Fall detection and physiological sign and sleep monitoring and early warning system and method

Abstract

The invention discloses a fall detection, physiological sign and sleep monitoring and early warning system and method, and relates to the field of health monitoring, the system comprises a millimeter wave radar, a function module and a data real-time monitoring and early warning module; the signal output end of the millimeter wave radar is connected with the signal input end of the function module. The signal output end of the function module is connected with the signal input end of the data real-time monitoring and early warning module. The function module obtains physiological sign data through the phase change original signal and obtains the sleep condition based on the physiological sign data; generating point cloud data based on the original signal, and further obtaining human body posture data; the data real-time monitoring and early warning module is used for processing, storing and visually displaying the physiological sign data and the human body posture data, and carrying out real-time detection and fall early warning on abnormal physiological sign data, sleep and human body postures. According to the invention, the safety of medical monitoring can be improved, and more efficient and reliable technical support is provided for intelligent medical treatment and health management.

Description

[0001] This application claims priority to the invention patent application No. 202510252980.2 filed by Huazhong University of Science and Technology with the State Intellectual Property Office on March 5, 2025. Technical Field

[0002] This application relates to the field of health monitoring, and in particular to a fall detection, physiological signs and sleep monitoring and early warning system and method. Background Technology

[0003] In recent years, with the rapid development of artificial intelligence, cloud computing, the Internet of Things, and big data technologies, intelligent health monitoring technologies have made continuous breakthroughs and have been widely applied in fields such as medical health, smart elderly care, and safety monitoring. Especially in scenarios involving home-based elderly care and medical care, falls and sudden physiological abnormalities (such as cardiovascular diseases) have become major health risks. How to efficiently and accurately detect fall events, while simultaneously achieving non-contact real-time monitoring of physiological signs and providing timely warnings, has become an important direction for research in health monitoring technologies.

[0004] Currently, there are various fall detection, physiological sign, and sleep monitoring solutions on the market, which can be broadly categorized into contact and non-contact methods. Contact methods mainly rely on wearable devices, such as smart bracelets, heart rate monitors, and sleep trackers. While these devices can measure basic physiological parameters, they have limitations in terms of wearing comfort, user compliance, and long-term convenience. Furthermore, elderly individuals may experience difficulties using these devices due to forgetfulness, aversion, or mobility issues. Non-contact methods primarily rely on technologies such as visible light cameras, infrared sensors, and lidar. While they can monitor human activity to some extent, they exhibit poor stability in complex environments (such as changes in lighting or obstructions) and also raise privacy concerns.

[0005] Millimeter-wave radar, as an emerging non-contact sensing technology, has shown broad application prospects in smart homes, smart healthcare, and security monitoring in recent years. Millimeter-wave radar possesses strong penetration and all-weather operation capabilities, and can accurately capture subtle human movement signals to achieve real-time monitoring of respiration, heart rate, and sleep. Furthermore, combined with advanced signal processing algorithms and deep learning models, millimeter-wave radar can efficiently identify changes in human posture, accurately detect falls, and provide abnormal warnings based on trends in physiological parameters. Compared to traditional detection technologies, millimeter-wave radar offers greater environmental adaptability and privacy protection advantages, providing a more intelligent, seamless, and non-invasive solution for health monitoring. However, current research only focuses on how to process millimeter-wave radar signals to achieve fall detection or physiological sign and sleep monitoring methods; research on integrated system structures using millimeter-wave radar for fall detection, physiological sign, and sleep monitoring is lacking. Summary of the Invention

[0006] The purpose of this application is to provide a fall detection, physiological sign and sleep monitoring early warning system and method to improve the safety of medical monitoring, and thus provide more efficient and reliable technical support for smart healthcare and health management.

[0007] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a fall detection, physiological signs and sleep monitoring and early warning system, including: millimeter wave radar, functional modules and real-time data monitoring and early warning modules; The signal output terminal of the millimeter-wave radar is connected to the signal input terminal of the functional module; the signal output terminal of the functional module is connected to the signal input terminal of the real-time data monitoring and early warning module. The functional module is configured to perform the following: obtaining physiological sign data through phase change based on the raw signal output by the millimeter-wave radar, and obtaining sleep status based on the physiological sign data; generating point cloud data based on the raw signal output by the millimeter-wave radar, and obtaining human posture data based on the point cloud data; the physiological sign data includes: respiratory data and heart rate data; the human posture data includes human movement trajectory and human posture recognition results. The real-time data monitoring and early warning module is configured to: process and store the physiological signs data and the human posture data, and visualize the processed physiological signs data, sleep status and human posture data. At the same time, it performs real-time detection and fall warning for abnormal physiological signs data, abnormal sleep and abnormal human posture.

[0008] Secondly, this application provides a fall detection, physiological sign, and sleep monitoring early warning method, applied to the aforementioned fall detection, physiological sign, and sleep monitoring early warning system. The fall detection, physiological sign, and sleep monitoring early warning method includes: Physiological sign data is obtained from the raw signal output by millimeter-wave radar through phase changes, and sleep status is obtained based on the physiological sign data; the physiological sign data includes: respiratory data and heart rate data; Point cloud data is generated based on the raw signal output by millimeter-wave radar, and human posture data is obtained based on the point cloud data; the human posture data includes human movement trajectory and human posture recognition results. The physiological signs data and human posture data are processed and stored, and the processed physiological signs data, sleep status and human posture data are visualized. At the same time, real-time detection and fall warning are performed for abnormal physiological signs data, sleep abnormalities and abnormal human posture.

[0009] According to the specific embodiments provided in this application, this application has the following technical effects: 1. This invention achieves non-contact, multi-dimensional, 24 / 7 health and safety monitoring while protecting user privacy. Utilizing millimeter-wave radar technology, it overcomes the shortcomings of traditional camera monitoring solutions, such as privacy violations and susceptibility to light variations. It also avoids the problems of frequent charging and discomfort associated with wearable devices. The system integrates fall detection, physiological sign monitoring, and sleep quality assessment, and can operate in complex environments such as darkness and smoke, providing users with seamless, 24 / 7 health protection.

[0010] 2. A deep learning network based on physical perception and attention mechanisms is proposed, significantly improving the accuracy of pose estimation and fall detection. Addressing the challenges of sparse, disordered radar point clouds lacking semantic information, this invention innovatively constructs a deep learning network: Physical consistency constraint: By introducing the radar Doppler velocity projection principle to construct a physical loss function, and incorporating the electromagnetic wave imaging mechanism as a dynamic constraint into the deep learning network, this ensures that the predicted human joint movements conform to physical laws, significantly improving the realism and accuracy of pose reconstruction. Soft mapping relationship establishment: Utilizing a cross-attention mechanism, a soft association between disordered point clouds and human key points is dynamically established, effectively solving the mapping problem from sparse point clouds to the human skeleton, achieving high-precision pose perception even without a large amount of semantic label data.

[0011] 3. Employing a multi-level signal processing algorithm significantly improves the robustness and anti-interference capability of physiological sign monitoring. This invention employs a sophisticated link design in the signal processing module: locking the chest cavity location through range-to-Fourier transform, combined with static clutter suppression, DC component removal, and adaptive target range cell selection, effectively filtering out environmental noise; in particular, utilizing phase difference enhancement of the heartbeat component and pulse denoising technology based on exponentially weighted moving average solves the problem of traditional radar being susceptible to respiratory harmonic interference when monitoring weak heartbeat signals, achieving stable and high-precision estimation of respiratory rate and heart rate.

[0012] 4. A dual verification mechanism of "posture + vital signs" has been established, greatly reducing the false alarm rate of fall detection. Unlike the traditional single-mode approach that relies solely on posture changes to determine falls, this invention, after identifying the fall posture through a deep learning network, simultaneously monitors the monitored person's breathing and heart rate in real time. This multimodal fusion early warning mechanism can effectively distinguish between a genuine fall and similar actions such as the user voluntarily lying down to rest, ensuring the extremely high accuracy of the early warning system.

[0013] 5. Real-time edge-side processing and intelligent sleep assessment: This system can perform complex preprocessing such as DC mean removal, MTI clutter suppression, and CFAR target detection at the edge, reducing data transmission pressure and ensuring real-time monitoring. Simultaneously, based on deep learning algorithms, it analyzes long-term vital sign data to accurately predict sleep quality and provides timely warnings when vital sign data or sleep quality deviate from normal thresholds, providing reliable data support for chronic disease management and emergency response. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of a fall detection, physiological signs and sleep monitoring and early warning system provided in an embodiment of this application.

[0016] Figure 2 This is a schematic diagram of the functional module structure provided in one embodiment of this application.

[0017] Figure 3 A flowchart of the vital signs algorithm processing provided in an embodiment of this application.

[0018] Figure 4 This is a schematic diagram showing the result of adaptive target bin selection based on vital sign signals provided in an embodiment of this application.

[0019] Figure 5 This is a comparison chart of minute-level error results for exhalation heart rate in vital signs signals provided in an embodiment of this application.

[0020] Figure 6 This is a comparison chart showing the second-level error results of the exhalation heart rate in the vital signs signal provided in an embodiment of this application.

[0021] Figure 7 A flowchart of human posture detection data processing provided in an embodiment of this application.

[0022] Figure 8 This is a structural diagram of a sparse reconstruction algorithm provided in an embodiment of this application.

[0023] Figure 9 This is a schematic diagram of the structure of a real-time data monitoring and early warning module provided in one embodiment of this application.

[0024] Figure 10 This is a schematic diagram of another structural component of a fall detection, physiological sign and sleep monitoring and early warning system provided in one embodiment of this application.

[0025] Figure 11 This is a basic flowchart of a fall detection, physiological signs and sleep monitoring and early warning system provided in one embodiment of this application.

[0026] Figure 12 A flowchart illustrating a human posture warning provided in an embodiment of this application.

[0027] Figure 13 This is a schematic diagram of the framework structure of a fall detection, physiological signs and sleep monitoring and early warning system provided in an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] In one exemplary embodiment, this application provides a fall detection, physiological sign and sleep monitoring and early warning system, such as Figure 1As shown, the system includes: millimeter-wave radar 100, functional module 200, and real-time data monitoring and early warning module 300.

[0031] The signal output terminal of the millimeter-wave radar 100 is connected to the signal input terminal of the functional module 200. The signal output terminal of the functional module 200 is connected to the signal input terminal of the real-time data monitoring and early warning module 300.

[0032] Functional module 200 is configured to perform the following: Based on the raw signal output by millimeter-wave radar 100, obtain physiological sign data through phase transformation, and determine sleep status based on the physiological sign data. It also generates point cloud data based on the raw signal output by millimeter-wave radar 100, and obtains human posture data based on the point cloud data. The physiological sign data includes respiratory data and heart rate data. The human posture data includes human movement trajectory and human posture recognition results.

[0033] The real-time data monitoring and early warning module 300 is configured to: process and store physiological sign data and human posture data, and visualize the processed physiological sign data, sleep status and human posture data. At the same time, it performs real-time detection and fall warning for abnormal physiological sign data, sleep abnormalities and abnormal human posture.

[0034] In another exemplary embodiment of this application, the functional module 200 includes a target locking unit, such as Figure 2 As shown, it may also include a target locking unit, a physiological sign monitoring unit 201, and a human posture detection unit 202.

[0035] The signal input terminal of the target locking unit serves as the signal input terminal of the functional module 200. The signal input terminal of the target locking unit is connected to the signal input terminals of the physiological sign monitoring unit 201 and the human posture detection unit 202, respectively. The signal output terminals of the physiological sign monitoring unit 201 and the human posture detection unit 202 are integrated into the signal output terminal of the functional module 200.

[0036] The target locking unit is configured to perform: a Fast Fourier Transform (FFT) on the raw signal output by the millimeter-wave radar 100 based on the range dimension to obtain the target's range information, and then lock the target's position through adaptive threshold filtering. Essentially, the target locking unit calculates the target's range information based on the range dimension FFT.

[0037] The physiological sign monitoring unit 201 is configured to perform the following after locking the target location: static clutter suppression and DC component removal, adaptive target distance unit selection, phase extraction and unwrapping, phase differential enhancement of the heartbeat component, pulse denoising based on exponential weighted moving average, and bandpass filtering on the signal corresponding to the target location in sequence to obtain the correlation matrix of the signal corresponding to the target location; extract the main frequency feature from the signal corresponding to the target location based on the correlation matrix, and use it as the physiological sign data.

[0038] The physiological data includes respiratory and heart rate, as well as signal waveforms. The specific signal processing procedure is as follows: Figure 3 As shown in Table 1, our parameter configuration during this process includes:

[0039] Interference signals introduced by static objects such as walls, ceilings, and furniture in radar echoes are processed by static clutter suppression for the range dimension signal. Simultaneously, the DC component (DC offset) in the signal is zeroed out to eliminate DC bias introduced by system hardware and the environment. The formula is as follows:

[0040] Considering that minute displacements of the human thoracic cavity may be distributed across multiple adjacent range cells, this step introduces an adaptive target bin selection strategy. The energy or coherence of the main target range-bin and its multiple adjacent range-bins are jointly evaluated, and the energy-weighted target range cell group is selected as follows: Figure 4 As shown. Complex phase information is extracted from the echo signal of the selected target range unit, and a phase sequence is obtained through the arctangent function. Due to periodic phase transitions, the phase sequence is unwrapped to obtain a continuously changing phase signal, thus accurately reflecting the micro-displacement of the chest cavity caused by respiration and heartbeat. The formula is as follows:

[0041] The unwrapped phase signal is then subjected to time-differencing to suppress low-frequency trend terms and enhance high-frequency components. This processing significantly enhances the minute displacement signal caused by heartbeats while reducing the impact of slow-varying interference on heart rate estimation. The formula is as follows:

[0042] Next, to address potential instantaneous spike noise in the phase signal, an exponentially weighted moving average (EWMA) filtering method is used to smooth the signal. By assigning exponentially decaying weights to historical samples, effective suppression of sudden noise is achieved while preserving the temporal characteristics of the vital signs signal. The formula is as follows:

[0043] Where, x[n]: original phase signal, y[n]: EWMA filter output, λ: exponential weighting coefficient (smoothing factor); After bandpass filtering, the correlation matrix of the signal is constructed and the dominant frequency features are extracted, thus achieving stable and high-precision estimation of human respiratory rate and heart rate. Experimental results show that the error in heart rate at the second and minute levels is maintained at approximately 1.8 bpm. Figure 5 and Figure 6 As shown in Table 2, the results are compared with those of existing models, enabling continuous monitoring of vital signs and early warning of abnormalities, and then estimating the sleep quality of the monitored person through deep learning algorithms.

[0044] The human posture detection unit 202 is configured to perform the following: receiving raw signals acquired by millimeter-wave radar; obtaining target dynamic feature information by performing static clutter suppression on the raw signals at the edge side; applying a Hanning window function to the target dynamic feature information, and then calculating the target's distance information based on FFT in the distance dimension; setting a distance threshold for the distance information to obtain effective distance units; calculating the target's velocity information based on FFT in the Doppler dimension; setting zeros for frequency points below the noise threshold for the target's velocity information to obtain candidate units with significant energy characteristics; using the CFAR detection algorithm to perform target detection and extraction on the candidate units to obtain reliable target points; processing the reliable target points based on the sparse reconstruction spatial spectrum estimation algorithm to obtain the target's azimuth and pitch angle information, and calculating the target's point cloud data in combination with the target's distance information; and inputting the point cloud data into a trained deep learning network to obtain the human posture data.

[0045] The processing flow of the human posture data is as follows: Figure 7 As shown in Table 3, the configurable parameters in this data processing process are as follows:

[0046] Step S1: Millimeter-wave radar signal reception and preprocessing. The human pose detection unit first receives the raw intermediate frequency echo signal acquired by the millimeter-wave radar front end. Given that the raw signal contains human target echoes, static environmental reflections (such as reflections from walls, ground, and fixed furniture), and system thermal noise, and that the static background accumulates strong energy in both the range and Doppler dimensions, interfering with dynamic target detection, this step specifically includes the following processing: DC removal: The raw radar signal is processed to remove the DC component (DC Mean Removal) to eliminate the constant DC bias component introduced by the coupling of radar hardware circuits and fixed strong reflection points. Where: x(n) is the original ADC signal, and N is the number of sampling points.

[0047] Static clutter suppression: Based on DC removal, the Moving Target Indication (MTI) algorithm is applied to suppress static background clutter that changes slowly on the time axis by performing inter-frame or inter-pulse differential processing on the radar signal, thereby retaining only the dynamic characteristics of human motion with Doppler frequency shift. Where: k represents the frame index, and n represents the ADC sampling point.

[0048] Step S2: Range FFT Feature Extraction. Perform a Fast Fourier Transform (Range FFT) on the signal preprocessed in Step S1 in the range dimension to map the time-domain signal to the range-frequency domain in order to calculate the radial range information of the target relative to the radar.

[0049] Windowing: Preferably, a Hanning window function is applied to the input signal before performing the FFT calculation. This process aims to smooth abrupt changes at signal truncation points, effectively suppress spectral leakage effects and sidelobe energy, thereby increasing the concentration of main lobe energy and enhancing the range resolution capability for weak targets in complex noise environments.

[0050] , where w(n) is the windowing function and m is the distance bin index.

[0051] Distance threshold filtering: Further, a distance threshold is set based on the actual physical dimensions of the indoor monitoring scenario. The system automatically discards invalid range bins that are less than the minimum monitoring distance or greater than the maximum monitoring distance. .

[0052] Step S3: Doppler Feature Extraction (Doppler FFT) Based on the distance domain data output in Step S2, the system further performs a Fast Fourier Transform (Doppler FFT) in the Doppler dimension to obtain the radial velocity information of the target.

[0053] Velocity spectrum calculation: The Doppler frequency shift is used to directly characterize the target's motion state, thereby distinguishing between dynamic human behavior and static background. T f The frame period.

[0054] Noise threshold decision: The system is equipped with a noise threshold decision mechanism to address low-amplitude disturbances in the Doppler spectrum caused by human body micro-movements or system thermal noise. For Doppler frequency points with amplitudes lower than the preset noise threshold, their amplitudes are set to zero or nonlinearly suppressed.

[0055] Step S4: CFAR-based target detection and point cloud optimization. In order to accurately extract effective human target points from the range-Doppler map, this embodiment adopts the constant false alarm rate (CFAR) detection algorithm.

[0056] Adaptive detection: The CFAR detector adaptively estimates the local noise power level by sampling the reference cells around the detected cell (CUT) and dynamically adjusts the detection threshold accordingly to ensure a constant false alarm probability under different background noise intensities. α is the threshold parameter, N ref Reference cell number, point cloud generation: Only retain valid target cells whose amplitude exceeds the CFAR detection threshold.

[0057] Step S5: Angle Estimation and 3D Point Cloud Construction After obtaining the target's distance index and velocity index, the system uses the receiving antenna array data to calculate the angle.

[0058] Sparse Reconstruction Angle Measurement: A spatial spectrum estimation algorithm based on sparse reconstruction is used to jointly estimate the target's azimuth and elevation angles. Compared to traditional beamforming techniques, this method achieves higher angular resolution with finite aperture antenna arrays and effectively overcomes the angular ambiguity problem. Results are shown below. Figure 8 As shown: .

[0059] Coordinate mapping: The calculated azimuth and pitch angles are spatially geometrically mapped with the distance and velocity information obtained in steps S2 and S3, and finally a human point cloud dataset containing three-dimensional coordinates, radial velocity and signal-to-noise ratio is constructed.

[0060] The point cloud data is input into the PIGT model, and the point cloud data is input into the neural network for deep learning based on feature extraction. The model is trained with a large amount of sample data to perform feature classification and recognition of human posture. After training, the output recognition result of the neural network is transmitted to the real-time data monitoring and early warning module to realize real-time perception and anomaly detection of human posture.

[0061] In this embodiment, the specific process of the deep learning network is as follows: Step S1: Data Acquisition and Preprocessing. Acquire continuous time-series millimeter-wave radar point cloud data; the point cloud data includes the three-dimensional spatial coordinates, Doppler radial velocity, and signal-to-noise ratio characteristics of each scattering point; construct a radar point cloud feature sequence.

[0062] Step S2: Radar Feature Encoding and Semantic Mapping. A neural network model containing a radar encoder and a spatiotemporal graph Transformer is constructed. A set of learnable joint query vectors is introduced to represent the latent semantic features of key points in the human skeleton. A cross-attention mechanism is used to calculate the interaction between the joint query vectors and the radar point cloud features, aggregating the disordered radar point cloud features into a sequence of human skeleton features with spatiotemporal semantics. The point cloud feature extraction module is expressed as follows:

[0063] Where: t is the time step index, which indicates which frame of data is being processed. i is the radar scattering point index, indicating which point in the current frame. It is a high-dimensional feature vector. This is the raw radar input vector. This is a multi-layer perceptron. Here, it acts as a feature extractor, mapping low-dimensional physical data to a high-dimensional feature space.

[0064] Step S3: Physical Consistency Constraint Construction During the model training phase, a physical guidance module is constructed to constrain the network, specifically including: Sub-step S3.1: Kinematic decoding. The decoder is used to predict the three-dimensional positions and three-dimensional velocity vectors of the human joints for each frame from the human skeleton feature sequence.

[0065] Sub-step S3.2: Soft association establishment. Based on the weights of the cross-attention mechanism, calculate the point cloud-joint association matrix, which represents the probability distribution of each radar scattering point belonging to a specific human joint. The cross-attention mechanism and the association matrix formula are as follows:

[0066] in: Let be the skeleton feature matrix of frame t. A learnable joint query matrix. Let be the radar feature matrix of frame t. Scaling factor () is the normalized exponential function.

[0067]

[0068] in, For association probability weights, For the predicted three-dimensional position of the joint, This represents the actual three-dimensional location of the radar point.

[0069] Sub-step S3.3: Doppler physical residual calculation. Based on the principle of radar Doppler imaging, calculate the theoretical projection velocity of the predicted joint three-dimensional velocity vector in the radar line-of-sight direction; according to the point cloud-joint correlation matrix, calculate the weighted error between the theoretical projection velocity and the true Doppler radial velocity in the original radar point cloud, as the physical constraint loss function.

[0070]

[0071] The physical constraint loss function, For the observed Doppler velocity, For the predicted joint three-dimensional velocity vector, For the location of the radar sensor, It is a unit direction vector.

[0072] Step S4: Joint Optimization and Behavior Classification. The human skeleton feature sequence is input into the classifier, which outputs the behavior category prediction result. A total loss function is constructed, which is a weighted sum of the behavior classification loss and the physical constraint loss. The total loss function is used to perform end-to-end iterative training on the neural network model until the model converges. The hyperparameter combinations are shown in Table 4, and the results are compared with those of existing models in Table 5. The loss calculation formula is as follows:

[0073] in: For total training losses, For classifying losses, These are the physical constraint weighting coefficients.

[0074]

[0075] In another exemplary embodiment of this application, such as Figure 9 As shown, the real-time data monitoring and early warning module 300 includes: a data receiving unit 301, a data display unit 303, a data storage unit 302, and an anomaly early warning unit 304.

[0076] The signal input terminal of the data receiving unit 301 serves as the signal input terminal of the real-time data monitoring and early warning module 300. The signal input terminal of the data storage unit 302 is connected to the signal output terminal of the data receiving unit 301. The signal output terminal of the data storage unit 302 is connected to the signal input terminals of the data display unit 303 and the anomaly early warning unit 304, respectively. The signal output terminals of the data storage unit 302, the data display unit 303, and the anomaly early warning unit 304 are integrated into the signal output terminal of the real-time data monitoring and early warning module 300.

[0077] The data receiving unit 301 includes multiple real-time data receiving interfaces. These interfaces are connected to the physiological sign monitoring units 201 and the human posture detection units 202, respectively.

[0078] Data storage unit 302 uses the InfluxDB time-series database to store all real-time physiological sign data and human posture data received from functional module 200, and uses the MySQL structured database to store static structured data.

[0079] The data display unit 303 is used to display physiological sign data, human posture data, and calculated sleep quality data received from the functional module 200 in the form of charts. The data display unit 303 is connected to a data interaction interface, which connects to an external smart device, such as a web client or a mini-program client.

[0080] The abnormality warning unit 304 judges the received physiological signs data, sleep quality data and human posture. If there is abnormal data, it will issue an alert on the web terminal and the mini-program terminal to remind users to check the status of the monitored personnel and equipment. If there is a prolonged abnormality, it will notify the contact person or emergency contact person set on the web terminal and the mini-program terminal via SMS or telephone.

[0081] In some embodiments, the real-time data receiving interface provides independent interface addresses for the physiological sign monitoring unit 201 and the human posture detection unit 202 to receive the corresponding data streams. Human posture data, physiological sign data, and sleep quality data are written via the real-time data receiving interface and sent to the backend via Hypertext Transfer Protocol (HTTP) requests using a POST request method. Upon receiving the HTTP request, the data receiving unit 301 parses the JSON format data, establishes connections with the MySQL structured database and the InfluxDB time-series database, performs format conversion and processing on the data to meet the storage requirements of the target tables, and writes it into the corresponding tables to achieve persistent data storage. In some embodiments, the InfluxDB time-series database includes user physiological characteristics data tables, user motion trajectory data tables, and point cloud data tables, etc. The MySQL structured database includes user information data tables, device information data tables, sleep quality data tables, alarm record data tables, and user access log data tables, etc.

[0082] In some embodiments, the data display unit 303 reads real-time data from the InfluxDB time-series database and provides the data for display to the web and mini-program clients via an interface address. Simultaneously, the data display unit 303 also reads static structured data from a MySQL structured database to support data maintenance and querying.

[0083] In some embodiments, the data display unit 303 provides users with data interaction interfaces (e.g., front-end UI interfaces) on both the web and mini-program sides, enabling users to view real-time data from each module and manage information about the equipment and the personnel being monitored. Furthermore, the front-end UI interface is built using the Vue framework (a JavaScript-based Vue framework and some plugins) to display the real-time data monitoring and early warning modules. The front-end UI interface can also manage front-end state through Vuex, sharing and managing data, states, functions, etc., shared by multiple components. Third-party libraries and utility classes / plugins are used to better display data; for example, Echarts displays data in chart form, and Canvas draws data to implement various functions. Alova is used to send various requests to the back-end server, including POST and GET requests.

[0084] For example, the front-end UI may include a homepage, a physiological signs and sleep module, a human posture module, and a device management module. The homepage displays the working status of all devices. The physiological signs module displays real-time physiological sign data for all monitored individuals, changes in physiological sign data over different time periods within a set date, and daily sleep patterns. The human posture module displays real-time point cloud data of the human postures of all monitored individuals, posture recognition results, and movement trajectories. The device management module provides users with an interactive interface for managing the millimeter-wave radar 100, functional modules 200, and information about the monitored individuals.

[0085] In some embodiments, the overall process of a human posture warning system is as follows: Figure 10 As shown, the anomaly warning unit 304 is configured as follows: Set a normal range threshold for respiratory rate and a normal range threshold for heart rate. If the respiratory rate in the respiratory data or the heart rate in the heart rate data is not within the normal range threshold for respiratory rate or heart rate, the physiological sign data is judged as abnormal data. The aforementioned physiological sign exceeding-limit early warning mechanism internally stores pre-set normal range thresholds for respiratory rate and heart rate. The monitoring module compares the real-time calculated respiratory rate and heart rate with these thresholds. If the real-time data falls outside these normal ranges, the system determines that the current physiological sign data is abnormal and triggers a corresponding level of health warning based on the degree of deviation.

[0086] Deep learning algorithms are used to calculate sleep quality-related data, and normal ranges for sleep time and sleep quality are set. When the sleep quality-related data differs significantly from the normal range threshold, the sleep quality data is judged to be abnormal.

[0087] The intelligent sleep quality assessment module is equipped with a long-term data analysis unit. It uses a trained deep learning algorithm to extract features from monitored breathing, heart rate, and body movement data to calculate sleep stages and sleep quality scores. The system has set benchmark ranges for standard sleep duration and sleep quality scores. When the deviation of the calculated sleep quality parameters from the benchmark range exceeds a preset anomaly threshold, the system determines that the sleep quality is abnormal and generates an analysis report.

[0088] Deep learning algorithms are used to identify the returned point cloud data. When the posture is in a falling state, the human posture data is determined to be abnormal.

[0089] To address the false alarm problem that may arise from single-posture recognition, this invention employs a dual-verification logic of "posture + vital signs" for accurate fall detection based on posture and vital signs fusion. Specifically: First, the deep learning algorithm performs real-time inference on the point cloud data transmitted from millimeter-wave radar to identify the current human posture. When the posture classification result output by the deep learning network is "fall," the system does not immediately trigger an alarm but instead enters a secondary confirmation stage. In this stage, the system continuously monitors the user's breathing and heart rate values ​​during the fall posture. If rapid breathing, a surge in heart rate, or weak or absent signals are detected, it is confirmed as a genuine fall event and an alarm is triggered immediately; if stable breathing and a resting heart rate are detected, it is determined to be the user's voluntary rest behavior. This multi-dimensional feature fusion ensures high accuracy in fall behavior recognition.

[0090] In another exemplary embodiment of this application, the structural composition and working principle of the fall detection, physiological signs and sleep monitoring and early warning system provided in this application are illustrated with specific application examples, wherein: like Figure 11 As shown, the real-time data monitoring and early warning module 300 provided in this application may include a backend, a database, a mini-program, and a web interface. The backend is used for data reception, processing, and transmission, and interacts with the functional module 200, the database, the mini-program, and the web interface.

[0091] Multiple radars are integrated into functional module 200, with each radar selecting an existing millimeter-wave radar 100. Based on this, functional module 200, built upon the millimeter-wave radar 100 hardware, implements different functions, serving as the lower layer of the real-time data monitoring and early warning system, providing real-time data to the system. After obtaining the millimeter-wave radar signal (i.e., the raw signal), the phase change of the signal is extracted. Respiration and heart rate signals are separated and extracted using filters or mode decomposition algorithms, and then respiration and heart rate are obtained. Subsequently, deep learning algorithms are used to calculate sleep quality. After obtaining multi-channel millimeter-wave radar 100 data, the obtained multi-channel millimeter-wave radar 100 data is converted into point cloud data, and human posture recognition and detection are performed on the point cloud data using deep learning algorithms.

[0092] The real-time data monitoring and early warning module 300 provides a real-time data receiving interface for data returned by the physiological sign monitoring, sleep quality monitoring, and human posture detection modules. This ensures that multiple radars do not interfere with each other. Real-time data is transmitted to the real-time data monitoring and early warning module 300 through the interface, processed, and stored in the database. The data is then displayed graphically, allowing users to easily view the physiological signs, sleep quality, and human posture data and historical records of the monitored personnel. This enables information management of the real-time status of the equipment and the monitored personnel. Simultaneously, when the real-time data monitoring and early warning module 300 receives abnormal data, it can process the abnormal information according to its specific characteristics and issue alerts on the mini-program and web platforms to remind users to check the status of the monitored personnel and equipment.

[0093] Based on the above description, the overall process of the fall detection, physiological signs and sleep monitoring early warning system provided in this embodiment is as follows: Based on the hardware of different millimeter-wave radars 100, functional module 200 receives the raw signals from the millimeter-wave radar 100, extracts the phase changes from the raw signals, and then separates and extracts the breathing and heartbeat signals through filters or mode decomposition algorithms. Finally, it uses deep learning algorithms to calculate the sleep quality. For multi-channel millimeter-wave radar 100 data (i.e., raw signals), it is first converted into point cloud data, and then human pose recognition and detection are performed through deep learning.

[0094] Then, the functional module 200 transmits the data back to the server in the backend through the data receiving interface provided by the real-time data monitoring and early warning module 300. After obtaining the real-time data, the backend processes the data, changes the data format, connects to the database, and stores the data in the corresponding table according to the different data.

[0095] The backend reads real-time data from the database and provides API interfaces to the system's frontend interface (e.g., the frontend UI) in the form of interface addresses, waiting for the frontend to call them. For example, the frontend refers to the web client and the mini-program client.

[0096] When the front-end opens the interface, it obtains real-time data through the API interface provided by the back-end, visualizes the data, and sends corresponding requests to the back-end when the user performs other operations. After obtaining the data, the operation results are displayed in charts.

[0097] When the system detects anomalies in the received data, it writes the anomalies to the corresponding table in the database. The front-end periodically queries the anomaly detection interface, and when an anomaly is detected, it promptly alerts the user to take action.

[0098] Based on the above description, the workflow for real-time monitoring and early warning of human posture and physiological signs based on millimeter-wave radar 100 provided in this embodiment is as follows: Figure 12 As shown.

[0099] Furthermore, the real-time data monitoring and early warning module 300 provided in this application can provide users with an interface to query the database for real-time physiological signs and human posture data of the monitored personnel, real-time equipment status, and related information data, and obtain physiological signs and human posture data transmitted back from various radars. The backend sets response status codes and response information, and the frontend determines the interface status based on the returned corresponding information. The backend processes and stores the various received data, storing them in different database tables according to data type, and reads real-time data from the database to provide various data acquisition API interfaces to the frontend.

[0100] The backend uses the Spring Boot framework to build the server, which is responsible for communicating with the functional module 200, the database, and the frontend, obtaining and processing data, and providing interfaces to the frontend. When the frontend calls the relevant interfaces, it returns data in JSON format.

[0101] Furthermore, the data storage unit 302 provided in this application uses InfluxDB time-series database and MySQL structured database as data storage tools, and sets up storage modes such as user physiological characteristic data table, user movement trajectory data table, point cloud data table, user information data table, device information data table, sleep quality data table, alarm record data table, and user access log data table to persistently record and store relevant information of the monitored personnel and device information. The backend writes real-time physiological characteristic data, sleep quality data, human posture data, and device information of the monitored personnel through data interaction with the database, and records the abnormal information in the database when abnormal data is detected.

[0102] Furthermore, such as Figure 13 The framework structure of the fall detection, physiological sign, and sleep monitoring and early warning system shown can, in practical applications, use smart devices providing web and mini-program interfaces as the access layer. Based on this, the backend Spring Boot framework can divide the entire business process into an entity layer, persistence layer, business layer, control layer, and some tooling or static resource layers. The entity layer defines class attributes, which are consistent with the fields in the database; one table corresponds to one entity class. The persistence layer is responsible for interacting with the database (i.e., data interaction), completing basic CRUD tasks through interfaces and configuration files. It can use the JDBC (Java Database Connectivity) library to connect to and manipulate the database, essentially implementing the persistence layer functionality itself. The business layer encapsulates one or more interfaces of the persistence layer into a service and provides the interface to the control layer. This functionality can be implemented using SQL statements based on the JDBC library. The control layer is responsible for the interaction between the front-end and the back-end. After receiving the front-end request, it calls the specific interface of the business layer and returns the specific data to the access layer. It can use JDBC and SQL statements to implement the functions of the persistence layer and the business layer as a whole, and provide interfaces for the front-end UI (i.e. the front-end UI interface) through the @RequestMapping and @GetMapping annotations and return data in JSON format.

[0103] In summary, the fall detection, physiological sign, and sleep monitoring and early warning system provided in this application utilizes millimeter-wave radars with different functions as data sources. A data receiving interface is provided at the back end of the real-time data monitoring and early warning module to receive real-time physiological sign data, human posture point cloud data, and radar equipment information transmitted from the radar. The received data is processed and calculated using algorithms, then stored in the corresponding data tables in the database. Simultaneously, real-time data is read from the database and provided to the front end via a data interface for data visualization. The real-time data monitoring and early warning module provides an interactive interface for users, enabling them to view the monitored person's status and historical data fluctuations in real time, and manage the device's operation. When the real-time data monitoring and early warning module detects an anomaly, it can promptly alert the user to take timely action, achieving 24 / 7, all-round monitoring of the monitored person.

[0104] Based on the same inventive concept, this application also provides a monitoring and early warning method applied to the aforementioned fall detection, physiological sign, and sleep monitoring and early warning system. The solution provided by this method is similar to the implementation scheme described in the aforementioned system. Therefore, the specific limitations of one or more fall detection, physiological sign, and sleep monitoring and early warning method embodiments provided below can be found in the limitations of the fall detection, physiological sign, and sleep monitoring and early warning system described above, and will not be repeated here. Based on this, the fall detection, physiological sign, and sleep monitoring and early warning method provided by this application includes: Step 1: Based on the raw signal output by millimeter-wave radar, physiological sign data is obtained through phase changes, and sleep status is determined based on this physiological sign data. Physiological sign data includes: respiratory data and heart rate data.

[0105] Step 2: Generate point cloud data based on the raw signal output from the millimeter-wave radar, and obtain human posture data based on the point cloud data. Human posture data includes human movement trajectory and human posture recognition results.

[0106] Step 3: Process and store physiological sign data and human posture data, and visualize the processed physiological sign data, sleep status and human posture data. At the same time, perform real-time detection and fall warning for abnormal physiological sign data, sleep abnormalities and abnormal human posture.

[0107] As one embodiment of this application, step 3, the process of real-time detection and fall warning of abnormal physiological signs, sleep abnormalities, and abnormal human posture, includes: When the respiratory rate in the respiratory data is outside the normal threshold range, or the heart rate in the heart rate data is outside the normal threshold range, the physiological signs are determined to be abnormal. The physiological sign exceedance warning mechanism stores pre-set normal range thresholds for both respiratory rate and heart rate. The monitoring module compares the real-time calculated respiratory rate and heart rate with these thresholds. If the real-time data falls outside these normal ranges, the system determines the current physiological sign data is abnormal and triggers a corresponding level of health warning based on the degree of deviation. When the difference between sleep quality-related data and the normal range threshold reaches a set value, a sleep abnormality is determined. The intelligent sleep quality assessment module is equipped with a long-term data analysis unit. It uses a trained deep learning algorithm to extract features from the monitored respiratory, heart rate, and body movement data to calculate sleep stages and sleep quality scores. The system sets a baseline range for standard sleep duration and sleep quality scores. When the deviation of the calculated sleep quality parameters from the baseline range exceeds a preset abnormality threshold, the system determines sleep quality is abnormal and generates an analysis report. The system determines whether the human posture changes within a set time period. If no change occurs, an abnormal posture is identified. A deep learning algorithm is used to identify the returned point cloud data to obtain a human posture prediction result. If the predicted posture indicates a fall, a fall warning is generated.

[0108] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0109] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0110] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0111] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0112] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A fall detection, physiological sign and sleep monitoring and early warning system, characterized in that, The fall detection, physiological signs and sleep monitoring and early warning system includes: millimeter wave radar, functional modules and real-time data monitoring and early warning modules; The signal output terminal of the millimeter-wave radar is connected to the signal input terminal of the functional module; the signal output terminal of the functional module is connected to the signal input terminal of the real-time data monitoring and early warning module. The functional module is configured to perform the following: obtaining physiological sign data through phase change based on the raw signal output by the millimeter-wave radar, and obtaining sleep status based on the physiological sign data; generating point cloud data based on the raw signal output by the millimeter-wave radar, and obtaining human posture data based on the point cloud data; the physiological sign data includes: respiratory data and heart rate data; the human posture data includes human movement trajectory and human posture recognition results. The real-time data monitoring and early warning module is configured to: process and store the physiological signs data and the human posture data, and visualize the processed physiological signs data, sleep status and human posture data. At the same time, it performs real-time detection and fall warning for abnormal physiological signs data, abnormal sleep and abnormal human posture.

2. The fall detection, physiological signs and sleep monitoring and early warning system according to claim 1, characterized in that, The functional module includes: a target locking unit, a physiological sign monitoring unit, and a human posture detection unit; the signal input terminal of the target locking unit serves as the signal input terminal of the functional module; the signal input terminal of the target locking unit is connected to the signal input terminals of the physiological sign monitoring unit and the human posture detection unit, respectively; the signal output terminals of the physiological sign monitoring unit and the human posture detection unit are integrated into the signal output terminal of the functional module. The target locking unit is configured to perform: perform a fast Fourier transform on the original signal output by the millimeter-wave radar based on the range dimension to obtain the target's range information, and lock the target's position through adaptive threshold filtering; The physiological sign monitoring unit is configured to perform the following after locking the target location: static clutter suppression and DC component removal, adaptive target distance unit selection, phase extraction and unwrapping, phase differential enhancement of the heartbeat component, pulse denoising based on exponential weighted moving average, and bandpass filtering on the signal corresponding to the target location in sequence to obtain the correlation matrix of the signal corresponding to the target location; extract the main frequency feature from the signal corresponding to the target location based on the correlation matrix and use it as the physiological sign data. The human posture detection unit is configured to perform the following: using a sparse reconstruction-based spatial spectrum estimation algorithm to obtain the azimuth and elevation angle information of the target based on the original signal output by the millimeter-wave radar; using a peak detection method to enhance the accuracy of the target's range information; and using a Doppler fast Fourier transform method to determine the target's velocity information to generate point cloud data of the target; and inputting the point cloud data into a deep learning model to extract features and classify human posture to obtain the human posture data.

3. The fall detection, physiological sign and sleep monitoring and early warning system according to claim 2, characterized in that, The human posture detection unit is configured to perform the following: receive the raw signal acquired by millimeter-wave radar; obtain target dynamic feature information by performing static clutter suppression on the raw signal at the edge; apply a Hanning window function to the target dynamic feature information; calculate the target distance information based on the target dynamic feature using FFT in the range dimension; set a range threshold for the distance information to obtain an effective range unit; and calculate the target velocity information based on FFT in the Doppler dimension using the effective range unit. For the target's velocity information, zeroing is applied to frequencies below the noise threshold to obtain candidate units with significant energy characteristics; the CFAR detection algorithm is used to extract reliable target points from the candidate units; the spatial spectrum estimation algorithm based on sparse reconstruction is used to process the reliable target points to obtain the target's azimuth and pitch information; and the target's point cloud data is calculated by combining the target's distance information. The point cloud data is input into a trained deep learning network to obtain the human posture data.

4. The fall detection, physiological signs and sleep monitoring and early warning system according to claim 3, characterized in that, The training process of the trained deep learning network includes the following steps: S1: Acquire continuous time series millimeter-wave radar point cloud data; the point cloud data includes the three-dimensional spatial coordinates, Doppler radial velocity, and signal-to-noise ratio characteristics of each scattering point; construct a radar point cloud feature sequence using the millimeter-wave radar point cloud data; S2: The original deep learning network is trained to obtain the trained deep learning network; the original deep learning network is a neural network model containing a radar encoder and a spatiotemporal graph Transformer. The radar encoder introduces a set of learnable joint query vectors to represent the latent semantic features of key points of the human skeleton. When the radar point cloud features are input into the radar encoder, the interaction information between the joint query vectors and the radar point cloud features is calculated. The interaction information is used to aggregate the radar point cloud features into human skeleton features with spatiotemporal semantics, and then input into the spatiotemporal graph Transformer for joint prediction.

5. The fall detection, physiological sign and sleep monitoring and early warning system according to claim 4, characterized in that, The total loss function of the deep learning network Represented as ;in, The physical constraint loss function, ; t is the sequence number of the time step, T is the total number of time steps, i is the sequence number of the point cloud, N is the total number of point clouds, k is the sequence number of the joint, and K is the total number of joints. The association probability weights in the association matrix are, For the observed Doppler velocity, For the predicted joint 3D velocity vector, The original 3D position of point cloud i. For the location of the radar sensor, Represents the square of the 2-norm; For classification loss function, These are the physical constraint weighting coefficients.

6. The fall detection, physiological signs and sleep monitoring and early warning system according to claim 4, characterized in that, The real-time data monitoring and early warning module includes: a data receiving unit, a data display unit, a data storage unit, and an anomaly early warning unit; the signal input terminal of the data receiving unit serves as the signal input terminal of the real-time data monitoring and early warning module; the signal input terminal of the data storage unit is connected to the signal output terminal of the data receiving unit; the signal output terminal of the data storage unit is connected to the signal input terminals of the data display unit and the anomaly early warning unit, respectively; the signal output terminals of the data storage unit, the data display unit, and the anomaly early warning unit are integrated into the signal output terminal of the real-time data monitoring and early warning module; the data receiving unit establishes a connection with the data storage unit, performs format conversion and processing on the data, so that the converted data meets the storage requirements of the data storage unit; The data storage unit is used to process and store the physiological signs data and the human posture data; The data display unit is used to visualize the processed physiological signs data, sleep status, and human posture data. The abnormality warning unit is used to calculate sleep quality using the physiological sign data, analyze the human posture data to provide fall warning, and realize real-time detection and fall warning of abnormal physiological sign data, sleep abnormalities, and abnormal human posture.

7. The fall detection, physiological signs and sleep monitoring and early warning system according to claim 6, characterized in that, Both the data display unit and the anomaly warning unit include a data interaction interface; the data interaction interface interacts with the peripheral smart terminal for data and warning; The peripheral smart terminal includes a web terminal and a mini-program terminal; it provides early warnings and reminds users to check the status of monitored personnel and devices on the web terminal and the mini-program terminal; and it sends early warning notifications via wireless communication when there are abnormalities in the status of monitored personnel and devices; the recipients of the early warning notifications are the contacts set on the web terminal and the mini-program terminal. The web client and the mini-program client provide a front-end UI interface for data interaction; the front-end UI interface is built using the Vue framework; the front-end UI interface includes a homepage, a physiological signs and sleep module, a human posture module, and a device management module; The homepage is used to display the working status of the device; the physiological signs and sleep module is used to display the real-time physiological signs data of all monitored personnel, the changes in physiological signs data at different time periods within a set date, and sleep status; the human posture module is used to display the real-time point cloud data, posture recognition results, and movement trajectory information of all monitored personnel; the device management module is used to provide a data interaction interface.

8. The fall detection, physiological signs and sleep monitoring and early warning system according to claim 6, characterized in that, The data storage unit includes an InfluxDB time-series database and a MySQL structured database; The InfluxDB time-series database stores the physiological signs data, sleep data, and human posture data received from the functional modules; the MySQL structured database stores the user's device data, personal information data, sleep report data, and historical statistical data; the display unit reads static structured data from the MySQL structured database to support data maintenance and querying.

9. A method for fall detection, physiological signs, and sleep monitoring and early warning, characterized in that, The fall detection, physiological signs, and sleep monitoring early warning system according to any one of claims 1-8, wherein the fall detection, physiological signs, and sleep monitoring early warning method comprises: Physiological sign data is obtained from the raw signal output by millimeter-wave radar through phase changes, and sleep status is obtained based on the physiological sign data; the physiological sign data includes: respiratory data and heart rate data; Point cloud data is generated based on the raw signal output by millimeter-wave radar, and human posture data is obtained based on the point cloud data; the human posture data includes human movement trajectory and human posture recognition results. The physiological signs data and human posture data are processed and stored, and the processed physiological signs data, sleep status and human posture data are visualized. At the same time, real-time detection and fall warning are performed for abnormal physiological signs data, sleep abnormalities and abnormal human posture.

10. The fall detection, physiological sign and sleep monitoring and early warning method according to claim 9, characterized in that, The process of real-time detection and fall warning for abnormal physiological data, sleep abnormalities, and abnormal human posture includes: When the respiratory rate in the respiratory data is outside the normal threshold range, or the heart rate in the heart rate data is outside the normal threshold range, the physiological sign data is determined to be abnormal. When the difference between the relevant sleep quality data and the normal range threshold reaches a set value, a sleep abnormality is determined. It is determined whether the human posture changes within a set time. If there is no change, the human posture is determined to be abnormal. A deep learning algorithm is used to identify the returned point cloud data. When it is determined that the person is in a fall state, the changes in respiratory and heart rates before and after the fall are obtained. The changes in respiratory and heart rate are used to make a second judgment on the fall behavior to obtain the human posture prediction result. If it is determined that the human posture prediction result belongs to a fall state, a fall warning is generated.

Images