Millimeter wave radar signal processing method and system for sleep posture monitoring

Abstract

The invention discloses a millimeter-wave radar signal processing method and system for sleep posture monitoring, and relates to the technical field of millimeter-wave radar wireless sensing, and the method comprises the steps: collecting a background signal when there is no person and an echo signal when there is a person through a millimeter-wave radar, and obtaining a human body self-reflection signal; performing distance dimension and Doppler dimension processing on each frame of signal to obtain a distance-Doppler spectrogram; extracting a high-resolution range profile feature and a micro-Doppler spectrum feature of each frame, and constructing a fusion feature vector; setting an attitude stability threshold to primarily screen the fusion feature vector, and eliminating attitude conversion interference data; constructing a similarity coefficient matrix corresponding to the reserved fusion feature vector; and setting a clustering threshold, classifying the similar coefficient matrix through a density clustering algorithm, and outputting a posture recognition result. According to the invention, non-contact, high-precision and all-weather sleep posture monitoring is realized through double-feature fusion and a double-anti-interference mechanism.

Description

Technical Field

[0001] This invention belongs to the field of millimeter-wave radar wireless sensing technology, specifically relating to a millimeter-wave radar signal processing method and system for sleep posture monitoring. Background Technology

[0002] Sleep posture is a core indicator for assessing sleep quality and health risks. For example, prolonged prone sleeping may lead to breathing difficulties, while frequent side sleeping may be associated with cervical and lumbar spine diseases. Accurate and non-disturbing monitoring of sleep posture is of great significance in scenarios such as the diagnosis of sleep apnea syndrome and the prevention of pressure ulcers.

[0003] Existing sleep posture monitoring technologies are mainly divided into two categories: one is contact monitoring, such as using devices like smart bracelets or chest straps to collect human motion data and determine posture through accelerometers. However, these devices need to be in direct contact with the human body, which can easily cause discomfort, affect sleep quality and the accuracy of monitoring data. Furthermore, turning over at night may cause the device to fall off, resulting in monitoring interruption. The other category is non-contact monitoring, such as camera-based visual recognition technology. Although it does not require contact with the human body, its monitoring accuracy drops significantly in dark environments at night, and there is a risk of privacy leakage. In addition, some monitoring devices based on ultrasound and infrared are greatly affected by ambient temperature and obstacles, making it difficult to achieve continuous and stable posture recognition.

[0004] Millimeter-wave radar technology offers a new approach to sleep monitoring due to its advantages such as non-contact operation, resistance to light interference, strong penetration, and no risk of privacy breaches. It achieves posture perception by emitting electromagnetic waves and utilizing the spatial and motion characteristics of the reflected echoes from the human body, without requiring direct contact and penetrating obstructions such as bedding. Existing millimeter-wave radar-based monitoring methods primarily focus on extracting vital signs such as respiration and heart rate, with fewer solutions specifically for sleep posture monitoring. Furthermore, these methods are susceptible to dynamic interference such as turning over during long-term monitoring. Therefore, there is an urgent need for a sleep posture monitoring method that can integrate multi-dimensional features and has strong anti-interference capabilities to meet the precise needs of clinical and home health monitoring. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, this invention provides a millimeter-wave radar signal processing method and system for sleep posture monitoring. The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides a millimeter-wave radar signal processing method for sleep posture monitoring, comprising: Step 1: using millimeter-wave radar to collect background signals of the area to be monitored in an unmanned scene and echo signals in a scene with people sleeping, and obtaining the human body's own reflection signal based on the echo signals and the background signals; Step 2: performing range-dimensional and Doppler-dimensional processing on each frame of the human body's own reflection signal to obtain the range-Doppler spectrum corresponding to each frame; Step 3: extracting the range image feature vector and the micro-Doppler spectrum feature vector based on the range-Doppler spectrum corresponding to each frame, and constructing the fusion feature vector corresponding to each frame; Step 4: performing initial screening on the fusion feature vector according to a preset posture stability threshold, retaining the fusion feature vector corresponding to stable posture data segments, and removing the fusion feature vector corresponding to posture transition interference data; Step 5: constructing a similarity coefficient matrix corresponding to all fusion feature vectors retained after initial screening; Step 6: classifying the similarity coefficient matrix according to a preset clustering threshold using a density clustering algorithm, and outputting the final sleep posture recognition result.

[0006] This invention also provides a millimeter-wave radar signal processing system for sleep posture monitoring, applicable to the millimeter-wave radar signal processing method for sleep posture monitoring described in any of the above embodiments, comprising: a signal acquisition and preprocessing module, used to acquire background signals of the area to be monitored in an unmanned scene and echo signals in a scene with sleeping persons using millimeter-wave radar, and obtain human body self-reflection signals based on the echo signals and the background signals; a signal processing module, used to perform range-Doppler dimension processing on each frame of the human body self-reflection signals to obtain a range-Doppler spectrum corresponding to each frame; and a feature extraction and fusion module, used to perform feature extraction and fusion on each frame of the human body self-reflection signals. The distance-Doppler spectrum feature vector and micro-Doppler spectrum feature vector are extracted from the corresponding distance-Doppler spectrum, and a fusion feature vector is constructed for each frame. The stability screening module is used to perform initial screening of the fusion feature vector according to a preset attitude stability threshold, retaining the fusion feature vector corresponding to the stable attitude data segment and removing the fusion feature vector corresponding to the attitude transition interference data. The similarity coefficient matrix construction module is used to construct the similarity coefficient matrix corresponding to all the fusion feature vectors retained after initial screening. The attitude recognition module is used to classify the similarity coefficient matrix according to a preset clustering threshold using a density clustering algorithm, and output the final sleep attitude recognition result.

[0007] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the millimeter-wave radar signal processing method for sleep posture monitoring as described in any of the above embodiments.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The millimeter-wave radar signal processing method for sleep posture monitoring of this invention extracts dual features of high-resolution range image and micro-Doppler spectrum to construct a fused feature vector, achieving accurate identification of various sleep postures and enriching the application dimensions of non-contact radar monitoring. Utilizing millimeter-wave radar technology, it requires no contact with the human body, can penetrate obstructions such as bedding for monitoring, is unaffected by nighttime lighting conditions, and does not collect human visual images, fundamentally avoiding the risk of privacy leaks.

[0009] 2. The millimeter-wave radar signal processing method for sleep posture monitoring of the present invention employs a dual anti-interference mechanism. First, a posture stability threshold is used to initially screen the feature vectors of consecutive frames, eliminating unstable data segments during posture transitions. Then, a density clustering algorithm is used to classify the similarity coefficient matrix, further filtering out outlier noise points. This mechanism significantly reduces the misjudgment rate of recognition results due to dynamic interference, ensuring the stability and reliability of continuous monitoring throughout the night.

[0010] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of a millimeter-wave radar signal processing method for sleep posture monitoring provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a sleep posture monitoring scenario provided in an embodiment of the present invention; Figure 3 These are schematic diagrams of different sleeping postures provided in embodiments of the present invention; Figure 4 This is a schematic diagram of a millimeter-wave radar signal processing system for sleep posture monitoring provided in an embodiment of the present invention. Detailed Implementation

[0012] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following describes in detail a millimeter-wave radar signal processing method and system for sleep posture monitoring proposed according to the present invention, in conjunction with the accompanying drawings and specific embodiments.

[0013] The foregoing and other technical contents, features, and effects of the present invention will be clearly presented in the following detailed description of specific embodiments in conjunction with the accompanying drawings. Through the description of the specific embodiments, a more in-depth and concrete understanding can be gained of the technical means and effects adopted by the present invention to achieve its intended purpose. However, the accompanying drawings are for reference and illustration only and are not intended to limit the technical solutions of the present invention.

[0014] In a first aspect, embodiments of the present invention provide a millimeter-wave radar signal processing method for sleep posture monitoring. Please refer to... Figure 1 , Figure 1 This is a schematic diagram of a millimeter-wave radar signal processing method for sleep posture monitoring provided by an embodiment of the present invention. Figure 1 As shown, the millimeter-wave radar signal processing method for sleep posture monitoring in this embodiment may include the following steps: Step 1: Use millimeter-wave radar to collect background signals of the area to be monitored in an unmanned environment and echo signals in a sleeping environment. Obtain the human body's own reflection signals based on the echo signals and background signals.

[0015] In an optional embodiment, step 1 includes: Step 1.1: When there is no one in the area to be monitored, transmit and receive millimeter-wave radar signals to collect background signals; Step 1.2: When someone is sleeping in the area to be monitored, transmit and receive millimeter-wave radar signals and collect the echo signals; Step 1.3: Subtract the background signal from the echo signal in the time domain to obtain the human body reflection signal containing only the human body's own motion information.

[0016] For example, in an area to be monitored, such as when no one is in bed, the millimeter-wave radar can be activated to transmit an FMCW (Frequency Modulated Continuous Wave) signal and receive the echo, thus acquiring a background signal matrix. Subsequently, when someone is sleeping in the area, the signal is transmitted and received again to acquire the echo signal matrix. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a schematic diagram of a sleep posture monitoring scenario provided in an embodiment of the present invention. The background signal matrix and the echo signal matrix are subtracted in the time domain to obtain the human body's own reflection signal, which contains only information about the human body's own movement. This processing can effectively suppress strong static clutter generated by stationary furniture, walls, etc., in the room.

[0017] Step 2: Process each frame of the human body's own reflected signal in terms of distance and Doppler dimensions to obtain the corresponding distance-Doppler spectrum for each frame.

[0018] In this embodiment, the obtained human body self-reflection signal is processed in frames. For example, the duration of each frame is set to 50 milliseconds, corresponding to a frame rate of 20 frames per second. Frames can overlap, for example, with 50% overlap, to ensure temporal continuity.

[0019] In this embodiment, the discrete sampling sequence of the human body's own reflection signal is represented as follows: ;in, The amplitude of the baseband signal. For transmitting signal carrier frequency, The distance between the radar and the human target. At the speed of light, For frequency modulation slope, For fast time dimension sampling rate, These are the micro-Doppler changes caused by human activity during sleep. For slow time dimension sampling rate, For the fast time dimension sampling sequence number, , The number of sampling points in the fast time dimension. For the slow time dimension, the sampling sequence number. , The number of sampling points in the slow time dimension. It is the imaginary unit.

[0020] In an optional embodiment, step 2 includes: Step 2.1: Perform a Fourier transform on each frame of the human body's own reflection signal along the fast time dimension to obtain a one-dimensional high-resolution range image corresponding to each frame.

[0021] In this embodiment, a Fourier transform (FFT) is performed on each frame of signal along the fast time dimension to obtain a one-dimensional high-resolution range profile (HRRP) corresponding to each frame, i.e., a range-slow time matrix. The rows of this matrix correspond to the range cells, and the columns correspond to the slow time dimension.

[0022] Specifically, the discrete sampling sequence is processed according to formula (2). Performing FFT along the fast time dimension can obtain a one-dimensional high-resolution range profile of the target. . in, For signal bandwidth, This is the sampling sequence number of the one-dimensional high-resolution range image. , This represents the maximum number of sampling points for the distance image.

[0023] Step 2.2: Perform Fourier transform on the one-dimensional high-resolution range image corresponding to each frame along the slow time dimension to obtain the range-Doppler spectrum corresponding to each frame.

[0024] In this embodiment, a Fourier transform is performed on the one-dimensional high-resolution range image of each frame along the slow time dimension to obtain the range-Doppler spectrum corresponding to each frame. The horizontal axis of this spectrum represents the Doppler frequency, reflecting the motion velocity, the vertical axis represents the distance, and the pixel value reflects the signal strength.

[0025] Specifically, according to formula (3), for a one-dimensional high-resolution range image Performing FFT along the slow time dimension yields the target's range-microDoppler sequence. in, The sampling sequence number of Doppler. , This represents the maximum number of Doppler sampling points.

[0026] Step 3: Extract the range image feature vector and micro-Doppler spectrum feature vector from the range-Doppler spectrum corresponding to each frame, and construct the fused feature vector corresponding to each frame.

[0027] In an optional embodiment, step 3 includes: Step 3.1: Process the distance-Doppler spectrum of each frame using a constant false alarm rate (CFAR) detection algorithm to extract the signal of the human target region.

[0028] For example, cell-averaged CFAR (CA-CFAR) can be used to set the number of protection cells and reference cells to detect the distance-Doppler cells where the human target is located and extract the signal of the human target area.

[0029] Step 3.2: Based on the constant false alarm rate (CFAR) detection results, construct the distance image feature vector, which includes the number of distance image points, the location of the strongest point, and the energy entropy.

[0030] The number of distance points is the number of distance cells that have crossed the threshold detected by CFAR; the location of the strongest point is the index of the distance cell with the strongest energy; the energy entropy is the entropy value calculated after normalizing the energy of all distance cells that have crossed the threshold, reflecting the degree of concentration of energy distribution.

[0031] Step 3.3: Perform a short-time Fourier transform on the slow-time dimension signal to generate a micro-Doppler spectrum. Extract the micro-Doppler spectrum feature vector based on the micro-Doppler spectrum. The micro-Doppler spectrum feature vector includes the number of spectral peaks, the main peak frequency, and the energy entropy. In this embodiment, based on the range-slow-time matrix obtained in step 2.1, perform a short-time Fourier transform (STFT) on the slow-time dimension signal of the range cell where the target is located detected by CFAR to generate a micro-Doppler spectrum.

[0032] The number of spectral peaks is the number of peaks in the spectrum whose energy exceeds a preset threshold; the main peak frequency is the Doppler frequency corresponding to the highest energy peak; and the energy entropy is the entropy value calculated after normalizing the energy of the spectrum.

[0033] Step 3.4: Concatenate and fuse the distance image feature vector with the micro-Doppler spectrum feature vector to form a fused feature vector.

[0034] In this embodiment, the fused feature vector formed by splicing and merging is used to characterize the sleep posture of the current frame.

[0035] Step 4: Based on the preset attitude stability threshold, perform initial screening on the fused feature vectors, retain the fused feature vectors corresponding to stable attitude data segments, and remove the fused feature vectors corresponding to attitude transition interference data.

[0036] In an optional embodiment, step 4 includes: Calculate the cosine similarity between the fused feature vectors of N consecutive frames and take the mean. If the mean cosine similarity is greater than or equal to the attitude stability threshold, then the current N consecutive frames of data are determined to be a stable attitude data segment and the corresponding fused feature vector is retained; otherwise, they are determined to be attitude transition interference data and the corresponding fused feature vector is removed. Where N is an integer greater than or equal to 3, for example, N can take the value 3. The formula for calculating cosine similarity is: (4); among which, Represents cosine similarity. Indicates the first The fused feature vector corresponding to the frame Indicates the first The corresponding fused feature vector, This represents the norm of the vector. For example, the attitude stability threshold can be set to 0.7. When the mean cosine similarity is ≥0.7, the current 3 frames of data are determined to be stable attitude data segments, and all the fused feature vectors corresponding to these 3 frames are retained; when the mean cosine similarity is <0.7, the current 3 frames of data are determined to be attitude transformation interference data, and all the fused feature vectors corresponding to these 3 frames are removed.

[0037] Step 5: Construct the similarity coefficient matrix corresponding to all fusion feature vectors retained after the initial screening.

[0038] In an optional embodiment, step 5 includes: Step 5.1: For all the fused feature vectors retained after the initial screening, calculate the similarity coefficient between any two fused feature vectors using the cosine similarity calculation formula; Step 5.2: Construct a similarity coefficient matrix based on all the calculated similarity coefficients.

[0039] In this embodiment, the rows and columns of the similarity coefficient matrix correspond to different data frames, and the matrix element values ​​represent the degree of similarity between the feature vectors of the corresponding two frames.

[0040] Step 6: Based on the preset clustering threshold, classify the similarity coefficient matrix using density clustering algorithm, and output the final sleep posture recognition result.

[0041] Specifically, in this embodiment, based on a preset clustering threshold and minimum clustering size, a density clustering algorithm is used to process the similarity coefficient matrix. The processing includes: using the similarity coefficients in the similarity coefficient matrix as the connection strength criterion between points to identify the core point set, and recursively expanding the density-connected data points to divide the fused feature vector into several independent clusters, each cluster corresponding to a sleep posture.

[0042] In an optional embodiment, step 6 includes: Step 6.1: Based on the similarity coefficient matrix, determine the fused feature vectors with similarity coefficients greater than or equal to the clustering threshold as adjacent to each other; Step 6.2: Based on the determination results of Step 6.1, determine the core point set according to the preset minimum clustering size; Step 6.3: Starting from the core points in the core point set, the fused feature vector is divided into several independent clusters by recursively expanding the density of connected data points. Each cluster corresponds to a sleep posture.

[0043] Understandably, after density clustering is completed, several independent clusters are obtained, such as cluster A, cluster B, cluster C, and cluster D. In order to determine the specific sleep posture corresponding to each cluster, this embodiment uses a method of manual calibration during the calibration phase and template matching during the runtime phase for mapping.

[0044] The manual calibration method during the calibration phase involves inviting testers to perform the following steps sequentially according to instructions during the initial system deployment: Figure 3 The four positions shown are supine, lateral (left / right side), prone, and curled up. Figure 3 This is a schematic diagram of different sleep postures provided in this embodiment of the invention. Each posture is maintained for 2 minutes, and the corresponding time period for each posture is recorded. Steps 1-6 are performed to process the collected data and obtain clustering results. The clustering results are compared with the manually recorded time periods: for example, if the time period of cluster A corresponds exactly to the manually recorded "supine" time period, then cluster A is labeled as "supine"; cluster B corresponds to "side-lying", so it is labeled as "side-lying", and so on. A mapping table between cluster IDs and posture names is established and stored in the system.

[0045] The template matching method during the operational phase, which requires the system to automatically identify posture names during actual monitoring, allows for the pre-establishment of a posture template library during the calibration phase. Specifically, this involves collecting fusion feature vectors from a large number of test subjects of different body types in standard postures, averaging all vectors for the same posture to obtain the template vector for that posture, such as supine template, lateral template, prone template, and curled-up template. After obtaining the clusters in step 6.3, the center point of each cluster is calculated, which is the average of all feature vectors within the cluster. Then, the cosine similarity between this center point and the four posture templates is calculated. The posture corresponding to the template with the highest similarity is the posture of that cluster. For example, if the center point of cluster A has the highest similarity to the supine template, then cluster A is determined to represent "supine".

[0046] In this embodiment, the final output is the posture recognition results corresponding to each time period during the entire night's sleep, which can be used for subsequent applications such as sleep quality assessment and sleep apnea syndrome screening.

[0047] This invention presents a millimeter-wave radar signal processing method for sleep posture monitoring. By extracting dual features from high-resolution range images and micro-Doppler spectra, a fused feature vector is constructed, enabling accurate identification of various sleep postures. This enriches the application dimensions of non-contact radar monitoring. Utilizing millimeter-wave radar technology, it requires no contact with the human body, can penetrate obstructions such as bedding for monitoring, is unaffected by nighttime lighting conditions, and does not collect human visual images, fundamentally avoiding the risk of privacy leaks.

[0048] Secondly, embodiments of the present invention provide a millimeter-wave radar signal processing system for sleep posture monitoring, applicable to the millimeter-wave radar signal processing method for sleep posture monitoring provided in the first aspect. Please refer to... Figure 4 , Figure 4 This is a schematic diagram of a millimeter-wave radar signal processing system for sleep posture monitoring provided in an embodiment of the present invention, as shown below. Figure 4 As shown, the millimeter-wave radar signal processing system for sleep posture monitoring in this embodiment includes: a signal acquisition and preprocessing module, a signal processing module, a feature extraction and fusion module, a stability screening module, a similarity coefficient matrix construction module, and a posture recognition module.

[0049] The signal acquisition and preprocessing module uses millimeter-wave radar to acquire background signals in unmanned scenarios and echo signals in scenarios with sleeping people in the monitored area. It then obtains the human body's own reflection signals based on the echo and background signals. The signal processing module processes each frame of the human body's own reflection signals in both range and Doppler dimensions to obtain the range-Doppler spectrum corresponding to each frame. The feature extraction and fusion module extracts range image feature vectors and micro-Doppler spectrum feature vectors from the range-Doppler spectrum corresponding to each frame and constructs a fused feature vector for each frame. The stability screening module performs initial screening of the fused feature vectors based on a preset attitude stability threshold, retaining fused feature vectors corresponding to stable attitude data segments and removing fused feature vectors corresponding to attitude transition interference data. The similarity coefficient matrix construction module constructs a similarity coefficient matrix corresponding to all fused feature vectors retained after initial screening. The attitude recognition module classifies the similarity coefficient matrix using a density clustering algorithm based on a preset clustering threshold, outputting the final sleep attitude recognition result.

[0050] It is understandable that the posture recognition module internally implements a hardware accelerator or embedded software algorithm for density clustering algorithms (such as DBSCAN) and stores a cluster-posture mapping table, which can convert the clusters obtained by clustering into specific posture names (supine, lateral, prone, curled up) and output them.

[0051] Optionally, the modules in the system can be integrated into the same digital signal processor (DSP) or field-programmable gate array (FPGA) chip, or they can be distributed between the embedded processor and the host computer to work together. The system can achieve continuous, non-contact, and high-precision sleep posture monitoring throughout the night.

[0052] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the millimeter-wave radar signal processing method for sleep posture monitoring as provided in the first aspect.

[0053] It is understood that a computer-readable storage medium can be any medium that contains, stores, transmits, propagates, or transports a computer program, including but not limited to: read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks, flash memory, solid-state drives (SSDs), registers and other volatile or non-volatile memories, as well as communication media (such as carrier waves for transmitting signals). The storage medium is preferably a non-transitory computer-readable medium used for persistently storing program instructions.

[0054] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not expressly listed. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device comprising said element. Terms such as "connected" or "linked" are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect. The orientations or positional relationships indicated by terms such as "upper," "lower," "left," and "right" are based on the orientations or positional relationships shown in the accompanying drawings and are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention.

[0055] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0056] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A millimeter-wave radar signal processing method for sleep posture monitoring, characterized in that, include: Step 1: Use millimeter-wave radar to collect background signals of the area to be monitored in an unmanned environment and echo signals in a sleeping environment. Obtain the human body's own reflection signal based on the echo signals and the background signals. Step 2: Perform distance-doppler processing on each frame of the human body's own reflection signal to obtain the distance-Doppler spectrum corresponding to each frame; Step 3: Extract the range image feature vector and micro-Doppler spectrum feature vector from the range-Doppler spectrum corresponding to each frame, and construct the fused feature vector corresponding to each frame; Step 4: Based on the preset attitude stability threshold, perform initial screening on the fused feature vectors, retain the fused feature vectors corresponding to stable attitude data segments, and remove the fused feature vectors corresponding to attitude transition interference data. Step 5: Construct the similarity coefficient matrix corresponding to all fused feature vectors retained after the initial screening; Step 6: Based on the preset clustering threshold, classify the similarity coefficient matrix using a density clustering algorithm, and output the final sleep posture recognition result.

2. The millimeter-wave radar signal processing method for sleep posture monitoring according to claim 1, characterized in that, Step 1 includes: Step 1.1: When no one is in the area to be monitored, transmit and receive millimeter-wave radar signals to collect the background signal; Step 1.2: When someone is sleeping in the area to be monitored, transmit and receive millimeter-wave radar signals to collect the echo signals; Step 1.3: Subtract the background signal from the echo signal in the time domain to obtain the human body self-reflection signal that contains only the human body's own motion information.

3. The millimeter-wave radar signal processing method for sleep posture monitoring according to claim 1, characterized in that, Step 2 includes: Step 2.1: Perform a Fourier transform on each frame of the human body's own reflection signal along the fast time dimension to obtain a one-dimensional high-resolution range image corresponding to each frame; Step 2.2: Perform a Fourier transform on the one-dimensional high-resolution range image corresponding to each frame along the slow time dimension to obtain the range-Doppler spectrum corresponding to each frame.

4. The millimeter-wave radar signal processing method for sleep posture monitoring according to claim 1, characterized in that, The discrete sampling sequence of the human body's own reflection signal is represented as follows: in, The amplitude of the baseband signal. For transmitting signal carrier frequency, The distance between the radar and the human target. At the speed of light, For frequency modulation slope, For fast time dimension sampling rate, These are the micro-Doppler changes caused by human activity during sleep. For slow time dimension sampling rate, For the fast time dimension sampling sequence number, For the slow time dimension, the sampling sequence number. It is the imaginary unit.

5. The millimeter-wave radar signal processing method for sleep posture monitoring according to claim 1, characterized in that, Step 3 includes: Step 3.1: Process the distance-Doppler spectrum of each frame using a constant false alarm rate (CFAR) detection algorithm to extract the signal of the human target region; Step 3.2: Based on the constant false alarm rate detection results, construct the distance image feature vector, which includes the number of distance image points, the location of the strongest point, and the energy entropy; Step 3.3: Perform a short-time Fourier transform on the slow-time dimension signal to generate a micro-Doppler spectrum. Extract the micro-Doppler spectrum feature vector based on the micro-Doppler spectrum. The micro-Doppler spectrum feature vector includes the number of spectral peaks, the main peak frequency, and the energy entropy. Step 3.4: The distance image feature vector and the micro-Doppler spectrum feature vector are concatenated and fused to form the fused feature vector.

6. The millimeter-wave radar signal processing method for sleep posture monitoring according to claim 1, characterized in that, Step 4 includes: calculating the cosine similarity between the fused feature vectors of N consecutive frames and taking the mean value; if the mean cosine similarity is greater than or equal to the attitude stability threshold, then the current N consecutive frames of data are determined to be the stable attitude data segment and the corresponding fused feature vector is retained; otherwise, they are determined to be the attitude transition interference data and the corresponding fused feature vector is removed. Where N is an integer greater than or equal to 3, and the formula for calculating the cosine similarity is: in, Represents cosine similarity. Indicates the first The fused feature vector corresponding to the frame Indicates the first The corresponding fused feature vector, The norm of a vector.

7. The millimeter-wave radar signal processing method for sleep posture monitoring according to claim 1, characterized in that, Step 5 includes: Step 5.1: For all the fused feature vectors retained after the initial screening, calculate the similarity coefficient between any two fused feature vectors using the cosine similarity calculation formula; Step 5.2: Construct the similarity coefficient matrix based on all the calculated similarity coefficients.

8. The millimeter-wave radar signal processing method for sleep posture monitoring according to claim 1, characterized in that, Step 6 includes: Step 6.1: Based on the similarity coefficient matrix, determine the fused feature vectors with similarity coefficients greater than or equal to the clustering threshold as adjacent to each other; Step 6.2: Based on the determination result of Step 6.1, determine the core point set according to the preset minimum clustering size; Step 6.3: Starting from the core points in the core point set, the fused feature vector is divided into several independent clusters by recursively expanding the density of connected data points. Each cluster corresponds to a sleep posture.

9. A millimeter-wave radar signal processing system for sleep posture monitoring, characterized in that, The millimeter-wave radar signal processing method for sleep posture monitoring as described in any one of claims 1-8 includes: The signal acquisition and preprocessing module is used to acquire background signals of the monitored area in an unmanned scene and echo signals in a sleeping scene using millimeter-wave radar, and to obtain the human body's own reflection signal based on the echo signals and the background signals. The signal processing module is used to process each frame of the human body's own reflected signal in the distance dimension and Doppler dimension to obtain the distance-Doppler spectrum corresponding to each frame. The feature extraction and fusion module is used to extract the range image feature vector and the micro-Doppler spectrum feature vector based on the range-Doppler spectrum corresponding to each frame, and to construct the fused feature vector corresponding to each frame. The stability screening module is used to perform initial screening on the fused feature vectors according to a preset attitude stability threshold, retaining the fused feature vectors corresponding to stable attitude data segments and removing the fused feature vectors corresponding to attitude transition interference data. The similarity coefficient matrix construction module is used to construct the similarity coefficient matrix corresponding to all fused feature vectors retained after the initial screening. The posture recognition module is used to classify the similarity coefficient matrix according to a preset clustering threshold value using a density clustering algorithm, and output the final sleep posture recognition result.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the millimeter-wave radar signal processing method for sleep posture monitoring as described in any one of claims 1-8.

Images