Fall warning method for old people and intelligent ground mat

Abstract

The invention is suitable for the field of ground mats, and discloses an old man falling alarm method and an intelligent ground mat, and the method comprises the steps: collecting the surface deformation data of the intelligent ground mat in real time, generating three-dimensional pressure distribution time sequence data, and generating a pressure deformation acceleration, a pressure distribution dispersion and a deformation region; when the pressure deformation acceleration, the pressure distribution dispersion and the deformation area respectively meet a first condition, a second condition and a third condition, judging that the user has a falling risk; inputting the three-dimensional pressure distribution time sequence data into a pre-trained LSTM neural network when it is judged that the falling risk exists, and determining that the user falls if it is judged that no periodic pressure fluctuation exists in three continuous time windows and the displacement variance of the pressure center point is lower than an autonomous movement threshold value; when it is confirmed that the user falls down, whether pressure distribution reconstruction conforming to the standing characteristic is detected within the preset time or not is judged, if not, a sound-light alarm device is started, an emergency state code is sent to the monitoring terminal, and the method can improve the detection precision and reduce the misjudgment rate.

Description

Technical Field

[0001] The present invention relates to the field of floor mats, and in particular to an elderly fall alarm method and an intelligent floor mat. Background Art

[0002] As the aging of society accelerates, the safety and care of elderly people living alone has become a major social issue. Statistics show that over 30% of people aged 65 and over fall each year, with 50% unable to recover on their own. Delayed medical attention can lead to serious consequences.

[0003] Traditional pressure mat monitoring systems rely solely on a single pressure threshold, unable to distinguish between different behaviors like falls and sitting down, resulting in a false alarm rate exceeding 40%. The rigid sensor arrays employed by these systems struggle to accurately capture changes in human posture and lack the ability to detect persistent falls.

[0004] Existing smart floor mat technology mostly uses static judgment logic, which only monitors instantaneous pressure exceeding the limit. Its fixed threshold setting cannot adapt to users of different weights and changes in ambient temperature, resulting in a 35% decrease in sensitivity when the material hardens in winter.

[0005] Therefore, the existing technology still needs to be improved and developed. Summary of the Invention

[0006] The first purpose of the present invention is to provide a method for warning elderly people when they fall, which aims to solve the technical problem that traditional pressure pad monitoring systems only use a single pressure threshold to judge whether an elderly person has fallen, which easily leads to false alarms.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for warning elderly people of falling, comprising: collecting surface deformation data of a smart floor mat in real time, and generating three-dimensional pressure distribution time series data including a timestamp, a deformation position, and a pressure value based on the collected surface deformation data; performing noise filtering on the three-dimensional pressure distribution time series data, and using a sliding window algorithm to extract the pressure deformation acceleration, the pressure distribution dispersion, and the deformation area in the processed three-dimensional pressure distribution time series data; when the pressure deformation acceleration, the pressure distribution dispersion, and the deformation area respectively meet the first condition, the second condition, and the third condition, it is determined that the user is at risk of falling, the first condition being that the pressure deformation acceleration exceeds the fall acceleration threshold, and the second condition being that the pressure distribution dispersion is lower than The third condition of the human standing feature threshold is that the area of ​​the deformation region meets the threshold of the human lying area; when it is determined that there is a risk of falling, the three-dimensional pressure distribution time series data is input into the pre-trained LSTM neural network, and according to the output results of the LSTM neural network, it is judged whether there are periodic pressure fluctuations in three consecutive time windows and whether the pressure center point displacement variance is lower than the autonomous movement threshold; if it is determined that there are no periodic pressure fluctuations in three consecutive time windows and the pressure center point displacement variance is lower than the autonomous movement threshold, it is confirmed that the user has fallen; when it is confirmed that the user has fallen, it is determined whether the pressure distribution reconstruction that meets the standing characteristics is detected within the preset time. If not, the sound and light alarm device is activated, and the emergency status code is sent to the monitoring terminal.

[0009] Preferably, the surface deformation data of the smart floor mat is collected in real time by a distributed piezoelectric sensor array arranged in the smart floor mat.

[0010] Preferably, the three-dimensional pressure distribution time series data is subjected to noise filtering processing, including: filtering out the noise data of the three-dimensional pressure distribution time series data, and dividing the filtered noise data into high-frequency noise, low-frequency noise and burst noise; using a Butterworth low-pass filter to eliminate high-frequency noise; using a moving average filtering algorithm to suppress low-frequency noise; and using a median filtering algorithm to eliminate burst noise.

[0011] Preferably, a sliding window algorithm is used to extract the pressure deformation acceleration, pressure distribution discreteness and deformation area in the processed three-dimensional pressure distribution time series data, including: dividing the three-dimensional pressure distribution time series data into data frames according to a preset window length, each frame of data includes a spatial dimension and a time dimension, the spatial dimension is the pressure value of each piezoelectric sensor in the distributed piezoelectric sensor array, and the time dimension is the preset window length; calculating the acceleration by the displacement difference of the pressure center points of adjacent windows; calculating the pressure mean of each piezoelectric sensor in each frame of data, and calculating the discreteness in the form of standard deviation based on the pressure mean of each piezoelectric sensor, and performing normalization processing to obtain the pressure distribution discreteness; calculating the area of ​​the continuous region with a pressure value greater than 50N as the deformation region through a connected domain analysis algorithm.

[0012] Preferably, when the pressure deformation acceleration, the pressure distribution discreteness and the deformation area meet the first condition, the second condition and the third condition respectively, the user is determined to be at risk of falling, which also includes: determining whether the user's weight change satisfies the weight change>5% or whether the ambient temperature change satisfies the ambient temperature change greater than 10 degrees Celsius; if so, adjusting the pressure deformation acceleration threshold, the human standing feature threshold and the human lying area threshold according to the fall acceleration threshold adjustment formula, the human standing feature threshold adjustment formula and the human lying area threshold adjustment formula, and the fall acceleration threshold adjustment formula, the human standing feature threshold adjustment formula and the human lying area threshold adjustment formula are respectively expressed as:

[0013]

[0014] D th =(0.3+0.002×(W-W0))×(1+α D (T-T0))

[0015] A th =(0.5+0.005×(W-W0))×(1+α A (T-T0))

[0016] Where W∈[40,120]kg, D th ∈[0.2,0.4],A th ∈[0.4,0.8]m 2 , a th represents the fall acceleration threshold, D th Indicates the threshold of human standing feature, A th represents the threshold of the lying area of ​​the human body, W represents the user's weight, W0 represents the reference weight, T represents the ambient temperature, T0 represents the reference temperature, α0 represents the reference acceleration, α a represents the correction coefficient of temperature on the fall acceleration threshold; α D Indicates the correction coefficient of temperature on the threshold of human standing characteristics; α A Indicates the correction coefficient of temperature on the threshold of human lying area.

[0017] Preferably, after determining that the user is at risk of falling, the method further includes: collecting data from a three-dimensional acceleration sensor arranged in the floor mat, the data from the three-dimensional acceleration sensor including the vertical acceleration value and the horizontal displacement; if the vertical acceleration peak value is greater than 2g and the horizontal displacement is greater than 20cm, the correction determination result is that there is no risk of falling.

[0018] Preferably, the emergency status code includes a first field, a second field and a third field, the first field is the device ID and geographic location code, the second field is the fall severity classification, and the third field is the event timestamp.

[0019] Preferably, after starting the sound and light alarm device and sending the emergency status code to the monitoring terminal, if it is detected that the pressure distribution of the floor mat returns to the baseline form for 10 seconds and a confirmation instruction sent by the monitoring terminal is received at the same time, the alarm status is released and a post-analysis report containing the event timeline and key judgment parameters is generated.

[0020] The second object of the present invention is to provide a smart floor mat, characterized in that the smart floor mat is used to execute the elderly fall alarm method as described above, and the smart floor mat includes a floor mat body, a control module, a wireless communication module, a distributed piezoelectric sensor array, a three-dimensional acceleration sensor and a power supply module. The distributed piezoelectric sensor array includes multiple piezoelectric sensors, and the power supply module is respectively connected to the control module, the wireless communication module, the piezoelectric sensor and the three-dimensional acceleration sensor. The wireless communication module, the piezoelectric sensor and the three-dimensional acceleration sensor are respectively connected to the control module. The multiple piezoelectric sensors are arranged inside the floor mat body with a grid density of 10cm×10cm. Each piezoelectric sensor independently collects pressure data on the surface of the floor mat and transmits it to the control module; the wireless communication module is used to send an emergency status code to the monitoring terminal.

[0021] Preferably, a manual alarm button is provided on the edge of the floor mat body, a fluorescent mark is provided on the manual alarm button, and the manual alarm button is connected to the wireless communication module and the power supply module respectively.

[0022] In this solution, three-dimensional pressure distribution time series data is generated by real-time acquisition of surface deformation data of the smart floor mat. These data are then subjected to noise filtering and a sliding window algorithm is used to extract key features, such as pressure deformation acceleration, pressure distribution discreteness, and deformation area. This can effectively eliminate interference and obtain core information, providing a data basis for accurately judging fall risks. When the relevant features meet specific conditions, a fall risk is determined to exist. The data is then input into a pre-trained LSTM neural network for further in-depth analysis from two dimensions: periodic pressure fluctuations and pressure center point displacement variance, further improving the accuracy of fall judgment. Once a fall is confirmed, it is further determined whether there is a pressure distribution reconstruction that meets the standing characteristics within the preset time. If not, the sound and light alarm device is activated and an emergency status code is sent to the monitoring terminal to notify the guardian in time, providing reliable protection for the life safety of the elderly, greatly improving the safety monitoring capabilities of the elderly, and reducing the risk of serious consequences caused by failure to detect falls in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0024] Figure 1 This is a flow chart of the elderly fall alarm method provided by an embodiment of the present invention;

[0025] Figure 2 This is a structural block diagram of the smart floor mat provided by an embodiment of the present invention. Description of the drawings:

[0027] 10. Control module; 20. Wireless communication module; 30. Piezoelectric sensor; 40. Three-dimensional acceleration sensor; 50. Power module; 60. Manual alarm button. DETAILED DESCRIPTION

[0028] The terms "first," "second," "third," "fourth," and the like (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0029] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 In an embodiment of the present invention, the elderly fall alarm method includes:

[0030] S101, collecting surface deformation data of the smart floor mat in real time, and generating three-dimensional pressure distribution time series data including a timestamp, deformation position, and pressure value based on the collected surface deformation data;

[0031] S102, performing noise filtering on the three-dimensional pressure distribution time series data, and using a sliding window algorithm to extract the pressure deformation acceleration, pressure distribution dispersion, and deformation area from the processed three-dimensional pressure distribution time series data;

[0032] S103: When the pressure deformation acceleration, pressure distribution dispersion and deformation area meet the first, second and third conditions respectively, it is determined that the user is at risk of falling. The first condition is: the pressure deformation acceleration exceeds the fall acceleration threshold (5-8m / s 2 ), the second condition is that the pressure distribution dispersion is lower than the human standing characteristic threshold (<0.3), and the third condition is that the deformation area meets the human lying area threshold (0.4-0.8m 2 );

[0033] S104. When a fall risk is determined, the three-dimensional pressure distribution time series data is input into a pre-trained LSTM neural network. Based on the output of the LSTM neural network, it is determined whether there are periodic pressure fluctuations within three consecutive time windows and whether the pressure center displacement variance is lower than the autonomous movement threshold.

[0034] S105: If it is determined that there is no periodic pressure fluctuation (frequency 0.5-2 Hz) within three consecutive time windows and the pressure center displacement variance is lower than the autonomous movement threshold (<10 cm 2 ), it is confirmed that the user has fallen;

[0035] S106. When it is confirmed that the user has fallen, it is determined whether a pressure distribution reconstruction that meets the standing characteristics is detected within a preset time. If not, the sound and light alarm device is activated and an emergency status code is sent to the monitoring terminal.

[0036] In this embodiment, in step S101, the surface deformation data of the smart floor mat is collected in real time by a distributed piezoelectric sensor array provided in the smart floor mat. The distributed piezoelectric sensor array includes a plurality of piezoelectric sensors arranged inside the floor mat body at a grid density of 10 cm × 10 cm. Each piezoelectric sensor records the pressure value, position coordinates and timestamp to form three-dimensional pressure distribution time series data. The timestamp is used to mark the specific moment of data collection, the deformation position clearly defines the specific area where the pressure acts, and the pressure value intuitively reflects the magnitude of the force. For example, when the sampling rate is 20 Hz, a frame containing a snapshot of the spatial distribution of the pressure of all sensor nodes is generated every 50 ms.

[0037] In this embodiment, a flexible pressure-sensitive film can be used instead of a piezoelectric sensor. The response range of the flexible pressure-sensitive film or the piezoelectric sensor is 50N to 1000N.

[0038] In this embodiment, each frame of data is in a matrix structure of M×NM×N (M is the number of rows, N is the number of columns), and the time dimension is superimposed to form a time series cube.

[0039] In this embodiment, through the SPI bus or I 2 C protocol to achieve synchronous acquisition of multi-sensor data.

[0040] In this embodiment, in step S102, the three-dimensional pressure distribution time series data is subjected to noise filtering, specifically including: filtering out the noise data of the three-dimensional pressure distribution time series data, and dividing the filtered noise data into high-frequency noise, low-frequency noise and burst noise; using a Butterworth low-pass filter to eliminate the high-frequency noise; using a moving average filtering algorithm to suppress the low-frequency noise; and using a median filtering algorithm to eliminate the burst noise.

[0041] In this embodiment, a sliding window algorithm is used to extract the pressure deformation acceleration, pressure distribution discreteness and deformation area in the processed three-dimensional pressure distribution time series data, including: dividing the three-dimensional pressure distribution time series data into data frames according to a preset window length, each frame of data includes a spatial dimension and a time dimension, the spatial dimension is the pressure value of each piezoelectric sensor in the distributed piezoelectric sensor array, and the time dimension is the preset window length; calculating the acceleration by the displacement difference of the pressure center points of adjacent windows; calculating the pressure mean of each piezoelectric sensor in each frame of data, and calculating the discreteness in the form of standard deviation based on the pressure mean of each piezoelectric sensor, and performing normalization processing to obtain the pressure distribution discreteness; calculating the area of ​​the continuous region with pressure value >50N as the deformation region through the connected domain analysis algorithm.

[0042] In this embodiment, the acceleration is expressed as:

[0043]

[0044] Where ΔS represents the displacement of the pressure center points of adjacent windows, and Δt represents the window step time.

[0045] In this embodiment, in step S103, for example, the pressure deformation acceleration threshold, the human standing feature threshold, and the human lying area threshold are set according to actual conditions. The pressure deformation acceleration exceeds the fall acceleration threshold, which is generally 5-8 m / s. 2 The pressure distribution dispersion is lower than the human standing characteristic threshold, which is generally less than 0.3, and the deformation area meets the human lying area threshold, which is generally 0.4-0.8m 2 The first condition is used to exclude slow sitting (acceleration < 3m / s 2 ) and other non-falling actions.

[0046] In the second condition, when standing, the pressure is concentrated on the feet, and the dispersion is low; when lying down, the pressure is dispersed, and the dispersion is even lower.

[0047] The third condition is used to exclude interference such as pets passing by or heavy objects falling: Pets passing by (area < 0.2m 2 ) or heavy objects falling (sudden change in area but no continuity).

[0048] When the pressure-deformation acceleration exceeds the fall acceleration threshold, it indicates an abnormal rate of pressure change, possibly due to a sudden fall. The pressure distribution dispersion is lower than the human standing characteristic threshold, indicating that the pressure distribution at this point is different from that of a normal standing position and more closely resembles that of a fall. The deformation area meets the lying area threshold, further confirming the likelihood of a fall. When these three conditions are met simultaneously, a preliminary fall risk assessment is made.

[0049] In this embodiment, in order to improve the detection accuracy, the pressure deformation acceleration threshold, the human standing feature threshold, and the human lying area threshold can be dynamically adjusted according to the user's weight. Specifically, it is determined whether the user's weight change satisfies the weight change of more than 5% or whether the ambient temperature change satisfies the ambient temperature change of more than 10 degrees Celsius; if so, the pressure deformation acceleration threshold, the human standing feature threshold, and the human lying area threshold are adjusted according to the fall acceleration threshold adjustment formula, the human standing feature threshold adjustment formula, and the human lying area threshold adjustment formula. The fall acceleration threshold adjustment formula, the human standing feature threshold adjustment formula, and the human lying area threshold adjustment formula are respectively expressed as follows:

[0050]

[0051] D th =(0.3+0.002×(W-W0))×(1+α D (T-T0))

[0052] A th =(0.5+0.005×(W-W0))×(1+α A (T-T0))

[0053] Where W∈[40,120]kg, D th ∈[0.2,0.4],A th ∈[0.4,0.8]m 2 , a th represents the fall acceleration threshold, D th Indicates the threshold of human standing feature, A th represents the threshold of the lying area of ​​the human body, W represents the user's weight, W0 represents the reference weight, T represents the ambient temperature, T0 represents the reference temperature, α0 represents the reference acceleration, α a represents the correction coefficient of temperature on the fall acceleration threshold; α D Indicates the correction coefficient of temperature on the threshold of human standing characteristics; α A Indicates the correction coefficient of temperature on the threshold of human lying area.

[0054] In this embodiment, in step S104, the temporal pressure change (LSTM analysis) and the spatial displacement characteristics (variance calculation) are combined to comprehensively judge the user status to avoid misjudgment by a single sensor.

[0055] When a healthy adult is at rest, the chest cavity will produce periodic pressure fluctuations of 0.2-0.3Hz due to breathing; when trying to stand up, slight movements of the limbs may produce fluctuations of 0.5-2Hz.

[0056] Each input sequence covers 3 consecutive seconds of data (60 time steps at a 20Hz sampling rate). The LSTM neural network captures temporal features through the LSTM hidden layer states and uses FFT (Fast Fourier Transform) to extract the dominant frequency. If no fluctuations between 0.5 and 2 Hz are detected within three consecutive time windows (each 1 second), it is considered to be a sign of no autonomous activity.

[0057] In this embodiment, the pressure center displacement variance is expressed as:

[0058]

[0059] Where x i ,y i represents the coordinates of the pressure center at the i-th time step, Represents the mean value of the pressure center coordinates within the window.

[0060] For example, the LSTM neural network input is continuous 3 seconds of pressure data (no respiratory fluctuations, displacement variance = 5cm 2 ); the LSTM neural network output is fluctuation probability = 0.1, variance flag = 1 (below the threshold), and the judgment result is that the user has fallen.

[0061] In this embodiment, by fusing pressure distribution data with acceleration data, the problem of false alarms caused by non-fall events (such as heavy objects falling or pets colliding) in traditional single-sensor solutions is solved. After determining that the user is at risk of falling, the present embodiment also includes: collecting data from a three-dimensional acceleration sensor installed in the floor mat. The data of the three-dimensional acceleration sensor includes vertical acceleration values ​​and horizontal displacements. If the peak value of the vertical acceleration is greater than 2g and the horizontal displacement is greater than 20cm, the correction judgment result is that there is no risk of falling.

[0062] It should be noted that during a normal fall, the peak vertical acceleration is usually ≤1.5g; when the head lands, the peak can reach 1.8-2.0g, but the duration is extremely short (<200ms); when the trunk or limbs land, the peak is even lower (0.5-1.2g).

[0063] In non-fall scenarios, heavy objects fall, such as a falling vase (peak force can reach 3-5g); pets jump, small and medium-sized dogs jump (peak force 2-3g, but the contact area is small); violent stampede, an adult stomps quickly (peak force 1.5-2.5g).

[0064] In this embodiment, if the vertical acceleration peak value is greater than 2g, it is very likely to be a non-human motion and a false positive should be excluded.

[0065] During a fall, the body's center of gravity typically moves less than 15 cm (limited by the range of motion of the body's joints), and there is no sustained sliding of the trunk after it touches the ground (limited by friction). This can occur with heavy objects sliding, such as pushing or pulling a box (displacement >30 cm); or with pets running, such as cats and dogs, passing quickly (displacement >50 cm / s).

[0066] In this embodiment, a horizontal displacement greater than 20 cm indicates active movement or object sliding, which contradicts the static nature of the fall.

[0067] In this embodiment, in step S106, the average reaction time of a healthy adult from falling to trying to get up is 10-20 seconds. At the same time, in order to avoid premature alarm (the user may recover on his own) or delayed rescue (the user cannot move), the preset time is set to 30s.

[0068] In this embodiment, the emergency status code includes a first field, a second field, and a third field. The first field is the device ID and geographic location code, the second field is the fall severity classification, and the third field is the event timestamp.

[0069] In this embodiment, after the sound and light alarm device is activated and the emergency status code is sent to the monitoring terminal, if it is detected that the pressure distribution of the floor mat returns to the baseline form for 10 seconds and a confirmation instruction is received from the monitoring terminal at the same time, the alarm status is released and a post-analysis report including the event timeline and key judgment parameters is generated.

[0070] An example post-mortem report is as follows:

[0071]

[0072]

[0073] In this embodiment, three-dimensional pressure distribution time series data is generated by real-time acquisition of surface deformation data of the smart floor mat. This data is then subjected to noise filtering and a sliding window algorithm is used to extract key features, such as pressure deformation acceleration, pressure distribution dispersion, and deformation area. This effectively eliminates interference and obtains core information, providing a data basis for accurately determining fall risk. When the relevant features meet specific conditions, a fall risk is determined to exist. The data is then input into a pre-trained LSTM neural network for further in-depth analysis from two dimensions: periodic pressure fluctuations and pressure center point displacement variance, further improving the accuracy of fall judgment. Once a fall is confirmed, a further determination is made as to whether a pressure distribution reconstruction that meets the standing characteristics has occurred within a preset time. If not, an audible and visual alarm device is activated and an emergency status code is sent to the monitoring terminal to promptly notify the guardian. This provides reliable protection for the life safety of the elderly, greatly improves the ability to monitor the safety of the elderly, and reduces the risk of serious consequences caused by untimely detection of falls.

[0074] An embodiment of the present invention also provides a smart floor mat, which is used to execute the elderly fall alarm method described above. The smart floor mat includes a floor mat body, a control module 10, a wireless communication module 20, a distributed piezoelectric sensor 30 array, a three-dimensional acceleration sensor 40 and a power module 50. The distributed piezoelectric sensor 30 array includes multiple piezoelectric sensors 30. The power module 50 is respectively connected to the control module 10, the wireless communication module 20, the piezoelectric sensor 30 and the three-dimensional acceleration sensor 40. The wireless communication module 20, the piezoelectric sensor 30 and the three-dimensional acceleration sensor 40 are respectively connected to the control module 10. The multiple piezoelectric sensors 30 are arranged inside the floor mat body with a grid density of 10 cm×10 cm. Each piezoelectric sensor 30 independently collects pressure data on the floor mat surface and transmits it to the control module 10; the wireless communication module 20 is used to send an emergency status code to the monitoring terminal.

[0075] Furthermore, a manual alarm button 60 is provided on the edge of the floor mat body. A fluorescent mark is provided on the manual alarm button 60. The manual alarm button 60 is connected to the wireless communication module 20 and the power module 50 respectively.

[0076] In this embodiment, in actual application, the smart floor mat can be placed on the ground in front of the elderly's bed. When the elderly get up, the smart floor mat begins to collect surface deformation data in real time. These data will be quickly converted into three-dimensional pressure distribution time series data containing timestamps, deformation positions and pressure values, recording in detail every detail of the elderly getting up from the bed and touching the floor mat.

[0077] The smart floor mat performs noise filtering on the collected data to remove any possible interference and ensure data accuracy. Then, using a sliding window algorithm, it accurately extracts key features such as pressure deformation acceleration, pressure distribution dispersion, and deformation area. When these features meet specific conditions, such as if the pressure deformation acceleration exceeds the fall acceleration threshold, indicating that the elderly person may have stood up too hastily or unbalanced; if the pressure distribution dispersion is lower than the human standing characteristic threshold, indicating abnormal pressure distribution when the elderly person is standing; and if the deformation area meets the human lying area threshold, indicating that the elderly person is suspected of abnormal posture changes, the smart floor mat will determine that the elderly person is at risk of falling.

[0078] Once a fall risk is determined, the smart mat feeds the three-dimensional pressure distribution time series data into a pre-trained LSTM neural network for further analysis. This analysis further confirms the elderly person's condition by determining whether periodic pressure fluctuations occur within three consecutive time windows and whether the variance of the pressure center displacement is below the autonomous movement threshold. If there are no periodic pressure fluctuations within three consecutive time windows and the variance of the pressure center displacement is below the autonomous movement threshold, it is generally confirmed that the elderly person has fallen.

[0079] After confirming a fall, the smart mat continuously monitors for pressure distribution reconstruction consistent with standing within a preset timeframe. If no detection occurs within this timeframe, indicating the elderly person is likely unable to stand up on their own, the mat immediately activates its built-in sound and light alarm, emitting a striking flash and a loud siren to attract the attention of those nearby. Simultaneously, it quickly transmits an emergency status code to a pre-assigned monitoring terminal, such as a family member's mobile phone or a monitoring device at a hospital nurse's station. This way, even if an elderly person falls while using the restroom alone at night, they can be detected in time, buying valuable time for rescue and effectively protecting their life and health.

[0080] In this embodiment, three-dimensional pressure distribution time series data is generated by real-time acquisition of surface deformation data of the smart floor mat. This data is then subjected to noise filtering and a sliding window algorithm is used to extract key features, such as pressure deformation acceleration, pressure distribution dispersion, and deformation area. This effectively eliminates interference and obtains core information, providing a data basis for accurately determining fall risk. When the relevant features meet specific conditions, a fall risk is determined to exist. The data is then input into a pre-trained LSTM neural network for further in-depth analysis from two dimensions: periodic pressure fluctuations and pressure center point displacement variance, further improving the accuracy of fall judgment. Once a fall is confirmed, a further determination is made as to whether a pressure distribution reconstruction that meets the standing characteristics has occurred within a preset time. If not, an audible and visual alarm device is activated and an emergency status code is sent to the monitoring terminal to promptly notify the guardian. This provides reliable protection for the life safety of the elderly, greatly improves the ability to monitor the safety of the elderly, and reduces the risk of serious consequences caused by untimely detection of falls.

[0081] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made by using the contents of the present invention description and drawings under the inventive concept of the present invention, or direct / indirect application in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for warning elderly people when they fall, characterized in that: The elderly fall alarm method comprises: Collect the surface deformation data of the smart floor mat in real time, and generate three-dimensional pressure distribution time series data including timestamp, deformation position and pressure value based on the collected surface deformation data; The three-dimensional pressure distribution time series data is subjected to noise filtering, and the sliding window algorithm is used to extract the pressure deformation acceleration, pressure distribution dispersion and deformation area from the processed three-dimensional pressure distribution time series data; When the pressure deformation acceleration, pressure distribution dispersion, and deformation area meet the first, second, and third conditions respectively, the user is judged to be at risk of falling. The first condition is that the pressure deformation acceleration exceeds the fall acceleration threshold, the second condition is that the pressure distribution dispersion is lower than the human standing characteristic threshold, and the third condition is that the deformation area meets the human lying area threshold. When a fall risk is determined, the three-dimensional pressure distribution time series data is input into a pre-trained LSTM neural network. Based on the output of the LSTM neural network, it is determined whether there are periodic pressure fluctuations in three consecutive time windows and whether the displacement variance of the pressure center point is lower than the autonomous movement threshold. If it is determined that there is no periodic pressure fluctuation within three consecutive time windows and the pressure center displacement variance is lower than the autonomous movement threshold, the user is confirmed to have fallen; When it is confirmed that the user has fallen, it is determined whether the pressure distribution reconstruction that meets the standing characteristics is detected within the preset time. If not, the sound and light alarm device is activated and an emergency status code is sent to the monitoring terminal.

2. The elderly fall alarm method according to claim 1, characterized in that: The surface deformation data of the smart floor mat is collected in real time through a distributed piezoelectric sensor array arranged in the smart floor mat.

3. The elderly fall alarm method according to claim 1, characterized in that: Perform noise filtering on the three-dimensional pressure distribution time series data, including: Screening out noise data of the three-dimensional pressure distribution time series data, and dividing the screened noise data into high-frequency noise, low-frequency noise and burst noise; A Butterworth low-pass filter is used to eliminate high-frequency noise; The moving average filtering algorithm is used to suppress low-frequency noise; The median filter algorithm is used to eliminate burst noise.

4. The elderly fall alarm method according to claim 1, characterized in that: The sliding window algorithm is used to extract the pressure deformation acceleration, pressure distribution dispersion and deformation area from the processed three-dimensional pressure distribution time series data, including: The three-dimensional pressure distribution time series data is divided into data frames according to a preset window length. Each frame of data includes a spatial dimension and a temporal dimension. The spatial dimension is the pressure value of each piezoelectric sensor in the distributed piezoelectric sensor array, and the temporal dimension is the preset window length. The pressure deformation acceleration is obtained by calculating the displacement difference of the pressure center points of adjacent windows; Calculate the mean pressure value of each piezoelectric sensor in each frame of data, calculate the dispersion in the form of standard deviation based on the mean pressure value of each piezoelectric sensor, and perform normalization to obtain the pressure distribution dispersion; The area of ​​the continuous region with a pressure value greater than 50 N is calculated using the connected domain analysis algorithm, and the calculated area is used as the deformation area.

5. The elderly fall alarm method according to claim 1, characterized in that: When the pressure deformation acceleration, pressure distribution dispersion, and deformation area meet the first, second, and third conditions respectively, the user is judged to be at risk of falling, which also includes: Determine whether the user's weight change satisfies the requirement of weight change > 5% or whether the ambient temperature change satisfies the requirement of ambient temperature change greater than 10 degrees Celsius; If so, the pressure deformation acceleration threshold, the human standing feature threshold, and the human lying area threshold are adjusted according to the fall acceleration threshold adjustment formula, the human standing feature threshold adjustment formula, and the human lying area threshold adjustment formula. The fall acceleration threshold adjustment formula, the human standing feature threshold adjustment formula, and the human lying area threshold adjustment formula are respectively expressed as: D th =(0.3+0.002×(W-W0))×(1+α D ·(T-T0)) A th =(0.5+0.005×(W-W0))×(1+α A ·(T-T0)) Where W∈[40,120]kg, D th ∈[0.2,0.4],A th ∈[0.4,0.8]m 2 , a th represents the fall acceleration threshold, D th Indicates the threshold of human standing feature, A th represents the threshold of the lying area of ​​the human body, W represents the user's weight, W0 represents the reference weight, T represents the ambient temperature, T0 represents the reference temperature, α0 represents the reference acceleration, α a represents the correction coefficient of temperature on the fall acceleration threshold; α D Indicates the correction coefficient of temperature on the threshold of human standing characteristics; α A Indicates the correction coefficient of temperature on the threshold of human lying area.

6. The elderly fall alarm method according to claim 1, characterized in that: After determining that the user is at risk of falling, the process also includes: collecting data from a three-dimensional acceleration sensor installed in the floor mat. The data from the three-dimensional acceleration sensor includes vertical acceleration values ​​and horizontal displacements. If the vertical acceleration peak is greater than 2g and the horizontal displacement is greater than 20cm, the correction judgment result is that there is no risk of falling.

7. The elderly fall alarm method according to claim 1, characterized in that: The emergency status code includes a first field, a second field, and a third field. The first field is a device ID and a geographic location code, the second field is a fall severity rating, and the third field is an event timestamp.

8. The elderly fall alarm method according to claim 1, characterized in that: After activating the sound and light alarm device and sending the emergency status code to the monitoring terminal, if it is detected that the pressure distribution of the floor mat has returned to the baseline form for 10 seconds and a confirmation command is received from the monitoring terminal at the same time, the alarm status will be released and a post-analysis report containing the event timeline and key judgment parameters will be generated.

9. A smart floor mat, characterized in that: The smart floor mat is used to execute the elderly fall alarm method as described in any one of claims 1 to 8. The smart floor mat includes a floor mat body, a control module, a wireless communication module, a distributed piezoelectric sensor array, a three-dimensional acceleration sensor and a power module. The distributed piezoelectric sensor array includes multiple piezoelectric sensors. The power module is respectively connected to the control module, the wireless communication module, the piezoelectric sensor and the three-dimensional acceleration sensor. The wireless communication module, the piezoelectric sensor and the three-dimensional acceleration sensor are respectively connected to the control module. The multiple piezoelectric sensors are arranged inside the floor mat body with a grid density of 10 cm×10 cm. Each piezoelectric sensor independently collects pressure data on the surface of the floor mat and transmits it to the control module; the wireless communication module is used to send an emergency status code to the monitoring terminal.

10. The smart floor mat according to claim 9, characterized in that: A manual alarm button is provided at the edge of the floor mat body. A fluorescent mark is provided on the manual alarm button. The manual alarm button is connected to the wireless communication module and the power supply module respectively.

Images