A method and system for fall detection and alarm combining headphones and a watch.

Abstract

This invention provides a method and system for fall detection and alarm combining headphones and a watch. The method includes: a data acquisition step: collecting various sensor data from the watch and headphones; a data cleaning step: filtering, denoising, and reconstructing the collected sensor data; a model training step: training the model on normal data and extracting features from data under non-fall conditions; and an alarm step: providing secondary confirmation and voice call services. The beneficial effects of this invention are: compared with existing technologies, this invention combines head posture monitoring via headphones with watch data, using an abnormal or non-normal approach to determine falls. This results in less model training and more accurate judgment, which is of great significance for real-time fall warning and fall prevention research.

Description

Technical Field

[0001] This invention relates to the field of monitoring technology, and in particular to a method and system for fall detection and alarm combining headphones and a watch. Background Technology

[0002] Currently, my country has 267 million people aged 60 and above, accounting for 18.9% of the total population. Around 2035, this number is projected to exceed 400 million, representing over 30% of the total population, marking the beginning of a severely aging society. As the aging population intensifies, the health needs of middle-aged and elderly individuals are increasing. Prevention and early warning are just as important as treatment. Dangerous events are random and uncertain, and traditional methods of monitoring heart rate and blood pressure are overly cumbersome, especially for the elderly. Furthermore, injuries from falls increase with age, and complications from falls are a significant cause of death for people over 65. Providing immediate assistance after a fall can effectively reduce the risk of complications.

[0003] The shortcomings of existing technologies are: current watch fall detection is not comprehensive, its detection of human posture is inaccurate, and it cannot detect falls in special postures. Headphones are mainly used to monitor head movements and cannot fully reflect body posture. Neither watches nor headphones can accurately detect falls. Summary of the Invention

[0004] To address the problems in the prior art, this invention provides a method and system for fall detection and alarm combining headphones and a watch, enabling more accurate fall assessment.

[0005] This invention provides a method for fall detection and alarm combining headphones and a watch, comprising the following steps:

[0006] Data acquisition steps: Collect various sensor data from the watch and headphones;

[0007] Data cleaning steps: The collected sensor data is screened, filtered, denoised, and reconstructed.

[0008] Model training steps: Train on normal data and extract features from data under non-fall conditions;

[0009] Alarm procedure: Provide secondary confirmation and voice call service.

[0010] As a further improvement of the present invention, in the data cleaning step, during the screening, filtering, noise reduction and recombination process, the segmented binary data of the watch and earphone are converted into digital signals that are easy to process and fall into the [0~10] range. The recombination algorithm arranges multiple data segments into a two-dimensional array for noise reduction operation.

[0011] As a further improvement of the present invention, the noise reduction step in the data cleaning step includes the following steps:

[0012] Step a: Calculate the maximum and minimum values ​​of the overall data, unify the values ​​to return to the preset range, and calculate the ratio that needs to be reduced;

[0013] Step b: Loop through the items in the array, take the 8 values ​​around each item, and use the sum of squared deviations (SA) to calculate the degree of deviation of the surrounding points from the center point;

[0014] Step c: If the deviation is greater than the set value, reduce the value by the ratio and then return to step b; if the deviation is less than the set value, then proceed to step d.

[0015] Step d: After the overall data processing is completed, store this part of the data and pass it into the model training step.

[0016] As a further improvement of the present invention, in the model training step, the data entering the model is initially preset to be an abnormal situation. By inputting data with correct posture, the prediction vector is continuously optimized through a correlation algorithm. If the matching degree with the vector is high, it is determined to be a normal posture. The steps are as follows:

[0017] Step 1: Let i be the row number used for traversal, i = 0, X(i) = 1, Y(i) = 10;

[0018] Step 2: Fix i = 1, traverse all other rows, and let j be the row number used for traversal (j != i, j != i, j != i); Step 3: j = 2, X(j) = 2, Y(j) = 20;

[0019] Step 4: Let the function d(a,b)=|ab|. The d() function calculates the absolute value of the difference between two input values. If a and b are generalized to vectors, then it calculates the difference between two vectors.

[0020] Step 5: Calculate d(xi,xj) = |x1-x2| = |1-2| = 1, and set it as a new variable Rx, Rx = 1;

[0021] Step 6: Calculate d(yi,yj)=|y1-y2|, and set it as a new variable Ry, Ry=8; Step 7: Fix j, traverse all other rows; set k as the row number used for traversal (k!=i AND k!=j); k=3, X(k)=3, Y(k)=30;

[0022] Step 8: Calculate d(xi,xk) = |x1 - x3| = |1 - 3| = 2;

[0023] Step 9: Calculate d(yi,yk)=|y1-y3|=|12-30|=18;

[0024] Step 10: Set variables A11 = 0, A12 = 0, A21 = 0, A22 = 0;

[0025] Step 11: Compare the values ​​of d(xi,xk) and Rx, and compare the values ​​of d(yi,yk) and Ry. If d(xi,xk) <= Rx and d(yi,yk) <= Ry, then add 1 to A11, and A1 = A11 + A12.

[0026] Step 12: Increment k by 1, repeat steps 7 to 11, and accumulate the A series variables;

[0027] Step 13: Now that we have the contingency table for the cases i=1, j=2, we can use the chi-square test to calculate the result. The formula is as follows:

[0028]

[0029] Step 14: For the current case of i=1, j=2, we can calculate that S(i,j)=0;

[0030] Step 15: Add 1 to j, j = 3, X(j) = 3, Y(j) = 30;

[0031] Step 16: Repeat steps 5 to 16, calculating S(i,j) in each step and summing them up; after calculation, we get the total S(i,j) = 3 when i = 1, i = 2, X(i) = 2;

[0032] Step 17: Repeat steps 3 to 17 until all i are exhausted; sum up all S(i,j) to get the final S(i,j) = 5.25. This value represents the correlation strength between the two variables under the current degrees of freedom. The larger the value, the stronger the correlation strength; record the correlation strength of the data, combine it with the correct posture data in reality, statistically analyze and build a risk matrix, and judge the risk value.

[0033] As a further improvement to the present invention, based on the risk matrix, the alarm step performs the following operations:

[0034] For low-risk cases, monitoring measures are implemented, including continued surveillance of changes in body posture within 10 minutes of the event; if abnormalities are observed, the risk level is raised to medium.

[0035] When a medium-risk event occurs, a 10-second countdown will be used. The countdown must be manually canceled on the watch. If the countdown is not canceled and the event is upgraded to a high-risk event after 10 seconds, the event will be considered a high-risk event.

[0036] In high-risk situations, contact medical staff via voice outbound calls and be required to receive a response.

[0037] This invention also provides a system for fall detection and alarm combining headphones and a watch, comprising a watch and headphones. The watch has functions for measuring heart rate, blood pressure, and mental state, and includes a gyroscope sensor, an orientation sensor, a gravity sensor, and a rotation vector sensor. The headphones provide sound reception and also include a gyroscope sensor and a rotation vector sensor. The system further includes:

[0038] Data acquisition module: Used to collect various sensor data from the watch and headphones;

[0039] Data cleaning module: used to filter, reduce noise, and reassemble the collected sensor data;

[0040] Model training module: used to train on normal data and extract features from data in non-fall scenarios;

[0041] Alarm module: Used to provide secondary confirmation and voice call services.

[0042] As a further improvement of the present invention, in the data cleaning module, during the screening, filtering, noise reduction and recombination process, the segmented binary data of the watch and earphones are converted into digital signals that are easy to process and fall into the [0~10] range. The recombination algorithm arranges multiple data segments into a two-dimensional array for noise reduction operation.

[0043] As a further improvement of the present invention, the noise reduction in the data cleaning module includes the following steps:

[0044] Step a: Calculate the maximum and minimum values ​​of the overall data, unify the values ​​to return to the preset range, and calculate the ratio that needs to be reduced;

[0045] Step b: Loop through the items in the array, take the 8 values ​​around each item, and use the sum of squared deviations (SA) to calculate the degree of deviation of the surrounding points from the center point;

[0046] Step c: If the deviation is greater than the set value, reduce the value by the ratio and then return to step b; if the deviation is less than the set value, then proceed to step d.

[0047] Step d: After the overall data processing is completed, store this part of the data and transfer it to the model training module.

[0048] As a further improvement of the present invention, in the model training module, the data entering the model is initially preset to be an abnormal situation. By inputting data with correct posture, the prediction vector is continuously optimized through a correlation algorithm. If the matching degree with the vector is high, it is determined to be a normal posture. The steps are as follows:

[0049] Step 1: Let i be the row number used for traversal, i = 0, X(i) = 1, Y(i) = 10;

[0050] Step 2: Fix i = 1, traverse all other rows, and let j be the row number used for traversal (j != i, j != i, j != i); Step 3: j = 2, X(j) = 2, Y(j) = 20;

[0051] Step 4: Let the function d(a,b)=|ab|. The d() function calculates the absolute value of the difference between two input values. If a and b are generalized to vectors, then it calculates the difference between two vectors.

[0052] Step 5: Calculate d(xi,xj) = |x1-x2| = |1-2| = 1, and set it as a new variable Rx, Rx = 1;

[0053] Step 6: Calculate d(yi,yj)=|y1-y2|, and set it as a new variable Ry, Ry=8; Step 7: Fix j, traverse all other rows; set k as the row number used for traversal (k!=i AND k!=j); k=3, X(k)=3, Y(k)=30;

[0054] Step 8: Calculate d(xi,xk) = |x1 - x3| = |1 - 3| = 2;

[0055] Step 9: Calculate d(yi,yk)=|y1-y3|=|12-30|=18;

[0056] Step 10: Set variables A11 = 0, A12 = 0, A21 = 0, A22 = 0;

[0057] Step 11: Compare the values ​​of d(xi,xk) and Rx, and compare the values ​​of d(yi,yk) and Ry. If d(xi,xk) <= Rx and d(yi,yk) <= Ry, then add 1 to A11, and A1 = A11 + A12.

[0058] Step 12: Increment k by 1, repeat steps 7 to 11, and accumulate the A series variables;

[0059] Step 13: Now that we have the contingency table for the cases i=1, j=2, we can use the chi-square test to calculate the result. The formula is as follows:

[0060]

[0061] Step 14: For the current case of i=1, j=2, we can calculate that S(i,j)=0;

[0062] Step 15: Add 1 to j, j = 3, X(j) = 3, Y(j) = 30;

[0063] Step 16: Repeat steps 5 to 16, calculating S(i,j) in each step and summing them up; after calculation, we get the total S(i,j) = 3 when i = 1, i = 2, X(i) = 2;

[0064] Step 17: Repeat steps 3 to 17 until all i are exhausted; sum up all S(i,j) to get the final S(i,j) = 5.25. This value represents the correlation strength between the two variables under the current degrees of freedom. The larger the value, the stronger the correlation strength; record the correlation strength of the data, combine it with the correct posture data in reality, statistically analyze and build a risk matrix, and judge the risk value.

[0065] As a further improvement to the present invention, based on the risk matrix, the alarm module performs the following operations:

[0066] For low-risk cases, monitoring measures are implemented, including continued surveillance of changes in body posture within 10 minutes of the event; if abnormalities are observed, the risk level is raised to medium.

[0067] When a medium-risk event occurs, a 10-second countdown will be used. The countdown must be manually canceled on the watch. If the countdown is not canceled and the event is upgraded to a high-risk event after 10 seconds, the event will be considered a high-risk event.

[0068] In high-risk situations, contact medical staff via voice outbound calls and be required to receive a response.

[0069] The beneficial effects of this invention are as follows: Compared with the prior art, this invention combines head posture monitoring via headphones with watch data, and uses an abnormal or non-normal approach to determine falls. This results in a smaller model training load and more accurate judgment, which is of great significance for research on real-time fall warning and fall prevention. Attached Figure Description

[0070] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0071] This invention discloses a method for fall detection and alarm combining headphones and a watch, comprising the following steps:

[0072] Data acquisition steps: Collect various sensor data from the watch and headphones;

[0073] Data cleaning steps: The collected sensor data is screened, filtered, denoised, and reconstructed.

[0074] Model training steps: Train on normal data and extract features from data under non-fall conditions;

[0075] Alarm procedure: Provide secondary confirmation and voice call service.

[0076] In the data cleaning step, during the screening, filtering, noise reduction, and reconstruction processes, the segmented binary data of the watch and earphones are converted into easily processed digital signals, ensuring they fall within the [0~10] range. The reconstruction algorithm arranges multiple data segments into a two-dimensional array for noise reduction. The noise reduction includes the following steps:

[0077] Step a: Calculate the maximum and minimum values ​​of the overall data, unify the values ​​to return to the preset range, and calculate the ratio that needs to be reduced;

[0078] Step b: Loop through the items in the array, take the 8 values ​​around each item, and use the sum of squared deviations (SA) to calculate the degree of deviation of the surrounding points from the center point;

[0079] Step c: If the deviation is greater than the set value (if the deviation is greater than the set value, it is considered that the deviation is too high), then the value is reduced by the ratio, and then the process returns to step b. If the deviation is less than the set value, then the process returns to step d.

[0080] Step d: After the overall data processing is completed, store this part of the data and pass it into the model training step.

[0081] In the model training step, the data entering the model is initially pre-defined as an abnormal situation. By inputting data with correct poses, the prediction vector is continuously optimized through a correlation algorithm. If the matching degree with the vector is high, it is determined to be a normal pose. The steps are as follows:

[0082] Step 1: Let i be the row number used for traversal, i = 0, X(i) = 1, Y(i) = 10;

[0083] Step 2: Fix i = 1, traverse all other rows, and let j be the row number used for traversal (j != i, j != i, j != i);

[0084] Step 3: j = 2, X(j) = 2, Y(j) = 20;

[0085] Step 4: Let the function d(a,b)=|ab|. The d() function calculates the absolute value of the difference between two input values. If a and b are generalized to vectors, then it calculates the difference between two vectors.

[0086] Step 5: Calculate d(xi,xj) = |x1-x2| = |1-2| = 1, and set it as a new variable Rx, Rx = 1;

[0087] Step 6: Calculate d(yi,yj)=|y1-y2|, and set it as a new variable Ry, Ry=8;

[0088] Step 7: Fix j, traverse all other rows; let k be the row number used for traversal (k!= i AND k!= j); k=3, X(k)=3, Y(k)=30;

[0089] Step 8: Calculate d(xi,xk) = |x1 - x3| = |1 - 3| = 2;

[0090] Step 9: Calculate d(yi,yk)=|y1-y3|=|12-30|=18;

[0091] Step 10: Set variables A11 = 0, A12 = 0, A21 = 0, A22 = 0. These are the cumulative counts that will be used below.

[0092] Step 11: Compare the values ​​of d(xi,xk) and Rx, and compare the values ​​of d(yi,yk) and Ry. There are four possible comparison results, as shown in the table below:

[0093]

[0094] If d(xi,xk) <= Rx and d(yi,yk) <= Ry, then A11 is incremented by 1. Other cases can be understood according to the table. A1 = A11 + A12; see the table for other similar row / column sums. N represents the total number of rows.

[0095] Step 12: Increment k by 1, repeat steps 7 to 11, and accumulate the A series variables;

[0096] Step 13: Now that we have the contingency table for the cases i=1, j=2, we can use the chi-square test to calculate the result. The formula is as follows:

[0097]

[0098] Step 14: For the current case of i=1, j=2, we can calculate that S(i,j)=0;

[0099] Step 15: Add 1 to j, j = 3, X(j) = 3, Y(j) = 30;

[0100] Step 16: Repeat steps 5 to 16, calculating S(i,j) in each step and summing them up; after calculation, we get the total S(i,j) = 3 when i = 1, i = 2, X(i) = 2;

[0101] Step 17: Repeat steps 3 to 17 until all i are exhausted; sum up all S(i,j) to get the final S(i,j) = 5.25. This value represents the correlation strength between the two variables under the current degrees of freedom; the larger the value, the stronger the correlation. Record the correlation strength of the data, and combine it with the correct posture (risk-free posture) data in reality to statistically construct a risk matrix and determine the risk value (low, medium, high):

[0102]

[0103] Based on the risk matrix, the alarm procedure shall be performed as follows:

[0104] For low-risk cases, monitoring measures are implemented, including continued surveillance of changes in body posture within 10 minutes of the event; if abnormalities are observed, the risk level is raised to medium.

[0105] When a medium-risk event occurs, a 10-second countdown will be used. The countdown must be manually canceled on the watch. If the countdown is not canceled and the event is upgraded to a high-risk event after 10 seconds, the event will be considered a high-risk event.

[0106] In high-risk situations, contact medical staff via voice outbound calls and be required to receive a response.

[0107] like Figure 1 As shown, this invention also discloses a system for fall detection and alarm combining headphones and a watch, including a watch and headphones. The watch has functions for measuring heart rate, blood pressure, and mental state, and includes a gyroscope sensor, an orientation sensor, a gravity sensor, and a rotation vector sensor, and can output the data from these sensors. The headphones provide sound reception, and also include a gyroscope sensor and a rotation vector sensor, and can output the data from these sensors. The system further includes:

[0108] Data acquisition module: Used to collect various sensor data from the watch and headphones;

[0109] Data cleaning module: used to filter, reduce noise, and reassemble the collected sensor data;

[0110] Model training module: used to train on normal data and extract features from data in non-fall scenarios;

[0111] Alarm module: Used to provide secondary confirmation and voice call services.

[0112] Compared with existing technologies, this invention combines head posture monitoring via headphones with watch data, and uses an abnormal or non-normal approach to determine falls. This results in a smaller model training load and more accurate judgment, which is of great significance for research on real-time fall warning and fall prevention.

[0113] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for fall detection and alarm combining headphones and a watch, characterized in that, The steps include: Data acquisition steps: Collect various sensor data from the watch and headphones; Data cleaning steps: The collected sensor data is screened, filtered, denoised, and reconstructed. Model training steps: Train on normal data and extract features from data under non-fall conditions; Alarm procedure: Provide secondary confirmation and voice call service; In the model training step, the data entering the model is initially pre-defined as an abnormal situation. By inputting data with correct poses, the prediction vector is continuously optimized through a correlation algorithm. If the matching degree with the vector is high, it is determined to be a normal pose. The steps are as follows: Step 1: Let i be the row number used for traversal, i = 0, X(i) = 1, Y(i) = 10; Step 2: Fix i = 1, traverse all other rows, and let j be the row number used for traversal (j != i, j != i, j != i); Step 3: j = 2, X(j) = 2, Y(j) = 20; Step 4: Let the function d(a,b)=|ab|. The d() function calculates the absolute value of the difference between two input values. If a and b are generalized to vectors, then it calculates the difference between two vectors. Step 5: Calculate d(xi,xj) = |x1-x2| = |1-2| = 1, and set it as a new variable Rx, Rx = 1; Step 6: Calculate d(yi,yj)=|y1-y2|, and set it as a new variable Ry, Ry=8; Step 7: Fix j, traverse all other rows; let k be the row number used for traversal (k!= i AND k!= j); k=3, X(k)=3, Y(k)=30; Step 8: Calculate d(xi,xk) = |x1 - x3| = |1 - 3| = 2; Step 9: Calculate d(yi,yk)=|y1-y3|=|12-30|=18; Step 10: Set variables A11 = 0, A12 = 0, A21 = 0, A22 = 0; Step 11: Compare the values ​​of d(xi,xk) and Rx, and compare the values ​​of d(yi,yk) and Ry. If d(xi,xk) <= Rx and d(yi,yk) <= Ry, then add 1 to A11, and A1 = A11 + A12. Step 12: Increment k by 1, repeat steps 7 to 11, and accumulate the A series variables; Step 13: Now we have the contingency table for cases i=1, j=2. We use the chi-square test to calculate the result, as shown in the formula below: Step 14: For the current case of i=1, j=2, we can calculate that S(i,j)=0; Step 15: Add 1 to j, j = 3, X(j) = 3, Y(j) = 30; Step 16: Repeat steps 5 to 16, calculating S(i,j) in each step and summing them up; after calculation, we get the total S(i,j) = 3 when i = 1, i = 2, X(i) = 2; Step 17: Repeat steps 3 to 17 until all i are exhausted; sum up all S(i,j) to get the final S(i,j) = 5.

25. This value represents the correlation strength between the two variables under the current degrees of freedom. The larger the value, the stronger the correlation strength; record the correlation strength of the data, combine it with the correct posture data in reality, statistically analyze and build a risk matrix, and judge the risk value.

2. The method according to claim 1, characterized in that, In the data cleaning step, during the screening, filtering, noise reduction, and recombination process, the segmented binary data of the watch and earphones are converted into digital signals that are easy to process and fall into the [0~10] range. The recombination algorithm arranges multiple data segments into a two-dimensional array for noise reduction.

3. The method according to claim 2, characterized in that, In the data cleaning step, noise reduction includes the following steps: Step a: Calculate the maximum and minimum values ​​of the overall data, unify the values ​​to return to the preset range, and calculate the ratio that needs to be reduced; Step b: Loop through the items in the array, take the 8 values ​​around each item, and use the sum of squared deviations (SA) to calculate the degree of deviation of the surrounding points from the center point; Step c: If the deviation is greater than the set value, reduce the value by the ratio and then return to step b; if the deviation is less than the set value, then proceed to step d. Step d: After the overall data processing is completed, store this part of the data and pass it into the model training step.

4. The method according to claim 1, characterized in that, Based on the risk matrix, the alarm procedure shall be performed as follows: For low-risk cases, monitoring measures are implemented, including continued surveillance of changes in body posture within 10 minutes of the event; if abnormalities are observed, the risk level is raised to medium. When a medium-risk event occurs, a 10-second countdown will be used. The countdown must be manually canceled on the watch. If the countdown is not canceled and the event is upgraded to a high-risk event after 10 seconds, the event will be considered a high-risk event. In high-risk situations, contact medical staff via voice outbound calls and be required to receive a response.

5. A system for fall detection and alarm combining headphones and a watch, characterized in that, The system includes a watch and headphones. The watch has functions for measuring heart rate, blood pressure, and mental state, and includes a gyroscope sensor, orientation sensor, gravity sensor, and rotation vector sensor. The headphones provide sound pickup and also include a gyroscope sensor and a rotation vector sensor. The system further includes: Data acquisition module: Used to collect various sensor data from the watch and headphones; Data cleaning module: used to filter, reduce noise, and reassemble the collected sensor data; Model training module: used to train on normal data and extract features from data under non-fall conditions; Alarm module: used to provide secondary confirmation and voice call services; In the model training module, the data entering the model is initially pre-defined as an abnormal situation. By inputting data with correct poses, the prediction vector is continuously optimized through a correlation algorithm. If the matching degree with the vector is high, it is determined to be a normal pose. The steps are as follows: Step 1: Let i be the row number used for traversal, i = 0, X(i) = 1, Y(i) = 10; Step 2: Fix i = 1, traverse all other rows, and let j be the row number used for traversal (j != i, j != i, j != i); Step 3: j = 2, X(j) = 2, Y(j) = 20; Step 4: Let the function d(a,b)=|ab|. The d() function calculates the absolute value of the difference between two input values. If a and b are generalized to vectors, then it calculates the difference between two vectors. Step 5: Calculate d(xi,xj) = |x1-x2| = |1-2| = 1, and set it as a new variable Rx, Rx = 1; Step 6: Calculate d(yi,yj)=|y1-y2|, and set it as a new variable Ry, Ry=8; Step 7: Fix j, traverse all other rows; let k be the row number used for traversal (k!= i AND k!= j); k=3, X(k)=3, Y(k)=30; Step 8: Calculate d(xi,xk) = |x1 - x3| = |1 - 3| = 2; Step 9: Calculate d(yi,yk)=|y1-y3|=|12-30|=18; Step 10: Set variables A11 = 0, A12 = 0, A21 = 0, A22 = 0; Step 11: Compare the values ​​of d(xi,xk) and Rx, and compare the values ​​of d(yi,yk) and Ry. If d(xi,xk) <= Rx and d(yi,yk) <= Ry, then add 1 to A11, and A1 = A11 + A12. Step 12: Increment k by 1, repeat steps 7 to 11, and accumulate the A series variables; Step 13: Now we have the contingency table for cases i=1, j=2. We use the chi-square test to calculate the result, as shown in the formula below: Step 14: For the current case of i=1, j=2, we can calculate that S(i,j)=0; Step 15: Add 1 to j, j = 3, X(j) = 3, Y(j) = 30; Step 16: Repeat steps 5 to 16, calculating S(i,j) in each step and summing them up; after calculation, we get the total S(i,j) = 3 when i = 1, i = 2, X(i) = 2; Step 17: Repeat steps 3 to 17 until all i are exhausted; sum up all S(i,j) to get the final S(i,j) = 5.

25. This value represents the correlation strength between the two variables under the current degrees of freedom. The larger the value, the stronger the correlation strength; record the correlation strength of the data, combine it with the correct posture data in reality, statistically analyze and build a risk matrix, and judge the risk value.

6. The system according to claim 5, characterized in that, In the data cleaning module, during the screening, filtering, noise reduction, and recombination process, the segmented binary data of the watch and earphones are converted into digital signals that are easy to process and fall into the [0~10] range. The recombination algorithm arranges multiple data segments into a two-dimensional array for noise reduction.

7. The system according to claim 6, characterized in that, In the data cleaning module, noise reduction includes the following steps: Step a: Calculate the maximum and minimum values ​​of the overall data, unify the values ​​to return to the preset range, and calculate the ratio that needs to be reduced; Step b: Loop through the items in the array, take the 8 values ​​around each item, and use the sum of squared deviations (SA) to calculate the degree of deviation of the surrounding points from the center point; Step c: If the deviation is greater than the set value, reduce the value by the ratio and then return to step b; if the deviation is less than the set value, then proceed to step d. Step d: After the overall data processing is completed, store this part of the data and transfer it to the model training module.

8. The system according to claim 5, characterized in that, Based on the risk matrix, the alarm module performs the following operations: For low-risk cases, monitoring measures are implemented, including continued surveillance of changes in body posture within 10 minutes of the event; if abnormalities are observed, the risk level is raised to medium. When a medium-risk event occurs, a 10-second countdown will be used. The countdown must be manually canceled on the watch. If the countdown is not canceled and the event is upgraded to a high-risk event after 10 seconds, the event will be considered a high-risk event. In high-risk situations, contact medical staff via voice outbound calls and be required to receive a response.

Images