Pulse wave extraction method and system based on video space phase

Abstract

The invention discloses a pulse wave extraction method and system based on a video space phase. The method comprises the following steps: selecting an optimal sub-region from a target object region as a region of interest according to image quality; performing complex controllable pyramid decomposition on the video sequence of the region of interest; extracting phase information according to a preset target vibration direction, and calculating a phase difference between each frame of each spatial position in the region of interest and the reference frame; combining the phase difference information on the time sequence to form a two-dimensional phase difference matrix; performing band-pass filtering processing on the phase difference matrix, and decomposing the filtered matrix into a low-rank matrix L and a sparse matrix S by using a robust principal component analysis algorithm; phase information is extracted from the low-rank matrix L, pulse wave signals are reconstructed, frequency domain transformation processing is conducted on the reconstructed pulse wave signals, and the pulse waveform and the heart rate of the user are obtained; aiming at poor stability of a non-contact pulse wave extraction method, the non-contact pulse wave extraction method improves the robustness and accuracy of non-contact pulse wave extraction.

Description

Technical Field

[0001] This application relates to the field of signal processing, and more specifically, to a method and system for extracting pulse waves based on video spatial phase. Background Technology

[0002] Non-contact pulse wave extraction technology, as a method of acquiring physiological signals without skin contact, has significant application value in health monitoring, smart wearable devices, and telemedicine. Traditional physiological signal acquisition methods, such as electrocardiography (ECG) or photoplethysmography (PPG), require attaching sensors that come into direct contact with the human body, causing discomfort and limiting continuous monitoring scenarios. In contrast, non-contact pulse wave extraction technology based on camera video analyzes subtle changes on the human skin surface, enabling non-invasive, imperceptible, and long-term pulse wave monitoring, offering significant convenience and application potential.

[0003] Existing non-contact pulse wave extraction methods primarily rely on the optical properties of skin brightness or color changes with blood volume in video images, estimating pulse information through time-domain or frequency-domain analysis of the video signal. However, these methods are highly sensitive to changes in the external environment and are easily affected by factors such as lighting variations, camera shake, and involuntary human movements in practical applications, leading to decreased stability and reliability of the extracted pulse signal. To improve the robustness of non-contact pulse wave extraction, some studies have attempted to suppress interference through methods such as region selection, signal decomposition, or motion compensation. However, these methods still mainly depend on brightness or color changes to characterize vascular pulsation. When strong motion interference or complex background changes are present, motion artifacts often overlap with the effective pulse signal in terms of frequency and amplitude, making it difficult for traditional filtering or separation techniques to effectively distinguish between the real pulse component and the interference signal.

[0004] Chinese invention patent application CN2019100467238 discloses a non-contact radial artery waveform detection method, which mainly solves the problems of accuracy and robustness in non-contact extraction of radial artery pulse waves at the wrist in real-world scenarios. However, the signal extraction method based on brightness has poor robustness and cannot effectively separate noise from the target blood vessel pulsation. Therefore, in the process of video-based vibration analysis, how to accurately characterize the minute vibrations caused by blood vessel pulsation and achieve stable extraction of effective pulse signals under multi-source interference conditions remains an urgent technical problem to be solved in this field. Summary of the Invention

[0005] To address the poor stability of pulse signals extracted by existing non-contact pulse wave extraction methods based on brightness or color changes, this application provides a pulse wave extraction method and system based on video spatial phase, which improves the robustness and accuracy of non-contact pulse wave extraction.

[0006] One aspect of this application provides a pulse wave extraction method based on video spatial phase, comprising: S1, acquiring a continuous video sequence of a target object region, selecting a frame as a reference frame, and selecting an optimal sub-region as a region of interest (ROI) from the target object region based on image quality; S2, performing complex controllable pyramid decomposition on the video sequence of the ROI to obtain the amplitude and phase information of the video frames; extracting phase information according to a preset target vibration direction, calculating the phase difference between each frame at each spatial location within the ROI and the reference frame to obtain phase difference information in the time series; combining the phase difference information in the time series to form a two-dimensional phase difference matrix, wherein the matrix rows represent time frames and the columns represent spatial pixels; S3, performing bandpass filtering on the phase difference matrix, and using a robust principal component analysis algorithm to decompose the filtered matrix into a low-rank matrix L and a sparse matrix S; S4, extracting phase information from the low-rank matrix L and reconstructing a pulse wave signal, and performing frequency domain transformation on the reconstructed pulse wave signal to obtain the user's pulse waveform and heart rate.

[0007] Furthermore, based on image quality, the optimal sub-region is selected as the region of interest from the target object region, including: acquiring a continuous RGB video sequence of the radial artery region of the target object's wrist. Where x and y are spatial coordinates, and t is time; RGB video sequence Each frame is converted to a luminance-chrominance space, and the luminance component is extracted as the input image sequence. ; Input image sequence Divided into N x M sub-regions Extract each sub-region The brightness time series of each pixel is processed by temporal bandpass filtering, and the signal-to-noise ratio (SNR) of each sub-region is calculated. The sub-region with the highest average SNR over K consecutive frames is selected as the region of interest.

[0008] Furthermore, the signal-to-noise ratio (SNR) of each sub-region after filtering is calculated using the following formula: ;in, This represents the maximum signal power of the filtered brightness time series within a preset physiological frequency range; This represents the average noise power of the filtered brightness time series outside the preset physiological frequency range.

[0009] Furthermore, constructing a two-dimensional phase difference matrix includes: performing multi-scale, multi-directional complex controllable pyramid decomposition on the video sequence of the region of interest in the spatial dimension to obtain the complex response at the l-th scale and in the k-th direction; extracting the phase component from the complex response; determining the target analysis direction based on the preset target vibration direction and obtaining the phase information of the target analysis direction; calculating the phase difference between each spatial position in the region of interest and the reference frame in each frame, using the phase of the reference frame as a reference; and arranging the phase differences in time and space to construct a two-dimensional phase difference matrix.

[0010] Furthermore, the target analysis direction is determined based on the preset target vibration direction, including: performing complex controllable pyramid decomposition on the first frame of the video sequence to extract the contour of the target object; selecting two calibration points A and B on the contour, calculating the angle θ between the line connecting calibration points A and B and the preset coordinate axis, and using it as the angle of the main axis of the blood vessel; among the K directions obtained from the complex controllable pyramid decomposition, calculating the angle difference between each direction angle and θ+90° and the angle difference between each direction angle and θ-90°, and selecting the direction with the smallest angle difference as the target analysis direction.

[0011] Furthermore, in the bandpass filtering of the phase difference matrix, the frequency range is 0.8 to 2 Hz; an infinite impulse response (IIR) filter is used for filtering.

[0012] Furthermore, the optimization objective of the robust principal component analysis algorithm is: ;in, The nuclear norm is used to constrain low-rank properties. Let denote the (1,1) norm, used to enhance sparsity, and λ be the weighting parameter used to balance the separation effect of low-rank matrices and sparse matrices.

[0013] Another aspect of this application provides a pulse wave extraction system based on video spatial phase, comprising: a preprocessing module for acquiring a continuous video sequence of a target region, selecting a reference frame, and selecting an optimal sub-region based on image quality indicators; a matrix construction module for performing complex controllable pyramid directional decomposition on the optimal sub-region, directionally extracting phase information, calculating the phase difference, and constructing a phase difference matrix; a signal separation module for performing bandpass filtering and robust principal component analysis decomposition on the phase difference matrix to obtain a low-rank matrix and a sparse matrix; and a reconstruction module for extracting the phase signal from the low-rank matrix and reconstructing the pulse wave signal, and obtaining the heart rate through frequency domain transformation.

[0014] Compared to existing technologies, the advantages of this application are: To address the problem that existing non-contact pulse wave extraction methods based on brightness or color changes suffer from aliasing of motion artifacts and effective pulse signals in frequency and amplitude when interference such as lighting changes, camera shake, or involuntary human movement occurs, resulting in the inability to effectively separate noise from the target blood vessel pulsation signal and thus reducing the stability and reliability of the extracted pulse signal, this application provides a pulse wave extraction method based on video spatial phase. By utilizing complex controllable pyramid decomposition to extract phase information from video frames, the phase difference of the target vibration direction is extracted based on the principal axis direction of the blood vessel, and a robust principal component analysis algorithm is used to decompose the phase difference matrix into a low-rank matrix representing the pulse signal and a sparse matrix representing the interference noise. This achieves stable extraction of the effective pulse signal under multi-source interference conditions, improving the robustness and accuracy of non-contact pulse wave extraction. Attached Figure Description

[0015] Figure 1 A technical roadmap provided for embodiments of this application; Figure 2 A flowchart of a method for extracting radial artery waveforms based on video spatial phase provided in this application embodiment; Figure 3 This is a schematic diagram of the vascular-directed decomposition method provided in this embodiment. Detailed Implementation

[0016] The present application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0017] Example 1 A method and system for extracting pulse waves based on video spatial phase, which achieves efficient separation of pulse signals and motion noise by converting motion information to the phase domain and combining it with matrix factorization techniques, is characterized by the following steps: S1, Video Acquisition and Preprocessing: The target for acquisition can be the radial artery of the wrist. After obtaining a continuous video sequence of the target area, one frame is selected as the reference frame, and the optimal sub-region is selected based on image quality indicators (such as signal-to-noise ratio or motion intensity) to reduce background noise interference from non-target areas.

[0018] Further, in S2, phase signal extraction and phase difference matrix construction are performed: First, the video sequence of the optimal sub-region selected in S1 is subjected to complex controllable pyramid directional decomposition to obtain the local amplitude and phase information of the video frames; second, phase information is extracted directionally based on the spatial characteristics of the vascular pulsation direction, and the phase difference between each frame of each sub-region and the reference frame is calculated to obtain the phase difference signal in the time series; the phase difference signals in the time series are combined to construct a two-dimensional data matrix. In this matrix, rows represent time frames and columns represent spatial pixels.

[0019] Further, in S3, phase-robust principal component analysis and signal separation are performed: First, the phase difference matrix obtained in S2 is analyzed... A bandpass filter matching the human heart rate is applied, followed by Robust Principal Component Analysis (RPCA) to decompose the matrix into low-rank matrices. sparse matrix : At this point, the low-rank matrix This corresponds to the tiny, stable, periodic pulse vibration signal in the video, a sparse matrix. This corresponds to noise generated by large-amplitude, random camera shake or human movement.

[0020] Further, in S4, the pulse wave is reconstructed and processed: as shown in S3, from the low-rank matrix... The phase signal is extracted and reconstructed into a preliminary pulse wave signal. The reconstructed pulse wave signal is then subjected to post-processing such as Fast Fourier Transform to finally obtain a high-precision, interference-resistant pulse waveform and accurate heart rate.

[0021] Compared to existing technologies, the advantages of this application are: Achieving non-contact, continuous monitoring: This application utilizes only ordinary or high-definition cameras to capture video of the target area, eliminating the need to attach any electrodes to the subject's skin, wear sensors, or use contact medical devices. This fundamentally eliminates the discomfort, skin irritation, poor compliance, and risk of cross-infection associated with traditional contact measurement methods.

[0022] Significantly improved robustness against motion interference: The introduction of robust principal component analysis (RPCA) enables efficient separation of low-rank components representing periodic physiological signals from sparse components representing large-amplitude, random motion noise in the phase domain. This core technological innovation solves the technical challenge of low signal-to-noise ratio of pulse signals in the presence of rigid or non-rigid motion noise using traditional non-contact methods.

[0023] High accuracy and stability of signal extraction: The phase signal is extracted and processed using a complex controllable pyramid. This phase signal is inherently insensitive to changes in ambient light intensity and color, avoiding the problem of amplitude-based signals being easily interfered with by light fluctuations, thus ensuring the accuracy and stability of signal extraction.

[0024] Improved computational efficiency and accuracy: By combining the optimal sub-region (ROI) selection technique in the preprocessing stage, the region with the strongest radial artery micro-vibration signal is effectively focused, reducing the computation of unnecessary pixels and improving the real-time performance and resource utilization efficiency of the overall algorithm.

[0025] Example 2 refer to Figure 1 and Figure 2 As shown in the figure, this application provides a method and system for pulse wave extraction based on video spatial phase, which includes four main modules: 1) video acquisition and preprocessing module; 2) complex controllable pyramid decomposition and phase difference matrix construction module; 3) anti-interference noise filtering module; and 4) pulse wave signal reconstruction module.

[0026] The video acquisition and preprocessing module consists of three steps: First, video sequence acquisition is performed, with the image acquisition unit acquiring a continuous RGB video sequence of the radial artery region of the target object's wrist. ,in For spatial coordinates, For time.

[0027] Secondly, color space processing is performed, converting the RGB video frames into a luminance-chrominance space. Typically, the luminance component Y or a linear combination thereof is selected as the input image sequence. This is to ensure the stability of subsequent phase extraction.

[0028] Specifically, color space processing aims to separate image luminance information (Y component) from chrominance information. Since subsequent phase decomposition algorithms primarily rely on luminance changes caused by minute pixel displacements, and the chrominance channel is prone to noise from ambient lighting fluctuations, this application focuses on using the Y component, i.e., the luminance information channel, as the input image sequence for subsequent signal processing. .

[0029] Finally, the optimal ROI is divided and selected to improve the signal-to-noise ratio of the pulse wave signal and reduce the interference of background noise.

[0030] Specifically, the image sequence processed by the brightness channel is first... Divided into N*M overlapping or non-overlapping local sub-regions Secondly, for each sub-region First, the brightness signals of all pixels within the region are subjected to time-domain bandpass filtering. After filtering, the signal quality index of each sub-region is calculated. In this embodiment, the signal-to-noise ratio (SNR) is preferably used as the evaluation index.

[0031] Specifically, SNR is calculated based on power spectral density (PSD) analysis of the signal. For each sub-region, the power spectrum of its average time-domain signal is calculated within the physiological frequency range. Find the maximum peak value Simultaneously, calculate the average noise power in the non-physiological frequency bands surrounding this frequency. .

[0032]

[0033] Finally, the sub-region with the highest average SNR value within K consecutive frames is selected. This is considered the optimal ROI R. Theoretically, this optimal ROI R contains the signal with the strongest radial artery vibration and the least interference from environmental noise and motion artifacts.

[0034] The complex controllable pyramid decomposition and phase difference matrix construction module further performs phase-based vibration analysis on the video sequence of the target region to extract minute vibration information caused by vascular pulsation. This process mainly includes complex controllable pyramid decomposition, phase signal extraction, vascular target direction constraint, and phase difference matrix construction.

[0035] Among them, let the optimal sub-region R video sequence obtained by module 1 be: In this embodiment, the video sequence undergoes multi-scale, multi-directional complex controllable pyramid decomposition in the spatial dimension, mapping the original video signal to a set of directionally selective complex subband responses. The decomposed... The first scale, the first The complex response in each direction can be expressed as:

[0036] in, This represents the amplitude components at the corresponding scale and direction. Indicates the corresponding phase component, The imaginary unit is represented by the above decomposition process, which allows for the separation and representation of minute structural vibrations at different spatial scales and in different directions in the video, providing a foundation for subsequent phase-based motion analysis.

[0037] Furthermore, phase information is extracted from the complex response. Phase component This is used to describe the minute displacement changes of local structures in a video over time. Based on the linear approximation relationship between phase and displacement, under conditions of minute motion, phase change can be regarded as a displacement representation in the corresponding direction. Therefore, in this embodiment, the phase signal is selected as the main representation of vascular pulsation vibration to reduce the influence of illumination changes and intensity fluctuations on the analysis results.

[0038] Furthermore, considering that the radial artery at the wrist has a relatively stable spatial orientation in the image, and that skin vibrations caused by vascular pulsation are mainly generated along a direction perpendicular to the vascular orientation, this embodiment introduces a target direction constraint mechanism in the multi-directional phase components. Specifically, based on the spatial distribution characteristics of blood vessels within the target area, the principal axis direction of the blood vessels is determined, and from the multiple directional sub-bands obtained by complex controllable pyramid decomposition, a direction perpendicular or approximately perpendicular to the principal axis direction of the blood vessels is selected. As the direction of the target vibration analysis, such as Figure 3As shown. The corresponding phase component is represented as: This directional constraint process can effectively suppress interference components introduced by overall motion or vibrations in directions unrelated to blood vessels, thereby enhancing the phase change characteristics related to vascular pulsation.

[0039] Specifically, the wrist contour is extracted from the first frame of the video through complex controllable pyramid processing. Calibration points A and B are selected on the contour. The direction of the straight line where A and B are located is the direction of the main axis of the blood vessel, and the direction perpendicular to the straight line where A and B are located is the direction of the filter vibration analysis.

[0040] After obtaining the phase signal in the target direction, this embodiment further constructs a phase difference representation to eliminate static phase bias and highlight time-varying characteristics. Let the phase corresponding to the t-th frame be... Select the initial frame or reference frame. Using the phase as a reference, the phase difference is defined as:

[0041] The aforementioned phase difference represents the phase change relative to the reference frame, directly corresponding to the minute displacement changes of the local structure in the time dimension. Furthermore, the phase difference signals at all spatial locations within the target region are arranged in the time dimension to form a phase difference matrix:

[0042] Where N represents the number of pixels or sub-regions selected within the target area. The phase difference matrix serves as the input data structure for subsequent anti-interference noise filtering and pulse wave signal decomposition modules, enabling robust extraction of vascular pulsation signals.

[0043] After obtaining the phase difference matrix under the target direction constraint, to further suppress interference introduced by non-pulse factors such as camera shake, overall limb movement, and environmental disturbances, this embodiment performs anti-interference noise filtering on the phase difference matrix to effectively separate vascular pulsation-related signals from interference components. First, after extracting the sub-region phase difference signal, time-domain bandpass filtering is performed, setting the frequency range to the human heart rate range (0.8-2 Hz), and using an Infinite Impulse Response (IIR) filter.

[0044] Specifically, during the heartbeat and circulation of blood throughout the body, the heart's contraction causes a sudden surge of blood to fill the capillaries in the wrist, while the heart's expansion reduces the blood flow in these capillaries. Under undisturbed conditions, the heartbeat causes uniform vibration intensity in the wrist skin. Theoretically, the waveforms of the pulse signals from each sub-module should be identical. Therefore, arranging the pulse signals from each sub-module into a matrix row-wise would theoretically result in a low-rank matrix. However, sudden involuntary muscle vibrations in the radial artery region of the wrist can also cause vibrations in the wrist skin, resulting in abrupt pulse changes in the pulse signal of a particular sub-module. This sudden interference increases the rank of the entire pulse matrix. Based on these vascular pulsation characteristics, this embodiment decomposes the bandpass-filtered phase difference matrix into a low-rank matrix and a sparse matrix.

[0045]

[0046] in, It is a low-rank matrix used to represent phase component changes related to vascular pulsation; It is a sparse matrix used to represent overall motion and background interference components, including but not limited to small camera shakes, slow wrist posture changes, and globally consistent motion.

[0047] Furthermore, in order to separate the two motions and obtain a more accurate pulse wave signal, this embodiment uses robust principal component analysis (PCA) to decompose the matrix to solve this problem.

[0048]

[0049] in, This indicates the calculation of the nuclear norm, used to constrain low-rank properties. This indicates the (1, 1) norm, used to enhance sparsity. The weighting parameter is used to balance the separation effect of the two types of components. This method can achieve robust separation of the target signal in the presence of outliers and non-Gaussian noise.

[0050] Specifically, this embodiment uses the augmented Lagrange multiplier method to solve the above problem, and constructs the augmented Lagrange function as follows:

[0051] in, It is a multiplier. This indicates that the dot product of the two is being sought. It is a positive parameter. Used to calculate ( The element scores and the square root of the total are then used to iterate the matrices A and E alternately using the exact Lagrange multiplier method until the termination condition is met.

[0052]

[0053]

[0054] make , They converge to , Then the updated matrix formula is:

[0055] Further, update parameters We can obtain:

[0056] in, >1 is a constant. It is a number that is greater than 0 but approaches 0.

[0057] Furthermore, after completing the robust decomposition, this embodiment will... This matrix represents the phase difference signal related to vascular pulsation. The signals contained in this matrix exhibit stable periodic variations over time, consistent with the physiological rhythm of the pulse wave. After the aforementioned anti-interference noise filtering process, the obtained pulse-related phase signal will serve as the input for the pulse wave signal reconstruction in the next module.

[0058] The low-rank matrix obtained by robust principal component analysis decomposition The pulse wave signal represents the phase change associated with the pulse wave signal. The pulse wave vectors in each row are summed and averaged; the final result is the pulse wave. To further calculate the heart rate and verify the accuracy of the pulse wave, this embodiment requires integrating the phase difference signal... Converted into time-domain pulse wave signal ,

[0059] At this time, the time domain signal It represents the minute vibrations caused by vascular pulsation and is the time-domain representation of the pulse wave signal.

[0060] Furthermore, in order to extract the periodic characteristics of the pulse wave signal, this embodiment uses frequency domain analysis to calculate the main frequency of the pulse wave signal, and then derives physiological parameters such as heart rate.

[0061] Specifically, the first pair of time-domain pulse wave signals Frequency domain transformation using Fast Fourier Transform:

[0062] in, Indicates Fourier transform, This represents the frequency domain pulse wave signal.

[0063] Furthermore, from the frequency domain signal In the middle, select the main frequency This frequency corresponds to the frequency of the pulse wave.

[0064] Heart rate It can be calculated using the following formula:

[0065] in, The dominant frequency in the frequency domain signal is represented by Hz, and the heart rate is also mentioned. That is, the number of heartbeats per minute.

[0066] Through this step, this embodiment realizes non-contact pulse wave extraction and heart rate calculation based on video.

[0067] The foregoing illustrative description of the present application and its embodiments is not restrictive and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. The accompanying drawings are only one embodiment of the present application, and the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present application, such designs should fall within the scope of protection of this application. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for extracting pulse waves based on video spatial phase, characterized in that, include: S1, acquire a continuous video sequence of the target object region, select a frame as a reference frame, and select the optimal sub-region as the region of interest from the target object region based on image quality; S2, perform complex controllable pyramid decomposition on the video sequence of the region of interest to obtain the amplitude and phase information of the video frames; Phase information is extracted based on the preset target vibration direction, and the phase difference between each frame and the reference frame at each spatial location within the region of interest is calculated to obtain phase difference information in the time series. The phase difference information in the time series is combined to form a two-dimensional phase difference matrix, where the rows of the matrix represent time frames and the columns represent spatial pixels; S3, bandpass filtering is performed on the phase difference matrix, and the filtered matrix is ​​decomposed into a low-rank matrix L and a sparse matrix S using the robust principal component analysis algorithm. S4 extracts phase information from the low-rank matrix L and reconstructs the pulse wave signal. The reconstructed pulse wave signal is then subjected to frequency domain transformation to obtain the user's pulse waveform and heart rate.

2. The pulse wave extraction method based on video spatial phase according to claim 1, characterized in that: Based on image quality, the optimal sub-region is selected as the region of interest from the target object region, including: Obtain a continuous RGB video sequence of the radial artery region of the target object's wrist. Where x and y are spatial coordinates, and t is time; RGB video sequence Each frame is converted to a luminance-chrominance space, and the luminance component is extracted as the input image sequence. ; Input image sequence Divided into N x M sub-regions Extract each sub-region The brightness time series of each pixel is processed by temporal bandpass filtering, and the signal-to-noise ratio (SNR) of each sub-region is calculated after filtering. The region of interest is the sub-region with the highest average SNR over K consecutive frames.

3. The pulse wave extraction method based on video spatial phase according to claim 2, characterized in that: The signal-to-noise ratio (SNR) of each sub-region after filtering is calculated using the following formula: ;in, This represents the maximum signal power of the filtered brightness time series within a preset physiological frequency range; This represents the average noise power of the filtered brightness time series outside the preset physiological frequency range.

4. The pulse wave extraction method based on video spatial phase according to claim 2, characterized in that: The two-dimensional phase difference matrix consists of: The video sequence of the region of interest is decomposed into a complex controllable pyramid in multiple scales and directions in the spatial dimension to obtain the complex response at the l-th scale and in the k-th direction; Extracting the phase component from the complex response; The target analysis direction is determined based on the preset target vibration direction, and the phase information of the target analysis direction is obtained; Using the phase of the reference frame as a reference, calculate the phase difference between each spatial location within the region of interest and the reference frame in each frame; The phase differences are arranged in time and space to form a two-dimensional phase difference matrix.

5. The pulse wave extraction method based on video spatial phase according to claim 4, characterized in that: Determine the target analysis direction based on the preset target vibration direction, including: Perform complex controllable pyramid decomposition on the first frame of the video sequence to extract the outline of the target object; Select two calibration points A and B on the contour, and calculate the angle θ between the line connecting calibration points A and B and the preset coordinate axis, which is taken as the angle of the main axis of the blood vessel. In the K directions obtained from the complex controllable pyramid decomposition, the angle difference between each direction and θ+90° and the angle difference between each direction and θ-90° are calculated, and the direction with the smallest angle difference is selected as the target analysis direction.

6. The pulse wave extraction method based on video spatial phase according to any one of claims 1 to 5, characterized in that: In the bandpass filtering process of the phase difference matrix, the frequency range is 0.8 to 2 Hz; An infinite impulse response (IIR) filter is used for filtering.

7. The pulse wave extraction method based on video spatial phase according to claim 6, characterized in that: The optimization objective of the robust principal component analysis algorithm is: ； in, The nuclear norm is used to constrain low-rank properties. Let denote the (1,1) norm, used to enhance sparsity, and λ be the weighting parameter used to balance the separation effect of low-rank matrices and sparse matrices.

8. A pulse wave extraction system based on video spatial phase, characterized in that, include: The preprocessing module acquires a continuous video sequence of the target object region, selects reference frames, and selects the optimal sub-region based on image quality indicators; The matrix construction module performs complex controllable pyramid directional decomposition on the optimal sub-region, extracts phase information in a directional manner, calculates the phase difference, and constructs the phase difference matrix. The signal separation module performs bandpass filtering and robust principal component analysis decomposition on the phase difference matrix to obtain a low-rank matrix and a sparse matrix. The reconstruction module extracts the phase signal from the low-rank matrix and reconstructs the pulse wave signal, obtaining the heart rate through frequency domain transformation.

Images