An infectious disease contact person tracking and risk early warning system based on the Internet of Things

Abstract

The application relates to the technical field of Internet of Things and public health epidemic prevention, and discloses an infectious disease contactor tracking and risk early warning system based on Internet of Things, which runs on a terminal comprising an inertial measurement unit, a temperature and pressure sensor and an audio envelope circuit, and comprises the following modules: a sensor scheduling module, which acquires a radio frequency broadcast packet and motion data to generate a hardware interrupt signal; a central coordination processor, which responds to the interrupt to control the temperature and pressure sensor and the audio envelope circuit to switch states; a data preprocessing module, which extracts low-frequency air pressure and high-frequency acoustic signals; a baseline module, which performs time alignment on the signals to generate a macro baseline; a connectivity decision module, which performs physical isolation decision based on the macro baseline to generate a connectivity flag bit; and a risk score module, which combines the motion data, the radio frequency broadcast packet and the flag bit to calculate a risk index to trigger early warning. The application reduces power consumption through state machine scheduling, uses air pressure and acoustic characteristics to exclude physical barrier interference, and realizes fine and high-precision evaluation.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things (IoT) and public health and epidemic prevention technology, specifically to an IoT-based system for tracking and risk warning of infectious disease contacts. Background Technology

[0002] In infectious disease control, tracing close contacts during the incubation and infectious periods is crucial for blocking pathogen transmission. Existing contact tracing systems typically rely on smartphones or wearable devices to send radio frequency (RF) broadcast packets, estimating physical distances between individuals by extracting received signal strength indicators. However, methods that solely rely on RF signals for spatial distance measurement cannot identify physical barriers in the environment. When two users wearing devices are in adjacent rooms separated by solid walls or sealed glass, RF signals can still penetrate or diffract, leading the system to often calculate a closer physical distance and record it as a contact event. Since respiratory infectious diseases are primarily transmitted through droplets or aerosols in shared airspace, these physical barriers effectively block the transmission route. The system's lack of awareness of physical connectivity leads to numerous false alarms.

[0003] To address the misjudgments caused by spatial isolation, some technologies attempt to introduce acoustic or environmental sensors to help determine whether devices are in the same enclosed space. However, this directly leads to the contradiction of excessive power consumption of terminal devices. On low-power IoT nodes, continuous audio sampling and high-frequency environmental data acquisition will quickly deplete the device's battery, failing to meet the battery life requirements for prolonged daily wear. Simultaneously, existing early warning systems often employ simple distance-plus-time threshold triggering mechanisms when calculating exposure risk, failing to deeply integrate the wearer's relative movement characteristics. For example, they cannot distinguish whether two people are walking side-by-side in the same direction or merely brushing past each other briefly in a corridor. Due to the lack of dynamic scheduling of sensor operating states and the failure to deeply integrate spatial connectivity, physical distance, and dynamic trajectories, existing systems struggle to accurately and precisely quantify the true epidemiological exposure risk. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an Internet of Things (IoT)-based infectious disease contact tracing and risk warning system. It solves the problem that existing contact tracing systems rely solely on radio frequency signal ranging, which cannot identify spatial physical barriers and thus causes a large number of false warnings. It also solves the high energy consumption problem caused by introducing multi-dimensional environmental sensors to assist in decision-making in low-power IoT nodes, as well as the technical problem that existing systems lack deep integration of individual relative motion characteristics, thus failing to accurately quantify the true exposure risk.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] An Internet of Things (IoT)-based system for tracing and risk warning of infectious disease contacts, the system operates on an IoT terminal including a low-power communication module, an inertial measurement unit, a temperature and pressure sensor, and an audio envelope circuit, comprising:

[0007] The sensor scheduling module acquires radio frequency broadcast packets and motion data output by the inertial measurement unit (IMU) via a low-power communication module, generating a hardware interrupt signal accordingly. The central coprocessor responds to the hardware interrupt signal, controlling the switching of operating states between the temperature and pressure sensors and the audio envelope circuit. The data preprocessing module processes the air pressure values ​​output by the temperature and pressure sensors to generate low-frequency air pressure data and acquires high-frequency acoustic envelope signals via the audio envelope circuit. The baseline module receives the low-frequency air pressure data and the high-frequency acoustic envelope signal, performs time axis alignment operations, and generates a macroscopic air pressure baseline and a macroscopic acoustic baseline. The connectivity decision module performs physical isolation decisions based on the macroscopic air pressure baseline and the macroscopic acoustic baseline, generating a connectivity flag. The risk integration module performs integration operations based on the motion data output by the IMU, the radio frequency broadcast packets, and the connectivity flag, generating a cumulative risk index and triggering an early warning operation.

[0008] To achieve multidimensional feature extraction and hardware state machine scheduling, the sensor scheduling module extracts the original received signal strength indication attached to the radio frequency broadcast packet, performs smoothing filtering, and calculates the first-order time derivative based on the discrete-time difference model to obtain the time gradient feature characterizing the dynamic approach trend of the device. Specifically, this feature is obtained by calculating the difference between the smoothed received signal strength indication at the current time and the previous time and then dividing it by the time span parameter.

[0009] Simultaneously, the sensor scheduling module performs vector magnitude calculation and fusion on the raw triaxial acceleration data output by the inertial measurement unit to generate a comprehensive acceleration amplitude. It also calculates the sample variance of the comprehensive acceleration amplitude to obtain the acceleration variance feature. This feature is specifically obtained by calculating the sum of squares of the differences between each comprehensive acceleration amplitude and its arithmetic mean, and then dividing by the difference between the total number of samples and one. The sensor scheduling module sets the RF trigger flag when the time gradient feature is greater than the trend judgment threshold, and sets the motion trigger flag when the acceleration variance feature is greater than the motion energy wake-up threshold, jointly generating a hardware interrupt signal. The central coprocessor maintains a finite state machine with deep sleep, sniffing and inspection, and high-frequency tracking states. The central coprocessor switches to deep sleep when both the RF trigger flag and the motion trigger flag are reset; switches to sniffing and inspection when only one flag is set; and switches the finite state machine into high-frequency tracking state only when both flags are simultaneously set, controlling the temperature and pressure sensor and the audio envelope circuit to switch from sleep to working state.

[0010] After completing the aforementioned sensor wake-up process, the data preprocessing module calculates the physical difference between the reference temperature value output by the temperature and pressure sensor and the thermodynamic reference temperature to generate a relative temperature bias. A polynomial fitting model is used to solve for the feedforward error compensation term, which is obtained by multiplying the first-order and second-order terms of the difference between the reference temperature value and the thermodynamic reference temperature by the corresponding first-order and second-order thermal drift sensitivity coefficients, and then summing them. The data preprocessing module combines the original air pressure value with the feedforward error compensation term to obtain a high-precision air pressure value. A low-pass filter is used to perform frequency domain isolation on the high-precision air pressure value, outputting a smooth air pressure value sequence. Abnormal interference is eliminated when the absolute difference between adjacent values ​​jumps, outputting low-frequency air pressure data. A microphone transducer for capturing ambient sound waves is connected to the front end of the audio envelope circuit. An analog bandpass filter receives the original analog AC voltage signal and outputs a high-frequency AC signal. A full-wave precision rectifier converts it into a unipolar pulsating DC signal, and a hardware integrator network smooths this pulsating DC signal, converting it into a low-frequency envelope analog voltage. The data preprocessing module downsamples to obtain the acoustic energy sequence and calculates the energy mutation rate based on the long-term energy baseline of the ambient background noise. This mutation rate is specifically obtained by calculating the difference between the current acoustic energy amplitude and the long-term energy baseline, and then dividing by the sum of the long-term energy baseline and the fault tolerance bias. Based on this, the data preprocessing module confirms and outputs the high-frequency acoustic envelope signal.

[0011] To eliminate interference from spatial physical isolation, the baseline module filters remote cooperating nodes through a low-power communication module, receiving remote air pressure and acoustic data. The baseline module performs a sliding cross-correlation operation on the local high-frequency acoustic envelope signal and the remote acoustic data to extract local extrema and lock the true sound wave propagation delay. The baseline module uses the true sound wave propagation delay to perform time-domain inverse compensation alignment on the remote data, and performs weighted fusion based on a fundamental weighting factor constructed from node spatial distance and data variance attenuation laws to generate macroscopic air pressure and macroscopic acoustic baselines. The connectivity decision module subtracts the scaled macroscopic air pressure baseline from the low-frequency air pressure data to obtain the local microenvironment net air pressure feature sequence, uses a bandpass filter for frequency band isolation, and performs short-time energy normalization. The connectivity decision module uses the true sound wave propagation delay to perform time-domain inverse translation alignment on the remote microenvironment net air pressure feature sequence, constructs a discrete model based on local arithmetic expectation, quantifies geometric similarity to generate transient air pressure waveform coherence measures, and applies sliding filtering to generate historical steady-state coherence measures. The connectivity decision module converts the time-domain aligned high-frequency acoustic signal to the frequency domain for processing, outputting a smooth high-frequency power spectral density estimate. It calculates the high-frequency acoustic envelope difference by quantizing the logarithmic difference of the relative energy and introducing a noise floor bias mechanism. Specifically, this difference is achieved by calculating the logarithmic values ​​of the local and remote high-frequency power spectral density values, adding the noise floor bias constant, and then summing the absolute values ​​of the difference between these two sets of logarithmic values ​​within a specified interval. The connectivity decision module uses hysteresis comparison logic to generate pressure connectivity Boolean indicators and acoustic connectivity Boolean indicators, respectively, and performs logical AND operations to generate connectivity flags reflecting the physical isolation status.

[0012] Based on this, the system calculates the local step frequency sequence and local heading angle sequence, and acquires the remote step frequency sequence and remote heading angle sequence through the communication module. The risk integration module performs dynamic time warping on the dual-end step frequency sequences to generate a synchronization index mapping table, and uses the mapping table to align the remote heading angle sequence with the local heading angle sequence on the time axis. The risk integration module calculates the normalized covariance measure, classifies the motion state based on the local variance of the triaxial acceleration magnitude, and calculates the relative exposure weight coefficient for dynamic accompaniment states using a logistic regression smoothing mapping function. The risk integration module uses the path loss model to map radio frequency broadcast packets to distance assessment labels, and combines the spatial physical distance calculated by the real sound wave propagation delay to generate an equivalent statistical distance. The risk integration module uses a preset spatial attenuation floor bias constant and a basic pathogen release rate constant to calculate the basic attenuation risk value, and multiplies the basic attenuation risk value, connectivity flag, and relative exposure weight coefficient to output the transient exposure risk increment. The risk score module performs cumulative summation on the transient exposure risk increment within a sliding time window to update the cumulative risk index, and triggers an early warning operation when the cumulative risk index exceeds the limit and the average of the equivalent statistical distance is less than the near contact baseline.

[0013] This invention provides an Internet of Things (IoT)-based system for tracing contacts and providing early warning of risks associated with infectious diseases. It offers the following advantages:

[0014] 1. The sensor scheduling module of this invention utilizes a low-power communication module and an inertial measurement unit to generate hardware interrupt signals, and the central coprocessor uses a finite state machine to wake up the temperature and pressure sensors and audio envelope circuits as needed. It only switches to high-frequency tracking mode when specific motion characteristics and radio frequency triggering conditions are met, avoiding power consumption caused by constantly turning on high-energy-consuming sensors and extending the battery life of IoT terminals.

[0015] 2. The system of this invention utilizes low-frequency air pressure data and high-frequency acoustic envelope signals to construct a macroscopic baseline, and generates connectivity flags based on the coherence of the air pressure waveform and the difference in acoustic envelope. This processing method can identify the blocking effect of physical barriers such as walls and glass on the spread of pathogens, and eliminates false alarms caused by only spatial proximity but no actual air connection through logical AND operations, making the warning results more consistent with the actual contact logic of epidemic prevention.

[0016] 3. The risk integration module of this invention achieves time axis alignment of motion states between different individuals by dynamically warping the two-end step frequency sequence, and makes a comprehensive judgment by combining the heading angle, path loss model, and physical distance calculated by acoustic delay. This cumulative risk calculation method, which combines spatial distance, connectivity, and relative exposure weight coefficients, can more objectively quantify contact risk and provide high-confidence data support for infectious disease prevention and control. Attached Figure Description

[0017] Figure 1 This is an overall system architecture diagram of an embodiment of the present invention;

[0018] Figure 2 This is an overall flowchart of an embodiment of the present invention;

[0019] Figure 3 This is a bar chart comparing the tracking and recognition error rates in different scenarios of specific application embodiments of the present invention;

[0020] Figure 4 The graph shows the cumulative risk index changing over time in a specific application embodiment of the present invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See attached document Figure 1 This invention provides an Internet of Things (IoT)-based system for tracking and warning of infectious disease contacts, which operates on an IoT terminal. The hardware structure of the IoT terminal includes a main processor, a central coprocessor, a low-power communication module, an ultra-wideband (UWB) ranging module, an inertial measurement unit, a temperature and pressure sensor, an audio envelope circuit, and an electroacoustic conversion module (e.g., a speaker or piezoelectric buzzer) for transmitting high-frequency acoustic signals.

[0023] The IoT terminal operates its data processing logic independently based on an edge computing architecture. Data acquisition, synchronization processing, and risk score calculation within the terminal are all executed locally on the central coprocessor and main processor. Its core judgment process does not rely on the computing power support of external cloud servers, and only outputs results when an alert is triggered.

[0024] An IoT-based infectious disease contact tracing and risk warning system may include:

[0025] The module includes a sensor scheduling module, a data preprocessing module, a baseline module, a connectivity decision module, and a risk integration module.

[0026] The sensor scheduling module communicates with the low-power communication module and the inertial measurement unit. The sensor scheduling module acquires radio frequency broadcast packets sent by surrounding devices and extracts the time gradient characteristics of the received signal strength indication.

[0027] The sensor scheduling module simultaneously extracts the acceleration variance characteristics output by the inertial measurement unit. Based on the time gradient characteristics and acceleration variance characteristics of the received signal strength indication, the sensor scheduling module generates a hardware interrupt signal.

[0028] The central coprocessor responds to hardware interrupt signals and controls the temperature and pressure sensors and audio envelope circuitry to switch between sleep and active states.

[0029] The data preprocessing module is communicatively connected to the temperature and pressure sensors and the audio envelope circuit. It acquires the raw air pressure measurement values ​​and temperature reference values, performs air pressure and temperature compensation and digital bandpass filtering, and outputs low-frequency air pressure data.

[0030] The data preprocessing module acquires high-frequency energy characteristic data from the inertial measurement unit. Based on this high-frequency energy characteristic data, the data preprocessing module removes physical friction interference terms from the low-frequency air pressure data.

[0031] The data preprocessing module acquires the high-frequency acoustic envelope signal through the audio envelope circuit. The audio envelope circuit directly outputs the energy envelope of the specified frequency band.

[0032] The baseline module receives low-frequency air pressure data and high-frequency acoustic envelope signals output by the data preprocessing module. The baseline module filters remote cooperating nodes based on a preset radio frequency threshold and acquires the air pressure and acoustic data broadcast by the remote cooperating nodes.

[0033] The baseline module performs sliding window cross-correlation calculations on the locally acquired high-frequency acoustic envelope signal and the acoustic data broadcast by the remote cooperating node to determine the actual sound wave propagation delay.

[0034] The baseline module uses the actual sound wave propagation delay to perform time axis alignment calculations on local and remote air pressure and acoustic data, generating macroscopic air pressure baselines and macroscopic acoustic baselines.

[0035] The connectivity decision module receives data output from the data preprocessing module, as well as the macroscopic pressure baseline and macroscopic acoustic baseline generated by the baseline module. The connectivity decision module then separates the macroscopic pressure baseline and macroscopic acoustic baseline through differential operations.

[0036] The connectivity decision module calculates the Pearson correlation coefficient of the differentially processed air pressure waveform and the power spectral density difference of the acoustic envelope.

[0037] The connectivity decision module performs a double Boolean decision based on the difference between the Pearson correlation coefficient of the air pressure waveform and the power spectral density of the acoustic envelope, generating a connectivity flag to characterize the physical isolation state.

[0038] The risk integration module acquires the gait frequency sequence and heading angle data output by the inertial measurement unit. It performs dynamic time warping and alignment on the two-terminal gait frequency sequence, calculates the heading angle covariance, and maps it to generate relative exposure weight coefficients.

[0039] The risk integration module maps radio frequency signal strength data to statistical distance.

[0040] The risk integration module performs multidimensional dynamic risk integration calculations based on statistical distance, relative exposure weight coefficients, and connectivity flags to generate a cumulative risk index.

[0041] When the risk score module determines that the cumulative risk index has reached a preset threshold, it triggers an early warning operation.

[0042] S100: Sensor dynamic scheduling and state switching. Specific steps are as follows:

[0043] The sensor scheduling module periodically acquires the environmental radio frequency (RF) signal status through the terminal's underlying hardware bus interface. Considering that indoor RF signals are susceptible to multipath interference and fluctuations, directly using absolute signal strength to determine physical distance can easily lead to false triggering. Based on these physical limitations, in this embodiment, the sensor scheduling module extracts the temporal gradient features of the RF signal to characterize the device's dynamic approach trend, thereby avoiding biased judgments based on a single transient extreme value. The specific underlying sampling and analysis mechanism is executed according to the following steps.

[0044] In step S101, the sensor scheduling module drives the low-power communication module to perform periodic beacon scanning. As a preferred method, the low-power communication module captures broadcast data packets sent by surrounding nodes within a set scanning window and extracts the raw received signal strength indication attached to the broadcast data packet. For the specific implementation mechanism of the low-power communication module listening to broadcast packets and obtaining the RF link received signal strength indication at the physical medium correlation layer, those skilled in the art can consult and adopt the standard procedures in the existing Bluetooth core specifications. Its underlying RF demodulation and energy measurement mechanisms are well-known technologies in the field and will not be elaborated here. Simultaneously, to coordinate with subsequent spatiotemporal baseline construction and acoustic delay estimation, when the IoT terminal is in a preset active state (such as the high-frequency tracking state in subsequent steps), the IoT terminal drives its internal electroacoustic conversion module to synchronously transmit near-ultrasonic beacon pulses in a specified frequency band (such as 18kHz to 22kHz) within the same clock cycle as sending the low-power Bluetooth broadcast data packet.

[0045] In step S102, the sensor scheduling module performs data cleaning and smoothing filtering on the original received signal strength indication in the discrete time series, and outputs a smoothed received signal strength indication. As a specific implementation supporting the sub-features of the claim, the sensor scheduling module uses an exponential moving average algorithm to suppress transient signal spikes and high-frequency noise interference caused by multipath fading. Specifically, the sensor scheduling module assigns preset weight coefficients to the currently captured original received signal strength indication and the smoothed received signal strength indication from the previous sampling time, and performs a weighted summation operation to iteratively generate the smoothed received signal strength indication at the current time. The preset weight coefficients are real numbers greater than 0 and less than 1, and their specific values ​​are dynamically calibrated based on the signal-to-noise ratio of the current radio frequency environment; when the environmental noise variance is large, a smaller weight coefficient is selected to enhance the smoothing effect.

[0046] In step S103, to ensure the alignment of time-series data and the consistency of operating conditions, the sensor scheduling module establishes a time-dimensional data sliding window in memory. The sensor scheduling module pushes the calculated smoothed received signal strength indications into this data sliding window in timestamp order.

[0047] In step S104, the sensor scheduling module calculates the first-order time derivative of the smoothed received signal strength indication to obtain the time gradient feature of the received signal strength indication. Before introducing the specific calculation, it should be clarified that this time gradient feature reflects the relative rate of change of the radio frequency signal over time, and its physical meaning is equivalent to the radial velocity of the target node relative to the local terminal. This embodiment uses a discrete-time difference model to calculate this time gradient feature. The sensor scheduling module extracts the latest smoothed value at the current moment of the data sliding window and the historical smoothed value at the beginning of the window and performs a difference operation. The specific calculation logic is as follows:

[0048] ;

[0049] In the formula, Indicates time The time gradient characteristics of the received signal strength indication, Indicates time Smooth received signal strength indication, Indicates time Smooth received signal strength indication, This represents the time span parameter of the data sliding window. To avoid system crashes caused by division by zero, the sensor scheduling module verifies and ensures that the above division operation is performed before execution. The value of is strictly greater than zero; at the same time, The value is set based on the radio frequency Doppler shift period generated by the walking speed of a typical person, and in this embodiment, it is preferably between 0.5 seconds and 2.0 seconds.

[0050] In step S105, after obtaining the relative rate of change, the sensor scheduling module compares the calculated temporal gradient feature with a preset proximity trend determination threshold. A positive and continuously increasing temporal gradient feature indicates that the radial physical distance between the external node and the local IoT terminal is decreasing. The proximity trend determination threshold is determined based on empirical data of the rate of change of radio frequency attenuation at a normal walking speed of a healthy individual and is pre-written into the terminal's non-volatile memory.

[0051] In step S106, when the time gradient characteristic value of the received signal strength indication is greater than the proximity trend determination threshold, the sensor scheduling module determines that the radio frequency dynamic proximity condition is met. The sensor scheduling module generates and sets a radio frequency trigger flag in a register. This radio frequency trigger flag is used to indicate that there is an active node with physical proximity behavior within the current sensing range. The sensor scheduling module outputs this radio frequency trigger flag to the logic determination unit of the coprocessor as a pre-determination parameter for generating a hardware interrupt signal.

[0052] Considering that in real-world scenarios of contact tracing for infectious diseases, relying solely on radio frequency (RF) signals can easily misjudge statically placed devices as being carried, such as a terminal left on a desk receiving an RF broadcast from a moving person. Based on these physical limitations, in this embodiment, the sensor scheduling module introduces the kinematic characteristics of the inertial measurement unit (IMU) as a secondary verification dimension. By extracting the statistical variance of the triaxial acceleration signals, the system can effectively isolate the static gravity component and accurately quantify the actual kinetic energy accumulation generated by the IoT terminal as it moves with the human body. The specific kinetic energy feature extraction mechanism is implemented according to the following steps.

[0053] In step S107, the sensor scheduling module wakes up the inertial measurement unit (IMU) via its internal integrated circuit bus and periodically reads the motion data output by the IMU (specifically, raw triaxial acceleration data). To balance the integrity of capturing human motion frequency characteristics with the constraint of ultra-low power consumption of the terminal, the sampling rate of the IMU is set between 25 Hz and 50 Hz as a preferred approach. For the specific implementation mechanisms of the analog-to-digital conversion, data storage and backup, and bus communication protocol of the IMU's underlying microelectromechanical system (MEMS) sensor, those skilled in the art can consult existing standard datasheets. Its hardware drivers and basic digital signal output are well-known technologies in the field and will not be elaborated upon here.

[0054] Step S108: Due to the highly random placement of the IoT terminal in the user's pocket or backpack, to eliminate signal attenuation caused by posture changes along a single coordinate axis, the sensor scheduling module performs vector magnitude calculation on the acquired raw three-axis acceleration data, fusing it to generate a comprehensive acceleration amplitude decoupled from the device's spatial posture. Specifically, the fusion calculation method involves extracting the raw acceleration values ​​along the X, Y, and Z axes of the Cartesian coordinate system at each sampling time, calculating the sum of the squares of these three-axis values, and then taking the square root. Essentially, this operation transforms the three-dimensional spatial vector into a single scalar by calculating the Euclidean norm, thereby providing posture-independent input parameters for subsequent motion feature extraction.

[0055] Step S109: To ensure the alignment and continuity of kinematic characteristics with the aforementioned radio frequency (RF) proximity trend over time, the sensor scheduling module constructs an inertial data sliding window synchronized with the RF signal analysis. The sensor scheduling module sequentially pushes the calculated comprehensive acceleration amplitude values ​​into the inertial data sliding window according to timestamps until the number of data samples within the window reaches a preset window capacity threshold. The window capacity threshold is determined based on the sampling rate and a preset analysis duration, preferably corresponding to a time domain length of 1 to 2 seconds to cover the complete human gait cycle.

[0056] Step S110: The sensor scheduling module calculates the sample variance of the comprehensive acceleration amplitude within the inertial data sliding window to obtain the acceleration variance characteristics characterizing the kinetic energy fluctuations of the equipment. Before introducing the mathematical derivation, it should be clarified that the physical purpose of this step is to filter out the fixed local gravitational acceleration constant and extract the dynamic alternating energy purely caused by human gait or torso undulations. The specific variance calculation logic is as follows:

[0057] ;

[0058] In the formula, This represents the calculated acceleration variance characteristics; This represents the total number of samples within the sliding window of inertial data; Indicates the first in the window The combined acceleration amplitude at each sampling time point; This represents the arithmetic mean of all combined acceleration amplitudes within the sliding window of inertial data. To avoid division by zero errors that could cause the algorithm to crash, the sensor scheduling module verifies and ensures the total number of samples before performing the above division operation. The value of is strictly greater than 1; at the same time, the above subtraction operation uses Bessel correction to obtain unbiased estimation results that are closer to the real physical process.

[0059] Step S111: After acquiring the statistical indicators reflecting energy fluctuations, the sensor scheduling module compares the calculated acceleration variance characteristics with a preset motion energy wake-up threshold. The motion energy wake-up threshold is calibrated based on an empirical value representing the lower limit of three-dimensional acceleration fluctuations generated under minute displacement conditions such as slow walking or standing up in a healthy adult; its value is typically set within the range of 0.01g. 2 Up to 0.05g 2 Between, where g is the standard gravitational acceleration.

[0060] In step S112, when the value of the acceleration variance feature is greater than the motion energy wake-up threshold, the sensor scheduling module determines that the current IoT terminal is in an active human-carried mobile state and sets the motion trigger flag in its internal register. This motion trigger flag and the aforementioned radio frequency trigger flag form the logical basis for intersection determination, used to block false system wake-ups caused by interference from a single sensor at the source.

[0061] The central coprocessor, acting as the underlying power management hub of the IoT terminal, is responsible for coordinating the power supply timing and operating frequency of various physical sensors. Considering that frequent high-frequency sampling by sensors will drastically shorten the terminal's standby lifespan, based on the aforementioned extracted RF trigger and motion trigger identifiers, in this embodiment, the central coprocessor constructs a deterministic finite state machine with three operating modes, aiming to dynamically adjust the hardware power level according to the environmental risk level. The specific hardware interrupt and state switching mechanism is executed according to the following steps.

[0062] In step S113, the central coprocessor maintains the finite state machine in its internal static random access memory. This state machine defines three sensor scheduling states through a preset state transition matrix: deep sleep state, sniffing and inspection state, and high-frequency tracking state. Before introducing specific logic switching, it should be clear that the operating principle of this finite state machine is to minimize the online time of high-power modules in exchange for the overall energy utilization efficiency of the system. For the specific low-level operations of how the central coprocessor writes register configurations through the internal integrated circuit bus to control the on / off state of the peripheral sensor power rails, those skilled in the art can refer to the register mapping manuals of each sensor chip. The basic power domain control instructions and communication protocols are well-known technologies in the field and will not be elaborated upon here.

[0063] In step S114, the logic determination unit of the central coprocessor synchronously acquires the RF trigger flag from the RF analysis module and the motion trigger flag from the inertial measurement module at the edge of the set clock cycle. Since the physical cycles of RF scanning and inertial sampling differ, the central coprocessor executes operating condition alignment logic here, ensuring causal correlation between the two determination flags in the time dimension by setting a common observation period window. Simultaneously, to avoid false triggering caused by bus level glitches, the central coprocessor performs hardware-level dejitter processing on the acquired dual-parameter signal, only determining it as a valid logic input if the flag bit remains stable for three consecutive clock cycles.

[0064] In step S115, when both the RF trigger flag and the motion trigger flag are in a reset state, it indicates that the terminal is neither physically approaching a target nor in a human-carried mobile state. Based on the above double negation condition, it is determined that there is no risk of infectious disease contact in the current environment. The central coprocessor pulls down the enable signal of the core sensor through the general-purpose input / output pin, switching the system to a deep sleep state. In the deep sleep state, except for the real-time clock and the interrupt timer that maintains basic wake-up, the main power rails of the remaining high-frequency sampling links are hard-cut off to minimize device noise and static leakage current.

[0065] In step S116, when only one of the RF trigger flag and the motion trigger flag is set, the central coprocessor determines that there is a potential contact trigger but has not yet constituted a definite tracking condition, and then switches the system state to sniffing and inspection state. As a preferred approach, in this state, the central coprocessor adjusts the frequency divider parameters to set the broadcast listening frequency of the Bluetooth Low Energy module and the sampling duty cycle of the inertial measurement unit to a low baseline value, typically 20% to 30% of the normal tracking mode. The physical purpose of this is to maintain continuous awareness of environmental changes with extremely low power consumption, thereby ensuring that the system can respond quickly when risks escalate.

[0066] Step S117: If and only if the RF trigger flag and the motion trigger flag are simultaneously set, it indicates that the user wearing the terminal is moving, and there are other nearby nodes also active within the RF sensing range. The central coprocessor determines that a valid contact event is about to occur and immediately triggers a hardware interrupt signal. The specific hardware triggering logic is as follows:

[0067] ;

[0068] In the formula, This indicates a hardware interrupt signal generated by the central coprocessor; Indicates radio frequency triggering identifier; Indicates a motion trigger flag; This represents the logical AND operation in digital logic circuits. This formula defines a strict physical admission mechanism for the system to enter high-power tracking mode, avoiding false wake-ups caused by environmental interference from a single sensor through logical weighting of multi-dimensional parameters. Simultaneously, to cover static co-occurrence scenarios where physical proximity transforms into resident interaction (such as sitting together for work or meals), the central coprocessor is equipped with a state-holding timer. Once a hardware interrupt signal is valid and triggers the system to enter high-frequency tracking state, as long as the RF trigger flag... The flag remains in the active state, even if subsequent movement triggers the flag. Once the system returns to the reset state, it will remain in high-frequency tracking mode until the state hold timer times out, thus ensuring continuous acquisition of high-risk close-range static contact data.

[0069] In step S118, when the hardware interrupt signal is valid or within the maintenance period of the state hold timer, the central coprocessor control system fully switches to high-frequency tracking state. As a preferred approach, the central coprocessor responds to the interrupt vector, fully activating the ultra-wideband ranging module via its internal direct memory access channel, and simultaneously controlling the temperature and pressure sensors and audio envelope circuitry to switch from sleep to operating state. In this state, the system increases the pulse repetition frequency of the ultra-wideband ranging module to the preset maximum operating frequency to perform precise physical ranging between nodes, while simultaneously initiating high-frequency monitoring of ambient air pressure and acoustic envelope. Through the aforementioned scheduling strategy based on strict dual-parameter logic gating, the system effectively intercepts invalid wake-up requests in the background environment, thereby establishing an ultra-low power operating baseline at the source of the hardware architecture.

[0070] S200: Multimodal data preprocessing and signal cleaning. Specific steps are as follows:

[0071] In three-dimensional contact tracking scenarios, microelectromechanical system (MEMS) barometers are susceptible to interference from ambient temperature fluctuations and dynamic heating of components on the device's mainboard when sensing minute pressure changes. Temperature gradients induce thermal stress deformation on the piezoresistive silicon diaphragm inside the sensor, causing baseline drift in the output pressure value as temperature changes. To overcome these physical limitations, in this embodiment, the central coprocessor introduces a synchronously sampled reference temperature as an environmental variable. A polynomial feedforward algorithm is used to eliminate thermal drift components at the lower level, ensuring the physical confidence of the relatively high-level solution. The specific thermal drift feedforward compensation mechanism is implemented according to the following steps.

[0072] In step S201, the central coprocessor synchronously reads the raw air pressure value output by the barometer and the reference temperature value output by the on-chip temperature sensor on the same clock trigger edge via the internal integrated circuit bus. The barometer and the on-chip temperature sensor are integrated into a single temperature-pressure sensor. To ensure absolute alignment of multi-source heterogeneous data in the time dimension and avoid phase delay errors introduced by asynchronous sampling, the central coprocessor uses a hardware timer to directly trigger the dual-channel analog-to-digital conversion, ensuring that the timestamp error between air pressure sampling and temperature sampling is limited to within microseconds. For the specific implementation mechanisms of the underlying analog-to-digital conversion, register configuration, and bus reading timing of the barometer and on-chip temperature sensor, those skilled in the art can refer to the datasheet provided by the chip manufacturer. The basic hardware drivers and digital signal extraction are well-known technologies in the field and will not be elaborated upon here.

[0073] Step S202: Before performing specific error compensation calculations, a relative temperature change model needs to be constructed. The central coprocessor obtains the thermodynamic reference temperature pre-programmed into the non-volatile memory during the device's manufacturing process and calculates the physical difference between the previously obtained reference temperature value and this thermodynamic reference temperature to generate a relative temperature bias. The physical purpose of this calculation process is to strip away the absolute temperature background environment and extract only the thermodynamic changes that cause microscopic deformation of the silicon film, thereby providing highly correlated bias correction input features for the feedforward network.

[0074] Step S203: Before introducing a specific compensation model, it must be clarified that the temperature drift of microelectromechanical system (MEMS) sensors typically exhibits significant nonlinear characteristics, and simple linear scaling cannot cover wide-temperature operating conditions. As a specific implementation method supporting the sub-features of the claim, the central coprocessor adopts a second-order polynomial fitting model to calculate the feedforward error compensation term based on the relative temperature bias. The specific error evaluation logic is as follows:

[0075] ;

[0076] In the formula, This represents the feedforward error compensation term obtained from the solution; This represents the reference temperature value obtained at the current sampling time; This indicates the thermodynamic reference temperature read from non-volatile memory; This represents the first-order thermal drift sensitivity coefficient of the barometer; This represents the second-order thermal drift nonlinearity coefficient of the barometer. Regarding the determination of the coefficients in the above formula, both the first-order thermal drift sensitivity coefficient and the second-order thermal drift nonlinearity coefficient are obtained during the equipment's factory calibration phase through environmental calibration in a full-temperature range high and low temperature chamber. As a preferred method, the calibration equipment collects static pressure deviations under multiple temperature gradients and uses the least squares method for curve fitting to determine the coefficients. and The optimal solution is obtained by taking the values ​​of these coefficients based on the physical batch characteristics of the sensor wafers. The characteristic parameters of each individual unit are obtained through full-temperature environmental calibration before the equipment leaves the factory, and then they are stored in read-only memory to achieve accurate compensation for individual differences.

[0077] In step S204, the central coprocessor performs a subtraction operation between the original air pressure value synchronously read along the same clock trigger edge and the feedforward error compensation term, performing thermal drift feedforward compensation to obtain a high-precision air pressure value after thermal drift feedforward compensation. This operation aims to reversely remove spurious pressure increments caused by thermal expansion effects from temperature-contaminated mixed air pressure signals. Through the above basic arithmetic subtraction operation, the system achieves direct blocking of environmental interference at the underlying logic level.

[0078] In step S205, after obtaining the high-precision barometric pressure value to eliminate baseline drift, the central coprocessor writes this high-precision barometric pressure value into a specific address space of the direct memory access buffer. This high-precision barometric pressure value is subsequently used by the main processor to calculate the absolute altitude or relative floor span of the terminal device, thereby constructing the intersection of the physical location of the contact person in vertical space from multiple dimensions. By introducing the above-mentioned deterministic temperature feedforward correction mechanism, the system effectively suppresses 3D positioning misjudgments caused by sudden environmental changes such as user body temperature conduction or moving from indoors to outdoors, ensuring the robustness of trajectory tracking in the vertical coordinate axis direction from the underlying hardware computing power level.

[0079] After completing the bottom-level temperature feedforward compensation of the barometer, the system acquires a high-precision barometric pressure sequence with a stable baseline. Considering that IoT terminals often face micro-environmental aerodynamic interference in actual wearable scenarios—for example, local airflow shear generated when a user walks, or physical compression of the terminal inside a clothing pocket—can induce high-frequency wind noise fluctuations and transient pressure spikes on the dustproof, waterproof, and breathable membrane outside the sensor. To extract the gradually varying barometric pressure trend representing the actual floor crossing from the interfered mixed signal, in this embodiment, the central coprocessor constructs a two-layer data cleaning mechanism in both the frequency and time domains. The specific filtering and anomaly removal process is executed according to the following steps.

[0080] In step S206, the central coprocessor reads the high-precision barometric pressure sequence after thermal drift feedforward compensation from the direct memory access buffer in a time-series manner. Based on the general laws of fluid mechanics and human kinematics, the barometric pressure changes caused by actual physical altitude changes are low-frequency, slowly varying signals, while environmental wind noise typically manifests as broadband, high-frequency noise. To isolate the signal characteristics caused by these different physical sources, the central coprocessor needs to introduce a digital filter to perform frequency domain isolation on the barometric pressure sequence to suppress high-frequency jitter caused by non-altitude changes.

[0081] Step S207, as a preferred approach, considering the limited internal static random access memory capacity and computing power of the central coprocessor, the system employs a first-order infinite impulse response low-pass filter to filter out high-frequency dynamic noise. This filter achieves smooth processing of abrupt signals through weighted fusion of historical outputs and current inputs. During the first iteration calculation, due to the lack of historical states, the system initializes the filter state variables to the first acquired high-precision air pressure value to prevent step oscillations during the cold start phase. For the regular sampling phase, the specific frequency domain filtering logic is as follows:

[0082] ;

[0083] In the formula, This represents the smoothed air pressure value after low-pass filtering at the current sampling time; This represents the high-precision air pressure value obtained at the current sampling moment after thermal drift feedforward compensation. This represents the smoothed air pressure value obtained at the previous sampling time. This represents the smoothing coefficient of the filter. The magnitude of this smoothing coefficient essentially determines the cutoff frequency of the low-pass filter. Given that the frequency of height changes when a healthy adult walks up or down stairs or rides an escalator is typically below 2Hz, this embodiment sets the filter's cutoff frequency between 1Hz and 2Hz. Combined with the discretized sampling time step, The value of is usually empirically set between 0.05 and 0.2. For the optimization of fixed-point multiply-accumulate operations of the underlying difference equations of digital filters in digital signal processors, those skilled in the art can refer to general digital signal processing guidelines. The basic register shift and truncation control are well-known techniques in the field and will not be elaborated here.

[0084] In step S208, after filtering out high-frequency background wind noise, the system also needs to address transient interference. When the device is placed in a clothing pocket or backpack, unavoidable physical friction and localized compression will generate transient pressure spikes within the sealed cavity. This type of interference is not only huge in amplitude but also extremely short in duration, making it difficult for simple low-pass filtering to completely eliminate it. To perform physical friction elimination, the central coprocessor performs a first-order difference operation on the continuously acquired smoothed air pressure values ​​within the time domain window to evaluate the slope of the instantaneous change in the air pressure signal.

[0085] In step S209, the central coprocessor calculates the absolute difference between the smoothed air pressure values ​​at two adjacent sampling times and compares it with a preset air pressure jump hard decision threshold. This air pressure jump hard decision threshold is calibrated based on the physical boundary of the maximum air pressure change rate under normal elevator rapid ascent or descent conditions or people running downstairs, and its value range is usually set between 0.2 Pa / ms and 0.5 Pa / ms. When the calculated absolute difference is greater than the air pressure jump hard decision threshold, the system determines that the current sampling point has been subjected to physical shock interference not caused by changes in environmental altitude. To ensure the continuity of the air pressure sequence in subsequent integration calculations, the central coprocessor triggers the underlying engineering fault tolerance mechanism. Specifically, the system discards the current smoothed air pressure value distorted by physical friction and forces the valid value of the previous normal sampling time to be used as the maintenance output of the current state. At the same time, to avoid deadlock anomalies caused by blockage of the sensor's actual physical channels, the system is configured with a fault tolerance counter. When the number of consecutive triggers of the maintenance output exceeds the preset fault tolerance limit (e.g., 10 consecutive sampling cycles), the system forcibly resets the filter and reinitializes the baseline state.

[0086] In step S210, after acquiring the cleaned low-frequency air pressure data, the central coprocessor does not directly use it as the sole basis for final height determination. To avoid misjudging local static air pressure fluctuations caused by the start / stop of indoor air conditioning ducts or changes in meteorological conditions as vertical movement of equipment, the central coprocessor synchronously calls the motion trigger flag extracted in the aforementioned steps. The system executes operating condition alignment logic here, using the motion trigger flag as an auxiliary gating condition for determining the floor crossing mode. Specifically, when the motion trigger flag is in the set state, the air pressure change sequence is directly sent to the main processor to perform floor calculation for walking up and down stairs; when the motion trigger flag is in the reset state, but the cleaned air pressure sequence exhibits a monotonous smooth step that lasts for more than the preset elevator rise and fall judgment time (e.g., more than 3 seconds), the system determines it as an elevator riding condition and also allows it to be sent to the main processor for floor calculation. By introducing the above multi-dimensional logical intersection, the system blocks the interference of short-term local hot and cold airflows or air pressure artifacts under static placement at the application layer, avoiding one-sided judgments based solely on a single air pressure extreme value.

[0087] In the context of infectious disease contact tracing, to eliminate false positives for close contact caused by radio frequency signals penetrating physical walls, the system introduces near-ultrasonic frequencies as beacons to assist in ranging and spatial coexistence determination. To reduce system power consumption during continuous processing of ambient audio, the system constructs a hardware envelope extraction link at the analog front end, avoiding high-frequency digital sampling and matrix operations at the underlying level. In this embodiment, the specific processing flow of acoustic envelope hardware extraction is executed according to the following steps.

[0088] In step S211, the microphone transducer captures ambient sound waves and outputs a raw analog AC voltage signal. The system guides this raw signal into a front-end analog bandpass filter. The passband of this filter is physically limited to a specific near-ultrasonic frequency band (18kHz to 22kHz in this embodiment) to block ambient noise interference and filter out the normal human voice frequency band. For the selection of operational amplifiers and matching of RC networks for the analog bandpass filter, those skilled in the art can consult standard analog circuit design manuals. The basic second-order active filter topology is well-known in the art and will not be elaborated here.

[0089] In step S212, the frequency-isolated high-frequency AC signal is fed into a full-wave precision rectifier. This rectifier utilizes the high open-loop gain of the operational amplifier located in the feedback loop to eliminate the effect of diode dead zone, accurately inverting the bipolar high-frequency AC signal into a unipolar pulsating DC signal, providing a unidirectional charge accumulation basis for subsequent energy integration.

[0090] In step S213, a unipolar pulsating DC signal flows into a hardware integrating network composed of passive resistors and capacitors. This integrating network acts as an analog low-pass filter, utilizing the charging and discharging inertia of the capacitors to smooth the rapid fluctuations of the high-frequency carrier wave, converting it into a low-frequency envelope analog voltage characterizing the energy fluctuations of the beacon. In this embodiment, based on the typical signal modulation rate of near-ultrasonic beacons, the time constant of this integrating network (i.e., the product of the equivalent resistance and the equivalent capacitance) is calibrated between 1 ms and 5 ms.

[0091] In step S214, the analog-to-digital converter inside the central coprocessor periodically samples the low-frequency envelope analog voltage at a preset extremely low sampling rate (set between 50Hz and 100Hz in this embodiment) to obtain a digitized acoustic energy sequence. This dimensionality reduction mechanism significantly reduces the dynamic power consumption at the system's underlying layer.

[0092] Step S215: To further improve system energy efficiency management and prevent false wake-ups, the central coprocessor compares the acquired acoustic energy sequence with a preset hardware wake-up threshold. This hardware wake-up threshold is set with fluctuations based on the long-term energy baseline of the background noise in the device's environment. To accurately quantify the aforementioned energy fluctuation states, the central coprocessor performs the following ratio evaluation calculation in the digital domain:

[0093] ;

[0094] In the formula, This represents the acoustic energy mutation rate during the current acoustic sampling period; This indicates the amplitude of the digital acoustic energy obtained from the current analog-to-digital conversion; This represents the long-term energy baseline of environmental background noise obtained through a long-term moving average algorithm. This represents the acoustic energy tolerance bias. The physical purpose of introducing the acoustic energy tolerance bias in the above algorithm logic is to avoid arithmetic overflow anomalies in the digital signal processor due to division by zero when the device is in an extremely quiet, enclosed microenvironment causing the long-term energy baseline to approach zero.

[0095] To avoid biased judgments caused by single extreme values ​​(such as sudden broadband transient noise from metal object collisions) arising from non-beacon-related factors, the system does not rely solely on the result of a single calculation. Instead, the system constructs a multi-dimensional judgment logic. Only when the calculated energy mutation rate exceeds the hardware wake-up threshold for multiple consecutive sampling periods (e.g., 3 to 5 consecutive periods) does the system determine that a valid near-field acoustic beacon event has been captured, outputting a high-frequency acoustic envelope signal and waking up the main processor via a dedicated interrupt pin to verify spatial coexistence characteristics. By constructing the aforementioned purely hardware-level acoustic detection and downsampling mechanism, the system achieves all-weather, ultra-low-power contact beacon monitoring, providing solid underlying hardware support for determining the physical isolation of high-risk contacts in three-dimensional space.

[0096] S300: Spatiotemporal baseline construction and synchronization of collaborating nodes. Specific steps are as follows:

[0097] Based on the aforementioned near-field acoustic detection mechanism, the system has completed the initial identification of high-risk coexistence events at the underlying level. To further decouple the local micro-environmental disturbances of individuals from the overall physical changes in the macroscopic space, the system needs to construct a macroscopic environmental baseline. In this embodiment, the system selects remote devices located beyond a safe distance as collaborating nodes and collects their sensor data to synthesize a background baseline. The specific remote background collaborating node filtering mechanism is implemented according to the following steps.

[0098] In step S301, the wearable device's radio frequency receiving module listens for Bluetooth Low Energy broadcast data packets sent by surrounding IoT nodes. When parsing the broadcast data packets, the system extracts the unique device identifier of each transmitting source and the corresponding instantaneous received signal strength indicator value. To ensure the temporal correlation of multi-source sensor data in the subsequent fusion calculation stage, the system synchronously obtains the local timestamp of the broadcast data packet arrival at this stage and parses the remote transmission timestamp carried in the data packet payload, combining it with the local real-time clock to calculate the relative time offset caused by network transmission. For the specific frame structure and media access control mechanism of the Bluetooth Low Energy protocol, those skilled in the art can consult relevant wireless communication standard documents; its physical layer data frame parsing process is well-known in the field and will not be elaborated here.

[0099] Step S302: Considering the widespread radio frequency multipath effects caused by wall reflections and human body obstruction in indoor building structures, the instantaneous radio frequency signal strength of a single acquisition will exhibit drastic random fluctuations. Directly using instantaneous values ​​for distance determination can easily lead to frequent oscillations in node state classification. As a preferred approach, the main processor unit performs time-domain smoothing on the radio frequency signal strength sequence of the same device identifier based on an exponentially weighted moving average algorithm. The underlying mathematical model of this smoothing operation is as follows:

[0100] ;

[0101] In the formula, Indicates that for the first The peripheral nodes in the current time slot The acquired smooth radio frequency signal strength; This indicates the instantaneous radio frequency signal strength currently measured at the physical layer; Indicates the previous time slot The retained smooth radio frequency signal strength historical state quantity; This represents the weight update coefficient. The physical purpose of the above algorithm formula is to utilize the integral inertia of historical states to forcibly suppress high-frequency fading spikes in the radio frequency channel. Regarding the calibration of the update coefficient, The value of is fixed between 0.15 and 0.25. This range is chosen based on the balance between the smoothness of the filter and the spatial hysteresis effect of the system response when the target device moves.

[0102] In step S303, the system uses a logarithmic distance path loss model to map the smoothed RF signal strength to an estimated interval of spatial physical distance. To avoid additional dynamic power consumption overhead when the main processor performs floating-point logarithmic operations, the system pre-configures a lookup table logic based on discrete intervals in the underlying registers. Specifically, this lookup table divides the expected dynamic RF receiving range of the system design into multiple consecutive quantization intervals, each of which is directly mapped to a corresponding distance evaluation tag, thereby transforming complex nonlinear algebraic operations into extremely low-power memory addressing operations. For the conventional physical diffusion radius of pathogen aerosols, the system sets a dual-layer RF discrimination threshold in the lookup table: an inner high-risk contact threshold (e.g., equivalent RF strength −65dBm) and an outer far-end safety threshold (e.g., equivalent RF strength −85dBm).

[0103] For candidate nodes whose smooth radio frequency signal strength is greater than or equal to the inner layer high-risk contact threshold, the system includes their device identifier in the near-end high-risk node set as the target tracking object for subsequent spatial connectivity determination and contact risk scoring; for candidate nodes whose smooth radio frequency signal strength is between the inner layer high-risk contact threshold and the outer layer far-end safety threshold, the system marks them as grayscale detached nodes, temporarily adds them to the observation queue and does not participate in baseline construction and risk scoring in the current calculation cycle.

[0104] In step S304, for candidate nodes whose smoothed radio frequency signal strength is lower than the far-end safety threshold of the outer layer, the system further introduces spatial dwell time evaluation logic. The reason for this feature is that while other wearers passing by randomly and quickly may be within the far-end safety range for a short period, their transient sensor data cannot objectively characterize the macroscopic stable environment of the current area. The system evaluates whether the smoothed radio frequency signal strength of the candidate node strictly remains below the far-end safety threshold of the outer layer within a preset observation window (e.g., 60 consecutive observation periods).

[0105] In step S305, the system formally adds the corresponding device identifier to the dynamic set of collaborating nodes only when the aforementioned continuous dwell conditions are met. Subsequently, the system initiates an environmental feature synchronization request to the collaborating nodes in this set through the non-connection broadcast extension field of the radio frequency link. The system thereby obtains remote air pressure data and remote acoustic data collected by the collaborating nodes, providing a reference source for subsequent suppression of micro-environmental interference. To verify and eliminate the potential for misalignment of operating conditions in multi-source heterogeneous data, after receiving the remote air pressure data and remote acoustic data, the system uses the relative time offset calculated in step S301 to perform timestamp compensation interpolation on the remote sequence, ensuring that the baseline data of external collaborating nodes and the local sensor sampling data are within a strictly aligned absolute time slice. By constructing the above-mentioned filtering mechanism based on multipath smoothing and temporal dwell, the system physically isolates high-risk nodes at extremely close range and interference from rapidly moving transient nodes, ensuring the purity of the macro-environmental baseline data.

[0106] Based on the aforementioned remote background collaborative node filtering logic, the system establishes the radio frequency arrival time slice of the effective beacon and approximates it as the absolute zero point of sound wave transmission. To overcome the distance measurement deviation caused by acoustic multipath effects (i.e., the superposition of reflected and refracted waves) in indoor architectural spaces, the system constructs a data-driven adaptive cross-correlation search logic to extract the true sound wave delay. In this embodiment, the specific processing flow for sound wave delay estimation is executed according to the following steps.

[0107] In step S306, the system performs acoustic delay estimation on all captured valid nodes (including near-end high-risk nodes and cooperative candidate nodes under residency assessment). Specifically, the main processor unit uses the arrival time of the extracted valid radio frequency beacon as the time reference zero point and opens an acoustic sliding sampling window of corresponding width in a circular buffer. The physical length of this sliding sampling window is limited to a preset time domain interval (30ms to 50ms in this embodiment) to cover the direct acoustic band within the expected maximum contact distance. Within this window, the system retrieves the previously output digitized acoustic envelope sequence to construct a local reception sequence. Based on the device identifier parsed in the radio frequency broadcast, the system extracts the reference beacon envelope template sequence matching the cooperative node and simultaneously acquires the acoustic sampling period parameters of the analog-to-digital converter.

[0108] In step S307, the system performs a normalized sliding cross-correlation operation on the local received sequence and the reference beacon envelope template sequence to quantify the waveform similarity at different time-sliding scales. The specific cross-correlation operation logic is as follows: the system calculates the dot product of the local energy deviation of the local received sequence at each sliding index offset and the standard energy deviation of the reference beacon sequence, using this as the numerator; the system calculates the product of the energy standard deviations of the two sequences, and adds a preset cross-correlation anti-overflow bias constant (a small positive real number) to this product to construct an error-proof denominator; the system divides the above numerator by this error-proof denominator and outputs the waveform cross-correlation coefficient sequence. This bias constant mechanism prevents processor arithmetic overflow anomalies induced by the energy standard deviation approaching zero at extremely low signal-to-noise ratios.

[0109] In step S308, as the discrete sliding index offset increases, the waveform cross-correlation sequence exhibits multiple local extrema. Based on the physical law that multipath reflected waves arrive later than line-of-sight direct waves, the system sets a dynamic waveform confidence threshold (calibrated between 0.65 and 0.80 in this embodiment) based on the real-time root mean square energy of the background ambient noise. The system performs a linear scan along the time axis from left to right in the cross-correlation sequence, extracting the target index corresponding to the first local peak that is greater than or equal to the dynamic waveform confidence threshold. The system multiplies this target index by the aforementioned acoustic sampling period parameter, completing the conversion from discrete points to physical time units and locking in the true sound wave delay. This early peak extraction mechanism eliminates interference from strong reflected echoes arriving later at the algorithmic level.

[0110] In step S309, the system converts the acquired actual sound wave delay into the spatial physical distance between the target cooperative node and the current device. To eliminate the influence of temperature on the speed of sound, the system uses the real-time air temperature value collected by the local environmental sensor for dynamic sound speed compensation. The specific physical mapping logic is as follows: the system adds the product of the standard sound wave propagation speed of an ideal gas at zero degrees Celsius (value 331.4 m / s), the sound speed temperature gradient compensation coefficient (value 0.6 m / (s·℃), and the real-time environmental temperature in degrees Celsius to calculate the actual compensated sound speed; the system multiplies the actual sound wave delay by this actual compensated sound speed to derive the spatial physical distance between the target cooperative node and the local device. For the control logic and analog-to-digital conversion mechanism of the local digital temperature sensor, those skilled in the art can refer to the standard digital bus communication protocol, and its underlying hardware driver is a well-known technology in the field, and will not be described in detail here.

[0111] Based on the aforementioned actual acoustic wave delay and corresponding spatial physical distance, the system performs phase alignment and weighted fusion on the multi-source heterogeneous sensor sequences to construct a background baseline reflecting the overall physical changes in the current macroscopic space. In this embodiment, the specific processing flow for generating the dynamically weighted macroscopic baseline is executed according to the following steps.

[0112] In step S310, the system continuously acquires remote air pressure data and remote acoustic data from cooperating nodes via a wireless communication link. To eliminate the absolute delay caused by spatial physical propagation, the system constructs a circular buffer queue with absolute timestamps in the underlying memory. The main processor calculates the historical mapping time by subtracting the actual acoustic delay of the corresponding cooperating node from the local current absolute timestamp, and adjusts the pointer accordingly to retrieve the remote sensor data of that historical time from the buffer queue. The above operation achieves strict alignment of the multi-node sensor data sequence on the local absolute time slice, and the system uses this actual acoustic propagation delay to perform time-domain reverse compensation alignment of the remote air pressure data and remote acoustic data. For the memory address out-of-bounds control logic of the circular buffer queue, those skilled in the art can consult relevant literature on embedded system data structures; its cache addressing mechanism is a well-known technology in the field and will not be elaborated here.

[0113] Step S311: The system checks the total number of currently valid collaborating nodes. When the number of valid nodes is zero, the system suspends the dynamic fusion calculation and maintains the macroscopic baseline state of the previous cycle within the current time slot. When there are valid nodes in the set, the system calculates a basic weight factor for each collaborating node to avoid local extrema causing the overall baseline to be skewed by the arithmetic mean. The specific weight evaluation logic is as follows: the system calculates a negative exponential decay term based on the spatial distance between nodes, in which a preset spatial decay damping coefficient (calibrated between 0.1 and 0.3 in this embodiment) is introduced to control the distance penalty sensitivity; at the same time, the system superimposes a preset variance tolerance bias constant (set to 10 in this embodiment) into the local statistical variance of the physical sequence within the current observation window. -4 The system calculates the inverse variance of the summed negative exponential decay term and then multiplies it by this inverse variance. The system then combines this negative exponential decay term with the inverse variance, which reflects the data variance decay pattern, to construct the basic weight factors. After obtaining the basic weight factors for all valid nodes, the system divides the basic weight factor of each collaborating node by the sum of the basic weight factors of all nodes to complete the normalized allocation of weight coefficients. This model ensures that nodes closer to the local node and with smaller data variance are assigned a larger weight share.

[0114] Step S312: Based on the weighting coefficients obtained from the normalization calculation above, the system performs a linear weighted summation on all aligned remote sensor data to generate a macroscopic environmental baseline. For the macroscopic pressure baseline and the macroscopic acoustic baseline, the system multiplies the remote pressure sample value or acoustic noise floor sample value of each cooperating node (after inverse compensation for real acoustic time delay) by its corresponding normalized weighting coefficient, and then linearly accumulates all weighted sample values. The scalar result of this summation is used as the macroscopic physical baseline value generated at the current absolute observation time locally.

[0115] In step S313, the system performs a multi-dimensional consistency check before outputting the macroscopic environmental baseline to avoid biased judgments caused by local extreme airflow. The system monitors the first-order time derivative of the generated macroscopic pressure baseline in real time over a continuous preset time period (e.g., 55 seconds). When the absolute value of this first-order time derivative is lower than the macroscopic atmospheric natural fluctuation threshold (in this embodiment, it is calibrated to be in the range of 0.5 Pa / s to 1.5 Pa / s based on the slow-change characteristics of the central air conditioning system), the system confirms the validity of the macroscopic pressure baseline and writes it into the baseline register. When the absolute value is detected to be greater than or equal to the above threshold, the system determines that a step change has occurred, such as a switch between indoor and outdoor scenes or the opening and closing of a passageway door. It then suspends the current baseline sequence and triggers the re-detection and baseline reconstruction of the cooperating nodes. This check mechanism ensures the effectiveness of the macroscopic environmental baseline in determining subsequent high-risk respiratory events.

[0116] S400: Cross-media spatial connectivity decision. The specific steps are as follows:

[0117] Based on the aforementioned dynamically generated macroscopic environmental baseline, the system filters out global common interference from the local original sensor sequence to extract micro-environmental net features characterizing local weak airflow and acoustic changes, thus avoiding false detections of through-wall contact. In this embodiment, the process of stripping micro-environmental net features is performed according to the following steps.

[0118] In step S401, the main processor unit performs time-domain differential calculations on the low-frequency air pressure data and the local raw acoustic noise floor sequence, comparing them with the macroscopic environmental baseline. To avoid differential residuals introduced by differences in the sensitivity of the hardware sensor transducers, the system introduces a sensor hardware sensitivity calibration coefficient. The specific differential calculation logic is as follows: the system subtracts the macroscopic air pressure baseline value scaled by the sensor hardware sensitivity calibration coefficient from the local raw air pressure reading at the same observation time, and uses the resulting algebraic difference as the local micro-environment net air pressure characteristic value at that time. The value of this calibration coefficient is determined based on the device's factory calibration parameters, and in this embodiment, it is distributed between 0.85 and 1.15. The above operation cancels out macroscopic common fluctuations and extracts independent disturbances in the local micro-environment. The local raw acoustic noise floor sequence uses a homogeneous differential processing link.

[0119] In step S402, to filter out high-frequency thermal noise and quantization noise, the digital signal processing unit uses an infinite impulse response bandpass filter to perform frequency band isolation on the differential sequence. Considering the physiological characteristics of human respiration, the passband cutoff frequency of this filter is limited to the range of 0.1Hz to 0.5Hz. The main processor performs recursive filtering operations using preset feedforward numerator coefficients and feedback denominator coefficients to extract the low-frequency envelope signal related to the respiratory rhythm. For the design method of the bottom-level poles and zeros of the infinite impulse response filter, those skilled in the art can consult standard digital signal processing literature; the Chebyshev or Butterworth approximation algorithms are well-known techniques in the field and will not be elaborated upon here.

[0120] In step S403, the system employs an overlapping sliding window mechanism to perform short-time energy framing on the net feature sequence after frequency band isolation. The system sets a fixed-length data frame and slides it along the time axis with a preset step size. For each sliding data frame, the system calculates the sum of squares of the signal amplitudes of all discrete sampling points within the window, divides it by the total number of sampling points contained in the window, performs mean averaging, and outputs the mean net atmospheric pressure energy of the sliding data frame. In this embodiment, the length of a single window is configured to correspond to the number of sampling points in 22 seconds of physical time, and the sliding step size is set to half the window length to ensure a 50% overlap rate.

[0121] Step S404: To ensure the robustness of subsequent judgments, the system performs maximum-minimum normalization processing on the mean net atmospheric pressure energy of each frame. To prevent the range from approaching zero in a quiet environment, which could induce division anomalies, the system introduces fault-tolerant bias logic. The specific normalization operation logic is as follows: The system extracts the maximum and minimum energy values ​​within the observation buffer of the past minute, including the current frame; the system subtracts the minimum energy value from the mean net atmospheric pressure energy of the current data frame as the numerator, and adds a preset noise floor bias constant (10 in this embodiment) to the difference between the maximum and minimum energy values. -6 The numerator is divided by the denominator to calculate the normalized net energy characteristic mapped to the interval [0, 1]. Thus, the system smooths out the differences in data magnitude under different operating conditions and outputs a local microenvironment characteristic sequence with a uniform scale.

[0122] Based on the aforementioned extracted local microenvironmental net characteristics and the synchronization data of remote collaborative nodes, the system verifies whether the local device and each remote node are in the same physically connected medium space. Utilizing the isomorphic propagation characteristics of airflow dynamics in a connected space, the system performs coherence measurement calculations on the multi-terminal air pressure waveforms to eliminate false contact misjudgments caused by solid walls blocking airflow exchange. In this embodiment, the specific processing flow for calculating the air pressure waveform coherence measurement is executed according to the following steps.

[0123] In step S405, for the high-risk near-end nodes in the aforementioned dynamic set, the main processor synchronously extracts the local microenvironment net air pressure feature sequence and the target near-end microenvironment net air pressure feature sequence within the corresponding observation time window from the underlying buffer queue (this target sequence is also obtained after differential stripping using the aforementioned generated macroscopic environmental baseline). To eliminate the inherent phase difference caused by spatial medium propagation, the system performs a time-domain reverse translation alignment on the far-end sequence using the previously acquired real acoustic delay. The system traces the data pointer of the far-end air pressure sequence backward along the time axis by the offset of the discrete sampling point corresponding to the real acoustic delay, so that the two sets of heterogeneous data, local and far-end, achieve strict synchronization at the spatiotemporal starting point.

[0124] Step S406: For the time-domain aligned local and remote net pressure sequences, the system calculates the local arithmetic expectation of both sequences within a preset integration time window to isolate long-term low-frequency drift and baseline DC bias. The integration time window is configured to include a data segment containing sampling points from the past 10 to 15 seconds, covering a typical human respiratory cycle. For the accumulation and addressing mechanism of the discrete mean calculation of the digital sequence, those skilled in the art can consult relevant literature on microcomputer principles; its underlying fixed-point arithmetic logic is well-known in the field and will not be elaborated upon here.

[0125] Step S407: Based on the aforementioned local arithmetic expectation, the system constructs a discrete integral model based on the Pearson correlation coefficient to quantify the geometric similarity between the two sets of air pressure waveforms. The specific coherence measurement calculation logic is as follows: The system uses the time-domain aligned local microenvironment net air pressure feature sequence and the remote microenvironment net air pressure feature sequence, respectively, by subtracting their corresponding local arithmetic expectation values ​​to generate two sets of centered sequences; the system calculates the cumulative sum of the point-by-point multiplications of these two sets of centered sequences within the same integration time window, using this sum as the numerator of the coherence operation; the system calculates the sum of squares of the elements of each of the two sets of centered sequences, performs square root operations on these sums, and then calculates their product, adding a preset air pressure waveform noise floor bias constant (valued at 10 in this embodiment) to this product. −8 The main processor divides the numerator by this error-proof denominator and outputs the coherence measure of the transient pressure waveform at the current discrete sampling time. This error-proof denominator mechanism effectively avoids illegal division instruction anomalies caused when the waveform variance approaches zero due to a quiet environment.

[0126] Step S408: To improve the steady-state reliability of connectivity verification and suppress occasional noise interference, the system applies a first-order exponential moving average filter to the continuously calculated transient pressure waveform coherence measure sequence. The specific filtering iteration update logic is as follows: the main processor extracts the transient pressure waveform coherence measure output at the current sampling moment and multiplies it by a preset moving average weight coefficient; simultaneously, the system extracts the historical steady-state coherence measure from the previous discrete moment and multiplies it by a complementary weight coefficient (i.e., the difference between the value 1 and the moving average weight coefficient); the system sums the above two multiplications and outputs the updated historical steady-state coherence measure at the current moment. Based on the slowly varying characteristics of the aerodynamic environment, this moving average weight coefficient is typically set to 0.1 to 0.2. Subsequently, the system compares the smoothed historical steady-state coherence measure with a preset spatial connectivity determination threshold (calibrated in the range of 0.65 to 0.75 in this embodiment). When the smoothed historical steady-state coherence measure is continuously greater than or equal to the connectivity determination threshold within a preset observation period (such as 5 consecutive data frames), the system confirms that the local and remote nodes are in the same aerodynamically connected space that is not blocked by physical walls, marks it as a high-confidence companion device, and grants open system data authorization for subsequent co-calculation of vital signs.

[0127] Based on the preliminary confirmation of the coherence measurement of the aforementioned air pressure waveform regarding the same source airflow disturbance, in order to eliminate misjudgments caused by low-frequency air pressure waves penetrating physical barriers (such as door gaps or air conditioning ducts), the system utilizes the line-of-sight attenuation and physical penetration loss characteristics of high-frequency acoustic wavelengths to perform a secondary verification from the acoustic energy spectrum dimension. In this embodiment, the specific processing procedure for high-frequency acoustic envelope difference verification is executed according to the following steps.

[0128] In step S409, for the near-end high-risk nodes identified through the initial screening based on barometric coherence, the main processor acquires the raw audio sequences collected by the local digital microphone and the remote device. The system uses a digital high-pass filter to extract the acoustic envelope of a preset high-frequency band (selected as 18kHz to 22kHz in this embodiment). The main processor calls the previously acquired real acoustic delay to perform a backtracking offset of the time-domain data pointer on the remote high-frequency acoustic sequence, strictly aligning the local and remote acoustic observation windows on the absolute time axis to ensure phase consistency in subsequent frequency domain analysis.

[0129] In step S410, for the aligned local and remote high-frequency acoustic sequences, the system transforms them to the frequency domain to calculate the power spectral density. The main processor performs a Fast Fourier Transform (FFT) and uses the Welch piecewise overlap averaging method to smooth the local spectrum. Specifically, the time-domain sequence is divided into data segments with a 50% overlap rate. A Hamming window is applied to each segment for weighting, and then the arithmetic mean of the periodograms of each segment is calculated to output a robust estimate of the high-frequency power spectral density. For the butterfly architecture of the FFT and the weighted accumulation logic of the Welch method, those skilled in the art can refer to standard digital signal processing specifications; the underlying computational mechanism is well-known in the field and will not be elaborated here.

[0130] Step S411: Based on the acquired power spectral density, the system constructs an integral evaluation model for the high-frequency power spectral density difference. By accumulating the relative logarithmic differences of the energy at each discrete frequency point within the target high-frequency band, the system extracts micro-environmental features reflecting the penetration loss of the entity. To prevent arithmetic overflow anomalies induced by silent audio points, the system introduces a background noise bias mechanism in the calculation of the logarithmic independent variable. The discrete mathematical expression for this difference verification is as follows:

[0131] ;

[0132] In the formula, Indicates the first Within the first sliding data frame, the local device and the... High-frequency acoustic envelope differences between remote collaborative nodes; Indicates the index of the sliding data frame; A device identifier index representing a remote collaborating node; Indicates the index of a discrete frequency unit; This indicates the index of the discrete frequency unit that represents the lower limit cutoff of the preset high-frequency band (corresponding to a physical frequency of 18kHz). This indicates the index of the discrete frequency unit that represents the upper limit cutoff of the preset high-frequency band (corresponding to a physical frequency of 22kHz). Indicates that the local device is in the Within the sliding data frame, the first High-frequency power spectral density values ​​on discrete frequency units; Indicates the first The remote collaboration node in the Within the sliding data frame, the first High-frequency power spectral density values ​​on discrete frequency units; This represents the acoustic energy spectrum noise floor bias constant. In this embodiment, The value is determined based on the microphone hardware calibration's baseline electrical noise limit, and is set to 10. −12 W / Hz.

[0133] In step S412, the system applies low-pass smoothing to the continuously calculated high-frequency acoustic envelope difference sequence to eliminate interference from transient spectral abrupt changes. The main processor compares the smoothed difference sequence with a preset transmission loss judgment threshold (calibrated in this embodiment within the integral energy range of 25dB to 35dB equivalent physical attenuation). Based on the initial screening results from the air pressure dimension, the system performs a joint judgment: when the smoothed air pressure waveform coherence measure is greater than the preset connectivity threshold, and the smoothed high-frequency acoustic envelope difference is less than the transmission loss judgment threshold, the system determines that the two-end devices are in the same physical space without physical isolation. This aerodynamic and acoustic positive interaction mechanism effectively isolates pseudo-contact through walls.

[0134] Based on the aforementioned low-frequency air pressure historical steady-state coherence measure and high-frequency acoustic envelope difference, the system needs to perform a final cross-medium spatial connectivity decision. In the general technical principle of multimodal data fusion, different physical field signals have natural complementary characteristics when propagating in space: air pressure disturbances can characterize macroscopic aerodynamic connectivity in space, but are prone to boundary misjudgments in complex HVAC duct environments; high-frequency acoustics is extremely sensitive to physical isolation, but acoustics alone is easily interfered with by local soft obstructions. By establishing a dual Boolean decision mechanism based on dual physical field weighting, the system can perform rigorous orthogonal verification of the evaluation results of the two independent dimensions at the logical level. In this embodiment, the specific processing flow of the physical isolation dual Boolean decision is executed according to the following steps.

[0135] Step S413, for the first in the dynamic set For each high-risk near-end node, the main processor extracts in real-time the historical steady-state coherence measure of low-frequency air pressure and the difference in high-frequency acoustic envelope after low-pass smoothing. To avoid frequent switching of spatial connectivity states due to small fluctuations in sensor readings near the judgment threshold, the system introduces hysteresis comparison logic before binarization. Specifically, the system sets non-overlapping on and off thresholds for each physical dimension, forming a margin of immunity to interference in maintaining the state. When a continuously input evaluation measure positively crosses the on threshold, the connectivity state is triggered to be valid; when the measure falls back to the off threshold, the connectivity state is triggered to be invalid; if the measure is in the interval between the two, the system forcibly maintains the judgment state of the previous discrete sampling time. For the specific difference equation and memory state register configuration in the timing logic judgment of hysteresis comparison, those skilled in the art can refer to the Schmitt trigger principle in digital circuits. Its underlying state transition mechanism is a well-known technology in this field and will not be elaborated here.

[0136] Step S414: Based on the steady-state trigger signal output by the aforementioned hysteresis comparison logic, the main processor generates connectivity indicators for independent physical dimensions. The technical purpose is to map continuous physical observations into digital Boolean variables that can be directly processed by the underlying logic gates. Specifically, for the low-frequency air pressure dimension, when the smoothed coherence measure is greater than the preset air pressure spatial connectivity judgment threshold (calibrated in this embodiment based on fluid dynamics experiments in the range of 0.65 to 0.75), the corresponding air pressure connectivity Boolean indicator is set to a valid state (e.g., outputting a Boolean value of 1); otherwise, it is set to an invalid state (e.g., outputting a Boolean value of 0). Correspondingly, for the high-frequency acoustic dimension, to ensure the consistency of logic polarity, the system adopts inverse comparison discrimination. That is, when the smoothed envelope difference is less than the preset transmission loss judgment threshold (set in this embodiment based on building material attenuation experiments to an equivalent value range of 25dB to 35dB), it indicates that there is no physical wall obstruction, and the acoustic connectivity Boolean indicator is set to a valid state; otherwise, if the attenuation is too large, it is set to an invalid state. By clarifying the direction of the numerical comparison and the triggering conditions, the system completes the two-dimensional state discretization.

[0137] In step S415, after obtaining the two-dimensional Boolean indicators, the system executes a joint decision mechanism. The main processor defines the connectivity flag between the local device and the remote collaborating node as the result of a hardware bitwise AND operation of the two Boolean indicators. This dual Boolean decision mechanism establishes strict physical boundary constraints: the system verifies the macroscopic connectivity of space through the consistency of air pressure fluctuations and verifies the isolation state of physical entities using the attenuation characteristics of high-frequency acoustic signals. Through the logical AND operation of the two, it ensures that connectivity is determined only when both aerodynamic and acoustic characteristics conform to the same spatial propagation law, thereby effectively excluding pseudo-contact scenarios such as through walls and windows. The system only outputs the connectivity flag when connectivity is confirmed in both physical dimensions.

[0138] In step S416, based on the connectivity flag bit of the output judgment, the system performs subsequent node lifecycle management and collaborative data authentication. When the connectivity flag bit is set to a valid high level (i.e., Boolean value 1), the main processor officially registers the target node as a highly trusted companion device and allocates a dedicated feature data buffer queue for it in the memory stack, allowing its corresponding data stream to be integrated into the local collaborative computing bus across devices. Conversely, once the connectivity flag bit is set to an invalid low level (i.e., Boolean value 0), the system blocks the node's communication authorization at the physical isolation level, forcibly discards the data frame passed by the corresponding node, and simultaneously reclaims the memory pointer it occupies. Through the above-mentioned strict closed-loop control of physical state, the system eliminates pseudo-contact nodes that are not coexisting in the same space from the root of hardware resources and data links, ensuring that the information source of subsequent joint calculation has reliable physical space consistency. For companion devices that establish physical space connectivity, the system further initiates a higher-order feature synchronization request to them through the communication link to obtain the step frequency sequence and absolute heading angle sequence of the built-in inertial measurement unit of the remote device within the historical observation window, for subsequent dynamic companion determination.

[0139] S500: Multidimensional dynamic risk integration and early warning triggering. Specific steps are as follows:

[0140] Based on the established physical spatial connectivity results, the system uses the Dynamic Time Warping (DTW) algorithm to align the time axes of the two-end devices, and then quantifies the statistical covariance of the heading angle to determine whether the target node belongs to a high-risk accompanying target. In this embodiment, the specific processing flow of step frequency sequence DTW alignment and covariance analysis is performed according to the following steps.

[0141] Step S501: For high-risk nodes near the established physical space connectivity, the main processor extracts the kinematic feature sequences of the built-in inertial measurement units of the local and remote devices within the historical observation window, specifically obtaining the step frequency sequence and absolute heading angle sequence per unit time. Due to the frequency drift of the system crystal oscillators of heterogeneous devices, the main processor performs linear resampling interpolation on the two sets of received sequences based on the relative offset obtained from the underlying wireless communication network (such as Bluetooth Low Energy broadcast synchronization timestamp or ultra-wideband underlying synchronization frame) or the local area network global timestamp, in order to unify the time reference of the underlying discrete data frames and avoid time-domain feature misalignment caused by sampling rate differences.

[0142] In step S502, the system performs dynamic time warping to find the optimal time alignment path for the two-end motion features. The main processor uses the local step frequency sequence as a reference template and the far-end step frequency sequence as a test sample, employing one-dimensional Euclidean distance as a local metric function to calculate point-by-point similarity and construct a two-dimensional cumulative distance matrix. During matrix optimization, the system introduces a Sakoe-Chiba global constraint window to limit the allowable offset of the time axis within the range of normal human reaction time delay, preventing extreme time distortions without physical meaning. For the dynamic programming state transition equations and the boundary setting logic of the global constraint window in the dynamic time warping algorithm, those skilled in the art can refer to standard digital sequence analysis theory; the underlying matrix calculation mechanism is a well-known technology in this field and will not be elaborated upon here.

[0143] In step S503, based on the optimal warping path output by the dynamic time warping operation, the main processor generates a synchronization index mapping table. This mapping table records the point-to-point alignment relationship between the local and remote step frequency time frames under nonlinear stretching. The system applies this index mapping table to the absolute heading angle sequences of both ends, and by reorganizing the memory data pointers of the remote heading angle sequence, forces it to maintain strict alignment with the local heading angle sequence in the time dimension, thereby eliminating the relative start-up lag when the two people are walking.

[0144] In step S504, the main processor extracts the aligned local and remote heading angle sequences and calculates the normalized covariance measure. To prevent the heading angle variance from approaching zero and causing a division-to-zero anomaly under stationary or straight-line uniform speed conditions, the system injects a background noise bias into the calculation of the covariance denominator. The specific calculation logic is as follows: the system calculates the difference sequences of the local and remote heading angles from their respective means within the observation window, and takes the discrete inner product of these two difference sequences as the covariance numerator; the system calculates the sum of squares of the two difference sequences, adds a preset heading angle noise bias constant to each sum of squares, and then takes the square root of the two sets of values ​​after adding the bias and multiplies them, taking the product as the covariance denominator; the main processor divides the covariance numerator by the denominator and outputs the normalized covariance measure. The value of the heading angle noise floor bias constant is determined based on the zero bias instability parameter calibrated by the gyroscope hardware at the factory. In this embodiment, it is set to a very small positive real number of 10 to the power of -8.

[0145] In step S505, the system performs a multi-dimensional feature joint determination to establish the companion status. The main processor compares the calculated normalized covariance measure with a preset motion consistency determination threshold (calibrated in the range of 0.75 to 0.85 in this embodiment). Simultaneously, the system extracts the normalized path cost output during the aforementioned dynamic time warping process. Only when the normalized path cost is lower than a preset step frequency dispersion threshold and the normalized covariance measure is greater than the aforementioned motion consistency determination threshold, does the system determine that the remote cooperating node has genuine companion behavior and triggers a high-dynamic-risk exposure warning flag in the memory state machine. This joint determination mechanism effectively avoids misjudgments caused by occasional unidirectional walking or local sensor jitter.

[0146] Based on the connectivity flags and normalized covariance measure of the aforementioned output, the system performs classification and dynamic weighting of exposure modes. Since the wake effect generated by human movement in dynamic companionship scenarios increases the risk of contact propagation, the system constructs a dynamic weighting mechanism to distinguish between companionship and static coexistence states. In this embodiment, the specific processing flow for dynamic weighting of relative exposure modes is executed according to the following steps.

[0147] Step S506: For high-risk nodes with established physical connectivity, the main processor performs joint classification of motion states based on the feature sequences of the inertial measurement units of both devices. The system extracts the triaxial acceleration magnitude sequence within the historical observation window and calculates its local variance, comparing this variance with a preset static judgment threshold. The acceleration magnitude variance effectively filters out the fixed offset of the gravity vector, purely reflecting the intensity of the device's motion. Based on the comparison results, when the acceleration magnitude variances of both devices are lower than the preset static judgment threshold (calibrated in the range of 0.02 to 0.05 in this embodiment), the system determines that both devices are in a static coexistence mode; if the magnitude variance of either device is greater than or equal to the threshold, the system classifies it into a dynamic companion state. For the state machine discrimination logic based on acceleration magnitude variance, those skilled in the art can refer to the anti-shake algorithm of wearable devices, whose underlying state classification mechanism is a well-known technology in the field and will not be elaborated here.

[0148] Step S507: For remote collaborative nodes identified as being in a static coexistence mode, the system assigns a basic exposure weight coefficient. Since the respiratory airflow diffusion model tends to be steady-state in a static environment, the system assigns it a constant-level baseline weight. As a preferred approach, this static basic weight coefficient is directly set to a standard normalized baseline constant, i.e., a value of 1.0. The technical purpose of this setting is to ensure that the risk exposure in static scenarios strictly follows a linear growth law over time.

[0149] In step S508, for nodes determined to be in a dynamic accompanying state, the main processor calls the aforementioned normalized covariance measure to calculate the relative exposure weight coefficient. To avoid discontinuous risk integrals caused by weight abrupt changes at threshold edges, the system uses an improved logistic regression smoothing mapping function for continuous assignment calculation. The specific calculation logic is as follows: the system calculates the difference between the preset dynamic exposure surge weight upper limit and the static basic exposure weight constant (i.e., the value 1.0), and uses it as the maximum adjustable increment; the system calculates the difference between the current normalized covariance measure and the preset motion consistency determination threshold, multiplies this difference by a preset smoothing scaling factor, and takes the negative value as the exponent of the natural constant base; the system adds the value 1 to the exponent of the natural constant to construct the denominator; finally, the main processor divides the aforementioned maximum adjustable increment by the denominator, and accumulates the resulting quotient with the static basic exposure weight constant to output the relative exposure weight coefficient for that node. In this algebraic operation structure, since the exponential function range of the natural constant is always greater than zero, the denominator constructed by adding it to the value 1 is always greater than 1. This avoids the arithmetic anomaly of the denominator approaching 0 in the division operation from the underlying logic.

[0150] In step S509, the system performs boundary constraints and calibrations on the key parameters of the above mapping function, taking into account the specific physical scenario. The dynamic exposure surge weight upper limit is used to characterize the risk gain in the accompanying state; in this embodiment, it is set within the numerical range of 2.5 to 3.5. The smoothing scaling factor is set to an empirical value of 10 to 15 to ensure that the weights have appropriate transitional continuity near the judgment threshold. By introducing this nonlinear mapping mechanism, when the normalized covariance measure is much lower than the motion consistency judgment threshold, the weighting coefficient smoothly falls back to near 1.0; when the measure approaches the positive correlation limit, the weighting coefficient asymptotically saturates to the surge upper limit.

[0151] In step S510, combining the above branch processing logic, the main processor outputs the final relative exposure weight coefficient for the remote cooperative node. This coefficient is directly written into the risk buffer register in the underlying memory and fed as an operand into the subsequent risk integral model. By explicitly distinguishing physical patterns and assigning differentiated smooth weights, the system establishes an evaluation closed loop based on motion state feedback, improving the quantitative accuracy of contact risk.

[0152] Based on the connectivity flag and relative exposure weight coefficients output above, the system constructs a multidimensional dynamic risk integral model, quantifying exposure risk through continuous accumulation over discrete time slices. In this embodiment, the specific processing flow for the multidimensional dynamic risk integral and early warning triggering is executed according to the following steps.

[0153] In step S511, the main processor synchronously acquires the multi-source feature input stream of the target's near-end high-risk node within a discrete time slice. Specifically, the system extracts the equivalent statistical distance, connectivity flag, and relative exposure weight coefficient. Addressing the differences in sensor sampling frequencies, the main processor performs temporal resampling and data frame alignment on the multi-source data based on the global timestamp provided by the operating system, ensuring that the feature data participating in subsequent calculations are in the same spatiotemporal cross-section. The equivalent statistical distance is generated by weighted fusion based on the distance evaluation label generated in step S303 using the logarithmic distance path loss model, combined with the aforementioned spatial physical distance calculated using the actual acoustic delay between the target and the near-end high-risk node.

[0154] In step S512, after data alignment is completed, the system calculates the transient exposure risk increment within the current discrete time slice. To prevent arithmetic overflow anomalies where the denominator is zero due to excessively close physical distances, the system introduces a spatial attenuation noise floor bias constant in the division operation. The specific logic for calculating the transient exposure risk increment is as follows: the system performs an exponential operation using the equivalent statistical distance within the current discrete time slice as the base and a preset spatial attenuation index as the power, and adds the result to the preset spatial attenuation noise floor bias constant to construct the basic attenuation denominator; the system extracts the basic pathogen release rate constant pre-stored in the local non-volatile memory as the numerator, divides it by the aforementioned basic attenuation denominator to obtain the basic attenuation risk value; the main processor performs a continuous multiplication operation on the basic attenuation risk value, the connectivity flag bit within the current discrete time slice (set to 1 when there is no physical barrier, otherwise set to 0), and the relative exposure weight coefficient, and outputs the product as the transient exposure risk increment for the near-end high-risk node.

[0155] In step S513, the system performs physical boundary constraints and calibration on the above parameters. The spatial attenuation index is typically calibrated to 2.0 to 3.0; the basic pathogen release rate constant is configured according to the adaptability of the prevalent pathogen, and in this embodiment, the normalized value is 1.0. The spatial attenuation noise floor bias constant is determined based on the equivalent physical thickness of the equipment casing, and in this embodiment, it is set to 0.01. The connectivity flag bit participates in the calculation as a multiplicative mask, directly filtering out invalid risk accumulation under physical isolation conditions through the multiplication attribute.

[0156] In step S514, the system performs cumulative summation on the transient exposure risk increments within the sliding time window. The main processor extracts the transient exposure risk increments corresponding to all discrete time slices included in the current sliding time window, adds them discretely, and uses the sum as the cumulative risk index for the near-end high-risk node at the current observation time. In this embodiment, the sliding window span is set to 15 to 30 minutes. For the underlying circular queue and sliding window summation algorithm, those skilled in the art can adopt a sliding window incremental update strategy, and its buffer maintenance mechanism is a well-known technology in the field, which will not be described in detail here.

[0157] In step S515, the system performs a multi-dimensional joint early warning judgment and reports it. The main processor extracts the discrete equivalent statistical distance sequence within the sliding time window and calculates the average equivalent statistical distance within the window. When the cumulative risk index is greater than the preset risk trigger threshold, and the above average equivalent statistical distance is less than the preset close-range contact baseline, the system determines that there is a high-risk exposure. When the above triggering conditions are met, the system encapsulates the local device identity, the high-risk remote node identity, and the over-limit event timestamp, and calls the underlying network protocol stack to upload it to the security gateway. This joint triggering mechanism effectively intercepts false alarms caused by local signal fluctuations or extremely short transient contact.

[0158] Specific application examples:

[0159] Consider the following typical complex daily scenario. Assume that both User A (a potential infected person) and User B (a healthy contact) are wearing IoT terminals (such as smart badges or wristbands) running the system of this invention.

[0160] Scenario 1: False contact between adjacent offices (avoiding false alarms caused by walls being penetrated)

[0161] Working conditions: User A and User B are seated in two adjacent offices, separated by a solid brick wall. The straight-line physical distance between the two is only 2 meters.

[0162] System response. Hardware scheduling: The radio frequency module detected a strong Bluetooth signal (the signal remained strong even after passing through walls due to the close proximity), and the two users' terminals occasionally got up to retrieve documents, triggering slight fluctuations in the inertial measurement unit (IMU). The system was awakened from deep sleep mode to high-frequency tracking mode.

[0163] Baseline and Connectivity Decision: The terminal begins collecting air pressure and high-frequency audio data. System calculations reveal that although the radio frequency distance is short, the coherence of the net air pressure waveforms in the microenvironment between the two parties is extremely low (airflow is blocked by the wall), and the difference in high-frequency acoustic envelopes is extremely large (near-ultrasound cannot penetrate the brick wall).

[0164] Judgment result: The double Boolean decision output of the cross-media space connectivity decision module is 0 (in a physical isolation state).

[0165] Points and warnings: The connectivity flag mask in the risk points module is 0. Therefore, even if two people sit still for 2 hours with a distance of 2 meters between them, the cumulative risk index is always 0, which successfully avoids the common wall penetration misjudgment in traditional Bluetooth solutions.

[0166] Scenario 2: High-risk companionship in corridors and elevators (dynamic weighting and precise early warning)

[0167] Working conditions: When leaving get off work, User A and User B meet in the corridor, walk side by side towards the elevator, and then take the elevator downstairs together.

[0168] System Response. Connectivity Confirmation: In an unobstructed corridor, the system detects a significant increase in low-frequency air pressure coherence and minimal high-frequency acoustic energy transmission loss, resulting in a double Boolean decision output of 1 (in the same physical space).

[0169] Gait alignment and accompaniment determination: The main processor acquires the IMU gait frequency sequences of both parties, eliminates the start time difference through the Dynamic Time Warping (DTW) algorithm, and calculates the heading angle covariance. The results show that the gait frequencies are highly consistent and the headings are parallel, and the system determines that the two have entered a dynamic accompaniment state.

[0170] Floor crossing identification: After entering the elevator, the temperature and pressure sensor collected a significant step change in air pressure (human body temperature interference has been eliminated through temperature feedforward compensation), confirming that both parties had experienced a vertical floor crossing, further confirming the fact that they were in a high-risk enclosed space together.

[0171] Points and Early Warning: The risk points module assigns a high relative exposure weight coefficient. Within 5 minutes of walking or riding an elevator together, the transient exposure risk increases exponentially. If the accumulated risk index quickly exceeds the preset threshold, the system will immediately trigger a local vibration warning and upload high-risk records related to the epidemic.

[0172] To verify the effectiveness of this system, comparative experiments were conducted in complex architectural environments, including open office areas, glass conference rooms, and rooms with solid wall partitions. The comparison object was a traditional single-mode low-power Bluetooth (BLE) tracking system.

[0173] Test evaluation dimensions Traditional single-mode RF solutions (relying solely on BLERSSI) The present invention integrates radio frequency, air pressure, acoustics, and inertial navigation. Improvement in performance False alarm rate of solid wall obstruction 82.4% 1.2% By utilizing both air pressure and high-frequency acoustics for dual determination, the possibility of false contact through walls is completely eliminated at the physical medium level. Dynamic companion recognition rate Unable to identify (only reflects distance) 96.5% By aligning the gait and heading angle of both ends using the DTW algorithm, the system effectively distinguishes between brief shoulder brushes and high-risk parallelism. High-frequency ranging effective power consumption The current level remains high (approximately 15mA). Extremely low (1.8mA, mean) By employing a three-level hardware wake-up state machine and a simulated envelope extraction circuit, the overall system energy consumption is significantly reduced. Distance measurement error in complex environments ±3.5 meters (severely affected by multipath effects) ±0.4 meters By introducing UWB assistance and near-ultrasound real time delay alignment, the distance jump caused by radio frequency fading is completely avoided.

[0174] See attached document Figure 3 , Figure 3 The paper presents a comparison of the wall penetration misjudgment and recognition error rates (percentage) of the fusion technology of single-mode radio frequency technology and the present invention in four different typical working scenarios.

[0175] Unobstructed close range: In an ideal unobstructed environment, both technologies perform well with error rates of 5% or less, but the fusion technology of this invention (2%) is still slightly better than the single-mode RF technology (5%).

[0176] Glass curtain wall obstruction: When moderate obstruction is introduced, the error rate of single-mode radio frequency technology rises sharply to 65%, while the technology of this invention remains highly stable with an error rate of only 3%.

[0177] Solid brick wall blockage: In extremely challenging environments with heavy obstruction, single-mode radio frequency technology suffers severe failures with a false positive rate as high as 85%; in contrast, the fusion technology of this invention exhibits extremely strong penetration and anti-interference capabilities, with the error rate suppressed to an extremely low 1%.

[0178] Dynamic accompaniment status: Under target movement and accompaniment interference conditions, the single-mode radio frequency error rate drops to 12%, while the present invention still maintains a low level of 4%.

[0179] Traditional single-mode radio frequency technology is prone to false positives when faced with physical obstructions (especially solid brick walls). The fusion technology proposed in this invention not only maintains an extremely low error rate of less than 5% in all scenarios, but also solves the problem of high false positives in complex obstruction environments.

[0180] See attached document Figure 4 In this figure, the dashed line represents the traditional radio frequency-based solution, and the solid line represents the multidimensional dynamic solution of this invention. The vertical dashed line at 15 minutes in the figure is the dividing line between the end of physical blocking and the beginning of tandem. The first 15 minutes are the physical blocking phase, and the last 15 minutes are the dynamic tandem phase.

[0181] In the initial stage (0-15 minutes, physical isolation period): the cumulative risk index of the traditional RF solution increases linearly from the start of the test, reaching 50 by the 15-minute mark; however, the multi-dimensional dynamic solution of this invention exhibits extremely high reliability in this stage, perfectly shielding the false signals caused by the isolation, and its cumulative risk index remains at 0. In the later stage (15-30 minutes, dynamic companionship period): after crossing the boundary, the system enters a dynamic companionship state, with a sharp increase in environmental complexity and interference. At this point, the risk indices of both solutions begin to rise. Although the risk index of this invention also increases in the later stages (eventually reaching 120), because it achieves zero risk accumulation in the first 15 minutes, its overall risk exposure level and anti-interference hysteresis performance are superior to the traditional solution, which is constantly accumulating risk.

Claims

1. An Internet of Things-based system for tracing and risk warning of infectious disease contacts, characterized in that, Operating in IoT terminals that include low-power communication modules, inertial measurement units, temperature and pressure sensors, and audio envelope circuits, including: The sensor scheduling module acquires radio frequency broadcast packets through the low-power communication module and acquires motion data output by the inertial measurement unit, thereby generating a hardware interrupt signal. The central coprocessor responds to the hardware interrupt signal and controls the temperature and pressure sensor and the audio envelope circuit to switch their operating states. The data preprocessing module processes the air pressure values ​​output by the temperature and pressure sensor to generate low-frequency air pressure data, and obtains high-frequency acoustic envelope signals through the audio envelope circuit. The baseline module receives the low-frequency air pressure data and the high-frequency acoustic envelope signal, performs time axis alignment operations, and generates a macroscopic air pressure baseline and a macroscopic acoustic baseline. The connectivity decision module performs a physical isolation decision based on the macroscopic pressure baseline and the macroscopic acoustic baseline, and generates a connectivity flag bit; The risk scoring module parses the radio frequency broadcast packet to generate contact tracking records, and performs integral calculations based on the motion data, the radio frequency broadcast packet, and the connectivity flag to generate a cumulative risk index and trigger an early warning operation.

2. The Internet of Things-based infectious disease contact tracing and risk warning system according to claim 1, characterized in that, The sensor scheduling module extracts the original received signal strength indication attached to the radio frequency broadcast packet, performs smoothing filtering on the original received signal strength indication, and calculates the first-order time derivative to obtain the time gradient characteristics; The motion data includes raw triaxial acceleration data. The sensor scheduling module performs vector magnitude calculation and fusion on the raw triaxial acceleration data to generate a comprehensive acceleration amplitude, and calculates the sample variance of the comprehensive acceleration amplitude to obtain acceleration variance features. The sensor scheduling module sets the radio frequency trigger flag when the time gradient feature is greater than the approach trend determination threshold, sets the motion trigger flag when the acceleration variance feature is greater than the motion energy wake-up threshold, and generates the hardware interrupt signal based on the radio frequency trigger flag and the motion trigger flag.

3. The Internet of Things-based infectious disease contact tracing and risk warning system according to claim 2, characterized in that, The central coprocessor maintains a finite state machine with deep sleep state, sniffing and inspection state, and high-frequency tracking state; When both the radio frequency trigger flag and the motion trigger flag are in a reset state, the central coprocessor switches the finite state machine to the deep sleep state; When only one of the radio frequency trigger flag and the motion trigger flag is set, the central coprocessor switches the finite state machine to the sniffing and inspection state. The central coprocessor switches the finite state machine into the high-frequency tracking state in response to the hardware interrupt signal only when the radio frequency trigger flag and the motion trigger flag are synchronously set, and controls the temperature and pressure sensor and the audio envelope circuit to switch from the sleep state to the working state.

4. The Internet of Things-based infectious disease contact tracing and risk warning system according to claim 1, characterized in that, The pressure value output by the temperature and pressure sensor includes the original pressure value, and the temperature and pressure sensor also outputs a reference temperature value. The data preprocessing module calculates the physical difference between the reference temperature value and the preset thermodynamic reference temperature to generate a relative temperature offset. Based on the relative temperature offset, it calculates the feedforward error compensation term and combines the original air pressure value with the feedforward error compensation term to obtain a high-precision air pressure value. The data preprocessing module uses a low-pass filter to perform frequency domain isolation on the high-precision air pressure values ​​and outputs a smooth air pressure value sequence. When the absolute difference between adjacent values ​​in the smooth air pressure value sequence jumps, abnormal interference is eliminated, and the low-frequency air pressure data is output.

5. The Internet of Things-based infectious disease contact tracing and risk warning system according to claim 1, characterized in that, The audio envelope circuit is connected to a microphone transducer at its front end for capturing ambient sound waves and outputting a raw analog AC voltage signal. The audio envelope circuit includes an analog bandpass filter, a full-wave precision rectifier, and a hardware integration network. The analog bandpass filter receives the original analog AC voltage signal and outputs a high-frequency AC signal; The full-wave precision rectifier converts the high-frequency AC signal into a unipolar pulsating DC signal. The hardware integral network smooths the unipolar pulsating DC signal and converts it into a low-frequency envelope analog voltage; The data preprocessing module downsamples the low-frequency envelope analog voltage to obtain an acoustic energy sequence, calculates the energy mutation rate based on a preset long-term energy baseline of environmental background noise, and confirms and outputs the high-frequency acoustic envelope signal based on the energy mutation rate.

6. The Internet of Things-based infectious disease contact tracing and risk warning system according to claim 2, characterized in that, The baseline module filters remote cooperating nodes through the low-power communication module and receives remote air pressure data and remote acoustic data broadcast by the remote cooperating nodes. The baseline module performs a sliding cross-correlation operation on the locally acquired high-frequency acoustic envelope signal and the remote acoustic data to extract local extrema and lock the true sound wave propagation delay. The baseline module uses the real sound wave propagation delay to perform time-domain reverse compensation and alignment on the remote air pressure data and the remote acoustic data, and calculates the node spatial distance in combination with the preset sound speed. Based on the basic weighting factor constructed by the node spatial distance and the data variance attenuation law, it performs weighted fusion to generate the macroscopic air pressure baseline and the macroscopic acoustic baseline.

7. The Internet of Things-based infectious disease contact tracing and risk warning system according to claim 6, characterized in that, The connectivity decision module subtracts the scaled macroscopic pressure baseline from the low-frequency air pressure data to obtain the local microenvironment net air pressure feature sequence, and performs frequency band isolation and short-time energy normalization processing on the local microenvironment net air pressure feature sequence using a bandpass filter. The connectivity decision module obtains the remote microenvironment net air pressure feature sequence through the low-power communication module, and performs time-domain reverse translation alignment on the remote microenvironment net air pressure feature sequence using the real sound wave propagation delay. The connectivity decision module constructs a discrete model based on the local arithmetic expectation of the above feature sequences, quantifies the geometric similarity between the local microenvironment net air pressure feature sequence and the aligned remote microenvironment net air pressure feature sequence, generates a transient air pressure waveform coherence measure, and applies sliding filtering to generate a historical steady-state coherence measure.

8. The Internet of Things-based infectious disease contact tracing and risk warning system according to claim 7, characterized in that, The connectivity decision module converts the high-frequency acoustic envelope signal and the time-domain aligned far-end acoustic data to the frequency domain for processing, outputs a smooth high-frequency power spectral density estimate, and calculates the high-frequency acoustic envelope difference by quantifying the relative logarithmic difference between the two and introducing a noise floor bias mechanism. The connectivity decision module uses hysteresis comparison logic to generate a barometric connectivity Boolean indicator for the historical steady-state coherence measure and an acoustic connectivity Boolean indicator for the high-frequency acoustic envelope difference. The connectivity decision module performs a logical AND operation on the barometric connectivity Boolean indicator and the acoustic connectivity Boolean indicator to jointly generate the connectivity flag bit that reflects the physical isolation status.

9. A system for tracking and risk warning of infectious disease contacts based on the Internet of Things according to claim 6, characterized in that, The risk integration module calculates the local step frequency sequence and the local heading angle sequence based on the motion data output by the inertial measurement unit. The risk integration module obtains the remote step frequency sequence and remote heading angle sequence of the remote cooperative node through the low-power communication module. The risk integration module performs dynamic time warping on the local step frequency sequence and the remote step frequency sequence to generate a synchronization index mapping table, and uses the mapping table to align the remote heading angle sequence with the local heading angle sequence on the time axis. The risk integral module calculates the normalized covariance measure of the aligned far-end heading angle sequence and the local heading angle sequence, jointly classifies the motion state based on the acceleration variance feature, and calculates the relative exposure weight coefficient through a logistic regression smoothing mapping function when the motion state is determined to be dynamic accompaniment.

10. A system for tracking and risk warning of infectious disease contacts based on the Internet of Things according to claim 9, characterized in that, The system is pre-loaded with distance assessment tags and a basic pathogen release rate constant for characterizing physical distance; The risk integration module uses a path loss model to map the original received signal strength indication attached to the radio frequency broadcast packet to the distance evaluation label, and combines it with the node spatial distance calculated from the actual sound wave propagation delay to generate a weighted equivalent statistical distance. The risk integration module calculates the basic attenuation risk value using a preset spatial attenuation noise floor bias constant and the basic pathogen release rate constant, and outputs the transient exposure risk increment by multiplying the basic attenuation risk value, the connectivity flag, and the relative exposure weight coefficient. The risk scoring module performs cumulative summation on the transient exposure risk increment within a sliding time window and updates the cumulative risk index. When the cumulative risk index exceeds the limit and the average value of the equivalent statistical distance is less than the preset close contact baseline, the warning operation is triggered.

Images