A home elderly fall risk early warning method

Abstract

The application discloses a home old people fall risk early warning method, which comprises the following steps: step S1, collecting point cloud data; step S2, generating a time sequence matrix; step S3, extracting multi-dimensional behavior time sequence features; step S4, constructing a database and calculating a deviation index; step S5, determining an early warning level and pushing to a child end APP; step S6, constructing a risk-intervention mapping table and generating health suggestions; and step S7, continuously iteratively updating the database. The home old people fall risk early warning method can analyze and compare multi-dimensional time sequence features of the old people through the millimeter wave radar fall monitoring technology, early warning can be given before the fall occurs, and the early warning can be pushed to the child end APP, so that early prediction and post-alarm are realized, and the fall risk of the old people living alone is reduced.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, and in particular, to a method for early warning of fall risk in elderly people living at home. Background Technology

[0002] How to effectively monitor and provide early warning of fall risks for elderly people living alone has become a research hotspot in the fields of rehabilitation engineering and smart elderly care.

[0003] Currently, existing fall detection methods can be divided into the following categories. One is wearable device detection, which collects human motion data through sensors such as accelerometers and gyroscopes, and uses threshold methods or machine learning algorithms to identify fall behavior. However, the elderly are prone to forgetting to wear them or resisting wearing them, and they require frequent charging, resulting in poor actual user compliance.

[0004] Secondly, there is camera-based visual detection, which uses RGB or depth cameras to capture images of human posture and uses computer vision algorithms to identify falls. However, cameras pose a risk of infringing on personal privacy, so they are not suitable for private places such as bedrooms and bathrooms. At the same time, the lighting conditions at night may also prevent the accurate capture of human posture images.

[0005] Third, fall detection based on millimeter-wave radar collects the distance-Doppler spectrum of actions such as falling, walking, sitting, and standing still, and identifies falls using a large amount of data and specific algorithms. However, existing millimeter-wave radar fall detection technologies are all post-event detection modes, and their behavior recognition dimensions are relatively simple, lacking individual baseline adaptation mechanisms. Summary of the Invention

[0006] Therefore, it is necessary to provide a method for early warning of fall risks for elderly people living at home, which can achieve pre-accident prediction and post-accident alert.

[0007] The technical solution adopted by this invention to solve its technical problem is: a method for early warning of fall risk for elderly people living at home, which specifically includes the following steps:

[0008] Step S1: Collect point cloud data. Continuously collect human activity data of the elderly in the home environment using millimeter-wave radar to generate a point cloud data sequence with distance, speed and angle information.

[0009] Step S2: Generate a time series matrix. Extract human skeleton feature sequences based on the accumulated point cloud data of T frames, and organize the skeleton feature sequences of consecutive T frames into a time series matrix F.

[0010] Step S3: Extract multi-dimensional behavioral temporal features from the temporal matrix;

[0011] Step S4: Construct a database and calculate the deviation index. Extract the multidimensional behavioral time-series features to form a database, and compare the current feature value with the feature value in the database to calculate the deviation index.

[0012] Step S5: Determine the warning level and push it to the child's APP. The warning level is determined based on the cumulative risk of the deviation index of each feature. The warning level is used to generate an output signal and the output signal is pushed to the child's APP.

[0013] Step S6: Construct a risk-intervention mapping table and generate health recommendations. The risk-intervention mapping table is constructed by associating the characteristic abnormality type with the corresponding intervention recommendation. Then, the executable health recommendations are generated by comparing the characteristic values ​​of the deviation with the characteristic values ​​on the risk-intervention mapping table.

[0014] Step S7: Continuously iterate and update the database. After each day ends, the feature data of that day is included in the database, and the mean and standard deviation in the database are automatically updated, thereby realizing the adaptive adjustment of the database.

[0015] Further, step S2 includes the following steps:

[0016] Step S21: Use a clustering algorithm to perform target detection on the point cloud of frame T, and separate the human target from the static background point cloud;

[0017] Step S22: Continuously track the human target by matching the correlation of point clouds in adjacent frames;

[0018] Step S23: Extract the three-dimensional spatial coordinates of 17 joints from the tracked human target point cloud, so as to construct a human skeleton model;

[0019] Step S24: Organize the human skeleton feature sequences of consecutive T frames into a time-series matrix F, F={f1, f2, ...,f t}

[0020] Further, in step S3, the multidimensional behavioral temporal features include the following steps:

[0021] Step S31: Gait feature extraction, extract the trajectory of the ankle and hip joints, and calculate the walking speed, stride length, cadence, and left-right cadence asymmetry in each walking sequence;

[0022] Step S32: Extract features of sitting-up ability, extract the height changes of the hip and shoulder joints, detect the transition features between sitting and standing, and calculate the sitting-up time, sitting-up speed, and dependence on hand support.

[0023] Step S33: Extract sleep and nighttime behavioral features, extracting changes in the location of human targets at night;

[0024] Step S34: Extract circadian rhythm features, extracting the spatial location distribution and activity intensity of human targets throughout the day.

[0025] Further, step S4 includes the following steps:

[0026] Step S41: First, enter the cold start phase, collect and store the user's multi-dimensional behavioral time-series feature data, and establish a database;

[0027] Step S42: Calculate the deviation index using the formula D = |x - μ| / σ.

[0028] Further, step S5 includes the following steps:

[0029] Step S51: When the deviation index exceeds the preset threshold, it is marked as a yellow warning;

[0030] Step S52: A red alert is triggered when the following three conditions are met simultaneously and remain abnormal for three consecutive days, as follows:

[0031] First, the walking speed decreased by more than 15% compared to the walking speed in the database;

[0032] Secondly, the duration of sitting up and sitting down was more than 30% longer than the duration of sitting up and sitting down in the database;

[0033] Third, the number of times someone gets out of bed at night is one or more times more than the number of times someone gets out of bed at night recorded in the database;

[0034] Fourth, the angle of the body leaning forward when standing up is more than 30° greater than the angle of the body leaning forward when standing up in the database;

[0035] Fifth, the amount of daytime activity decreased by more than 40% compared to the amount of daytime activity in the database;

[0036] Step S53: Generate an output signal based on the warning level and push the output signal to the child's APP.

[0037] Further, step S6 includes the following steps:

[0038] Step S61: Construct a risk-intervention mapping table by associating the characteristic anomaly type with the corresponding intervention recommendation;

[0039] Step S62: Based on the currently triggered combination of abnormal features, retrieve a matching intervention suggestion from the risk-intervention mapping table;

[0040] Step S63: Combine the warning level, abnormal feature description and intervention recommendations to generate a health report in natural language.

[0041] Furthermore, in step S1, the millimeter-wave radar is installed at a height of 2.2m on the interior wall of the bedroom, living room, and bathroom, and covers the main activity areas of the bedroom, living room, and bathroom at a downward angle of 30°.

[0042] Furthermore, the millimeter-wave radar continuously acquires data at 20 FPS, and each frame output contains a data set of M point clouds.

[0043] Furthermore, physical cancel buttons are installed on the walls of the main activity areas of the bedroom, living room, and bathroom. The physical cancel buttons are located one meter above the ground on the wall. The physical cancel buttons allow the elderly to cancel false alarms and feed back false alarm events to the database for optimizing the alarm threshold.

[0044] Furthermore, in step S23, the 17 joint points are the head, neck, right shoulder, right elbow, right wrist, left shoulder, left elbow, left wrist, hip center, right hip, right knee, right ankle, left hip, left knee, left ankle, spine, and chest.

[0045] The beneficial effects of this invention are: the fall risk warning method for elderly people living at home provided by this invention can analyze and compare the multi-dimensional temporal characteristics of the elderly through the millimeter-wave radar fall monitoring technology, issue a warning before a fall occurs, and push it to the children's APP, thereby realizing pre-event prediction and post-event alarm, so as to reduce the fall risk of elderly people living alone. Attached Figure Description

[0046] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0047] Figure 1 This is a flowchart of the fall risk warning method for elderly people living at home according to the present invention;

[0048] Figure 2 yes Figure 1 The diagram shows the logic of a fall risk warning method for elderly people living at home. Detailed Implementation

[0049] The present invention will now be described in detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0050] Please see Figure 1 and Figure 2 This invention provides a method for early warning of fall risk for elderly people living at home, which includes the following steps:

[0051] Step S1: Collect point cloud data. Continuously collect human activity data of the elderly person in a home environment using millimeter-wave radar to generate a point cloud data sequence containing distance, speed, and angle information. In this embodiment, the millimeter-wave radar uses an AWR1843 chip, operates at 77GHz, and is configured as a three-transmit, four-receive antenna array.

[0052] Furthermore, the millimeter-wave radar is installed 2.2m high on the interior wall of the bedroom, living room, and bathroom, and covers the main activity areas of the bedroom, living room, and bathroom at a 30° downward angle.

[0053] Furthermore, the millimeter-wave radar continuously acquires data at 20 FPS, with each frame output containing a dataset of M point clouds. Furthermore, each point cloud dataset includes range, radial velocity, azimuth, elevation, and signal-to-noise ratio information. In this embodiment, 200 < M < 500. Preferably, M = 350.

[0054] Step S2: Generate a temporal matrix. Extract the human skeleton feature sequence based on the accumulated point cloud data of T frames, and organize the human skeleton feature sequences of consecutive T frames into a temporal matrix F. In this embodiment, the human skeleton extraction employs a joint regression model based on a graph attention network, thereby improving the joint prediction accuracy under partial occlusion conditions.

[0055] Specifically, step S2 includes the following steps:

[0056] Step S21: Use a clustering algorithm to perform target detection on the T-frame point cloud, separating the human target from the static background point cloud.

[0057] Step S22: Continuously track human targets by matching the correlation of point clouds in adjacent frames.

[0058] Step S23: Extract the three-dimensional spatial coordinates of 17 joints from the tracked human target point cloud, thereby constructing a human skeleton model. In this embodiment, the 17 joints are the head, neck, right shoulder, right elbow, right wrist, left shoulder, left elbow, left wrist, hip center, right hip, right knee, right ankle, left hip, left knee, left ankle, spine, and chest. The 17 joints are numbered as follows: head is 0, neck is 1, right shoulder is 2, right elbow is 3, right wrist is 4, left shoulder is 5, left elbow is 6, left wrist is 7, hip center is 8, right hip is 9, right knee is 10, right ankle is 11, left hip is 12, left knee is 13, left ankle is 14, spine is 15, and chest is 16.

[0059] Step S24: Organize the human skeleton feature sequences of consecutive T frames into a temporal matrix F, F={f1, f2, ...,ft}. In this embodiment, each frame ft contains the three-dimensional coordinates of each joint point.

[0060] Step S3: Extract multidimensional behavioral temporal features from the temporal matrix.

[0061] The multidimensional behavioral temporal features include the following steps:

[0062] Step S31: Gait feature extraction, extract the trajectory of the ankle and hip joints, and calculate the walking speed, stride length, cadence, and left-right cadence asymmetry in each walking sequence.

[0063] Step S32: Extract features of sitting-up ability, extract the height changes of the hip and shoulder joints, detect the transition features between sitting and standing, and calculate the sitting-up duration, sitting-up speed, and dependence on hand support.

[0064] Step S33: Extract sleep and nighttime behavioral features, extracting changes in the location of human targets at night. Specifically, extract the number of times the person gets out of bed at night, the duration of time out of bed, the time to return to bed after getting out of bed, and the frequency of minute movements during bed rest.

[0065] Step S34: Extract circadian rhythm features, extracting the spatial distribution and activity intensity of the human target throughout the day. Specifically, extract the human target's daily wake-up time, first time out of bed, daytime activity level, and cumulative daytime sitting time.

[0066] Step S4: Construct a database and calculate the deviation index. Extract the multidimensional behavioral time-series features to form a database, and compare the current feature value with the feature value in the database to calculate the deviation index.

[0067] Specifically, step S4 includes the following steps:

[0068] Step S41: First, the cold start phase begins, during which the user's multidimensional behavioral time-series characteristic data is collected and stored to establish a database. In this embodiment, the cold start phase lasts for 30 days.

[0069] Step S42: Calculate the deviation index using the formula D = |x - μ| / σ. Where x is the current (e.g., today or this week) feature value; μ is the mean of the feature in the database; σ is the standard deviation of the feature in the database; the larger the D value, the greater the deviation from the daily pattern.

[0070] Step S5: Determine the warning level and push it to the child's APP. The warning level is determined based on the composite risk accumulation of the deviation index of each feature. The warning level is used to generate an output signal and the output signal is pushed to the child's APP.

[0071] Specifically, step S5 includes the following steps:

[0072] Step S51: When the deviation index exceeds a preset threshold, it is marked as a yellow warning. In this embodiment, when D > 2.0, it is marked as a yellow warning.

[0073] Step S52: A red alert is triggered when three or more of the following conditions are met simultaneously and remain abnormal for three consecutive days, as detailed below:

[0074] First, the walking speed decreased by more than 15% compared to the walking speed in the database;

[0075] Secondly, the duration of sitting up and sitting down was more than 30% longer than the duration of sitting up and sitting down in the database;

[0076] Third, the number of times someone gets out of bed at night is one or more times more than the number of times someone gets out of bed at night recorded in the database;

[0077] Fourth, the angle of the body leaning forward when standing up is more than 30° greater than the angle of the body leaning forward when standing up in the database;

[0078] Fifth, the amount of daytime activity decreased by more than 40% compared to the amount of daytime activity in the database.

[0079] Step S53: Generate an output signal based on the warning level and push the output signal to the children's app. If it is a yellow warning, push the signal to the children's app and suggest that the children pay attention to the elderly person's health; if it is a red warning, push the signal to the children's app and suggest that the children take the elderly person to see a doctor for evaluation.

[0080] Step S6: Construct a risk-intervention mapping table and generate health recommendations. The risk-intervention mapping table is constructed by associating the characteristic abnormality type with the corresponding intervention recommendation. Then, the executable health recommendations are generated by comparing the characteristic values ​​of the deviation with the characteristic values ​​on the risk-intervention mapping table.

[0081] Specifically, step S6 includes the following steps:

[0082] Step S61: Construct a risk-intervention mapping table by associating characteristic abnormality types with corresponding intervention recommendations. For example, if a decrease in walking speed is observed, it is recommended to perform 30 minutes of lower limb strength training daily; if difficulty sitting up is observed, it is recommended to install handrails by the bedside; if an increase in nighttime ambulation is observed, it is recommended to consult a urologist.

[0083] Step S62: Based on the currently triggered abnormal feature combination, retrieve a matching intervention suggestion from the risk-intervention mapping table.

[0084] Step S63: Combine the warning level, abnormal feature description and intervention recommendations to generate a health report in natural language.

[0085] Step S7: Continuously iteratively update the database. Specifically, at the end of each day, the feature data of that day is incorporated into the database, and the mean and standard deviation in the database are automatically updated, thereby achieving adaptive adjustment of the database. In this embodiment, the database update adopts an exponentially weighted moving average method, which can smooth the impact of short-term fluctuations on the database.

[0086] Specifically, taking gait speed as an example, the average gait speed of all walking segments throughout the day was extracted daily, resulting in 30 gait speed samples over 30 days. The mean μ was calculated to be 0.72 m / s, and the standard deviation σ was 0.08 m / s. Taking the data from the elderly person on day 45 as an example, the gait speed was 0.58 m / s, and the deviation index D... 步速 =|x - μ| / σ=|0.58- 0.72| / 0.08=1.75, the sitting-up time increased from 2.1s in the database to 3.0s, and the deviation index D 起坐时长 =1.5, the number of times the elderly person gets out of bed at night increased from 0.8 times / night in the database to 2.2 times / night. The simultaneous presence of these three abnormalities triggered a red light warning, which was then pushed to the children's app, displaying "The elderly person's walking speed has decreased by 22%, the difficulty in getting up and sitting up has worsened, and the frequent getting up at night significantly increases the risk of falls. It is recommended to seek medical evaluation within this week and consider installing handrails in the bathroom."

[0087] Furthermore, physical cancel buttons are installed on the walls of the main activity areas of the bedroom, living room, and bathroom. The physical cancel buttons are located one meter above the ground on the wall. The physical cancel buttons allow the elderly to cancel false alarms and feed back false alarm events to the database for optimizing the alarm threshold.

[0088] The fall risk warning method for elderly people living alone provided by this invention can analyze and compare the multi-dimensional temporal characteristics of the elderly through the millimeter-wave radar fall monitoring technology, issue a warning before a fall occurs, and push it to the children's APP, thereby realizing pre-event prediction and post-event alarm, so as to reduce the fall risk of elderly people living alone.

[0089] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the scope of the present invention. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for early warning of fall risk for elderly people living at home, characterized in that: The method for early warning of fall risk for elderly people living at home includes the following steps: Step S1: Collect point cloud data. Continuously collect human activity data of the elderly in the home environment using millimeter-wave radar to generate a point cloud data sequence with distance, speed and angle information. Step S2: Generate a time series matrix. Extract human skeleton feature sequences based on the accumulated point cloud data of T frames, and organize the skeleton feature sequences of consecutive T frames into a time series matrix F. Step S3: Extract multi-dimensional behavioral temporal features from the temporal matrix; Step S4: Construct a database and calculate the deviation index. Extract the multidimensional behavioral time-series features to form a database, and compare the current feature value with the feature value in the database to calculate the deviation index. Step S5: Determine the warning level and push it to the child's APP. The warning level is determined based on the cumulative risk of the deviation index of each feature. The warning level is used to generate an output signal and the output signal is pushed to the child's APP. Step S6: Construct a risk-intervention mapping table and generate health recommendations. The risk-intervention mapping table is constructed by associating the characteristic abnormality type with the corresponding intervention recommendation. Then, the executable health recommendations are generated by comparing the characteristic values ​​of the deviation with the characteristic values ​​on the risk-intervention mapping table. Step S7: Continuously iterate and update the database. After each day ends, the feature data of that day is included in the database, and the mean and standard deviation in the database are automatically updated, thereby realizing the adaptive adjustment of the database.

2. The method for early warning of fall risk for elderly people living at home as described in claim 1, characterized in that: Step S2 includes the following steps: Step S21: Use a clustering algorithm to perform target detection on the point cloud of frame T, and separate the human target from the static background point cloud; Step S22: Continuously track the human target by matching the correlation of point clouds in adjacent frames; Step S23: Extract the three-dimensional spatial coordinates of 17 joints from the tracked human target point cloud, so as to construct a human skeleton model; Step S24: Organize the human skeleton feature sequences of consecutive T frames into a time-series matrix F, F={f1, f2, ..., f t } 3. The method for early warning of fall risk for elderly people living at home as described in claim 2, characterized in that: In step S3, the multidimensional behavioral temporal features include the following steps: Step S31: Gait feature extraction, extract the trajectory of the ankle and hip joints, and calculate the walking speed, stride length, cadence, and left-right cadence asymmetry in each walking sequence; Step S32: Extract features of sitting-up ability, extract the height changes of the hip and shoulder joints, detect the transition features between sitting and standing, and calculate the sitting-up time, sitting-up speed, and dependence on hand support. Step S33: Extract sleep and nighttime behavioral features, extracting changes in the location of human targets at night; Step S34: Extract circadian rhythm features, extracting the spatial location distribution and activity intensity of human targets throughout the day.

4. The method for early warning of fall risk for elderly people living at home as described in claim 3, characterized in that: Step S4 includes the following steps: Step S41: First, enter the cold start phase, collect and store the user's multi-dimensional behavioral time-series feature data, and establish a database; Step S42: Calculate the deviation index using the formula D = |x - μ| / σ.

5. The method for early warning of fall risk for elderly people living at home as described in claim 4, characterized in that: Step S5 includes the following steps: Step S51: When the deviation index exceeds the preset threshold, it is marked as a yellow warning; Step S52: A red alert is triggered when the following three conditions are met simultaneously and remain abnormal for three consecutive days, as follows: First, the walking speed decreased by more than 15% compared to the walking speed in the database; Secondly, the duration of sitting up and sitting down was more than 30% longer than the duration of sitting up and sitting down in the database; Third, the number of times someone gets out of bed at night is one or more times more than the number of times someone gets out of bed at night recorded in the database; Fourth, the angle of the body leaning forward when standing up is more than 30° greater than the angle of the body leaning forward when standing up in the database; Fifth, the amount of daytime activity decreased by more than 40% compared to the amount of daytime activity in the database; Step S53: Generate an output signal based on the warning level and push the output signal to the child's APP.

6. The method for early warning of fall risk for elderly people living at home as described in claim 5, characterized in that: Step S6 includes the following steps: Step S61: Construct a risk-intervention mapping table by associating the characteristic anomaly type with the corresponding intervention recommendation; Step S62: Based on the currently triggered combination of abnormal features, retrieve a matching intervention suggestion from the risk-intervention mapping table; Step S63: Combine the warning level, abnormal feature description and intervention recommendations to generate a health report in natural language.

7. The method for early warning of fall risk for elderly people living at home as described in claim 1, characterized in that: In step S1, the millimeter-wave radar is installed at a height of 2.2m on the interior wall of the bedroom, living room, and bathroom, and covers the main activity areas of the bedroom, living room, and bathroom at a downward angle of 30°.

8. The method for early warning of fall risk for elderly people living at home as described in claim 7, characterized in that: The millimeter-wave radar continuously acquires data at 20 FPS, and each frame output contains a data set of M point clouds.

9. The method for early warning of fall risk for elderly people living at home as described in claim 7, characterized in that: Physical cancel buttons are also installed on the walls of the main activity areas of the bedroom, living room, and bathroom. The physical cancel buttons are located one meter above the ground on the wall. The physical cancel buttons allow the elderly to cancel false alarms and feed back false alarm events to the database to optimize the alarm threshold.

10. The method for early warning of fall risk for elderly people living at home as described in claim 2, characterized in that: In step S23, the 17 joint points are the head, neck, right shoulder, right elbow, right wrist, left shoulder, left elbow, left wrist, hip center, right hip, right knee, right ankle, left hip, left knee, left ankle, spine, and chest.

Images