Multi-radar vital sign monitoring system applied to fixed crowd places

Abstract

The invention discloses a multi-radar vital sign monitoring system applied to a fixed crowd place. The multi-radar vital sign monitoring system comprises a horizontal layer 24 GHz radar network, a vertical layer 60 GHz radar array and an auxiliary sensor module. Through three-dimensional collaborative deployment of heterogeneous radars, an improved variational mode decomposition algorithm and a joint probability data association tracking technology are combined to realize separation of breathing, heart rate and behavior characteristics of multiple targets in a dense scene. The system adopts a space-time diagram convolutional network to extract individual gait periodic features, establishes a unique biological feature coding file through dynamic gesture track matching, and completes non-contact identity binding. A multi-band radar chip and a dual-mode communication interface are integrated in hardware design, and a lightweight AI model is deployed in software architecture to realize localized real-time analysis. The embodiment covers nursing home tumble monitoring, hospital postoperative monitoring and kindergarten safety management and control scenes, the radar distance and the beam direction can be dynamically adjusted, and the problems of privacy disclosure and target confusion existing in a traditional monitoring system are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of radar monitoring, and in particular to a fixed-site vital sign monitoring system based on a multi-radar array combination, which is applicable to scenarios such as nursing homes, hospitals, schools, etc. that require long-term tracking of individual health and behavioral characteristics. Background Art

[0002] In the prior art, a single millimeter-wave radar is limited by angle resolution and signal processing capabilities, and it is difficult to distinguish multiple targets in a dense scenario (such as overlapping breathing signals of multiple people). Traditional camera monitoring has a risk of privacy leakage, while wearable devices have comfort and battery life problems. Recent research has proposed a multi-radar collaboration scheme, but core problems such as heterogeneous radar data fusion and individual feature binding have not been solved. Summary of the Invention

[0003] The present invention proposes a multi-radar vital sign monitoring system applied to a fixed population site, aiming to achieve multi-target tracking and continuous health monitoring through heterogeneous data fusion and feature binding. The system architecture is shown in Figure 1 and mainly includes a three-dimensional heterogeneous radar array and a core module.

[0004] In the three-dimensional heterogeneous radar array, the horizontal layer: deploy 24GHz FMCW radar (with strong penetration), the spacing is dynamically adjusted according to the site density (1.5 - 3m) to form a coverage network; the vertical layer: embed 60GHz millimeter-wave radar in the ceiling (with high resolution) to improve the detection accuracy of the pitch angle, and combine MIMO technology to achieve three-dimensional positioning; auxiliary sensors: infrared thermal imaging module (for detecting abnormal body temperature) and barometric pressure sensor (for identifying fall shock waves).

[0005] The core module includes a signal preprocessing unit, a feature binding engine, and an edge computing node. Signal preprocessing unit: decompose the multi-target mixed signal based on the improved WOA-VMD algorithm to suppress environmental noise; feature binding engine ( Figure 2 ) : establish a unique ID file for each individual through the gait cycle (0.8 - 1.2Hz) and gesture features (dynamic trajectory matching); edge computing node: deploy a lightweight AI model to achieve local real-time analysis (delay < 500ms).

[0006] The hardware design of this system needs to consider radar selection and communication interfaces. Radar selection: use the TIAWR1843 chip (4 transmitters and 3 receivers, bandwidth 4GHz) for the horizontal layer, and the Infineon BGT60LTR11AIP (60GHz, bandwidth 7GHz) for the vertical layer. The heterogeneous radars achieve nanosecond-level collaboration through a time synchronization module (PTP protocol); communication interface: support dual-mode redundancy of LoRaWAN (long-distance and low-power) and Wi-Fi 6 (high-speed data transmission).

[0007] The software design of this system includes a data flow pipeline and an AI algorithm model. Data flow pipeline: Raw signal → VMD decomposition → Multi-target tracking (JPDA algorithm) → Feature extraction (Wavelet transform + CNN) → Archive update; The AI algorithm model needs to complete multiple tasks such as gait recognition, gesture parsing, and anomaly detection. Gait recognition: Based on the spatio-temporal graph convolutional network (ST-GCN), with an accuracy rate > 92%; Gesture parsing: Adopt the YOLO-Hand model, supporting 12 preset gestures (such as waving for help); Anomaly detection: The LSTM network predicts the trends of respiration / heart rate, and a deviation from the threshold by ±20% triggers an alarm.

[0008] According to the system design, this system can be applied to specific demand scenarios such as fall monitoring in nursing homes, postoperative care in hospitals, and safety management in kindergartens.

[0009] Application scenario 1: Fall monitoring in nursing homes. 1) Radar array deployment: In the bedroom, a corner triangle topology is adopted (coverage blind area < 5%), and a hexagonal layout in the activity area; 2) Function: Determine a fall through the combined judgment of a sudden change in gait speed (decrease > 40%) and a change in body position height (1.5m → 0.3m), and link to the first aid system.

[0010] Application scenario 2: Postoperative care in hospitals. 1) Radar frequency band combination: 24GHz monitors respiration (error < 0.2 times / minute), and 60GHz detects fine hand movements (such as pressing the call button); Data fusion: When the heart rate is abnormal, automatically retrieve the thermal imaging to verify the body temperature.

[0011] Application scenario 3: Safety management in kindergartens. 1) Feature binding: Associate the child ID through a preset gesture (such as raising the hand to sign in) when entering the kindergarten; Electronic fence: Use radar point clouds to demarcate dangerous areas (such as windows), triggering an audible and visual alarm.

[0012] The technical advantages of this system are reflected in three aspects. First, there is good privacy protection: Only feature vectors are stored, and the original images are not retained; Second, high robustness: The multi-radar redundant design ensures that the system still maintains a coverage rate > 90% when a single node fails; Finally, low power consumption: Adopt dynamic power adjustment, and the standby power consumption < 5W. Description of the Drawings

[0013] Figure 1 : System three-dimensional deployment schematic diagram (horizontal + vertical radar + thermal imaging module).

[0014] Figure 2 : Working flow chart of the feature binding engine (signal decomposition → feature extraction → ID association).

[0015] Figure 3 : Application scenario example (topology of the bedroom and activity hall in a nursing home). Detailed Implementation Manner

[0016] Taking a nursing home as an example, illustrate the installation method and process of the multi-radar vital sign monitoring system ( Figure 3 ). 1) Install 6 24GHz radars (at a height of 1.2m on the wall) and 3 60GHz radars (suspended 0.5m below the ceiling); 2) Register the gait baseline for each elderly person through a preset gesture (such as crossing hands); 3) Deploy a lightweight ST-GCN model (model size <50MB) at the edge node, and update the feature library every 30 minutes.

Claims

1. A multi-radar vital signs monitoring system for fixed crowd places, characterized in that: It includes a horizontal 24GHz radar network, a vertical 60GHz radar array and auxiliary sensors, and achieves multi-target tracking through heterogeneous data fusion and feature binding.

2. The system according to claim 1, characterized in that: The horizontal layer radar adopts a honeycomb or linear topology, and the vertical layer radar is installed at a depression angle of 30°-60° to form three-dimensional coverage.

3. The system according to claim 1, characterized in that: The feature binding engine generates a unique biometric code through gait cycle, gesture trajectory and vital sign parameters.

4. The system according to claim 1, characterized in that: The improved WOA-VMD algorithm is used to perform modal decomposition of mixed signals and combined with the JPDA algorithm to achieve multi-target tracking.

5. A radar array deployment method, characterized in that: The radar spacing is dynamically adjusted according to the density of people in the venue. The calculation formula is: D=k⋅c2f⋅N (where k is the environmental coefficient, c is the speed of light, f is the radar frequency, and N is the expected number of people to be monitored).

Images