several sub-regions, and candidate physiological feature signals corresponding to each sub-region are extracted, including: dividing the target detection region into several sub-regions based on human geometric region division rules or equal grid division rules; extracting corresponding initial candidate physiological feature signals from each sub-region; adjusting the mean of each initial candidate physiological feature signal to zero through zero-mean processing; performing bandpass filtering and normalization processing on the zero-mean processed signal to obtain standard candidate physiological feature signals; By retrieving historical physiological feature signals stored in the time-series memory, and constraining and correcting the candidate physiological feature signals, high-quality region signals are obtained, including: Based on the historical physiological feature signals, a time-series prediction model is constructed to characterize the continuous evolution of physiological feature signals over time. Based on the time-series prediction model, the current physiological feature signal corresponding to the current time slice is predicted. Each candidate physiological feature signal is matched with the current physiological feature signal to obtain a similarity score. Based on the similarity score, the high-quality region signal is determined. A baseline physiological feature signal is selected from the historical physiological feature signals as a template signal; a template matching algorithm is used to perform similarity matching between each candidate physiological feature signal and the template signal to obtain a similarity score; based on the similarity score, the high-quality region signal is determined. The high-quality regional signal is input into the backbone network; the high-quality regional signal and the historical physiological feature signal are concatenated according to the time dimension to construct a time sequence that integrates temporal context; multi-level feature extraction and mapping transformation are performed on the time sequence to obtain the reconstructed signal, and parameter prediction is performed based on the reconstructed signal to obtain the physiological parameters of the object to be detected.

2. The method for continuous and stable remote physiological signal measurement in complex dynamic scenarios as described in claim 1, characterized in that, Based on the similarity score, the high-quality region signal is determined, including: Based on the similarity score, the candidate physiological feature signal with the highest similarity score is selected as the high-quality region signal; or Based on the similarity score, the weights corresponding to each candidate physiological feature signal are determined; Based on the weights, the candidate physiological feature signals are weighted and fused to obtain the high-quality region signal.

3. The method for continuous and stable remote physiological signal measurement in complex dynamic scenarios as described in claim 1, characterized in that, The parameter prediction based on the reconstructed signal to obtain the physiological parameters of the object to be detected includes: The reconstructed signal is subjected to frequency domain transformation within a preset time window to obtain the signal spectrum distribution. Based on the signal spectrum distribution, the effective physiological frequency band is determined; Within the effective physiological frequency band, identify the dominant frequency component with the largest energy or amplitude; Based on the correspondence between the dominant frequency component and the physiological parameters, the physiological parameter values ​​corresponding to the object to be detected are calculated.

4. A remote, continuous, and stable measurement system for physiological signals in complex dynamic scenarios, characterized in that, The system, used to implement the method for continuous and stable remote physiological signal measurement in complex dynamic scenarios as described in any one of claims 1 to 3, comprises: A video data acquisition unit is used to acquire video data of the object to be detected within the current time slice, wherein the video data includes the target detection area; The candidate physiological feature signal extraction unit is used to divide the target detection area into several sub-regions and extract the candidate physiological feature signals corresponding to each sub-region. A high-quality region signal filtering unit is used to call up historical physiological feature signals stored in the time-series memory bank, constrain and correct the candidate physiological feature signals, and obtain high-quality region signals. The physiological parameter generation unit is used to input the high-quality regional signal into the backbone network; to concatenate the high-quality regional signal with the historical physiological feature signal according to the time dimension to construct a time sequence that integrates the temporal context; to perform multi-layer feature extraction and mapping transformation on the time sequence to obtain the reconstructed signal; and to perform parameter prediction based on the reconstructed signal to obtain the physiological parameters of the object to be detected.

5. The remote physiological signal continuous and stable measurement system in complex dynamic scenarios as described in claim 4, characterized in that, The physiological parameter generation unit is also used for: The reconstructed signal is subjected to frequency domain transformation within a preset time window to obtain the signal spectrum distribution. Based on the signal spectrum distribution, the effective physiological frequency band is determined; Within the effective physiological frequency band, identify the dominant frequency component with the largest energy or amplitude; Based on the correspondence between the dominant frequency component and the physiological parameters, the physiological parameter values ​​corresponding to the object to be detected are calculated.

6. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and running on the processor, characterized in that, When the processor executes the computer-readable instructions, it implements the method for continuous and stable remote physiological signal measurement in complex dynamic scenarios as described in any one of claims 1 to 3.

7. A readable storage medium having computer-readable instructions stored thereon, characterized in that, When the computer-readable instructions are executed by the processor, they implement the method for continuous and stable remote physiological signal measurement in complex dynamic scenarios as described in any one of claims 1 to 3.

Images