Fall monitoring and help calling method for home elderly

Abstract

The invention is suitable for the technical field of fall detection and help calling, and provides a fall detection and help calling method for the elderly at home, and the method comprises the steps: collecting the multi-source health monitoring data of the elderly through a monitoring camera and a sensor, and carrying out the preprocessing based on the collected data; respectively extracting walking posture features, getting-up and sitting-down posture features, daily activity posture features, motion features, acceleration features, gyroscope features and heart rate features of the elderly; inputting the extracted features into a trained abnormal action detection model for action detection; when the old people have abnormal behavior actions, performing multi-dimensional analysis on the abnormal behavior actions of the old people by utilizing a clustering algorithm; and feeding back the analysis result of the abnormal behavior action of the old person in real time according to a preset help calling rule. According to the invention, the aims of comprehensively monitoring the health state of the elderly, timely discovering abnormity, deeply analyzing reasons and quickly responding to rescue can be fulfilled.

Description

Technical Field

[0001] The present invention relates to the technical field of fall detection and help calling, and in particular to a fall detection and help calling method for elderly people at home. Background Art

[0002] As the global aging process accelerates, the number of elderly people continues to rise, and the health and safety of the elderly are receiving increasing attention. Falling is a common accident for the elderly, which may cause serious physical injuries, such as fractures, craniocerebral injuries, etc., and even endanger their lives. It also puts a burden on families and equipment. Elderly fall technology uses advanced technical means to monitor the daily activities of the elderly in real time, detect fall events and issue alarms, so that rescue measures can be taken quickly to reduce the risks and injuries caused by falls.

[0003] In the prior art, many fall detection systems rely on a single type of sensor, and the detection only focuses on the body movements and states of the elderly themselves, while ignoring the impact of environmental factors on fall detection, making the detection dimension not comprehensive enough. In addition, the behavior of the elderly in daily life is relatively complex. In addition to the common fall postures, there are many ambiguous situations between normal movements and falls. The existing detection has a weak ability to recognize these complex movements, and it is difficult to accurately distinguish whether it is a change in normal behavior or a precursor to an impending fall, resulting in errors in detection accuracy. When the elderly fall, many systems only issue a simple alarm notification to inform family members or relevant personnel that the elderly have fallen, but lack detailed information about the specific circumstances of the fall. This makes it difficult for rescuers to quickly and comprehensively understand the situation on the scene after receiving the alarm, making it difficult to make adequate rescue preparations, which may delay the best time for rescue.

[0004] In view of this, a fall detection and help calling method for elderly people at home is needed. Summary of the invention

[0005] The embodiment of the present application provides a method for fall detection and calling for help for elderly people at home, which is used to solve the problem that fall detection is not accurate enough and calling for help is not timely.

[0006] The first aspect of the embodiment of the present application provides a fall detection and help calling method for elderly people at home, comprising:

[0007] Collect multi-source health monitoring data of the elderly through surveillance cameras and sensors, and perform preprocessing based on the collected data;

[0008] According to the preprocessed data, the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly are extracted respectively;

[0009] Inputting the elderly's walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features into a trained abnormal motion detection model for motion detection;

[0010] When the elderly have abnormal behaviors, clustering algorithms are used to conduct multi-dimensional analysis of their abnormal behaviors.

[0011] According to the preset help-calling rules, the analysis results of the elderly’s abnormal behavior and actions are fed back in real time.

[0012] Furthermore, the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly are extracted from the preprocessed data, including:

[0013] Walking posture characteristics:

[0014]

[0015] Where: t is the frame number in the video image, SA(t) is the stride of the tth frame, and are the frame numbers of the right and left feet landing respectively, for The coordinates of the center of gravity of the right foot landing on the ground, for The coordinates of the center of gravity of the frame where the left foot lands; θ knee (t) is the knee bending angle at the tth frame, is the coordinate of the left knee joint point in frame t, is the coordinate of the hip joint point in the tth frame; α toe (t) is the swing angle of the footsteps in the tth frame, (x toe (t), y toe (t)) is the coordinate of the trajectory point of the left foot tip during the swinging process in the tth frame, t start is the frame number at the start of the swing, (x toe (t start ), y toe (t start )) is t start The coordinates of the trajectory points of the left foot's toe during the swinging process.

[0016] Furthermore, the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly are extracted from the preprocessed data, including:

[0017] Standing up and sitting down posture characteristics:

[0018] Δyjoint (t) = y joint (t)-y joint (t start ′)

[0019]

[0020] Where: t start ′ is the starting frame detected in the video when the old man starts to stand up from a sitting position, y joint (t) is the vertical displacement of each joint point in frame t, Δy joint (t) is the change in vertical displacement of each joint point in frame t, y joint (t start ′) is t start ′ is the vertical displacement of each joint point in the frame; L is the sum of the horizontal movement distances of the shoulder and elbow joint points, t end ' is the frame number of the start and end, x joint (t) is the horizontal displacement of each joint point in frame t, x joint (t start ′) is t start 'The horizontal displacement of each joint point in the frame; is the descending speed of the body center of gravity before contacting the seat, t contact For the moment when the buttocks touch the seat, (t contact ) is t contact The coordinates of the body's center of gravity in the vertical direction, t start ″The coordinates of the body's center of gravity in the vertical direction, It is the time from the beginning of sitting down to the time when the buttocks touch the seat.

[0021] Furthermore, the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly are extracted from the preprocessed data, including:

[0022] Posture characteristics of daily activities:

[0023]

[0024] Where: θ back (t) is the bending angle of the neck and back at frame t, (x neck (t), y neck (t)) is the coordinate of the key node of the neck in frame t, (x waist (t), y waist (t)) is the coordinate of the waist key node of frame t, α waist (t) is the waist twist angle in frame t, is the starting frame of bending, (x waist (t)′,y waist (t)′) is the horizontal coordinate of the waist during the bending process of frame t, for The horizontal coordinate of the frame waist; L arm (t) is the arm extension length in frame t, (x wrist (t)″,y wrist (t)″) is the coordinate of the wrist joint node in frame t, (x shoulder (t), y shoulder (t)) is the coordinate of the shoulder joint node in frame t, θ shoulder (t) is the shoulder elevation angle, t reach，tart The starting frame when the reaching action begins. t reach，tart Frame the coordinates of the shoulder joint node.

[0025] Furthermore, the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly are extracted from the preprocessed data, including:

[0026] Sports Features:

[0027]

[0028]

[0029] in: and are the coordinates of the body center of gravity in frame t-1 and frame t, respectively, x (t) and v y (t) are the horizontal and vertical velocities of frame t, respectively, and Δt is the time interval between frames; v x (t-1) and v y (t-1) are the horizontal and vertical speeds of frame t-1, a x (t) and a y (t) are the accelerations in the horizontal and vertical directions of frame t, is the average speed in N frames, is the velocity variance in N frames; is the curvature change rate of frame t, k(t) and k(t-1) are the curvatures of the fitting curves of frame t and frame t-1.

[0030] Furthermore, the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly are extracted from the preprocessed data, including:

[0031] Acceleration characteristics and gyroscope characteristics:

[0032]

[0033]

[0034] Where: f is the frequency, A x (f) is the frequency domain vector representation, is the power spectrum density of the low frequency band of 0-5Hz, f s is the sampling frequency, for Power spectral density over frequency range; is the ratio of low-frequency power spectral density to total power spectral density, and are the starting and ending time of each impact event, T impact is the impact duration; M is the sampling point, is the mean angular velocity, ω x (t) is the angular velocity of frame t, is the variance of the angular velocity, is the root mean square error of the angular velocity.

[0035] Furthermore, the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly are extracted from the preprocessed data, including:

[0036] Heart rate features:

[0037]

[0038] Among them: SDNN is the standard deviation of the interval between adjacent heartbeats, RR i+1 and RR i are the i+1th and i adjacent heartbeat intervals of K heartbeat intervals respectively.

[0039] Furthermore, the walking posture characteristics, standing up and sitting posture characteristics, daily activity posture characteristics, movement characteristics, acceleration characteristics, gyroscope characteristics and heart rate characteristics of the elderly are input into the trained abnormal action detection model for action detection, including:

[0040] Normalizing the data corresponding to the elderly person's walking posture characteristics, standing up and sitting posture characteristics, daily activity posture characteristics, movement characteristics, acceleration characteristics, gyroscope characteristics, and heart rate characteristics;

[0041] Based on the normalized loading of each feature, the trained Gaussian mixture model is correspondingly constructed, and the probability of the feature vectors corresponding to the walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features in the model is calculated to determine the abnormal indicator variable;

[0042] Construct a hierarchical model with the elderly's behavior state as the target layer, and walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features as the criterion layer, and obtain the weight vectors corresponding to the walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features;

[0043] Obtaining a comprehensive abnormality index by performing weighted summation according to the abnormality index variable and the weight vector;

[0044] The comprehensive abnormal index is compared with a preset comprehensive threshold value. If it is greater than the threshold value, it is determined that the elderly person has abnormal behavior. If it is less than or equal to the threshold value, it is determined that the elderly person's behavior is normal.

[0045] Furthermore, when the elderly have abnormal behaviors, a clustering algorithm is used to perform multi-dimensional analysis on the elderly's abnormal behaviors, including:

[0046] Collect feature data before and after the elderly's abnormal movements, and perform dimensionality reduction processing on the collected feature data;

[0047] Use clustering algorithm to cluster the processed data, determine the number of clusters, initialize the cluster center and iterate clustering, and determine the primary category of abnormal behavior by analyzing the clustering results;

[0048] Divide the data into the first-level categories, select relevant features for each subset to train a decision tree model, and perform second-level category prediction;

[0049] Based on the results of the first-level and second-level categories, the cause of the abnormal behavior is determined in combination with feature data and environmental information, the time of occurrence is determined based on the timestamp, and the location of occurrence is determined using camera image analysis and home map coordinate conversion.

[0050] Furthermore, the analysis result of the abnormal behavior of the elderly is fed back in real time according to the preset rescue rules, including:

[0051] Develop detailed emergency call rules that include different abnormal behavior trigger conditions based on the individual situation of the elderly, and set the priority of each type of abnormal behavior;

[0052] After receiving the results of the analysis of the abnormal behavior of the elderly, the analysis results are matched with the preset rescue rules;

[0053] If the analysis results meet the distress call rules, an alert containing details of the abnormal behavior is sent to the guardian, and the smart device is controlled to respond;

[0054] The time, content, and recipient information of each feedback are updated and recorded in real time.

[0055] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0056] The present invention collects multi-source health monitoring data of the elderly through monitoring cameras and sensors, and performs preprocessing based on the collected data; according to the preprocessed data, the elderly's walking posture characteristics, standing up and sitting down posture characteristics, daily activity posture characteristics, movement characteristics, acceleration characteristics, gyroscope characteristics and heart rate characteristics are respectively extracted, and the extraction of these characteristics provides a specific quantitative basis for subsequent behavior analysis and health assessment, and can more carefully characterize the physical state of the elderly; the elderly's walking posture characteristics, standing up and sitting down posture characteristics, daily activity posture characteristics, movement characteristics, acceleration characteristics, gyroscope characteristics and heart rate characteristics are input into a trained abnormal action detection model for action detection, and the elderly's behavior can be monitored in real time or near real time. The model can accurately identify abnormal situations in the elderly's behavior by learning the characteristic patterns of a large amount of normal and abnormal behavior data; the clustering algorithm can classify abnormal behaviors with similar characteristics, and analyze the abnormal behaviors of the elderly from multiple dimensions; according to the preset call for help rules, it can ensure that an alarm can be issued in time when the elderly have abnormal behaviors, so as to achieve the purpose of comprehensive monitoring of the elderly's health status, timely detection of abnormalities, in-depth analysis of causes and rapid response to rescue.

[0057] Other advantages, objectives, and features of the present invention will be set forth in part in the following description, and in part will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 The present invention is a flowchart of an embodiment of a method for detecting a fall and calling for help for elderly people at home. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0060] The fall detection and help calling method for the elderly at home in this embodiment is used to improve the accuracy of fall detection and the timeliness of help calling for the elderly. The implementation method in this embodiment can be implemented in the system, can be implemented in the server, and can also be implemented in the terminal, and is not specifically limited.

[0061] Embodiment 1

[0062] See also Figure 1 , an embodiment of a fall detection and help calling method for elderly people at home in the present invention comprises the following steps:

[0063] S11. Collect multi-source health monitoring data of the elderly through monitoring cameras and sensors, and perform preprocessing based on the collected data;

[0064] In this embodiment, the multi-source health monitoring data includes monitoring camera data, acceleration sensor data, gyroscope sensor data and heart rate sensor data, wherein the monitoring camera data includes the body posture and movement trajectory of the elderly, obtains the position and posture information of various parts of the elderly body such as the head, limbs, trunk, etc., determines the range of activities of the elderly in the indoor environment, such as whether they are in the bedroom, living room or bathroom, and collects the interaction between the elderly and the surrounding objects, such as whether the furniture is used normally, whether they are tripped or collided with objects. Acceleration sensor data includes measuring the acceleration of the elderly body in the three-dimensional space of x, y, and z axes; calculating the speed of change of acceleration, that is, the first-order derivative of acceleration. Gyroscope sensor data is to detect the rotation of the elderly body, including the rotation of the head, trunk and limbs; according to the data obtained by the gyroscope, calculate the variance or standard deviation of the posture change of the body over a period of time. Heart rate sensor data includes real-time monitoring of the elderly's heart rate. The resting heart rate of a normal adult is generally 60-100 times / minute, and the heart rate of the elderly may be slightly lower; it also includes analyzing heart rate variability, which reflects the changes in adjacent heartbeat intervals.

[0065] S12. Extracting the walking posture features, standing and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly according to the preprocessed data;

[0066] Walking posture characteristics:

[0067]

[0068] Where: t is the frame number in the video image, SA(t) is the stride of the tth frame, and are the frame numbers of the right and left feet landing respectively, for The coordinates of the center of gravity of the right foot landing on the ground, for The coordinates of the center of gravity of the frame where the left foot lands; θ knee (t) is the knee bending angle at the tth frame, is the coordinate of the left knee joint point in frame t, is the coordinate of the hip joint point in the tth frame; α toe (t) is the swing angle of the footsteps in the tth frame, (x toe (t), y toe (t)) is the coordinate of the trajectory point of the left foot tip during the swinging process in the tth frame, t start is the frame number at the start of the swing, (x toe (t start ), y toe (t start )) is t start The coordinates of the trajectory points of the left foot's toe during the swinging process.

[0069] Standing up and sitting down posture characteristics:

[0070] Δy joint (t) = y joint (t)-y joint (t start ′)

[0071]

[0072] Where: t start ′ is the starting frame detected in the video when the old man starts to stand up from a sitting position, y joint (t) is the vertical displacement of each joint point in frame t, Δy joint (t) is the change in vertical displacement of each joint point in frame t, y joint (t start ′) is t start ′ is the vertical displacement of each joint point in the frame; L is the sum of the horizontal movement distances of the shoulder and elbow joint points, t end ' is the frame number of the start and end, x joint (t) is the horizontal displacement of each joint point in frame t, x joint (t start ′) is t start 'The horizontal displacement of each joint point in the frame; is the descending speed of the body center of gravity before contacting the seat, t contact For the moment when the buttocks touch the seat, (t contact ) is t contact The coordinates of the body's center of gravity in the vertical direction, t start ″The coordinates of the body's center of gravity in the vertical direction, It is the time from the beginning of sitting down to the time when the buttocks touch the seat.

[0073] Posture characteristics of daily activities:

[0074]

[0075] Where: θ back (t) is the bending angle of the neck and back at frame t, (x neck (t), y neck (t)) is the coordinate of the key node of the neck in frame t, (x waist (t), y waist (t)) is the coordinate of the waist key node of frame t, α waist (t) is the waist twist angle in frame t, is the starting frame of bending, (x waist (t)′,y waist (t)′) is the horizontal coordinate of the waist during the bending process of frame t, for The horizontal coordinate of the frame waist; L arm (t) is the arm extension length in frame t, (x wrist (t)″,y wrist (t)″) is the coordinate of the wrist joint node in frame t, (x shoulder (t), y shoulder (t)) is the coordinate of the shoulder joint node in frame t, θ shoulder (t) is the shoulder elevation angle, t reach，tart The starting frame when the reaching action begins. t reach，tart Frame the coordinates of the shoulder joint node.

[0076] Sports Features:

[0077]

[0078] in: and are the coordinates of the body center of gravity in frame t-1 and frame t, respectively, x (t) and v y (t) are the horizontal and vertical velocities of frame t, respectively, and Δt is the time interval between frames; v x (t-1) and v y (t-1) are the horizontal and vertical speeds of frame t-1, a x (t) and a y (t) are the accelerations in the horizontal and vertical directions of frame t, is the average speed in N frames, is the velocity variance in N frames; is the curvature change rate of frame t, k(t) and k(t-1) are the curvatures of the fitting curves of frame t and frame t-1.

[0079] Acceleration characteristics and gyroscope characteristics:

[0080]

[0081]

[0082] Where: f is the frequency, A x (f) is the frequency domain vector representation, is the power spectrum density of the low frequency band of 0-5Hz, f s is the sampling frequency, for Power spectral density over frequency range; is the ratio of low-frequency power spectral density to total power spectral density, and are the starting and ending time of each impact event, T impact is the impact duration; M is the sampling point, is the mean angular velocity, ω x (t) is the angular velocity of frame t, is the variance of the angular velocity, is the root mean square error of the angular velocity.

[0083] Heart rate features:

[0084]

[0085] Among them: SDNN is the standard deviation of the interval between adjacent heartbeats, RR i+1 and RR i are the i+1th and i adjacent heartbeat intervals of K heartbeat intervals respectively.

[0086] S13. Input the elderly's walking posture characteristics, standing and sitting posture characteristics, daily activity posture characteristics, movement characteristics, acceleration characteristics, gyroscope characteristics and heart rate characteristics into the trained abnormal motion detection model for motion detection;

[0087] In this embodiment, step S13 further includes the following:

[0088] 1. Normalize the data corresponding to the elderly's walking posture characteristics, standing up and sitting posture characteristics, daily activity posture characteristics, movement characteristics, acceleration characteristics, gyroscope characteristics and heart rate characteristics;

[0089] Since the numerical range and dimension of different features may be different, all features need to be normalized first. The normalized features are combined into a feature vector in a certain order.

[0090] 2. Load the corresponding trained Gaussian mixture model based on each feature after normalization, and calculate the probability of the feature vectors corresponding to the walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features in the model to determine the abnormal indicator variable;

[0091] Load the trained Gaussian mixture model, which is trained on a large amount of normal behavior data and contains multiple Gaussian distribution components, each of which represents the probability distribution of a feature vector under a normal behavior mode. Each Gaussian distribution is parameterized by a mean vector, a covariance matrix, and a weight. For the newly constructed feature vector containing various characteristics of the elderly, calculate its probability in the GMM. Set a probability threshold, which can be determined by experimenting on the validation data set. If the probability density of the Gaussian mixture model is less than the set threshold, it is determined that an abnormal behavior has occurred; otherwise, the elderly's behavior is considered normal.

[0092] Each feature is normalized separately, and the method is the same as the normalization step in the feature fusion method to ensure that the value range of each feature is in the appropriate range for subsequent comparison. Load the GMM trained for walking posture features. The model is trained based on normal walking posture data and is also composed of multiple Gaussian distribution components. The parameters of each component are the mean vector, covariance matrix and weight in the walking posture feature space. Extract the current elderly walking posture feature vector, such as a vector containing only step length, stride, knee bending angle and footstep swing angle, and calculate its probability in the walking posture GMM according to the probability density function formula of the above Gaussian mixture model. Set the probability threshold of the walking posture feature. If the calculated probability is less than the threshold, it is determined that there is an abnormality in the walking posture. According to the same steps as the walking posture feature comparison, load the GMM trained for the standing and sitting posture features, calculate the probability of the current elderly standing and sitting posture feature vectors in the model, set the probability threshold of the standing and sitting posture features, and determine whether there is an abnormality in the standing and sitting posture. Load the corresponding GMM, calculate the probability, set the threshold, and make an abnormal judgment on the daily activity posture. Load the motion feature GMM for the motion feature, calculate the probability, set the threshold, and determine whether the motion state is abnormal for the motion feature. Similarly, determine the threshold and indicator variable for the acceleration feature, gyroscope and heart rate feature.

[0093] 3. Construct a hierarchical model with the elderly's behavior state as the target layer, and walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features as the criterion layer, and obtain the weight vectors corresponding to the walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features;

[0094] Construct a hierarchical model: Take the behavior state of the elderly as the target layer, and each feature, namely walking posture, standing up and sitting down posture, daily activity posture, movement, acceleration, gyroscope, and heart rate, as the criterion layer to establish a hierarchical model. Through expert experience or analysis of a large amount of historical data, compare the relative importance of each feature in judging whether the elderly's behavior is abnormal, and construct a judgment matrix. Use the eigenvector method to solve the maximum eigenroot of the judgment matrix and its corresponding eigenvector, and normalize the eigenvector to obtain the weight vector of each feature.

[0095] 4. Obtain a comprehensive anomaly index by performing weighted summation based on the anomaly index variable and the weight vector;

[0096] 5. Compare the comprehensive abnormal index with the preset comprehensive threshold. If it is greater than the threshold, it is determined that the elderly person has abnormal behavior. If it is less than or equal to the threshold, it is determined that the elderly person's behavior is normal.

[0097] According to the weight vector and indicator variable of each feature, the comprehensive abnormality index S is calculated:

[0098]

[0099] Among them: i traverses all features, w i is the weight corresponding to the i-th feature, I i is the ith eigenvector.

[0100] A comprehensive threshold is set, where the threshold is determined through experiments and data analysis. If the comprehensive abnormal index is greater than the threshold, it is determined that the elderly has abnormal behavior; if it is less than or equal to the threshold, the elderly's behavior is considered normal.

[0101] Through the above steps, on the basis of comparing each feature separately, through adaptive threshold adjustment and feature weight determination based on hierarchical analysis method, it is possible to more accurately judge whether the elderly have abnormal behavior, reduce misjudgment and missed judgment, and improve the reliability and practicality of detection.

[0102] S14. When the elderly have abnormal behaviors, use clustering algorithms to conduct multi-dimensional analysis of the elderly’s abnormal behaviors;

[0103] In this embodiment, step S14 includes the following:

[0104] 1. Collect feature data before and after the elderly's abnormal movements, and perform dimensionality reduction processing on the collected feature data;

[0105] Considering that there may be more feature dimensions, in order to improve clustering efficiency and effect, the features can be selected or reduced in dimension first.

[0106] 2. Use the clustering algorithm to cluster the processed data, determine the number of clusters, initialize the cluster center and iterate the clustering, and determine the primary category of abnormal behavior by analyzing the clustering results;

[0107] Based on the initial understanding and experience of abnormal behavior, the appropriate number of clusters is determined through some clustering effectiveness indicators. K initial cluster centers are randomly selected, and K samples can be randomly selected from the feature data set as the initial centers. Each feature vector is assigned to the category to which the nearest cluster center belongs, and the new cluster center of each category is calculated. This process is repeated until the cluster center no longer changes significantly. The clustering results are analyzed, and a primary category label of abnormal behavior is determined for each cluster based on the feature distribution and actual meaning of the samples in each cluster, combined with medical knowledge and daily life experience.

[0108] 3. Divide the data into the first-level categories, select relevant features for each subset to train a decision tree model, and perform second-level category prediction;

[0109] According to the clustering results, the feature data set is divided into subsets corresponding to the first-level categories. Each subset contains all sample feature vectors and corresponding annotation information belonging to the first-level category. For each first-level category subset, feature selection is performed again to select feature subsets with high correlation with the second-level category division under the first-level category. For example, for "fall-related abnormalities", body tilt angle, arm support action features, body contact part features, etc. are selected as input features of the decision tree. Then these feature subsets and corresponding second-level category annotation information, such as falling forward, falling backward, and falling to the side, are used to train the decision tree model. Feature selection indicators such as information gain, gain rate, and Gini index are used to select the optimal feature subset, and the decision tree algorithm is used for model training. By recursively dividing the data set, a decision tree structure is constructed. Each internal node represents a feature test condition, each branch represents a value or value range of the feature, and the leaf node represents the final second-level category prediction result.

[0110] Use the trained decision tree model to predict the secondary category of the unlabeled samples under the corresponding primary category. According to the branch path of the decision tree, starting from the root node, judge in sequence according to the characteristic values ​​of the sample, and finally reach the leaf node to obtain the secondary category prediction result of the sample.

[0111] 4. Based on the results of the primary and secondary categories, combine the feature data and environmental information to determine the cause of the abnormal behavior, determine the time of occurrence based on the timestamp, and determine the location of occurrence using camera image analysis and home map coordinate conversion.

[0112] Based on the determined primary and secondary categories of abnormal behavior actions, combined with various feature data extracted previously and possible environmental information, such as ground humidity and light intensity obtained through environmental sensors, and the distribution of surrounding objects analyzed from camera images, the cause of the abnormal behavior actions is further inferred.

[0113] The exact time when the abnormal behavior occurred is determined based on the timestamp information of the data collection. For example, the data collection system collects and records various sensor data and video image frames at a fixed time interval, such as once per second. When an abnormal behavior is detected, the time when the abnormal behavior occurred can be determined by recording the timestamp of that moment, such as "10:30:20 on December 25, 2024".

[0114] The location where the abnormal behavior occurred is determined by combining camera image analysis with the coordinate mapping relationship of the home map. First, the position coordinates of the elderly in the image are determined by analyzing the camera image. For example, the pixel coordinates of the center of the elderly's body in the image are obtained by the target detection algorithm, with the upper left corner of the image as the origin. Then, according to the pre-established camera view and the coordinate conversion model of the home map, the image coordinates are converted into actual physical coordinates on the home map, such as two-dimensional plane coordinates with a corner of the room as the origin, so as to determine the specific location of the elderly in the home environment, such as "2 meters in front of the sofa in the living room".

[0115] The above steps use the clustering decision tree algorithm to conduct a comprehensive analysis of the abnormal behaviors of the elderly, which can more accurately determine the specific cause, time and location of the abnormal behaviors, provide important basis and support for subsequent rescue and care measures, and ensure the life safety and health of the elderly.

[0116] S15. Provide real-time feedback on the analysis results of the elderly's abnormal behavior according to the preset rescue rules.

[0117] In this embodiment, step S15 also includes the following:

[0118] 1. Develop detailed emergency call rules that include different abnormal behavior trigger conditions based on the individual situation of the elderly, and set the priority of each type of abnormal behavior;

[0119] Develop detailed rules for calling for help based on the elderly's health conditions, living habits, and possible abnormal behaviors. For example: For fall-related abnormalities, if you fall forward and the angle of fall exceeds 60 degrees, or fall backward and your head hits the ground, you should call for help immediately. For syncope-related abnormalities, if the heart rate drops by more than 30 / minute and lasts for more than 30 seconds, you need to call for help urgently. For sudden illness-related abnormalities, such as heart attack symptoms, you should contact emergency personnel immediately. For accidental slip-related abnormalities, if the elderly cannot get up on their own within 5 minutes after slipping, send out a distress signal. Set priorities for different types of abnormal behaviors. For example, sudden illness-related abnormalities usually have the highest priority, followed by falls that may cause serious injuries, and then syncope and accidental slips.

[0120] 2. After receiving the results of the analysis of the abnormal behavior of the elderly, the analysis results are matched with the preset rescue rules;

[0121] Receive the analysis results of abnormal behavior of the elderly obtained by clustering decision tree algorithm or other analysis methods, including the primary category of abnormal behavior, such as falling, fainting, sudden illness, accidental slipping, etc., the secondary category such as falling forward, fainting caused by heart attack, etc., the specific cause, time and location of occurrence. Match the analysis results with the preset rescue rules. If the analysis results show that the elderly fainted due to a heart attack, check whether the symptoms meet the rescue rules for a heart attack; if the analysis results show that the elderly fell forward and the fall angle reached 70 degrees, check whether the rescue conditions for falling are met.

[0122] 3. If the analysis results meet the distress call rules, an alert containing details of the abnormal behavior is sent to the guardian, and the smart device is controlled to respond;

[0123] If the analysis results match the rules that require calling for help, an alarm message will be immediately sent to the guardian at home through the internal communication system. The alarm message should include the type, location and time of the elderly's abnormal behavior. At the same time, the relevant smart devices will be controlled to respond. If the situation is more urgent and meets the preset external call for help conditions, a distress signal will be sent to community medical staff through the community medical emergency call system. The distress signal should contain detailed health information, location and urgency of the elderly. For serious sudden illnesses or abnormal behaviors that may endanger life, the emergency center will be called directly and automatically, and accurate location information and description of the elderly's condition will be provided.

[0124] 4. Update and record the time, content, and recipient information of each feedback in real time.

[0125] Record each feedback, including feedback time, feedback content, person or system receiving feedback, etc. These records are helpful for subsequent analysis of the effect of the call for help and changes in the health status of the elderly. After receiving the call for help and handling the elderly's abnormal situation, the guardian or professional medical staff will feedback the processing results to the system so that the system can update the elderly's health records and adjust the call for help rules.

[0126] The above-mentioned embodiment collects and pre-processes multi-source health data of the elderly through monitoring cameras and sensors, and extracts multiple types of posture, movement, and physiological characteristics on this basis. These characteristics are input into the abnormal action detection model for real-time monitoring. When abnormal behavior is found, the clustering algorithm is used to deeply analyze the cause. Finally, the analysis results are fed back in real time according to the preset call for help rules, so as to achieve the purpose of comprehensive monitoring of the health status of the elderly, timely detection of abnormalities, in-depth analysis of the cause and rapid response to rescue.

[0127] Those of ordinary skill in the art will appreciate that the units of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0128] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored. In addition, each functional unit in each embodiment of the present invention can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units.

[0129] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nlyMemory), random access memory (RAM, RandomAccessMemory), mobile hard disk, magnetic disk or optical disk, etc., which can store program code.

[0130] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A method for detecting falls and calling for help for elderly people at home, characterized in that: include: Collect multi-source health monitoring data of the elderly through surveillance cameras and sensors, and perform preprocessing based on the collected data; According to the preprocessed data, the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly are extracted respectively; Inputting the elderly's walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features into a trained abnormal motion detection model for motion detection; When the elderly have abnormal behaviors, clustering algorithms are used to conduct multi-dimensional analysis of their abnormal behaviors. According to the preset help-calling rules, the analysis results of the elderly’s abnormal behavior and actions are fed back in real time.

2. The fall detection and help calling method for elderly people at home according to claim 1 is characterized in that: The method extracts the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly from the pre-processed data, including: Walking posture characteristics: Where: t is the frame number in the video image, SA(t) is the stride of the tth frame, and are the frame numbers of the right and left feet landing respectively, for The coordinates of the center of gravity of the right foot landing on the ground, for The coordinates of the center of gravity of the frame where the left foot lands; θ knee (t) is the knee bending angle at the tth frame, is the coordinate of the left knee joint point in frame t, is the coordinate of the hip joint point in the tth frame; α toe (t) is the swing angle of the footsteps in the tth frame, (x toe (t),y toe (t)) is the coordinate of the trajectory point of the left foot tip during the swinging process in the tth frame, t start is the frame number at the start of the swing, (x toe (t start ),y toe (t start )) is t start The coordinates of the trajectory points of the left foot's toe during the swinging process.

3. The fall detection and help calling method for elderly people at home according to claim 2, characterized in that: The method extracts the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly from the pre-processed data, including: Standing up and sitting down posture characteristics: Δy joint (t)=y joint (t)-y joint (t start ′) Where: t start ′ is the starting frame detected in the video when the old man starts to stand up from a sitting position, y joint (t) is the vertical displacement of each joint point in frame t, Δy joint (t) is the change in vertical displacement of each joint point in frame t, y joint (t start ′) is t start ′ is the vertical displacement of each joint point in the frame; L is the sum of the horizontal movement distances of the shoulder and elbow joint points, t end ' is the frame number of the start and end, x joint (t) is the horizontal displacement of each joint point in frame t, x joint (t start ′) is t start 'The horizontal displacement of each joint point in the frame; is the descending speed of the body center of gravity before contacting the seat, t contact For the moment when the buttocks touch the seat, t contact The coordinates of the body's center of gravity in the vertical direction, t start ″The coordinates of the body's center of gravity in the vertical direction, It is the time from the beginning of sitting down to the time when the buttocks touch the seat.

4. The fall detection and help calling method for elderly people at home according to claim 3, characterized in that: The method extracts the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly from the pre-processed data, including: Posture characteristics of daily activities: Where: θ back (t) is the bending angle of the neck and back at frame t, (x neck (t),y neck (t)) is the coordinate of the key node of the neck in frame t, (x waist (t),y waist (t)) is the coordinate of the waist key node of frame t, α waist (t) is the waist twist angle in frame t, is the starting frame of bending, (x waist (t)′,y waist (t)′) is the horizontal coordinate of the waist during the bending process of frame t, for The horizontal coordinate of the frame waist; L ar m(t) is the arm extension length in frame t, (x wrist (t)″,y wrist (t)″) is the coordinate of the wrist joint node in frame t, (x shoulder (t),y shoulder (t)) is the coordinate of the shoulder joint node in frame t, θ shoulder (t) is the shoulder elevation angle, t reach,tart The starting frame when the reaching action begins. t reach,tart Frame the coordinates of the shoulder joint node.

5. The method for detecting a fall and calling for help for an elderly person at home according to claim 4, characterized in that: The method extracts the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly from the pre-processed data, including: Sports Features: in: and are the coordinates of the body center of gravity in frame t-1 and frame t, respectively, x (t) and v y (t) are the horizontal and vertical velocities of frame t, respectively, and Δt is the time interval between frames; v x (t-1) and v y (t-1) are the horizontal and vertical speeds of frame t-1, a x (t) and a y (t) are the accelerations in the horizontal and vertical directions of frame t, is the average speed in N frames, is the velocity variance in N frames; is the curvature change rate of frame t, k(t) and k(t-1) are the curvatures of the fitting curves of frame t and frame t-1.

6. The method for detecting a fall and calling for help for an elderly person at home according to claim 5, characterized in that: The method extracts the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly from the pre-processed data, including: Acceleration characteristics and gyroscope characteristics: Where: f is the frequency, A x (f) is the frequency domain vector representation, is the power spectrum density of the low frequency band of 0-5Hz, f s is the sampling frequency, for Power spectral density over frequency range; is the ratio of low-frequency power spectral density to total power spectral density, and are the starting and ending time of each impact event, T impact is the impact duration; M is the sampling point, is the mean angular velocity, ω x (t) is the angular velocity of frame t, is the variance of the angular velocity, is the root mean square error of the angular velocity.

7. The method for detecting a fall and calling for help for an elderly person at home according to claim 5, characterized in that: The method extracts the walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features and heart rate features of the elderly from the pre-processed data, including: Heart rate features: Among them: SDNN is the standard deviation of the interval between adjacent heartbeats, RR i+1 and RR i are the i+1th and i adjacent heartbeat intervals of K heartbeat intervals respectively.

8. The method for detecting a fall and calling for help for an elderly person at home according to claim 1, characterized in that: The step of inputting the elderly's walking posture features, standing up and sitting posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features into a trained abnormal motion detection model for motion detection includes: Normalizing the data corresponding to the elderly person's walking posture characteristics, standing up and sitting posture characteristics, daily activity posture characteristics, movement characteristics, acceleration characteristics, gyroscope characteristics, and heart rate characteristics; Based on the normalized loading of each feature, the trained Gaussian mixture model is correspondingly constructed, and the probability of the feature vectors corresponding to the walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features in the model is calculated to determine the abnormal indicator variable; Construct a hierarchical model with the elderly's behavior state as the target layer, and walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features as the criterion layer, and obtain the weight vectors corresponding to the walking posture features, standing up and sitting down posture features, daily activity posture features, movement features, acceleration features, gyroscope features, and heart rate features; Obtaining a comprehensive abnormality index by performing weighted summation according to the abnormality index variable and the weight vector; The comprehensive abnormal index is compared with a preset comprehensive threshold value. If it is greater than the threshold value, it is determined that the elderly person has abnormal behavior. If it is less than or equal to the threshold value, it is determined that the elderly person's behavior is normal.

9. The fall detection and help calling method for elderly people at home according to claim 1, characterized in that: When the elderly have abnormal behaviors, a clustering algorithm is used to perform multi-dimensional analysis on the elderly's abnormal behaviors, including: Collect feature data before and after the elderly's abnormal movements, and perform dimensionality reduction processing on the collected feature data; Use clustering algorithm to cluster the processed data, determine the number of clusters, initialize the cluster center and iterate clustering, and determine the primary category of abnormal behavior by analyzing the clustering results; Divide the data into the first-level categories, select relevant features for each subset to train a decision tree model, and perform second-level category prediction; Based on the results of the first-level and second-level categories, the cause of the abnormal behavior is determined in combination with feature data and environmental information, the time of occurrence is determined based on the timestamp, and the location of occurrence is determined using camera image analysis and home map coordinate conversion.

10. The method for detecting a fall and calling for help for an elderly person at home according to claim 9, characterized in that: The analysis result of the abnormal behavior of the elderly is fed back in real time according to the preset rescue rules, including: Develop detailed emergency call rules that include different abnormal behavior trigger conditions based on the individual situation of the elderly, and set the priority of each type of abnormal behavior; After receiving the results of the analysis of the abnormal behavior of the elderly, the analysis results are matched with the preset rescue rules; If the analysis results meet the distress call rules, an alert containing details of the abnormal behavior is sent to the guardian, and the smart device is controlled to respond; The time, content, and recipient information of each feedback are updated and recorded in real time.

Images