Non-contact sleeping posture detection and vital sign monitoring method based on millimeter wave radar

Abstract

The invention discloses a non-contact sleeping posture detection and vital sign monitoring method based on a millimeter-wave radar, and the method employs a multi-transmitting and multi-receiving millimeter-wave radar, employs an innovative signal processing algorithm and a data processing algorithm, and carries out sleep monitoring through comprehensive consideration of frequency spectrum information and point cloud information. Comprising the steps of extraction of vital sign data such as respiration and heartbeat and judgment of in-bed, body movement and sleeping posture states. Breathing and heartbeat components are effectively extracted through cross-correlation entropy operation, then the signal-to-noise ratio of the breathing and heartbeat components is increased, then fast Fourier transform is carried out on cross-correlation entropy, and it is proved that the breathing and heartbeat frequency is effectively extracted.

Description

Technical Field

[0001] The present invention relates to the technical field of sleep monitoring, and particularly to a non-contact sleeping posture detection and vital sign monitoring method based on a millimeter-wave radar. Background Art

[0002] Research shows that sleeping postures are closely related to sleep quality and respiratory health. Improper sleeping postures may cause airway obstruction or affect blood circulation. Especially for patients with sleep apnea syndrome, Parkinson's disease, and other chronic diseases, sleeping posture detection will help improve sleep quality and reduce health risks. For example, the supine position is likely to exacerbate the breathing difficulties of patients with sleep apnea syndrome, while the lateral sleeping position can relieve airway obstruction. Therefore, real-time identification of sleeping posture changes can not only help individuals improve sleep quality but also provide strong support for the management of specific diseases.

[0003] Millimeter-wave radars with the FMCW system have extremely high application potential in the field of non-contact sleeping posture detection and can achieve the discrimination of sleeping postures through complex signal processing algorithms and pattern recognition techniques. However, in existing sleeping posture detection algorithms, most methods focus on the identification of dynamic processes, such as the detection of actions like lying down and turning over, but do not cover the detection of static sleeping postures such as supine and lying positions. This is because there are some technical difficulties in detecting human static sleeping postures based on FMCW radars. Specifically, in the signal processing process of FMCW radars, a static target elimination algorithm usually needs to be implemented to eliminate static clutter interference, such as the echo signals of static targets like walls and beds. However, in the deep sleep state, most of the human body also remains static, and only the local micro-movements caused by breathing and heartbeat are relatively obvious. The echo signals of the rest of the body will significantly weaken after the static target elimination, resulting in the loss of human posture information. Therefore, it is a difficult problem to extract what features and how to extract features when performing sleeping posture detection.

[0004] In addition, breathing and heartbeat, as basic vital sign information of the human body, can directly reflect the physiological state and health level of the human body. And with the increasing attention to people's physical health, the monitoring demand for vital signs has become stronger. Therefore, the real-time monitoring of vital signs has extremely high practical value.

[0005] In this regard, traditional contact sensors have gradually revealed problems such as inconvenient use and poor compliance. In contrast, FMCW radars can capture the tiny vibrations caused by heartbeat and breathing, thereby achieving high-precision and non-contact vital sign monitoring. In addition, the transmitted signal in the millimeter wave band has strong penetration ability and anti-interference ability, enabling the millimeter-wave radar with the FMCW system to perform continuous monitoring without affecting the normal life of users. This feature is particularly important in the health monitoring of special populations such as newborns and burn patients.

[0006] In summary, the millimeter-wave radar based on the FMCW system has opened up a non-contact and unobtrusive technical path in the field of health monitoring, providing a highly promising solution for daily monitoring and clinical assistance. SUMMARY OF THE INVENTION

[0007] In view of this, the purpose of the present invention is to provide a non-contact sleeping posture detection and vital sign monitoring method based on a millimeter-wave radar, which uses the millimeter-wave radar to achieve non-contact sleep monitoring.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The non-contact sleeping posture detection and vital sign monitoring method based on a millimeter-wave radar provided by the present invention includes the following steps:

[0010] Step 1: Set the millimeter-wave radar to continuously transmit signals to the area where the bed is located, receive the echo signals, and generate spectrum data and point cloud data through signal processing algorithms;

[0011] Step 2: Perform clustering processing on the point cloud data obtained in Step 1 to obtain the number of point clouds contained in the cluster and the position of the cluster centroid in the effective area where the bed is located, and obtain the presence of detection targets in the bed area;

[0012] Step 3: Count the number of point clouds at the edge position of the effective area and identify the body movement signals of the detection target in each partition;

[0013] Step 4: Identify the sleeping posture state of the detection target according to the height of the cluster centroid and the number of point clouds in each partition, and the sleeping posture state includes a flat state and a side-lying state;

[0014] Step 5: After locating the human body in the bed area and determining that it is in the bed in Step 2, extract the current scene background noise according to the spectrum data obtained in Step 1, then calculate the signal-to-noise ratio with the range image data, and perform moving target detection on the range image to extract the range gate data where the target is located;

[0015] Step 6: Extract the slow time dimension information of the target range gate where no body movement occurs in Step 2, extract the phase of the target range gate data and perform phase unwrapping to obtain a phase signal containing heartbeat and respiration information, and use an IIR filter to extract the respiration signal and the heartbeat signal respectively; adopt a time-frequency change algorithm based on cross-correlation entropy to perform frequency detection on the extracted respiration and heartbeat signals, and also extract the respiration signal with the highest intensity within the respiration frequency range; finally calculate the real-time heart rate and respiration, and give an alarm for the situation of apnea;

[0016] Step 7: Conduct sleep analysis, score each item of data according to its ideal range respectively, and summarize through the scoring weights; and give sleep quality assessment and improvement suggestions in the sleep analysis report, and push them to the user through the cloud platform;

[0017] Further, the sleep analysis in Step 7 is performed by combining the number of body movements, changes in sleeping postures, average heart rate, average breathing rate, and the number of apnea events.

[0018] Further, in Step 2, the detection target is determined by the in-bed detection algorithm, and the in-bed detection algorithm is specifically carried out according to the following steps:

[0019] Determine the bed parameters and spatial position of the target user, so as to construct the effective area of the human body in the bed,

[0020] First, use the DBSCAN algorithm to cluster the point cloud in the effective area;

[0021] Then, according to Calculate the cluster centroid, where m i =(x i ,y i ,z i ) represents the spatial position of a certain point cloud in the three-dimensional Cartesian coordinate system, and k represents the number of point clouds contained in the cluster;

[0022] Finally, if the number of point clouds is higher than the empirical threshold and the cluster centroid is within the effective area, it will be judged that the user is in bed for subsequent processing.

[0023] Further, in Step 3, the body movement signals of the detection target in each partition are detected by the body movement detection algorithm, and the body movement detection algorithm is specifically carried out according to the following steps:

[0024] Divide the effective area into multiple partition parts, which respectively correspond to the head, abdomen, and limbs longitudinally; when the target is in a resting state, obtain the point cloud data at the chest and abdomen; when the user has limb movements or body tremors, obtain the point cloud data of the limb part; determine the body movement signal of the target according to the number of point cloud data.

[0025] Further, the sleeping posture state in Step 4 is determined by the sleeping posture discrimination algorithm, and the sleeping posture discrimination algorithm is specifically carried out according to the following steps:

[0026] Divide the effective area where the bed is located into an upper area and a lower area; the lower area is the human body area when the target lies flat, and the upper area is the human body area when the target lies on its side;

[0027] Count the number of point clouds in the upper area and the lower area;

[0028] Determine the sleeping posture state according to the change amount of the point cloud data in the upper layer area and the lower layer area.

[0029] Further, in step 5, an adaptive noise extraction algorithm is used to extract the current scene background noise from the spectrum data. The specific steps of the adaptive noise extraction algorithm are as follows:

[0030] First, process the range image signals of the bed area and the non-bed area. After processing the point cloud data, the targets with motion information will be screened out, and then the stationary signals will be screened out. Extract the clutter and background noise therein, filter out the background noise through smoothing filtering, and if the signal strength of the range gate of the suspected target is greater than the preset signal-to-noise ratio, it is confirmed as a target.

[0031] Further, the time-frequency change algorithm based on cross-correlation entropy in step 6 is specifically carried out according to the following steps:

[0032] Calculate the cross-correlation entropy of the multi-channel data;

[0033] Perform Fourier transform on the cross-correlation entropy to obtain the cross-correlation entropy spectrum;

[0034] Extract the respiration and heartbeat signals from the cross-correlation entropy spectrum.

[0035] Further, the following steps are also included in step 1:

[0036] Step 1.1: Select a sawtooth wave FMCW as the transmitted signal s TX (t) = cos(2πf c t + πkt 2 + Φ0). When the electromagnetic wave emitted by the radar encounters a target during the scattering process, it will be reflected back; it is received through multiple antennas, and the received signal of a certain channel can be expressed as s RX (t) = s TX (t - τ) = cos(2πf c (t - τ) + πk(t - τ) 2 + Φ0); Mix the received signal and the transmitted signal, and then obtain the difference frequency signal s IF (t) = Acos(2πkτt + Φ IF ); Sample the difference frequency signals of each receiving channel respectively to form a three-dimensional radar data block;

[0037] Step 1.2: Select a fast time-slow time sampling matrix of a certain receiving channel, perform FFT on the fast time dimension and perform non-coherent accumulation on the slow time dimension to obtain a one-dimensional range image spectrum for subsequent extraction of vital sign signals;

[0038] Step 1.3: Perform static clutter filtering, 2D-FFT, and CFAR detection on each receiving channel respectively. Then, based on the information of the range-Doppler spectrum, perform angle estimation in the azimuth and elevation directions respectively to complete ranging, velocity measurement, and angle measurement, and further obtain point cloud data for subsequent judgment of being in bed, body movement, and sleeping posture.

[0039] Furthermore, the sleeping posture state in step 4 is determined in the following manner:

[0040] The data characteristics of the target user lying on the side are calculated according to the following formula:

[0041] z > bed height + shoulder width / 2;

[0042] where z represents the height in the clustering centroid coordinate m = (x, y, z);

[0043] When the height is greater than the threshold, the target user is in the side-lying state;

[0044] Furthermore, the side-lying is judged according to the side-lying probability, and the side-lying probability is calculated according to the following formula:

[0045]

[0046] In the formula, the parameters k and b are used to adjust the slope and offset of the calculation formula; the variable x is the difference between the centroid height and the set threshold, that is, x = z - (bed height + shoulder width / 2).

[0047] The beneficial effects of the present invention are as follows:

[0048] The non-contact sleeping posture detection and vital sign monitoring method based on millimeter-wave radar provided by the present invention uses a multi-transmitter and multi-receiver millimeter-wave radar, adopts innovative signal processing algorithms and data processing algorithms, comprehensively considers spectrum information and point cloud information for sleep monitoring, including the extraction of vital sign data such as breathing and heartbeat and the discrimination of being in bed, body movement, and sleeping posture states. Utilizing the penetration ability and anti-interference ability of millimeter waves, continuous health monitoring of users can be achieved without contacting the body, providing a non-intrusive and efficient monitoring experience for special groups such as newborns, burn patients, etc. This method comprehensively uses spectrum information and point cloud information as the discrimination basis, and based on the regional distribution of point clouds and the position of point cloud centroids, can accurately distinguish whether the user is in bed in a complex scene, and effectively adapt to dynamic and static scenes through body movement detection and sleeping posture discrimination, overcoming the problem that static human posture information is easily lost, ensuring the comprehensiveness and accuracy of detection results. In addition, the system supports real-time data processing, instantaneously feedbacks the sleep state through a cloud platform, integrates multi-dimensional health indicators such as heart rate and respiratory rate for sleep quality assessment, and provides personalized improvement suggestions, with both real-time and interactivity, providing technical support for home health management and medical assistance.

[0049] The cross-correlation entropy method utilizes the second-order and higher-order statistical information of signals, enabling more accurate quantification of target information. As a result, better imaging effects and parameter estimation performance of micro-motion targets can be achieved in non-Gaussian backgrounds. Meanwhile, by performing cross-correlation processing on multi-channel signals, it can effectively suppress random noise and clutter components, further improving the imaging quality of micro-motion targets. Finally, by fusing multiple groups of cross-correlation entropy results, the imaging effect of the target and the output signal-to-noise ratio can be further enhanced, outperforming the traditional Fourier transform method in non-Gaussian backgrounds. Considering the above advantages, the cross-correlation entropy method can more effectively extract micro-motion signals such as breathing and heartbeat. Even in the presence of strong background noise, it can still maintain a relatively high output signal-to-noise ratio and measure more accurate vital signs.

[0050] Other advantages, objectives, and features of the present invention will, to some extent, be described in the subsequent specification. And to some extent, they will be obvious to those skilled in the art based on the study of the following text, or can be learned from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] To make the objectives, technical solutions, and beneficial effects of the present invention clearer, the following drawings are provided for illustration.

[0052] Figure 1 This is the effective area of the human body on the bed in this embodiment.

[0053] Figure 2 This is the segmentation of the effective area of the bed body in this embodiment.

[0054] Figure 3 This is the algorithm block diagram in this embodiment.

[0055] Figure 4 This is the frequency-time function of the intermediate-frequency signal in this embodiment.

[0056] Figure 5 This is the evaluation of the side-lying probability based on the centroid height in this embodiment.

[0057] Figure 6 This is the comparison diagram of the time-frequency transformation effects in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] The following further illustrates the present invention in conjunction with the drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it. However, the described embodiments shall not be construed as limitations on the present invention.

[0059] Embodiment 1

[0060] As Figure 1As shown in the figure, the non-contact sleeping posture detection and vital sign monitoring method based on millimeter-wave radar provided in this embodiment is specifically implemented as follows:

[0061] Step 1: Mount the millimeter-wave radar at a specified height at the head of the bed. After the radar is powered on, it will continuously emit electromagnetic signals to the area where the bed is located. In this embodiment, the echo signals of multiple receiving channels are used to generate spectral data and point cloud data through signal processing algorithms (range-dimensional FFT, velocity-dimensional FFT, CFAR detection, angle estimation); subsequently, based on the point cloud data, judgments on being in bed, body movement, and sleeping posture are made, and vital sign signals are extracted based on the spectral data;

[0062] Step 2: Cluster the point cloud data obtained in Step 1. In this embodiment, the number of point clouds contained in the cluster within the effective area and the position of the cluster centroid are comprehensively considered. An in-bed detection algorithm is used to determine whether there is a user (detection target) in the bed area. When the user is in bed, judgments on body movement, sleeping posture, and extraction of vital sign signals will be made;

[0063] Step 3: Since the radar is more sensitive to moving targets, when the target is in a resting state, the point clouds mainly gather at the chest, while when the user has limb movement or body tremors, the point clouds will spread outwards; in this embodiment, by counting the number of point clouds at the edge position of the effective area, a body movement detection algorithm is used to identify whether there are large body movements such as turning over and small body movements such as coughing of the target; when body movement is detected, the sleeping posture discrimination will not be performed, and the vital sign data during the time of body movement needs to be excluded;

[0064] Step 4: In this embodiment, the height of the cluster centroid obtained in Step 2 and the number of point clouds in each partition obtained in Step 3 are comprehensively considered. According to the change of the above data in the lying flat and side-lying states, a sleeping posture discrimination algorithm is used to determine whether the user is currently in a lying flat sleeping posture or a side-lying sleeping posture; since the chest movement in the side-lying sleeping posture is in the tangential direction of the radar and the radar is not very sensitive to this, which has a certain impact on the subsequent apnea discrimination, therefore, performing apnea judgment on the basis of distinguishing the sleeping posture can reduce false alarms;

[0065] Step 5: After locating the human body in the bed area in Step 2 and determining that the person is in bed, for the spectral data obtained in Step 1, an adaptive noise extraction algorithm with high signal-to-noise ratio is used to extract the background noise of the current scene, then the signal-to-noise ratio is calculated with the range image data, and moving target detection is performed on the range image to extract the range gate data of the target with high signal-to-noise ratio;

[0066] Step 6: Extract the slow-time dimension (Doppler dimension) information of the target's distance from the door where no body movement occurred in Step 2. Extract the phase of the data in the distance bin where the target is located and perform phase unwrapping to obtain a phase signal containing heartbeat and respiration information. Use an IIR filter to extract the respiration signal and the heartbeat signal respectively; adopt a time-frequency variation algorithm based on cross-correlation entropy to detect the frequencies of the extracted respiration and heartbeat signals. Also use the adaptive noise extraction algorithm used in Step 1 to extract the respiration signal with the highest intensity within the respiration frequency range; finally, calculate the real-time heart rate and respiration, and give an alarm for the case of apnea;

[0067] The formulas for calculating the real-time heart rate and respiration are as follows:

[0068] Heartbeat rate = 60 * f heart , Respiration rate = 60 * f breath

[0069] where f heart , f breath are the heartbeat frequency and respiration frequency extracted from the respiration and heartbeat signals through the time-frequency variation of cross-correlation entropy.

[0070] Step 7: Combine the number of body movements, changes in sleeping postures, average heart rate, average respiration frequency, and the number of apnea episodes for sleep analysis. Score respectively according to the ideal ranges of each item of data, and summarize through scoring weights; and give a sleep quality assessment and improvement suggestions in the sleep analysis report, and push them to the user through the cloud platform;

[0071] Further, for Step 1, there are the following sub-steps:

[0072] Step 1.1: Select a sawtooth wave FMCW as the transmitted signal s TX (t) = cos(2πf c t + πkt 2 + Φ0). When the electromagnetic wave emitted by the radar encounters a target during scattering, it will be reflected back; receive through multiple antennas, and the received signal of a certain channel can be expressed as s RX (t) = s TX (t - τ) = cos(2πf c (t - τ) + πk(t - τ) 2 + Φ0); Mix the received signal and the transmitted signal, and then obtain the difference-frequency signal s IF (t) = Acos(2πkτt + Φ IF ); Sample the difference-frequency signals of each receiving channel respectively to form a three-dimensional radar data block;

[0073] Step 1.2: Select the fast time - slow time sampling matrix of a certain receiving channel, perform FFT on the fast time dimension and perform non - coherent accumulation on the slow time dimension to obtain a one - dimensional range image spectrum for subsequent extraction of vital sign signals;

[0074] Step 1.3: Perform static clutter filtering, 2D - FFT, and CFAR detection on each receiving channel respectively, and then perform angle estimation in the azimuth and elevation directions based on the information of the range - Doppler spectrum to complete ranging, velocity measurement, and angle measurement, and then obtain point cloud data for subsequent judgment of in - bed status, body movement, and sleeping posture;

[0075] As Figure 1 shown, Figure 1 is the effective area of the human body in bed; The in - bed detection algorithm described in step 2 is detailed as follows:

[0076] According to the bed length, bed width, and bed height parameters input by the user on the Web platform, delimit the spatial position of the bed body in the three - dimensional Cartesian coordinate system, and then construct the effective area of the human body in bed above the bed body based on empirical information, with the bed width, bed length, and a certain height above the bed height as boundaries. This area covers the entire surface of the bed body and the space above it, as Figure 1 shown. To avoid misjudgment when someone passes by the bedside, this embodiment comprehensively considers the number of point clouds contained in the cluster in the effective area and the position of the cluster centroid for in - bed detection.

[0077] First, use the DBSCAN algorithm to perform clustering processing on the point clouds in the effective area; Then, according to calculate the cluster centroid, where m i =(x i ,y i ,z i ) represents the spatial position of a certain point cloud in the three - dimensional Cartesian coordinate system, and k represents the number of point clouds contained in the cluster; Finally, if the number of point clouds is higher than the empirical threshold and the cluster centroid is within the effective area, it will be judged that the user is in bed for subsequent processing;

[0078] As Figure 2 shown, Figure 2 is the block division of the effective area; The body movement detection algorithm described in step 3 is detailed as follows:

[0079] Further divide the effective area into as Figure 2The six parts shown correspond to the head, abdomen, and lower limbs longitudinally; the radar is more sensitive to moving targets. When the target is in a resting state, the point cloud mainly gathers in the chest cavity, while when the user has limb movement or body tremors, the point cloud will spread outwards; among them, the number of point clouds in area 6 changes significantly during body movement. Therefore, in this embodiment, the number of point clouds in area 6 is accumulated within a certain period of time, and an empirical threshold is set based on the statistical characteristics of various test samples to identify whether the target has large body movements such as turning over and small body movements such as coughing.

[0080] Further, the sleeping posture discrimination algorithm described in step 4 is described in detail as follows:

[0081] The number of point clouds in areas 1, 2, 4, and 5 are respectively counted. The lower layer area can just contain the lying human body. When the user lies flat, there is usually no point cloud or only a very small number of point clouds in the upper layer area. Therefore, when the user lies on the side, the number of point clouds in the upper layer area (areas 4 and 5) increases significantly; in addition, since the chest movement in the lying flat state is in the radial direction of the radar, and the chest movement in the side-lying state is in the tangential direction of the radar, however, the radar has poor perception ability for tangential movement, and the upper layer area takes away some point clouds. Therefore, when the user lies on the side, the number of point clouds in the lower layer area (areas 1 and 2) decreases significantly; according to the above data change characteristics, in this embodiment, an empirical threshold is set based on the statistical characteristics of various test samples, and a scoring weight is set according to the reliability of the data in different areas, and the point cloud data of the four parts are comprehensively considered to give the score score1 for this item.

[0082] The height of the clustering centroid obtained in step 2 will also change significantly due to different sleeping postures. Obviously, the side-lying height should be higher than the lying flat height; according to this change characteristic, in this embodiment, the threshold is set to threshold = bed height + shoulder width / 2 based on physical meaning and the statistical characteristics of various test samples. When the centroid height is greater than the threshold, it can be considered to a large extent that the user is in a side-lying state.

[0083] For this reason, in this embodiment, a scoring function is defined based on the sigmod function, and the difference between the height and the threshold is used as the dependent variable to calculate the score score2 for this item.

[0084] This embodiment comprehensively considers the number of point clouds in each part and the height of the clustering centroid. By setting a scoring weight weight, the two scores are combined to obtain score = (1 - weight) · score1 + score2. Score can be regarded as the probability that the user is in a side-lying state. Therefore, the closer score is to 1, the more likely the user is to be in a side-lying state.

[0085] Further, the following sub-steps are included in step 5:

[0086] After locating the human body in the bed area in step 2 and determining that the human body is in bed, an adaptive noise extraction algorithm is performed on the spectrum data obtained in step 1 for the distance gate where the human target is located: First, the distance image signals of the bed area and the non-bed area are processed. After point cloud data processing, targets with motion information will be screened out, and then the stationary signals will be filtered out. The clutter and background noise existing in them are extracted. The background noise can be filtered out through smoothing filtering. When the signal intensity of the distance gate of the suspected target is greater than the preset signal-to-noise ratio, it is confirmed as a target, and the signals of this distance gate will be used for subsequent signal processing;

[0087] Further, the time-frequency change algorithm based on cross-correlation entropy described in step 6 is described in detail as follows:

[0088] The traditional method of Fourier transform spectrum estimation for respiration and heartbeat signals is easily interfered by noise, and it is difficult to obtain high signal-to-noise ratio respiration and heartbeat signals under certain hardware conditions. In this embodiment, when processing respiration and heartbeat signals, the concept of cross-correlation entropy is combined, and the cross-correlation entropy of multi-channel data is calculated. In this process, since the respiration and heartbeat are strongly correlated in different channels, performing the cross-correlation entropy operation can effectively extract the respiration and heartbeat components, thereby increasing their signal-to-noise ratio. Then, the fast Fourier transform is performed on the cross-correlation entropy, and it is proved that the respiration and heartbeat frequencies can be effectively extracted.

[0089] Embodiment 2

[0090] As Figure 3 shown Figure 3 is the algorithm block diagram; The method for realizing non-contact sleep monitoring using a multi-transmitter and multi-receiver millimeter-wave radar proposed in this embodiment is as follows:

[0091] Step 1: A 60 GHz frequency-modulated continuous wave radar is used and installed at a height of 1.5 m from the ground. After the radar is powered on and working, it will continuously transmit a sawtooth wave FMCW signal to the area where the bed is located, which can be specifically expressed as:

[0092] s TX (t) = cos(2π·∫f(t)dt) = cos(2πf c t + πkt 2 + Φ0) t ∈ [0, T c (1)

[0093] In the formula, f c is the starting frequency; k is the frequency modulation slope, satisfying k = BT c ; B is the signal bandwidth; T c is the frequency sweep period; Φ0 is the random initial phase.

[0094] Assume that a target is within the radar detection range and the distance between its centroid and the radar is R. According to the fact that the propagation speed of electromagnetic waves in air is the speed of light c, the time delay of the echo signal can be expressed as:

[0095] τ = 2R / c (2)

[0096] Then the echo signal can be expressed as the transmitted signal with a time lag:

[0097] s RX (t) = s TX (t - τ) = cos(2πf c (t - τ) + πk(t - τ) 2 + Φ0) (3)

[0098] Mix the transmitted signal and the echo signal expressed by Equation (1) and Equation (3) (i.e., multiply them). According to the product-to-sum formula, the mixed-frequency output signal is expressed as follows:

[0099]

[0100] Then the frequency components contained in the mixed-frequency output signal are:

[0101]

[0102] f M2 = 2f c + k(2t - τ) (6)

[0103] Equation (5) and Equation (6) indicate that the frequency difference f M1 and the frequency sum f M2 are far apart, and it is easy to obtain the intermediate-frequency signal with a frequency of f IF = f M1 through low-pass filtering, which contains the distance information and velocity information of the target. According to the transmitted signal and the echo signal, the frequency-time function of the intermediate-frequency signal is obtained as Figure 4 shown. In the range of t ∈ [τ, T c , the intermediate-frequency signal has a constant frequency (the frequency difference between the transmitted signal and the echo signal), and its waveform is a sine function; as Figure 4 shown, Figure 4 is the frequency-time function of the intermediate-frequency signal.

[0104] Then the intermediate-frequency signal can be expressed as:

[0105]

[0106] Select the fast-time - slow-time sampling matrix of a certain receiving channel, perform FFT on the fast-time dimension and non-coherent accumulation on the slow-time dimension to obtain a one-dimensional range image spectrum for subsequent extraction of vital sign signals.

[0107] Static clutter filtering, 2D-FFT, and CFAR detection are respectively performed on each receiving channel, and then angle estimation is performed in the azimuth and elevation directions based on the information of the range-Doppler spectrum to complete ranging, velocity measurement, and angle measurement, and then point cloud data is obtained for subsequent judgment of being in bed, body movement, and sleeping posture.

[0108] Step 2: First, use the DBSCAN algorithm to cluster the point cloud within the effective area. After the DBSCAN algorithm obtains the measurement set D = {x1, x2,..., x n}, without specifying the number of classifications, two parameters ε and MinPts are used to represent the correlation degree between a data point and other data points. Among them, ε is the radius of a circle centered on a data point, and its function is to define a threshold to determine whether there are other data within the circle drawn by this point. MinPts is the minimum number of data points contained within the circle with a radius of ε. The DBSCAN algorithm defines all measurement data as core points, boundary points, and noise points according to the data density. The detailed steps of its algorithm are as follows:

[0109] ① Initialize the DBSCAN algorithm: Select appropriate density threshold ε and minimum clustering number M;

[0110] ② Randomly select an unvisited point p and mark it as visited;

[0111] ③ Calculate the three-dimensional Euclidean distance to search for whether there are M data points within the circle centered on point p with a density threshold ε as the radius. If so, construct the data points within the circle as a candidate set;

[0112] ④ Traverse each data point in the candidate set. If the number of data points within the circle with a radius of ε for each data point satisfies being greater than M, then add these points to the cluster.

[0113] ⑤ Repeat steps ③ and ④ until all data points in the data set have been visited.

[0114] After obtaining the clustering result, for the clusters whose point cloud quantity meets the threshold requirement, calculate the centroid using the following formula:

[0115]

[0116] where m i = (x i , y i , z i) represents the spatial position of a certain point cloud in a three-dimensional Cartesian coordinate system; k represents the number of point clouds contained in the cluster. It can be determined whether the user is in bed by judging whether the centroid of the cluster with the number of point clouds meeting the threshold requirement is within the effective area.

[0117] Step 3: Count the number of point clouds in area 6 and accumulate them within consecutive K frames. Determine whether the total number of point clouds is higher than the empirical threshold for body movement. Then, the data feature of body movement can be expressed as:

[0118]

[0119] Step 4: Calculate the absolute number of point clouds in areas 1 and 2 and the relative proportion of point clouds in areas 4 and 5, and also accumulate them within consecutive K frames. According to the previous analysis, when the user changes from lying flat to lying on the side, the number of point clouds in areas 1 and 2 should decrease, and the proportion of areas 4 and 5 should increase. Then, the data feature of the user lying on the side can be expressed as:

[0120]

[0121] In this embodiment, the empirical thresholds ThodZone1 / 2 / 4 / 5 are set based on the statistical characteristics of various test samples, and the scoring weights score1_1 / 2 / 4 / 5 are set according to the reliability of the data in different areas. Finally, the score score1 of this item is given by comprehensively considering the point cloud data of the four parts: score1 = score1_1 + score1_2 + score1_4 + score1_5.

[0122] The centroid coordinates m = (x, y, z) obtained in step 2, where the centroid height z will change significantly with different sleeping postures. Then, the data feature of the user lying on the side is:

[0123] z > bed height + shoulder width / 2 (11)

[0124] When the height is greater than the threshold, it can be considered to a great extent that the user is in the side-lying state. For this reason, this embodiment defines a scoring function based on the sigmod function to evaluate the probability that the user is in the side-lying state. The score score2 of this item is calculated with the difference between the height and the threshold as the dependent variable x, and the calculation formula is expressed as follows:

[0125]

[0126] In the formula, the parameters k and b are used to adjust the slope and bias of the calculation formula; the variable x is the difference between the centroid height and the set threshold, that is, x = z - (bed height + shoulder width / 2). The variation law of the side-lying probability with the difference between the height and the threshold is as Figure 5 shown, Figure 5 to evaluate the side-lying probability according to the centroid height.

[0127] In this embodiment, the point cloud quantity of each part and the height of the clustering centroid are comprehensively considered. By setting the scoring weight weight, the two scores are combined to obtain score = (1 - weight)·score1 + score2. score can be regarded as the probability that the user is in the side-lying state. Therefore, the closer score is to 1, the more likely the user is in the side-lying state.

[0128] Step 5: If it is recognized that the human target is in bed and the corresponding in-bed flag bit is valid, then the distance dimension signal is differentiated with respect to time to increase the signal-to-noise ratio of the vital sign signal. The formula is as follows:

[0129] fifo MTI [i] = X[i + 1] - X[i] (13)

[0130] This differential operation can screen out the target of the motion information. Then, according to the boundary range of the bed, the unmanned area at the edge of the bed is circled, and the background noise intensity NoisePower is extracted through smoothing filtering. When the signal intensity of the distance gate of the suspected target is greater than the set signal-to-noise ratio, it is confirmed as the target. At this time, the index SubjectIdx of the distance gate signal corresponding to the maximum signal-to-noise ratio is extracted, and the corresponding signal X[SubjectIdx] is the reflected signal at the target chest cavity. The specific formula is as follows:

[0131] SubjectIdx = argmax|X[Index]|, SNR > beta·NoisePower (14)

[0132] In the formula, SNR is the signal-to-noise ratio, NoisePower is the power of the extracted background noise, and beta is a scaling coefficient determined by the empirical value after statistical comparison of various data.

[0133] Step 6: The time-frequency change algorithm based on cross-correlation entropy is established on the basis of auto-correlation entropy. The following is the definition of auto-correlation entropy:

[0134] V(t, t + τ) = E[k σ (x(t) - x * (t))] (15)

[0135] In the formula, k σ (x, y) represents the kernel function, which is usually represented by a Gaussian function, that is:

[0136]

[0137] For the discrete time signal X[n], its auto-correlation entropy V[m] is expressed as follows:

[0138]

[0139] Suppose the autocorrelation entropy of the k-th antenna of the radar is V k [m], k = 1, 2, ..., M, then the cross-correlation entropy between the k-th channel and the l-th channel has the following formula:

[0140] V k,l [m] = E[V k [n]·V l [n - m]], k ≠ l (18)

[0141] Then, perform Fourier transform on the cross-correlation entropy to obtain the cross-correlation entropy spectrum:

[0142]

[0143] Compared with directly performing Fourier transform on x(t), the cross-correlation entropy method can effectively extract respiratory and heartbeat signals from multi-channel data, greatly improving its signal-to-noise ratio, and can even effectively extract respiratory and heartbeat frequencies when SNR < 1. As Figure 6 shown Figure 6 is a comparison diagram of the effects of two time-frequency transforms under the condition of SNR = 0.5. Under the condition of SNR = 0.5, the cross-correlation entropy method is significantly better than the traditional method.

[0144] After extracting the respiratory and heartbeat frequencies, convert them into respiratory and heartbeat rates:

[0145] Respiratory rate = 60 * respiratory frequency Heartbeat rate = 60 * heartbeat frequency (20)

[0146] Step 7: This embodiment combines the number of body movements, sleep posture changes, average heart rate, average respiratory rate, and the number of apnea episodes for sleep analysis, scores are given respectively according to the ideal ranges of each item of data, and finally summarized through the scoring weights.

[0147] In the sleep analysis report, display images of the in-bed state, body movement state, and sleep posture state changing over time, give a sleep quality assessment according to each index and its score, and give targeted sleep improvement suggestions according to the abnormal indexes of the data, and push them to the user through the cloud platform.

[0148] The above-described embodiments are only preferred embodiments given to fully illustrate the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention. The protection scope of the present invention is subject to the claims.

Claims

1. A non-contact sleeping posture detection and vital sign monitoring method based on millimeter-wave radar, characterized in that: It includes the following steps: Step 1: Set the millimeter-wave radar to continuously emit signals to the area where the bed is located, receive the echo signals, and generate spectral data and point cloud data through signal processing algorithms; Step 2: Perform clustering processing on the point cloud data obtained in Step 1 to obtain the number of point clouds contained in the clusters in the effective area where the bed is located and the positions of the cluster centroids, and obtain the detection targets existing in the bed area; Step 3: Count the number of point clouds at the edge positions of the effective area, and identify the body movement signals of the detection targets in each partition; Step 4: Identify the sleeping posture state of the detection target according to the height of the cluster centroid and the number of point clouds in each partition, and the sleeping posture state includes the lying flat state and the side-lying state; Step 5: After positioning the human body in the bed area and determining that it is in the bed in Step 2, extract the current scene background noise according to the spectral data obtained in Step 1, then calculate the signal-to-noise ratio with the range image data, and perform moving target detection on the range image to extract the range gate data where the target is located; Step 6: Extract the slow time dimension information of the range gates of the targets without body movement in Step 2, extract the phase of the range gate data where the target is located and perform phase unwrapping to obtain the phase signal containing heartbeat and respiration information, and use an IIR filter to extract the respiration signal and the heartbeat signal respectively; adopt the time-frequency change algorithm based on cross-correlation entropy to perform frequency detection on the extracted respiration and heartbeat signals, and also extract the respiration signal with the highest intensity within the respiration frequency range; finally, calculate the real-time heart rate and respiration, and give an alarm for the situation of apnea; Step 7: Conduct sleep analysis, score respectively according to the ideal ranges of each item of data, and summarize through scoring weights; and give sleep quality evaluation and improvement suggestions in the sleep analysis report, and push them to the user through the cloud platform.

2. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter-wave radar according to claim 1, characterized in that: The sleep analysis in Step 7 is performed by combining the number of body movements, sleeping posture changes, average heart rate, average respiration frequency and the number of apnea times.

3. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter wave radar according to claim 1, characterized in that: In Step 2, the detection target is determined by the in-bed detection algorithm, and the in-bed detection algorithm is specifically carried out according to the following steps: Determine the bed parameters and spatial positions of the target user, so as to construct the effective area where the human body is in the bed, First, use the DBSCAN algorithm to perform clustering processing on the point clouds in the effective area; Then, according to calculate the cluster centroid, where m i =(x i , y i , z i ) represents the spatial position of a certain point cloud in the three-dimensional Cartesian coordinate system, and k represents the number of point clouds contained in the cluster; Finally, if the number of point clouds is higher than the empirical threshold and the cluster centroid is within the effective area, it will be judged that the user is in the bed and subsequent processing will be carried out.

4. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter-wave radar according to claim 1, characterized in that: In Step 3, the body movement signals of the detection targets in each partition are detected by the body movement detection algorithm, and the body movement detection algorithm is specifically carried out according to the following steps: Divide the effective area into multiple partition parts, which respectively correspond to the head, abdomen and limbs longitudinally; when the target is in a resting state, obtain the point cloud data at the chest and abdomen; When the user has limb movement or body tremor, obtain the point cloud data of the limb parts; determine the body movement signal of the target according to the number of point cloud data.

5. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter-wave radar according to claim 1, characterized in that: The sleeping posture state in Step 4 is judged by the sleeping posture discrimination algorithm, and the sleeping posture discrimination algorithm is specifically carried out according to the following steps: Divide the effective area where the bed is located into an upper layer area and a lower layer area; the lower layer area is the human body area when the target lies flat, and the upper layer area is the human body area when the target lies on its side; Count the number of point clouds in the upper area and the lower area; Determine the sleeping posture state according to the change amount of the point cloud data in the upper area and the lower area.

6. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter wave radar according to claim 1, characterized in that: In step 5, an adaptive noise extraction algorithm is used to extract the current scene background noise from the spectrum data. The adaptive noise extraction algorithm is specifically carried out according to the following steps: First, process the range image signals of the bed area and the non-bed area. After processing the point cloud data, the targets with motion information will be screened and excluded, and then the stationary signals will be filtered out. Extract the clutter and background noise therein, and filter out the background noise through smoothing filtering. When the signal intensity of the range gate of the suspected target is greater than the preset signal-to-noise ratio, it is confirmed as a target.

7. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter-wave radar according to claim 1, wherein: The time-frequency change algorithm based on cross-correlation entropy in step 6 is specifically carried out according to the following steps: Calculate the cross-correlation entropy of the multi-channel data; Perform Fourier transform on the cross-correlation entropy to obtain the cross-correlation entropy spectrum; Extract the respiration and heartbeat signals from the cross-correlation entropy spectrum.

8. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter wave radar according to claim 1, characterized in that: The following steps are also included in step 1: Step 1.1: Select a sawtooth wave FMCW as the transmitted signal s TX (t) = cos(2πf c t + πkt 2 + Φ0), when the electromagnetic wave emitted by the radar encounters a target during the scattering process, it will be reflected back; received through multiple antennas, and the received signal of a certain channel can be expressed as s RX (t) = s TX (t - τ) = cos(2πf c (t - τ) + πk(t - τ) 2 + Φ0); Mix the received signal and the transmitted signal, and then obtain the difference frequency signal s through low-pass filtering IF (t) = Acos(2πkτt + Φ IF ); Sample the difference frequency signals of each receiving channel respectively to form a three-dimensional radar data block; Step 1.2: Select the fast time-slow time sampling matrix of a certain receiving channel, perform FFT on the fast time dimension and perform non-coherent accumulation on the slow time dimension to obtain a one-dimensional range image spectrum for subsequent extraction of vital sign signals; Step 1.3: Perform static clutter filtering, 2D-FFT, and CFAR detection on each receiving channel respectively, and then perform angle estimation in the azimuth and elevation directions based on the information of the range-Doppler spectrum to complete ranging, velocity measurement, and angle measurement, and then obtain the point cloud data for subsequent judgment of being in bed, body movement, and sleeping posture.

9. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter wave radar according to claim 5, characterized in that: The sleeping posture state in step 4 is determined in the following manner: The data characteristics of the target user lying on the side are calculated according to the following formula: z > bed height + shoulder width / 2; Where z represents the height in the clustering centroid coordinate m = (x, y, z); When the height is greater than the threshold, the target user is in the side-lying state.

10. The non-contact sleeping posture detection and vital sign monitoring method based on millimeter-wave radar according to claim 9, wherein: The side-lying is judged according to the side-lying probability, and the side-lying probability is calculated according to the following formula: In the formula, the parameters k and b are used to adjust the slope and bias of the calculation formula; the variable x is the difference between the centroid height and the set threshold, that is, x = z - (bed height + shoulder width / 2).

Images