A method and system for home-based safety monitoring of elderly people based on behavior perception

By fusing radar echoes and plantar pressure signals, extracting limb swing cycles and trunk displacement vectors, and calculating phase difference sequences and their gradient changes, the accuracy and timeliness issues of gait asynchrony recognition in existing technologies are solved, enabling early warning of falls in elderly people living at home.

CN123119618APending Publication Date: 2026-11-17GUANGXI LVFA TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611499722.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately identify gait asynchrony in elderly individuals living at home by aligning and integrating radar echo sensing and plantar pressure sensing at a multi-dimensional temporal level without intrusion, resulting in short fall warning windows and frequent false alarms and missed alarms.

Method used

By fusing ultra-wideband radar echoes and plantar pressure distribution signals, the limb swing cycle and trunk displacement vector are extracted, the phase difference sequence and its gradient are calculated, and the number and duration of desynchronization segments are combined to output a fall warning signal.

Benefits of technology

It enables early detection and accurate warning of gait coordination abnormalities in the elderly, reduces the probability of false alarms, and improves the timeliness and accuracy of warnings.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123119618A_ABST
    Figure CN123119618A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for home-based safety monitoring of elderly people based on behavior perception, belonging to the field of intelligent monitoring technology. The method includes: acquiring continuous motion echo signals and plantar pressure distribution signals of the subject within a target area; extracting the limb swing period and trunk displacement vector from the continuous motion echo signals, and extracting the bipedal support phase time series from the plantar pressure distribution signals; aligning the limb swing period, trunk displacement vector, and bipedal support phase time series in chronological order to form multiple behavioral segments within sliding windows; calculating the phase difference sequence between the limb swing period and trunk displacement vector for each behavioral segment, determining whether the gradient of the phase difference sequence exceeds a first threshold, and marking behavioral segments exceeding this threshold as out-of-synchronization segments. This method integrates multi-source signals for gait coordination analysis, enabling early detection of out-of-synchronization signs and improving the accuracy and timeliness of fall warnings.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, specifically to a method and system for home-based safety monitoring of elderly people based on behavior perception. Background Technology

[0002] Fall monitoring for elderly people living at home is a challenge in the field of safety supervision. Current technologies commonly employ a single type of sensor for motion perception, such as using ultra-wideband radar to acquire distance and Doppler information of human movement, or using piezoresistive membranes laid on the ground to collect plantar pressure data. While ultra-wideband radar can penetrate clothing and non-metallic obstructions and is insensitive to light conditions, extracting the overall speed and displacement trend of trunk movements, its ability to distinguish small-scale rapid swings of the limbs is limited, making it difficult to precisely characterize the coordination of different body parts during the gait cycle. Plantar pressure sensing directly reflects the contact state between the feet and the ground and the process of center of gravity transfer, accurately distinguishing between the support and swing phases. However, it cannot perceive the dynamic information of the trunk and upper limbs. When an elderly person experiences dizziness or sudden weakness in the lower limbs, leading to a mismatch between trunk and limb movements, changes in plantar pressure alone often present a delayed imbalance characteristic, making it difficult to provide early warning when a fall is imminent.

[0003] Some studies have attempted to combine radar with pressure sensors, but most employ a weighted fusion approach after independent threshold judgments, failing to establish timing alignment and phase coordination analysis at the signal level. This results in the underutilization of the time-varying coupling relationship between the two types of heterogeneous signals. In actual gait instability, the originally stable phase difference between the limb swing rhythm and the trunk center of gravity displacement rapidly becomes disordered, causing a loss of coordination between swinging and stepping. This desynchronization phenomenon may not be significant in a single limb swing frequency or trunk velocity curve, but it is clearly reflected in the gradient of the phase difference between the two. Existing methods lack quantitative assessment tools for this limb-trunk phase coordination, often triggering alarms only after a fall has occurred or is about to hit the ground, resulting in short warning windows and frequent false alarms and missed alarms.

[0004] The current challenge is to find a way to align and integrate radar echo sensing and plantar pressure sensing at a multi-dimensional temporal level without disturbing elderly people living at home, and to identify gait desynchronization as early as possible based on the dynamic changes in the phase coordination between limb swing and trunk displacement. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for home-based safety monitoring of elderly people based on behavior perception. By fusing ultra-wideband radar echoes and plantar pressure distribution signals, the limb swing period and trunk displacement vector are extracted from the continuous motion echo signal, and the bipedal support phase time series is extracted from the plantar pressure distribution signal. Behavioral segments are constructed by aligning them according to the acquisition sequence. Within the behavioral segments, the phase difference sequence and its change gradient of the limb swing period and trunk displacement vector are calculated. The change gradient is used to identify out-of-synchronization segments, and the number and duration of out-of-synchronization segments per unit time are combined to output a fall warning signal, so as to achieve early detection and accurate warning of gait coordination abnormalities in the elderly.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a home-based safety monitoring method for elderly people based on behavior perception. Through multi-dimensional sensing signal fusion and gait desynchronization analysis, it achieves a non-contact, forward-looking assessment of fall risk, including: collecting continuous motion echo signals and plantar pressure distribution signals of the tested object within a target area; extracting the limb swing period and trunk displacement vector based on the Doppler frequency shift sequence of the continuous motion echo signals in the time domain, and extracting the bipedal support phase time series based on the plantar pressure distribution signal; aligning the limb swing period, trunk displacement vector, and bipedal support phase time series according to the acquisition sequence to form multiple behavioral segments within a continuous sliding window; for each behavioral segment, calculating the phase difference sequence between the limb swing period and trunk displacement vector, and determining whether the gradient of the phase difference sequence within the sliding window exceeds a first threshold. If it does, the corresponding behavioral segment is marked as a desynchronization segment; counting the number and duration percentage of the desynchronization segments per unit time, and outputting a fall warning signal when the number exceeds a second threshold and the duration percentage exceeds a third threshold.

[0007] As a preferred technical solution of the present invention, continuous motion echo signals and plantar pressure distribution signals of the measured object within the target area are collected. Specifically, this is achieved by transmitting a linear frequency modulated continuous wave at a fixed repetition frequency using an ultra-wideband radar, receiving the echo signals reflected from the target area, and extracting the phase change sequences at each range gate after mixing and Fourier transform as the continuous motion echo signal. Simultaneously, a piezoresistive sensor array laid on the surface of the target area collects the resistance change values ​​at the contact points between the sole and the ground at a fixed sampling frequency, mapping these resistance change values ​​to plantar pressure distribution signals. Through the spatiotemporal synchronous acquisition of radar and pressure sensors, joint sensing of large-scale trunk displacement and small-scale limb micro-movements of the human body, as well as direct measurement of the plantar contact state, is achieved, providing a stable and reliable multi-source data foundation for gait phase analysis.

[0008] As a preferred embodiment of the present invention, the extraction method of the limb swing period and trunk displacement vector is as follows: Frequency components higher than a first frequency threshold are separated from the Doppler frequency shift sequence as limb swing Doppler components; the interval between two consecutive zero-crossing points of the limb swing Doppler components is taken as the limb swing period; frequency components lower than a second frequency threshold are separated from the Doppler frequency shift sequence as trunk motion Doppler components; and the trunk motion Doppler components are phase-unwrapped and restored to the trunk displacement vector. Through frequency domain segmentation processing, the periodic characteristics representing rapid lower limb swing and the displacement characteristics representing slow movement of the body's center of gravity are simultaneously calculated from a single radar echo signal, without the need for any sensors worn on the human body, thus avoiding interference with the daily life of the elderly by invasive monitoring.

[0009] As a preferred embodiment of the present invention, the extraction method of the bipedal support phase time series is as follows: the piezoresistive sensor array is divided into a left foot region and a right foot region, and the total pressure in the left foot region and the total pressure in the right foot region are calculated respectively; when the total pressure in the left foot region is greater than a fourth threshold and the total pressure in the right foot region is greater than the fourth threshold, the current sampling time is marked as the bipedal support time; all sampling intervals continuously marked as bipedal support times are arranged in chronological order to form the bipedal support phase time series. This process utilizes the physical characteristic of simultaneous pressure on both feet to accurately define the stable support phase in the gait cycle, providing an accurate gait phase reference for subsequent limb-trunk phase difference analysis.

[0010] As a preferred technical solution of the present invention, the behavioral segment is constructed as follows: using the sampling time of the plantar pressure distribution signal as the reference time axis, the limb swing cycle and the trunk displacement vector are interpolated to the corresponding time on the reference time axis according to their respective signal acquisition times; a window of fixed duration is slid along the reference time axis with a fixed stride length, and the aligned limb swing cycle sequence, trunk displacement vector sequence, and bipedal support phase marker sequence within the window are encapsulated into a behavioral segment. Through time alignment and sliding window truncation, the multimodal time-series data output by heterogeneous sensors are standardized and organized on a unified time scale, so that each behavioral segment contains complete gait phase coupling information, facilitating subsequent continuous analysis and real-time processing.

[0011] As a preferred technical solution of the present invention, the specific method for determining whether the gradient of the phase difference sequence exceeds the first threshold is as follows: For each moment within the behavioral segment, the difference between the phase value of the limb swing period and the phase value of the trunk displacement vector corresponding to that moment is taken as the instantaneous phase difference; the instantaneous phase differences of all moments within the behavioral segment are arranged in chronological order to form a phase difference sequence, and the absolute value of the difference between two adjacent instantaneous phase differences in the phase difference sequence is calculated to obtain a sequence of absolute difference values; the intervals in the sequence of absolute difference values ​​where three or more consecutive absolute difference values ​​are greater than the fifth threshold are counted, and the gradient of change within the interval is marked as exceeding the first threshold. By setting a continuous point count requirement, single-point phase jumps caused by instantaneous noise or random fluctuations can be effectively filtered out, ensuring that the detected phase desynchronization is caused by continuous gait coordination degradation, and significantly reducing the probability of false alarms.

[0012] As a preferred technical solution of the present invention, the specific method for marking the corresponding behavior segment as a desynchronization segment is as follows: record the total duration covered by all intervals within the behavior segment marked as having a gradient change exceeding the first threshold; when the proportion of the total duration to the total duration of the behavior segment exceeds a sixth threshold, mark the behavior segment as a candidate desynchronization segment; when the behavior segments output by two consecutive sliding windows are both candidate desynchronization segments, the latter behavior segment is finally marked as the desynchronization segment. This solution introduces a candidate confirmation mechanism and continuity determination, confirming a valid desynchronization event only when the gait desynchronization phenomenon persists within a certain time domain, thereby enhancing the system's ability to distinguish between brief posture adjustments and true balance loss of control.

[0013] As a preferred technical solution of the present invention, the specific method for statistically analyzing the number and duration percentage of desynchronization segments per unit time is as follows: The number of all behavioral segments marked as desynchronization segments per unit time is counted as the number of desynchronization segments; the sum of the durations of each desynchronization segment per unit time is accumulated, and the sum of the durations is divided by the unit time length to obtain the duration percentage. This statistical method comprehensively measures the severity of gait desynchronization from two dimensions: frequency and density. This ensures that fall risk assessment not only focuses on the frequency of abnormal events but also on the cumulative effect of abnormal states over time, improving the accuracy and timeliness of early warnings.

[0014] As a preferred embodiment of the present invention, the specific method for outputting a fall warning signal is as follows: It is determined whether the number of desynchronization segments within a unit time is greater than a second threshold, and whether the duration percentage is greater than a third threshold. When both conditions are met simultaneously, a warning trigger flag is generated. After reading the warning trigger flag, a fall warning message containing a target area identifier and a timestamp is sent to the bound mobile terminal via a wireless communication module, and the local audible and visual alarm is activated. This method uses the dual exceedance of desynchronization frequency and density as the warning trigger condition, enabling proactive warnings before an elderly person's balance ability continuously deteriorates and a fall is imminent. This provides caregivers with valuable intervention time, and the dual-channel alarm system (audio and visual, and remote message) ensures that the warning information is received and processed promptly.

[0015] This invention also provides a home-based elderly safety monitoring system based on behavior perception, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the behavior perception-based home-based elderly safety monitoring method. This system integrates the multi-sensor capabilities of radar and pressure arrays, gait phase desynchronization analysis algorithms, and multi-level early warning logic into a single device. It can be deployed in the home environment for 24 / 7 unobtrusive monitoring of the elderly, proactively generating early warnings in the early stages of balance function decline before a fall occurs. This transforms passive, reactive alarms into proactive, pre-emptive intervention, providing reliable technical support for the safety of elderly people living at home.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By acquiring continuous motion echo signals and plantar pressure distribution signals of the target object within the target area, and extracting the limb swing period and trunk displacement vector based on the Doppler frequency shift sequence of the continuous motion echo signals in the time domain, and extracting the time series of the bipedal support phase based on the plantar pressure distribution signals, the three types of features are aligned according to the acquisition time sequence and formed into behavioral segments within multiple continuous sliding windows. This scheme will match and fuse the macroscopic limb rhythm and trunk spatial displacement information acquired by radar with the plantar ground contact support time information sensed by the piezoresistive array on a unified time axis, so that each behavioral segment contains both periodic features reflecting the speed of limb swing and displacement vector features representing the overall direction of body movement, while also including support phase markers of simultaneous foot contact. Compared to using radar alone, which can only acquire general body motion parameters, or using pressure sensors alone, which can only acquire foot strike status, the joint construction of multi-source temporal features can fully present the spatiotemporal coordination structure among upper limb swing, trunk displacement, and lower limb support in the gait cycle. This allows subtle coordination abnormalities caused by factors such as decreased muscle strength and loss of balance to be exposed in the multi-dimensional representation space of behavioral segments. When early gait instability occurs in the elderly, it often manifests as a unchanged limb swing rhythm but a lag in trunk displacement, or a sudden change in trunk displacement direction without timely adjustment by the limbs. This combination of multi-dimensional features within behavioral segments can detect motion incoordination phenomena that cannot be detected by a single signal source, providing a rich analytical foundation for subsequent synchronous recognition.

[0017] After obtaining behavioral segments, the phase difference sequence between the limb swing period and the trunk displacement vector is calculated for each segment. The gradient of this phase difference sequence within a sliding window is then determined to exceed a first threshold. Behavioral segments exceeding the first threshold are marked as desynchronized segments. The number and duration percentage of desynchronized segments per unit time are statistically analyzed. When the number exceeds a second threshold and the duration percentage exceeds a third threshold, a fall warning signal is output. The phase difference between the limb swing period and the trunk displacement vector directly quantifies the temporal coordination between the upper and lower limbs and the trunk. During normal walking, the two maintain a relatively stable phase relationship with a gentle gradient. When the body becomes unbalanced, slips, or is about to fall, the limbs instinctively attempt to regain balance through rapid swinging, while the trunk continues to undergo unexpected displacement due to inertia. The phase difference between the two changes continuously within a short period, forming a significant gradient. Using the gradient of the phase difference sequence as a desynchronization criterion allows for the mechanistic capture of the unique limb-trunk decoupling phenomenon before a fall, avoiding misjudgments caused by individual differences and the diversity of daily movements when relying solely on amplitude or velocity thresholds. Further employing a statistical method based on the continuous variation range of phase difference within a sliding window to determine desynchronization segments can filter out transient interference, marking only the time intervals where phase coordination significantly deteriorates. Post-hoc, a dual-threshold joint judgment is applied to the number of desynchronization segments and the proportion of their cumulative duration on a unit time scale. This considers both the frequency of desynchronization events and the proportion of their continuous accumulation, enabling early warning to be triggered when elderly individuals experience sustained and frequent deterioration in gait coordination. This allows for the output of alarm signals before a fall actually occurs, gaining more response time and reducing false alarms caused by occasional movements or brief center-of-gravity adjustments. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a flowchart of a home-based safety monitoring method for elderly people based on behavior perception. Figure 2 This is a flowchart of Doppler frequency shift separation and extraction of limb swing period and trunk displacement; Figure 3 This is a comparison chart of phase change sequences at different distance gates; Figure 4 These are Doppler frequency shift sequences and time-domain variation curves of Doppler components in limb swing and trunk movement; Figure 5 This is a diagram showing the sum of pressure in the left and right areas of the sole and the support threshold of both feet; Figure 6 This is a schematic diagram showing the absolute value of the phase difference sequence difference and the gradient of change within a behavior segment. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of 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, 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.

[0021] See Figure 1 This invention provides a home-based safety monitoring method for elderly people based on behavior perception, comprising: acquiring continuous motion echo signals and plantar pressure distribution signals of the subject within a target area; extracting the limb swing period and trunk displacement vector based on the Doppler frequency shift sequence of the continuous motion echo signals in the time domain, and extracting the bipedal support phase time series based on the plantar pressure distribution signal; aligning the limb swing period, trunk displacement vector, and bipedal support phase time series according to the acquisition sequence to form multiple behavior segments within a continuous sliding window; for each behavior segment, calculating the phase difference sequence between the limb swing period and the trunk displacement vector, and determining whether the gradient of the phase difference sequence within the sliding window exceeds a first threshold; if it exceeds, marking the corresponding behavior segment as a desynchronized segment; counting the number and duration percentage of the desynchronized segments per unit time; when the number exceeds a second threshold and the duration percentage exceeds a third threshold, outputting a fall warning signal.

[0022] In practical implementation, the ultra-wideband radar transmits linear frequency modulated continuous waves at a fixed repetition frequency. The expression for the transmitted signal of a single linear frequency modulated continuous wave is as follows: in, Indicates the transmission of a signal. Indicates the amplitude of the transmitted signal. This represents the starting frequency of a linear frequency modulated continuous wave. Indicates the frequency modulation bandwidth. Indicates the duration of a single frequency modulation cycle. This represents the time variable. Ultra-wideband radar receives echo signals reflected from the target area. The echo signal has a time delay relative to the transmitted signal; the expression for the echo signal is: , Indicates the amplitude of the echo signal. Indicates echo delay, , This indicates the distance between the target object and the ultra-wideband radar. This indicates the speed at which electromagnetic waves propagate through the air. It refers to the speed at which a signal is transmitted. With echo signal The signal is mixed, and then the mixed signal is passed through a low-pass filter to remove high-frequency components, resulting in an intermediate frequency signal. The expression for the intermediate frequency signal is: , This indicates the amplitude of the intermediate frequency (IF) signal. (For the IF signal...) Perform a Fourier transform to obtain the spectrum, and then search for the frequency corresponding to the peak in the spectrum. ,frequency satisfy According to frequency Calculate the distance . Distance along the distance dimension The range of values ​​is divided into multiple range gates, each corresponding to a continuous range. From the Fourier transform results, the phase values ​​within the range corresponding to each range gate are obtained. The phase values ​​acquired by a range gate in multiple consecutive frequency modulation cycles are arranged chronologically to form a phase change sequence for that range gate. The phase change sequence corresponding to the range gate containing the micro-motion information of the measured object is selected as the continuous motion echo signal. The ultra-wideband radar continuously repeats the above process to acquire the continuous motion echo signal at a fixed repetition frequency.

[0023] Simultaneously, a piezoresistive sensor array is laid on the surface of the target area. This array contains multiple piezoresistive sensitive elements, each whose resistance changes when subjected to pressure. The piezoresistive sensor array acquires data at a fixed sampling frequency. The acquisition circuit applies a constant voltage or current to each piezoresistive sensitive element, measuring the voltage or current value across each element. This value is then converted from analog to digital to obtain a digitized resistance change. The resistance change value of each piezoresistive sensitive element is input into a pre-calibrated mapping relationship. This mapping relationship is established by applying a known pressure value to the piezoresistive sensitive element and recording the corresponding resistance change value. Using the resistance change value as input, the corresponding pressure value is output. The pressure values ​​corresponding to all piezoresistive sensitive elements at the same sampling time are combined according to the row and column positions of the piezoresistive sensitive elements in the piezoresistive sensor array to form a pressure distribution matrix, thus obtaining the plantar pressure distribution signal. The ultra-wideband radar and the piezoresistive sensor array are synchronized by the same clock source. In each synchronization trigger signal cycle, the ultra-wideband radar completes the transmission and reception of a linear frequency modulated continuous wave, while the piezoresistive sensor array simultaneously completes the acquisition of the resistance change value of all piezoresistive sensitive units, so that the continuous motion echo signal and the plantar pressure distribution signal are synchronized in time.

[0024] See Figure 3In the figure, the horizontal axis represents time in seconds, and the vertical axis represents phase in radians. Five curves are plotted, corresponding to the phase change sequences of the five range gates set by the ultra-wideband radar. The phase sequences of range gate 1, range gate 2, range gate 4, and range gate 5 are represented by dashed and dotted lines, while the range gates containing the micro-motion information of the measured object are represented by solid lines.

[0025] The phase curve of the range gate containing the micro-motion information of the measured object, as shown in the figure, exhibits obvious oscillating characteristics, with an amplitude range significantly larger than that of the phase curves of the other four range gates. The phase of this range gate changes periodically over time, reflecting the continuous motion state of the target object's micro-motions. The phase curves of the other range gates have smaller amplitudes and gentler fluctuations, indicating that they do not contain significant micro-motion information of the measured object within their corresponding distance ranges. Within the time interval shown in the figure, the phase curve of the range gate containing the micro-motion information of the measured object shows a trend of alternating rapid fluctuations and slow changes, reflecting the dynamic characteristics of the target object's behavior. The time-domain variation trend of this phase curve can be used to further calculate the Doppler frequency shift sequence, thereby extracting the limb swing period and trunk displacement vector.

[0026] In specific implementation, please refer to Figure 2 The Doppler frequency shift sequence in the time domain is obtained from the continuous motion echo signal. The continuous motion echo signal is a phase change sequence at a selected range gate, and the phase change sequence is denoted as... , This represents the discrete-time index, where the time interval between adjacent sampling times is... .right Perform discrete difference operations to obtain the differential phase sequence. Doppler shift The relationship with the differential phase is as follows: in, Indicates the first Doppler frequency shift values ​​corresponding to each sampling time. This represents the repetition period of a linear frequency modulated continuous wave transmitted by an ultra-wideband radar. Arranged in chronological order, they form a Doppler shift sequence.

[0027] Frequency components above a first frequency threshold are separated from the Doppler shift sequence and identified as the limb-stance swing Doppler components. The first frequency threshold is determined based on the typical frequency range of limb-stance swing in human kinematics; typically, the Doppler shift caused by limb-stance swing is distributed in the higher frequency band. A cutoff frequency of 3 Hz is set as the first frequency threshold. The Doppler shift sequence is input into a digital high-pass filter with a cutoff frequency of 3 Hz. The digital high-pass filter employs an infinite impulse response filter structure with a filter order of 4, and the filter coefficients are obtained using the Butterworth design method. The digital high-pass filter performs point-by-point filtering on the Doppler shift sequence, and the output sequence is the limb-stance swing Doppler component.

[0028] The interval between two consecutive zero-crossings of the limb-truncation Doppler component is taken as the limb-truncation period. Zero-crossing points where the signal amplitude changes from negative to positive in the limb-truncation Doppler component are detected, and the time corresponding to each positive zero-crossing is recorded. The time difference between two consecutive positive zero-crossings is taken as one limb-truncation period, and this process is repeated to obtain a sequence of limb-truncation periods.

[0029] Frequency components below a second frequency threshold are isolated from the Doppler shift sequence as the trunk motion Doppler components. The second frequency threshold is determined based on the typical speed range of overall human trunk movement, with the Doppler shift caused by trunk movement concentrated in the low-frequency band. The second frequency threshold is set to 1 Hz. The Doppler shift sequence is input into a digital low-pass filter with a cutoff frequency of 1 Hz. The digital low-pass filter adopts an infinite impulse response filter structure with a filter order of 4, and the filter coefficients are obtained using the Butterworth design method. The digital low-pass filter processes the Doppler shift sequence, and the output sequence is the trunk motion Doppler component.

[0030] After phase unwrapping of the torso motion Doppler components, the torso displacement vector is restored. The entanglement phase is obtained by accumulating over time. : in, Indicates the first The entanglement phase at each moment, Indicates the first Doppler component values ​​of trunk motion at each moment, Indicates the sampling time interval. Indicates the initial phase. For Perform unwrapping operations to eliminate discontinuities caused by periodic phase jumps: sequentially compare the wrapping phase differences at adjacent time points. When the absolute value of the difference is greater than At that time, from All subsequent entanglement phase values ​​from the start of the time step are uniformly subtracted or added. The integer multiples of the term ensure a smooth and continuous transition between adjacent phase differences. After unwrapping, an unwrapped phase sequence is obtained. .

[0031] The torso displacement vector is restored from the unwrapped phase sequence. Based on the phase-displacement conversion relationship in the radar Doppler effect, the displacement value is... The formula for calculation is: in, Indicates the first The trunk displacement value at each moment. This indicates the speed at which electromagnetic waves propagate through the air. This indicates the center frequency of the carrier wave in the ultra-wideband radar transmitted signal. This represents the phase value after untangling. Arranged in chronological order, a sequence of trunk displacement vectors is obtained.

[0032] See Figure 4 In the figure, the horizontal axis represents time in seconds, ranging from 0 to 8 seconds; the vertical axis represents Doppler frequency shift in Hertz, ranging from approximately -1 to 1 Hertz. The figure shows three curves: the Doppler frequency shift sequence (solid line, blue), the limb swing Doppler component (dashed line, orange), and the trunk motion Doppler component (dotted line, green).

[0033] The Doppler frequency shift sequence is the raw Doppler frequency shift data obtained after phase difference operation on continuous motion echo signals. The curve shows relatively frequent fluctuations with significant amplitude changes, and the frequency components cover both high and low frequency ranges. The limb swing Doppler component is obtained by high-pass filtering the Doppler frequency shift sequence with a cutoff frequency of 3 Hz. This curve mainly reflects the higher frequency components generated by limb swing, showing multiple periodic peaks and troughs with short periods and relatively regular waveforms. The trunk motion Doppler component is obtained by processing the raw Doppler frequency shift sequence with a low-pass filter with a cutoff frequency of 1 Hz. It shows a relatively gentle trend with a longer waveform period, reflecting the low-frequency characteristics generated by the overall trunk motion.

[0034] As can be seen from the figure, there are significant differences in the amplitude and fluctuation rate of the limb swing Doppler component and the trunk motion Doppler component at different time points, and their frequency characteristics are clearly separated. The frequent occurrence of peaks and troughs in the limb swing Doppler component indicates the periodic movement of the limbs; the trunk motion Doppler component changes slowly, consistent with the slow movement characteristics of trunk displacement. The overall curve exhibits alternating fluctuations over time, reflecting the dynamic behavior of the limb and trunk movements of the subject during the monitoring period.

[0035] In practical implementation, after acquiring the plantar pressure distribution signal, the piezoresistive sensor array is divided into left foot and right foot regions. The piezoresistive sensor array consists of several piezoresistive sensitive elements arranged in rows and columns. Each piezoresistive sensitive element has unique position coordinates, represented as... ,in Indicates row index, , This represents the total number of rows. Indicates column index, , This represents the total number of columns. The left and right foot regions are divided as follows: using the vertical line from the geometric center of the piezoresistive sensor array as the boundary, the column indices satisfy... All piezoresistive sensitive cells are assigned to the left foot region, and the column index satisfies All piezoresistive sensitive units are assigned to the right foot region, among which This indicates the floor function.

[0036] At each sampling time, the plantar pressure distribution signal is acquired. The plantar pressure distribution signal is a pressure value matrix, and the matrix is ​​located at the first... Line 1 The elements of the column represent coordinates. The pressure value measured by the piezoresistive sensing unit is denoted as ,in Indicates the first Each sampling time point. The method for calculating the total pressure in the left foot region is as follows: traverse the coordinates corresponding to all piezoresistive sensitive units within the left foot region, and calculate the pressure value at each coordinate. By summing the results, we obtain the total pressure in the left foot area. The method for calculating the total pressure in the right foot region is as follows: traverse the coordinates corresponding to all piezoresistive sensitive units within the right foot region, and calculate the pressure value at each coordinate. By summing the results, we obtain the total pressure in the right foot area. .

[0037] The fourth threshold is set by obtaining the weight value of the subject being tested. Weight value In kilograms, the acceleration due to gravity is taken as... Calculate the weight of the object being measured in meters per square second. Set the fourth threshold to gravity. With a proportionality coefficient The product of the products, the proportionality coefficient The value is 0.25. (Proportionality coefficient) The value of 0.25 is based on the following: In a normal bipedal standing posture, the vertical ground reaction force borne by the sole of one foot is generally not less than 25% of the body weight. When the pressure on the sole of one foot is lower than this value, it indicates that the foot on that side is not providing effective support. The fourth threshold is used... This indicates the fourth threshold. The calculation formula is: in, This represents the fourth threshold. This represents the proportionality coefficient. This indicates the weight value of the person being measured. It represents the acceleration due to gravity.

[0038] At each sampling time The total pressure in the left foot area With the fourth threshold Compare and sum the pressure in the right foot area. With the fourth threshold Comparison. When and When established simultaneously, the sampling time will be... Mark the time as a bipedal support moment; otherwise, do not mark it. Iterate through all sampling moments to obtain a bipedal support marker sequence that corresponds one-to-one with each sampling moment. Each element in the marker sequence takes the value of either a bipedal support identifier or a non-bipedal support identifier.

[0039] Extract sampling intervals consecutively marked with two-foot support moments from the two-foot support marker sequence. Scan the two-foot support marker sequence; when the first two-foot support marker is encountered, record the start time of the sampling interval as [time value missing]. Continue scanning until a non-bipedal support marker appears in the marker sequence. The bipedal support moment immediately preceding the non-bipedal support marker is taken as the end time of the sampling interval. This yields a bipedal support phase, the time period of which is denoted as: Repeat the above scanning process until the complete bipedal support marker sequence has been traversed, extracting all consecutive bipedal support phase time periods. Arrange all bipedal support phase time periods in chronological order to obtain a bipedal support phase time series. Each element in the bipedal support phase time series is a time period with a start time and an end time.

[0040] See Figure 5 In the graph, the horizontal axis represents time in seconds, and the vertical axis represents pressure in Newtons (N). The curve contains three lines: the total pressure in the left foot region (blue solid line), the total pressure in the right foot region (orange dashed line), and the fourth threshold (red dashed line).

[0041] As shown in the figure, the total pressure in the left and right foot regions alternates, fluctuating between 0 and 450 Newtons. The alternating peak pressures of the left and right feet indicate the alternating weight-bearing characteristics of the subject during normal walking. Both pressure curves are above the fourth threshold (approximately 170 Newtons) at most points in time, but there are moments during the alternation that fall below the fourth threshold, reflecting changes in unilateral foot support pressure. The pressure curves in the figure cross this threshold multiple times, indicating a clear alternating support state between the two feet, and the time period of the two-foot support phase can be extracted by the moments when the pressure is simultaneously above this threshold. The overall pressure waveform shows a clear periodicity, with a period of approximately 1 second, consistent with the periodic pressure distribution pattern of normal human gait.

[0042] In practical implementation, the sampling time of the plantar pressure distribution signal is used as the reference time axis. The sampling frequency of the piezoresistive sensor array determines the interval between adjacent sampling times on the reference time axis, denoted as . Each value in the limb swing period sequence corresponds to an acquisition time, and each value in the trunk displacement vector sequence corresponds to an acquisition time. The acquisition times of the limb swing period sequence and the trunk displacement vector sequence are determined by the signal processing time of the ultra-wideband radar. Each limb swing period value in the limb swing period sequence is mapped to the corresponding time on the reference time axis through linear interpolation according to its acquisition time, generating a limb swing period value corresponding to each sampling time on the reference time axis. Similarly, each trunk displacement vector value in the trunk displacement vector sequence is mapped to the corresponding time on the reference time axis through linear interpolation according to its acquisition time, generating a trunk displacement vector value corresponding to each sampling time on the reference time axis. Each bipedal support phase time segment in the bipedal support phase time sequence is converted into a bipedal support phase marker sequence on the reference time axis. The marker sequence assigns a first identifier to each reference time axis sampling point covered by the bipedal support phase time segment and a second identifier to the remaining sampling points.

[0043] Set a sliding window of fixed duration on the baseline timeline; the window duration is denoted as [missing information]. The window sliding step size is denoted as The window starts at time zero on the reference timeline and moves... Duration, excerpt The time interval is defined as follows: The interpolated limb swing period values ​​corresponding to all sampling points on the reference time axis within each time interval are arranged chronologically to form a limb swing period sequence; the interpolated trunk displacement vector values ​​are arranged chronologically to form a trunk displacement vector sequence; and the marker values ​​corresponding to the bipedal support phase marker sequence are arranged chronologically to form a bipedal support phase marker sequence. The limb swing period sequence, trunk displacement vector sequence, and bipedal support phase marker sequence within a time interval are combined into a single behavior segment. Multiple consecutive behavior segments are formed by continuously sliding a sliding window.

[0044] For each behavioral segment, the limb swing phase and trunk displacement vector phase are extracted at each sampling time within the segment. The limb swing phase is obtained as follows: at the beginning of the behavioral segment, the initial value of the limb swing phase is set to zero. Starting from the second sampling time of the behavioral segment, based on the limb swing phase value of the previous sampling time and the limb swing period value of the current sampling time, the limb swing phase value of the current sampling time is calculated using a phase accumulation method. The recursive relationship used in the calculation is that the limb swing phase of the current sampling time equals the limb swing phase of the previous sampling time plus... The product of the time step and the current sampling time is divided by the limb swing period value at the current sampling time. The time step is the sampling interval of the reference time axis. The method for obtaining the trunk displacement vector phase is as follows: Perform a Hilbert transform on the trunk displacement vector sequence within the behavioral segment to obtain an analytic signal. Extract the phase value of the analytic signal at each sampling moment as the trunk displacement vector phase. Take the difference between the limb swing phase and the trunk displacement vector phase at each sampling moment within the behavioral segment to obtain the instantaneous phase difference, calculated as follows: in, Indicates the first action segment within the action segment The instantaneous phase difference at each sampling moment, This represents the sequence number of the sampling time within the action segment. Indicates the first The phase value of limb swing at each sampling time. Indicates the first The phase values ​​of the torso displacement vector at each sampling time are obtained. The instantaneous phase differences at all sampling times within the behavior segment are arranged in chronological order to obtain the phase difference sequence.

[0045] Calculate the absolute value of the difference between two adjacent instantaneous phase differences in the phase difference sequence to obtain the absolute value difference sequence. The nth... The calculation method for the nth element is to take the nth element in the phase difference sequence. The instantaneous phase difference minus the first The absolute value of the difference obtained from each instantaneous phase difference, where From 1 to , The length of the phase difference sequence is given. A fifth threshold is defined as follows: Motion data from multiple healthy elderly individuals walking normally without falling are collected. The limb swing period and trunk displacement vector are extracted using the same method described above, and behavioral segments are constructed. The absolute value of the phase difference sequence difference within each behavioral segment is calculated. The distribution of the absolute values ​​of the differences across all normal walking behavioral segments is statistically analyzed, and the 95th percentile of this distribution is taken as the fifth threshold. The fifth threshold is denoted as […]. Scan the sequence of absolute differences; when three or more consecutive absolute differences are greater than the fifth threshold, the search continues. In cases where the gradient exceeds the first threshold, the corresponding phase difference sequence interval is marked as the interval where the gradient change exceeds the first threshold. A gradient change exceeding the first threshold indicates that the instantaneous phase difference fluctuates drastically within that interval. The first threshold is used to determine whether the gradient change of the phase difference sequence within the sliding window is in an abnormal state, and is characterized by a joint judgment rule for multiple consecutive sampling points in the absolute difference value sequence. Specifically, when three or more consecutive absolute difference values ​​in the absolute difference value sequence are greater than the fifth threshold... When the gradient of the phase difference sequence change within that interval exceeds the first threshold, the determination rule translates the physical meaning of the first threshold into an executable comparison operation, ensuring that the first threshold and the fifth threshold are compared. A clear equivalence relationship is formed between them. Fifth threshold The determination method involves collecting motion data from multiple healthy elderly individuals during normal walking without falls. Limb swing cycles and trunk displacement vectors are extracted using the same method to construct behavioral segments. The absolute value of the phase difference sequence within each behavioral segment is calculated. The distribution of the absolute values ​​of these differences across all normal walking behavioral segments is statistically analyzed, and the 95th percentile of this distribution is used as the fifth threshold. The first threshold is determined by the fifth threshold. It adapts to changes in the fifth threshold. The data, obtained from statistical analysis of normal walking data of healthy elderly individuals, reflects the typical distribution range of the absolute value of the phase difference sequence difference under normal gait. Therefore, the abnormal gradient judgment criterion represented by the first threshold is directly related to the normal gait characteristics of the monitored subject. When the fluctuation amplitude of the normal gait phase difference of the monitored subject is large, the fifth threshold... The threshold increases accordingly, and the criteria for determining the first threshold are relaxed accordingly; when the fluctuation amplitude of the normal gait phase difference of the monitored object is small, the fifth threshold... As the threshold decreases accordingly, the criteria for determining the first threshold tighten. This adaptive characteristic allows the first threshold to adapt to the gait differences of different individuals, avoiding potential misjudgments that might occur when a fixed threshold is used for different monitored subjects. Simultaneously, the requirement that the absolute values ​​of three or more consecutive differences are greater than the fifth threshold effectively filters out single-point phase jumps caused by instantaneous noise or accidental fluctuations, ensuring that detected phase desynchronization is caused by persistent deterioration in gait coordination, thus reducing the probability of false alarms.

[0046] Record the total number of sampling moments covered by all intervals within the behavior segment marked as having a gradient exceeding the first threshold, and multiply the total number of sampling moments by the sampling interval of the reference time axis. This gives the total duration covered by the marked interval. Divide the total duration by the total duration of the action segment. The proportion of the total duration of a fall to the total duration of a behavior segment is calculated. A sixth threshold of 0.3 is set. This threshold is based on statistical analysis of measured data from fall occurrence and walking instability processes. Desynchronization typically accounts for a significant proportion of the time in a behavior segment; 0.3 indicates that the proportion of desynchronization time must exceed 30% to qualify as a valid candidate for desynchronization. When the calculated proportion exceeds the sixth threshold of 0.3, the current behavior segment is marked as a candidate desynchronization segment. Continuously output behavior segments are judged sequentially. When two consecutive sliding windows output behavior segments are both marked as candidate desynchronization segments, the subsequent behavior segment is officially marked as a desynchronization segment.

[0047] See Figure 6 In the graph, the horizontal axis represents time in seconds, and the vertical axis represents the absolute value of the phase difference between two adjacent instantaneous phase differences in the phase difference sequence, in radians. The solid curve represents the change of the absolute value of the difference sequence over time, and the dashed line represents the fifth threshold. Its value is approximately 0.15 radians. The red shaded area in the figure indicates the interval in the phase difference sequence where the gradient change exceeds the first threshold.

[0048] As observed in the figure, within the time interval of approximately 4.5 to 5.8 seconds, multiple consecutive sampling points of the absolute difference sequence are above the fifth threshold of 0.15 radians, and this interval is covered by red shading, indicating that the phase difference sequence change gradient exceeds the first threshold at this moment. Within the red shading area, the absolute difference fluctuates drastically, with the peak repeatedly approaching and reaching 0.30 radians, significantly higher than the normal fluctuation level. Outside this time period, the absolute difference curve remains below the fifth threshold for most of the time, with small fluctuation amplitude, exhibiting a normal state. When three or more consecutive absolute differences are greater than the fifth threshold, the interval is marked as the interval where the change gradient exceeds the first threshold. Here, the red shading area represents the abnormal change gradient period determined by this judgment rule, indicating that the synchronization of the limb swing phase and the trunk displacement vector phase fluctuates significantly during this period, possibly corresponding to a desynchronization event in the behavioral segment.

[0049] In practice, a fixed time interval is set as the unit of time for statistics, and the unit of time is denoted as... , The value is 60 seconds. Starting from the start of the real-time monitoring process, the timeline is arranged according to... The duration is divided into continuous statistical periods. Within each statistical period, all behavioral segments appearing within that period are counted through iteration, and the segments marked as desynchronization segments are selected. For each segment marked as desynchronization, the start and end times are recorded, and the duration is calculated as the end time minus the start time. The total duration of all desynchronization segments within a statistical period is obtained by summing the durations of all segments marked as desynchronization. .

[0050] The number of desynchronized segments per unit time is recorded as follows: , This equals the number of behavioral segments marked as out of synchronization within the statistical period. The method for counting out-of-synchronization segments per unit time is as follows: at the end of each statistical period, obtain the number of behavioral segments marked as out of synchronization within that period. .

[0051] The sum of the durations of all desynchronized segments within the statistical period. ,Will Divide by unit time length The duration percentage is obtained, and the duration percentage is denoted as... Duration percentage The calculation formula is: in, Indicates the percentage of duration. This represents the sum of the durations of all behavior segments marked as out of synchronization within a unit of time, expressed in seconds. This represents the duration of a statistical unit of time, with a value of 60 seconds.

[0052] A second and a third threshold are set. The second threshold is used to determine whether the number of out-of-synchronization segments is abnormal, and its value is 5. The basis for setting the second threshold to 5 is: by collecting movement data from multiple elderly people in their daily living environment, the distribution of the number of behavioral segments marked as out-of-synchronization segments within 60 seconds under normal activity conditions is statistically analyzed. Under normal circumstances, this number usually does not exceed 3. Setting the second threshold to 5 can effectively distinguish frequent out-of-synchronization situations in a short period of time while maintaining tolerance for slow behavioral changes. The third threshold is used to determine whether the duration ratio of out-of-synchronization segments is abnormal, and its value is 0.4. The basis for setting the third threshold to 0.4 is: by analyzing datasets containing fall events, the time ratio of out-of-synchronization segments within 60 seconds before a fall is concentrated between 0.5 and 0.9. Under normal activity conditions, this ratio is usually below 0.2. Setting the third threshold to 0.4 can balance false negatives and false negatives.

[0053] At the end of each statistical period, the number of desynchronized segments will be... Compare with the second threshold of 5 to determine Is it greater than 5? Also, what is the percentage of duration? Compare with the third threshold of 0.4 to determine Is it greater than 0.4? Greater than 5 and When both conditions are met (greater than 0.4), a warning trigger flag is generated. This flag is a binary state variable and is set to an active state. When either condition is not met, no warning trigger flag is generated, and the binary state variable remains in an inactive state.

[0054] The system reads the warning trigger flag and, upon detecting that the flag is valid, executes the fall warning signal output process. The processor calls the communication interface of the wireless communication module, a radio frequency transceiver component with network access capabilities, whose communication interface follows a pre-defined communication protocol. A fall warning message is constructed, containing a target area identifier and a timestamp field. The target area identifier, a unique coded string, is pre-stored in memory to distinguish different monitoring areas. The timestamp field displays the system time corresponding to the generation of the warning trigger flag, provided by a real-time clock. The fall warning message is then transmitted wirelessly via the wireless communication module to a pre-bound mobile terminal, whose network address is pre-stored in memory.

[0055] Simultaneously, the processor controls the local audible and visual alarm to activate. The processor writes a high-level signal to the digital output port connected to the audible and visual alarm, driving the alarm's LEDs to flash at a fixed frequency and activating the alarm's buzzer to sound an alarm. The audible and visual alarm continues to operate until the processor receives a reset command or automatically stops after a preset alarm duration.

[0056] A behavior-aware-based home-based safety monitoring system for the elderly includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement all the steps of the aforementioned behavior-aware-based home-based safety monitoring method for the elderly. The memory is connected to the processor via a data bus, and the processor reads instructions from the memory and executes them. During the execution of the computer program, the processor acquires signals collected by ultra-wideband radar and piezoresistive sensor arrays through a data interface, and completes the extraction of limb swing cycles, trunk displacement vector extraction, bipedal support phase time series extraction, behavior segment construction, phase difference sequence calculation, desynchronization segment marking, unit time statistics, threshold comparison, and early warning output.

[0057] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for home-based safety monitoring of elderly people based on behavior perception, characterized in that, include: Collect continuous motion echo signals and plantar pressure distribution signals of the object under test within the target area; Based on the Doppler frequency shift sequence of the continuous motion echo signal in the time domain, the limb swing period and trunk displacement vector are extracted, and the time sequence of the bipedal support phase is extracted based on the plantar pressure distribution signal. The limb swing cycle, the trunk displacement vector, and the bipedal support phase time sequence are aligned according to the acquisition time sequence to form multiple behavioral segments within a continuous sliding window. For each behavioral segment, calculate the phase difference sequence between the limb swing period and the trunk displacement vector, and determine whether the gradient of the phase difference sequence changes within the sliding window exceeds a first threshold. If it does, mark the corresponding behavioral segment as a desynchronized segment. The number and duration percentage of the desynchronization segments are counted per unit time. When the number exceeds a second threshold and the duration percentage exceeds a third threshold, a fall warning signal is output.

2. The method for home-based safety monitoring of elderly people based on behavior perception according to claim 1, characterized in that, The acquisition of continuous motion echo signals and plantar pressure distribution signals of the object under test within the target area is specifically as follows: The ultra-wideband radar transmits linear frequency modulated continuous waves at a fixed repetition frequency, receives the echo signals reflected from the target area, and extracts the phase change sequence on each range gate after mixing and Fourier transform as the continuous motion echo signal. Simultaneously, a piezoresistive sensor array laid on the surface of the target area collects the resistance change value at the contact position between the sole of the foot and the ground at a fixed sampling frequency, and maps the resistance change value into a sole pressure distribution signal.

3. A method for home-based safety monitoring of elderly people based on behavior perception according to claim 2, characterized in that, The step of extracting the limb swing period and trunk displacement vector based on the Doppler frequency shift sequence of the continuous motion echo signal in the time domain is specifically as follows: Frequency components higher than the first frequency threshold are separated from the Doppler frequency shift sequence as limb swing Doppler components, and the interval between two consecutive zero crossings of the limb swing Doppler components is taken as the limb swing period. Frequency components below the second frequency threshold are separated from the Doppler frequency shift sequence as trunk motion Doppler components, and the trunk motion Doppler components are phase unwrapped and restored to trunk displacement vector.

4. A method for home-based safety monitoring of elderly people based on behavior perception according to claim 3, characterized in that, The extraction of the bipedal support phase time series based on the plantar pressure distribution signal specifically involves: The piezoresistive sensor array is divided into left foot region and right foot region, and the total pressure in the left foot region and the total pressure in the right foot region are calculated respectively. When the total pressure in the left foot region is greater than the fourth threshold and the total pressure in the right foot region is greater than the fourth threshold, the current sampling time is marked as the time of bipedal support. All sampling intervals continuously marked as bipedal support moments are arranged in chronological order to form the bipedal support phase time series.

5. A method for home-based safety monitoring of elderly people based on behavior perception according to claim 4, characterized in that, The process of aligning the limb swing period, the trunk displacement vector, and the bipedal support phase time series according to the acquisition sequence to form multiple behavioral segments within a continuous sliding window is specifically as follows: Using the sampling time of the plantar pressure distribution signal as the reference time axis, the limb swing period and the trunk displacement vector are interpolated to the corresponding time on the reference time axis according to their respective signal acquisition times; A fixed-duration window is slid along the reference time axis with a fixed step size, and the aligned limb swing cycle sequence, trunk displacement vector sequence, and bipedal support phase marker sequence within the window are encapsulated into a behavior segment.

6. A method for home-based safety monitoring of elderly people based on behavior perception according to claim 5, characterized in that, For each behavioral segment, the phase difference sequence between the limb swing period and the trunk displacement vector is calculated, and it is determined whether the gradient of the phase difference sequence within the sliding window exceeds a first threshold. Specifically: For each moment within a behavioral segment, the difference between the phase value of the limb swing period and the phase value of the trunk displacement vector at that moment is taken as the instantaneous phase difference; The instantaneous phase differences at all moments within the action segment are arranged in chronological order to form a phase difference sequence. The absolute value of the difference between two adjacent instantaneous phase differences in the phase difference sequence is calculated to obtain a sequence of absolute difference values. In the sequence of absolute difference values, if three or more consecutive absolute difference values ​​are greater than the fifth threshold, the intervals in which the change gradient exceeds the first threshold are identified.

7. A method for home-based safety monitoring of elderly people based on behavior perception according to claim 6, characterized in that, If the condition is exceeded, the corresponding behavior segment will be marked as a desynchronized segment, specifically: Record the total duration covered by all intervals within the behavior segment that are marked as having a gradient change exceeding the first threshold; When the proportion of the total duration to the total duration of the behavior segment exceeds the sixth threshold, the behavior segment is marked as a candidate desynchronization segment; If two consecutive sliding windows output behavior segments that are both candidate desynchronization segments, the latter behavior segment will be ultimately marked as the desynchronization segment.

8. A method for home-based safety monitoring of elderly people based on behavior perception according to claim 1, characterized in that, The number and duration percentage of the desynchronized segments within the statistical unit time are as follows: The number of all behavioral segments marked as out-of-synchronization segments within a unit of time is taken as the number of out-of-synchronization segments; The sum of the durations of each desynchronized segment within a unit time is accumulated, and the sum of the durations is divided by the unit time length to obtain the duration percentage.

9. A method for home-based safety monitoring of elderly people based on behavior perception according to claim 8, characterized in that, When the number exceeds the second threshold and the duration percentage exceeds the third threshold, a fall warning signal is output, specifically: Determine whether the number of desynchronized segments per unit time is greater than a second threshold, and determine whether the duration ratio is greater than a third threshold. When both conditions are met, a warning trigger flag is generated. After reading the warning trigger flag, the system sends a fall warning message containing the target area identifier and timestamp to the bound mobile terminal via the wireless communication module, and controls the local audible and visual alarm to sound.

10. A home-based elderly safety monitoring system based on behavior perception, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the behavior-aware home-based elderly safety monitoring method according to any one of claims 1 to 9.