CN111325946A - Fall detection method and system based on edge calculation - Google Patents

Fall detection method and system based on edge calculation Download PDF

Info

Publication number
CN111325946A
CN111325946A CN202010061335.XA CN202010061335A CN111325946A CN 111325946 A CN111325946 A CN 111325946A CN 202010061335 A CN202010061335 A CN 202010061335A CN 111325946 A CN111325946 A CN 111325946A
Authority
CN
China
Prior art keywords
time
falling
edge
fall
smv
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010061335.XA
Other languages
Chinese (zh)
Inventor
吴江
王嘉乐
刘琴
彭雪薇
王金燕
徐伟强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Sci Tech University ZSTU
Zhejiang University of Science and Technology ZUST
Original Assignee
Zhejiang Sci Tech University ZSTU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang Sci Tech University ZSTU filed Critical Zhejiang Sci Tech University ZSTU
Priority to CN202010061335.XA priority Critical patent/CN111325946A/en
Publication of CN111325946A publication Critical patent/CN111325946A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
    • G08B21/02—Alarms for ensuring the safety of persons
    • G08B21/04—Alarms for ensuring the safety of persons responsive to non-activity, e.g. of elderly persons
    • G08B21/0438—Sensor means for detecting
    • G08B21/0446—Sensor means for detecting worn on the body to detect changes of posture, e.g. a fall, inclination, acceleration, gait
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
    • G08B21/02—Alarms for ensuring the safety of persons
    • G08B21/04—Alarms for ensuring the safety of persons responsive to non-activity, e.g. of elderly persons
    • G08B21/0407—Alarms for ensuring the safety of persons responsive to non-activity, e.g. of elderly persons based on behaviour analysis
    • G08B21/043—Alarms for ensuring the safety of persons responsive to non-activity, e.g. of elderly persons based on behaviour analysis detecting an emergency event, e.g. a fall
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/12—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Gerontology & Geriatric Medicine (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Emergency Management (AREA)
  • Business, Economics & Management (AREA)
  • Computing Systems (AREA)
  • Social Psychology (AREA)
  • Psychology (AREA)
  • Psychiatry (AREA)
  • Medical Informatics (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

The invention discloses a falling detection method and system based on edge calculation, wherein the method comprises the steps that a user terminal collects triaxial acceleration data of user actions in real time; the method comprises the steps that triaxial acceleration data collected in real time are transmitted to an edge side, and whether a user falls is judged through a falling judgment model of the edge side; and if the user is judged to fall down, the edge side sends alarm information to the cloud platform. The system comprises an intelligent terminal, an edge detection computing end and a cloud platform; the intelligent terminal comprises a three-axis acceleration sensor and is used for acquiring three-axis acceleration data of the action of the wearer and sending the three-axis acceleration data to the edge detection computing end; and the edge detection calculation end calculates and judges whether the intelligent terminal wearer falls or not by using the falling judgment model, and sends alarm information to the cloud platform if the intelligent terminal wearer falls. The method has the advantages that the falling judgment is placed at the edge side close to the user instead of the cloud side, so that the response speed of the falling behavior of the old people can be improved, and the bandwidth bottleneck caused by the accumulation of cloud side data can be avoided.

Description

Fall detection method and system based on edge calculation
Technical Field
The invention belongs to the field of medical detection, and particularly relates to a falling detection method and system based on edge calculation.
Background
The aging of population is an important characteristic of the current society in China, and with the increase of the age of the old, the physical function level of the old is gradually declined, and the health risk is gradually increased. Especially, in the single-living homes of a great number of old people, once the old people fall down, if the old people cannot be found in time and corresponding rescue measures are taken, serious physical injuries such as fracture, bleeding, nerve injury, paralysis and the like can be caused. According to statistics, the fall is the leading cause of the unexpected death of the old aged over 65 years old. Moreover, the old people who experience the fall injury often reduce the outgoing activities due to fear of falling again, so that psychological concerns and social isolation further increase the falling risk of the old people. For society, the falling of the old people brings huge economic burden, and according to statistics, the direct medical cost of China caused by the falling of the old people every year reaches more than 50 hundred million yuan. Therefore, if the falling behavior of the old people can be detected in time at the first time when the old people fall, the old people can be effectively treated at the first time, the serious injury caused by the fact that the old people fall and are not found in time is avoided, and huge economic benefits and social benefits are brought in an increasingly aging society.
The current human body fall detection devices are roughly divided into three categories: the first type is a device actively triggered by a user, and needs the old to have conscious consciousness after falling down to trigger a button to alarm; the second type mainly carries out detection and identification through a camera, so that the user experience is better, but the detection range is limited, and the device is generally limited indoors; the third category is alarm devices, which are triggered primarily by sensors in the device, and the range of motion of the wearer is relatively wide. The third type of fall detection device has less limitation on the activity range, and gradually becomes the mainstream along with the development of wearable technology and internet of things technology, and the most important way in the device at present is to complete three-dimensional axis building of acceleration through a three-axis sensor or gyroscope, then process data according to different algorithms, and finally judge the fall. In the detection algorithm, the most direct method is to adopt a threshold method, namely, when the acceleration in three axial directions exceeds a certain threshold value, the user is judged to fall. The fall detection based on the threshold is easy to realize, the calculation efficiency is high, but the fault tolerance of different individuals is poor, and the detection precision is low. With the development of artificial intelligence technology, more complex algorithms are more and more apt to be used for improving the detection accuracy of the human body falling behavior.
In recent years, some researchers use a deep learning algorithm for human body fall detection, for example, chinese patent No. 201610509618.X, and design and implement a method for detecting old people fall based on machine learning and a detection system thereof, wherein a dictionary learning is used for constructing fall feature vectors, and a random forest classifier is used for fall determination. In the chinese patent with the patent application number of 201510232988.9, a fall detection system based on machine learning is designed, which is composed of a fall detector and a cloud platform, and a fall detection algorithm used by the fall detector next time is generated on the cloud platform according to a fall detection sample. Because a deep learning algorithm needs a large number of training samples, and a cloud server often serves many users and is far away from the users, great challenges are brought to transmission bandwidth and timeliness, and therefore, for medical applications with high real-time processing urgency, the bottleneck effect of the cloud server end becomes a problem which needs to be solved urgently.
Disclosure of Invention
The invention aims to solve the problem of bottleneck effect of a cloud server end in medical application at present, and provides a method and a system for fall detection based on edge calculation.
In order to achieve the purpose, the invention adopts the following technical scheme:
an edge calculation based fall detection method comprising the steps of:
s1: the user terminal collects triaxial acceleration data of user actions in real time;
s2: the method comprises the steps that triaxial acceleration data collected in real time are transmitted to an edge side, and whether a user falls is judged through a falling judgment model of the edge side;
s3: and if the user is judged to fall down, the edge side sends alarm information to the cloud platform.
Preferably, the fall decision model is a thresholding model, and the definition feature quantity is composed of three parameters: combining an acceleration semaphore vector SMV, a continuous occurrence quantity N of suspected falling data and an acceleration vector A in the vertical direction of the human body; the method comprises the following calculation and judgment steps:
s21: setting a threshold value (T) of the characteristic quantitySMV,TN,TG) Wherein, TSMVThreshold value of SMV to distinguish between falling and non-falling actions, TNA threshold value, T, for the amount of persistent occurrences, N, of suspected fall dataGA threshold value of an acceleration vector A in the vertical direction of the human body;
s22: calculating the value of real-time characteristic quantity (SMV, N, A) according to the triaxial acceleration data of the user action collected in real time;
s23: and comparing the value of the real-time characteristic quantity with a characteristic quantity threshold value, and judging whether the action falls: if SMV is satisfied at the same time>TSMV，N<TNAnd A<TGThen, the user is determined to fall.
Preferably, step S21 includes:
obtaining three-axis acceleration sample data of falling and non-falling actions of the user, and calculating SMV and T according to the three-axis acceleration sample dataSMVThe median of the smallest falling SMV and the largest SMV in most non-falling behaviors; t isNIs 50, TGIs 0.2G.
The other technical scheme adopted by the invention is as follows:
a fall detection method based on edge calculation is characterized in that a fall judgment model is a recurrent neural network model, the size of an input layer is set to be 45, the number of nodes of a hidden layer is set to be 13, and the size of an output layer is set to be 1; the method comprises the following calculation and judgment steps:
s201: training a recurrent neural network;
s202: inputting the real-time triaxial acceleration data into an input layer of the recurrent neural network, and judging whether the action falls down by using the trained recurrent neural network.
Preferably, step S201 includes:
s2011: initializing a weight matrix randomly between 0 and 1, and inputting triaxial acceleration sample data acquired by a sensor into an input layer;
s2012: computing
s1＝Ux1+Wh0
h1＝f(s1)
y1＝g(Vh1)
In the above formula, x1Is the input unit of the 1 st sampling time, h0Is an initial hidden layer matrix, s1Is an intermediate variable at the 1 st sampling instant, h1Is the hidden layer matrix at the 1 st sampling instant, y1Is the output layer result at the 1 st sampling moment, and U, V and W are weight matrixes used for updating errors; f is sigmoid function
Figure BDA0002374597680000031
g is the softmax function
Figure BDA0002374597680000032
As time goes on, h1And (3) taking the memory state as the upper layer to participate in the next fall behavior prediction:
s2＝Ux2+Wh1
h2＝f(s2)
y2＝g(Vh2)
in the above formula, x2Is the input unit of the 2 nd sampling instant, s2Is an intermediate variable at the 2 nd sampling instant, h2Is the hidden layer matrix at the 2 nd sampling instant, y2Is the output layer result at the 2 nd sampling time, and U, V and W are weight matrixes used for updating errors; and so on, calculating:
st＝Uxt+Wht-1
ht＝f(st)
yt＝g(Vht)
in the above formula, xtIs the input unit at the t-th sampling instant, stIs an intermediate variable at the t-th sampling instant, ht-1Is the hidden layer matrix at the t-1 th sampling time, htIs the hidden layer matrix at the t-th sampling instant, ytThe output layer result at the t-th sampling moment, and U, V and W are weight matrixes used for updating errors;
s2013: each time an output layer result y is obtainedtThen, the output layer error e is calculatedoAnd hidden layer error eh：
e0(t)＝ot-yt
eh(t)＝dh(e0(t)TV,t)
dh＝xst(1-st)
In the above formula, otIs an actual class label with a value of 0 or 1, e0(t) output layer error at time t, eh(t) hidden layer error at time t, dhIs a hidden layer error update formula; x is an input unit xtA vectorized representation of;
s2014: updating the weight matrix each time by using the output layer error and the hidden layer error:
Figure BDA0002374597680000041
Figure BDA0002374597680000042
Figure BDA0002374597680000051
in the above formula, U (t), V (t), W (t) are weight matrices for updating errors at time t, U (t +1), V (t +1), and W (t +1) are weight matrices for updating errors at time t +1, z is a time mark, eh(t-z)，xt-zAnd h (t-z-1) respectively represent the hidden layer error at time t-z, the input cell at time t-z, and the hidden layer matrix at time t-z, α is the learning rate, β is the normalization parameter.
S2015: will output layer error eoAs an indicator of the effectiveness of the training algorithm, when eoIs less than a threshold TeWhen so, the training is finished.
Preferably, the alarm information includes the name, sex, age, medical record number, current location, and the like of the elderly who are currently falling.
In order to achieve the purpose of the invention, the invention also adopts the following technical scheme:
a fall detection system based on edge computing comprises an intelligent terminal, an edge detection computing end and a cloud platform; the intelligent terminal comprises a three-axis acceleration sensor and is used for acquiring three-axis acceleration data of the action of the wearer and sending the three-axis acceleration data to the edge detection computing end; the edge detection computing end computes the received triaxial acceleration data by using a falling judgment model and judges whether the intelligent terminal wearer falls or not, and if the intelligent terminal wearer falls, alarm information is sent to the cloud platform; the cloud platform is used for receiving the falling alarm information sent by the edge detection computing end and maintaining the basic data of the intelligent terminal wearer.
Preferably, the edge detection computing end comprises an intelligent gateway, a fall detection power unit and a rule engine; the intelligent gateway is used for receiving the triaxial acceleration data sent by the intelligent terminal and sending the triaxial acceleration data to the fall detection power measuring unit; the fall detection power unit processes the triaxial acceleration data according to an algorithm selected by the rule engine, judges whether the intelligent terminal wearer falls or not, and sends a judgment result to the intelligent gateway; and if the judgment result is that the intelligent gateway falls down, the intelligent gateway sends alarm information to the cloud platform.
Preferably, the cloud platform comprises a cloud management server, communicates with the edge detection computing terminal, and maintains basic information, medical information and fall data of all monitored intelligent terminal wearers in a responsible area.
Preferably, the cloud management server is further configured to configure a rules engine to select a fall detection method.
Compared with the prior art, the invention has the beneficial effects that: by the edge calculation-based falling detection method and system, falling judgment is placed at the edge side close to a user instead of a remote cloud, so that the response speed of falling behaviors of the old people can be increased, and bandwidth bottleneck caused by data accumulation of the cloud can be avoided. The terminal side and the edge side of the invention can be arranged in the families of the elderly living alone and the institutions such as the nursing home, the cloud platform is arranged in the hospitals and the institutions such as the health care committee, the medical intervention and treatment can be carried out in time by detecting the falling behavior of the elderly at the first time, the health prevention and treatment of the elderly are protected to the greatest extent, the life health and safety are maintained, and the medical expenses caused by the falling occurrence and the non-detection of the elderly are greatly saved.
Drawings
Fig. 1 is a flowchart of a fall detection method based on edge calculation according to a first embodiment of the present invention;
fig. 2 is a flow chart of fall determination of a thresholding model according to a first embodiment of the invention;
fig. 3 is a block diagram of a fall detection system based on edge calculation according to a first embodiment of the present invention;
FIG. 4 is a block diagram of a recurrent neural network model according to a second embodiment of the present invention;
fig. 5 is a fall determination flowchart of the recurrent neural network model according to the second embodiment of the present invention;
fig. 6 is a flowchart of training a recurrent neural network model according to a second embodiment of the present invention.
Detailed Description
In order to more clearly illustrate the embodiments of the present invention, the following description will explain the embodiments of the present invention with reference to the accompanying drawings. It is obvious that the drawings in the following description are only some examples of the invention, and that for a person skilled in the art, other drawings and embodiments can be derived from them without inventive effort.
The first embodiment is as follows:
a fall detection method based on edge calculation in this embodiment, as shown in fig. 1, includes the following steps:
s1: the user terminal collects triaxial acceleration data of user actions in real time;
s2: the method comprises the steps that triaxial acceleration data collected in real time are transmitted to an edge side, and whether a user falls is judged through a falling judgment model of the edge side;
s3: and if the user is judged to fall down, the edge side sends alarm information to the cloud platform.
The alarm information includes the name, sex, age, medical record number, current position, etc. of the old who falls over at present.
In step S2, the fall determination model is a threshold method model, and the definition feature quantity is composed of three parameters: combining an acceleration semaphore vector SMV, a continuous occurrence quantity N of suspected falling data and an acceleration vector A in the vertical direction of the human body; as shown in fig. 2, the following calculation and determination steps are included:
s21: setting a threshold value (T) of the characteristic quantitySMV,TN,TG) Wherein, TSMVThreshold value of SMV to distinguish between falling and non-falling actions, TNA threshold value, T, for the amount of persistent occurrences, N, of suspected fall dataGA threshold value of an acceleration vector A in the vertical direction of the human body;
s22: calculating the value of real-time characteristic quantity (SMV, N, A) according to the triaxial acceleration data of the user action collected in real time;
s23: and comparing the value of the real-time characteristic quantity with a characteristic quantity threshold value, and judging whether the action falls: if SMV is satisfied at the same time>TSMV，N<TNAnd A<TGThen, the user is determined to fall.
Step S21 specifically includes:
obtaining three-axis acceleration sample data of falling and non-falling actions of the user, and calculating SMV and T according to the three-axis acceleration sample dataSMVThe median of the smallest falling SMV and the largest SMV in most non-falling behaviors; t isNIs 50, TGIs 0.2G.
Under the above parameters, the actual measurement fall detection decision results of the threshold method model are shown in table 1, and the comprehensive obtained fall detection probability is 93.1%.
TABLE 1 threshold value method model detection judgment actual measurement results of different behaviors
Figure BDA0002374597680000071
Figure BDA0002374597680000081
As shown in fig. 3, the system for detecting a fall based on edge computing, which applies the above-mentioned decision algorithm, includes an intelligent terminal, an edge detection computing terminal, and a cloud platform. The intelligent terminal can be wearable equipment, wherein a three-axis acceleration sensor is integrated, and the three-axis acceleration sensor is used for collecting three-axis acceleration data of actions of a wearer and sending the three-axis acceleration data to the edge detection calculation end. The user wears wearable acceleration sensor on one's body, gathers acceleration data, and data pass through wireless protocol and convey to the intelligent gateway of edge detection calculation end. Wireless protocols that may be employed include Zigbee, WIFI, bluetooth BLE, NB-IoT, and the like.
The edge detection computing end utilizes a falling judgment model to compute the received triaxial acceleration data and judge whether the intelligent terminal wearer falls or not, and if the intelligent terminal wearer falls, alarm information is sent to the cloud platform. The edge detection computing end comprises an intelligent gateway, a falling detection power unit and a rule engine; the intelligent gateway is used for receiving the triaxial acceleration data sent by the intelligent terminal and sending the triaxial acceleration data to the fall detection power measuring unit; the fall detection power unit processes the triaxial acceleration data according to an algorithm selected by the rule engine, judges whether the intelligent terminal wearer falls or not, and sends a judgment result to the intelligent gateway; and if the judgment result is that the intelligent gateway falls down, the intelligent gateway sends alarm information to the cloud platform.
At the edge side, the intelligent gateway receives data from the three-axis acceleration sensor at the terminal side through a wireless protocol consistent with that at the terminal side and sends the data to the fall detection power measuring unit. And operating a fall detection judgment algorithm in the fall detection power unit, wherein the specifically adopted judgment algorithm is configured by a rule engine. The fall detection power unit judges whether the intelligent gateway falls or not according to the acceleration data sent by the intelligent gateway, once the intelligent gateway falls, the fall detection power unit sends the result to the intelligent gateway, and the intelligent gateway sends alarm information to the cloud platform through the Internet.
The cloud platform is used for receiving the falling alarm information sent by the edge detection computing end and maintaining the basic data of the intelligent terminal wearer.
Specifically, the cloud platform comprises a cloud management server which is communicated with the edge detection computing terminal and is used for maintaining basic information, medical information and falling data of all monitored intelligent terminal wearers in a responsible area. The cloud management server is further used for configuring the rule engine and selecting a fall detection method.
Example two:
the difference between the present embodiment and the first embodiment is: a falling judgment model of a falling detection method based on edge calculation is a recurrent neural network model. The model of the recurrent neural network is shown in fig. 4, and the recurrent neural network is composed of an input layer, a hidden layer and an output layer, wherein each layer is connected by corresponding weight matrix U, V and W.
The core of the algorithm is the number of neurons, i.e. the number of hidden layer nodes m. If the number m of hidden layer nodes is too small, the finally obtained result cannot be distinguished; if the number m of nodes in the hidden layer is too large, the algorithm training time is increased, and an overfitting phenomenon is easily generated. While the number of hidden layer nodes m depends on the input layer size n and the output layer size l, the relationship between the three parameters can be described by the following empirical formula:
Figure BDA0002374597680000091
m＝log2n
Figure BDA0002374597680000092
where m is the number of hidden layer nodes, n is the number of input nodes, and l is the number of output nodes. In the present invention, 15 consecutive triaxial acceleration data are used as one input unit to output a result between 0 and 1, so the input layer size is 45 and the output layer size is 1. And selecting the number of the nodes of the hidden layer as 13 for training. The whole process represents that the algorithm judges the probability of falling currently according to the past 15 pieces of acceleration data, and the probability is between 0 and 1.
After the sizes of the layers are determined, the fall determination of the present embodiment is started, as shown in fig. 5, including the following steps:
s201: training a recurrent neural network;
s202: inputting the real-time triaxial acceleration data into an input layer of the recurrent neural network, and judging whether the action falls down by using the trained recurrent neural network.
Specifically, as shown in fig. 6, step S201 includes:
s2011: initializing a weight matrix randomly between 0 and 1, and inputting triaxial acceleration sample data acquired by a sensor into an input layer;
s2012: computing
s1＝Ux1+Wh0
h1＝f(s1)
y1＝g(Vh1)
In the above formula, x1Is the input unit of the 1 st sampling time, h0Is an initial hidden layer matrix, s1Is an intermediate variable at the 1 st sampling instant, h1Is the hidden layer matrix at the 1 st sampling instant, y1Is the output layer result at the 1 st sampling moment, and U, V and W are weight matrixes used for updating errors; f is sigmoid function
Figure BDA0002374597680000101
g is the softmax function
Figure BDA0002374597680000102
As time goes on, h1As a memory of the previous layerThe state participates in the prediction of the next fall behavior:
s2＝Ux2+Wh1
h2＝f(s2)
y2＝g(Vh2)
in the above formula, x2Is the input unit of the 2 nd sampling instant, s2Is an intermediate variable at the 2 nd sampling instant, h2Is the hidden layer matrix at the 2 nd sampling instant, y2Is the output layer result at the 2 nd sampling time, and U, V and W are weight matrixes used for updating errors; and so on, calculating:
st＝Uxt+Wht-1
ht＝f(st)
yt＝g(Vht)
in the above formula, xtIs the input unit at the t-th sampling instant, stIs an intermediate variable at the t-th sampling instant, ht-1Is the hidden layer matrix at the t-1 th sampling time, htIs the hidden layer matrix at the t-th sampling instant, ytThe output layer result at the t-th sampling moment, and U, V and W are weight matrixes used for updating errors;
s2013: each time an output layer result y is obtainedtThen, the output layer error e is calculatedoAnd hidden layer error eh：
e0(t)＝ot-yt
eh(t)＝dh(e0(t)TV,t)
dh＝xst(1-st)
In the above formula, otIs an actual class label with a value of 0 or 1, e0(t) output layer error at time t, eh(t) hidden layer error at time t, dhIs a hidden layer error update formula; x is an input unit xtA vectorized representation of;
s2014: updating the weight matrix each time by using the output layer error and the hidden layer error:
Figure BDA0002374597680000111
Figure BDA0002374597680000112
Figure BDA0002374597680000113
in the above formula, U (t), V (t), W (t) are weight matrices for updating errors at time t, U (t +1), V (t +1), and W (t +1) are weight matrices for updating errors at time t +1, z is a time mark, eh(t-z)，xt-zH (t-z-1) represents a hidden layer error at the time t-z, an input unit at the time t-z and a hidden layer matrix at the time t-z, α is a learning rate, β is a normalization parameter;
s2015: will output layer error eoAs an indicator of the effectiveness of the training algorithm, when eoIs less than a threshold TeWhen so, the training is finished.
In particular, end of training e is identifiedoIs determined by the absolute value ofeThe sampling rate can be 0.001, when the threshold value is reached, the falling data of the sample is converted into the classification label 1, the non-falling data is converted into the classification label 0, and the training can be finished.
The actual measurement falling detection judgment result of the recurrent neural network model is shown in table 2, and the falling detection probability obtained by synthesis is more than 95%.
TABLE 2 recurrent neural network model test judgment actual measurement results of different behaviors
Figure BDA0002374597680000121
Figure BDA0002374597680000131
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.

Claims (10)

1. A fall detection method based on edge calculation, comprising the steps of:
s1: the user terminal collects triaxial acceleration data of user actions in real time;
s2: the method comprises the steps that triaxial acceleration data collected in real time are transmitted to an edge side, and whether a user falls is judged through a falling judgment model of the edge side;
s3: and if the user is judged to fall down, the edge side sends alarm information to the cloud platform.
2. The edge calculation-based fall detection method according to claim 1, wherein the fall decision model is a thresholding model, and the defined feature quantity is composed of three parameters: combining an acceleration semaphore vector SMV, a continuous occurrence quantity N of suspected falling data and an acceleration vector A in the vertical direction of the human body; the method comprises the following calculation and judgment steps:
s21: setting a threshold value (T) of the characteristic quantitySMV,TN,TG) Wherein, TSMVThreshold value of SMV to distinguish between falling and non-falling actions, TNA threshold value, T, for the amount of persistent occurrences, N, of suspected fall dataGA threshold value of an acceleration vector A in the vertical direction of the human body;
s22: calculating the value of real-time characteristic quantity (SMV, N, A) according to the triaxial acceleration data of the user action collected in real time;
s23: and comparing the value of the real-time characteristic quantity with a characteristic quantity threshold value, and judging whether the action falls: if SMV is satisfied at the same time>TSMV，N<TNAnd A<TGThen, the user is determined to fall.
3. An edge calculation-based fall detection method as claimed in claim 2, wherein the step S21 comprises:
obtaining three-axis acceleration sample data of falling and non-falling actions of the user, and calculating SMV and T according to the three-axis acceleration sample dataSMVThe median of the smallest falling SMV and the largest SMV in most non-falling behaviors; t isNIs 50, TGIs 0.2G.
4. The edge calculation-based fall detection method according to claim 1, wherein the fall judgment model is a recurrent neural network model, the size of the input layer is set to 45, the number of nodes of the hidden layer is set to 13, and the size of the output layer is set to 1; the method comprises the following calculation and judgment steps:
s201: training a recurrent neural network;
s202: inputting the real-time triaxial acceleration data into an input layer of the recurrent neural network, and judging whether the action falls down by using the trained recurrent neural network.
5. An edge calculation-based fall detection method as claimed in claim 4, wherein the step S201 comprises:
s2011: initializing a weight matrix randomly between 0 and 1, and inputting triaxial acceleration sample data acquired by a sensor into an input layer;
s2012: computing
s1＝Ux1+Wh0
h1＝f(s1)
y1＝g(Vh1)
In the above formula, x1Is the input unit of the 1 st sampling time, h0Is an initial hidden layer matrix, s1Is an intermediate variable at the 1 st sampling instant, h1Is the hidden layer matrix at the 1 st sampling instant, y1Is the output layer result at the 1 st sampling moment, and U, V and W are weight matrixes used for updating errors; f is sigmoid function
Figure FDA0002374597670000021
g is the softmax function
Figure FDA0002374597670000022
As time goes on, h1And (3) taking the memory state as the upper layer to participate in the next fall behavior prediction:
s2＝Ux2+Wh1
h2＝f(s2)
y2＝g(Vh2)
in the above formula, x2Is the input unit of the 2 nd sampling instant, s2Is an intermediate variable at the 2 nd sampling instant, h2Is the hidden layer matrix at the 2 nd sampling instant, y2Is the output layer result at the 2 nd sampling time, and U, V and W are weight matrixes used for updating errors; and so on, calculating:
st＝Uxt+Wht-1
ht＝f(st)
yt＝g(Vht)
in the above formula, xtIs the input unit at the t-th sampling instant, stIs an intermediate variable at the t-th sampling instant, ht-1Is the hidden layer matrix at the t-1 th sampling time, htIs the hidden layer matrix at the t-th sampling instant, ytThe output layer result at the t-th sampling moment, and U, V and W are weight matrixes used for updating errors;
s2013: each time an output layer result y is obtainedtThen, the output layer error e is calculatedoAnd hidden layer error eh：
e0(t)＝ot-yt
eh(t)＝dh(e0(t)TV,t)
dh＝xst(1-st)
In the above formula, otIs the actual class label, has a value of 0 or 1,e0(t) output layer error at time t, eh(t) hidden layer error at time t, dhIs a hidden layer error update formula; x is an input unit xtA vectorized representation of;
s2014: updating the weight matrix each time by using the output layer error and the hidden layer error:
Figure FDA0002374597670000031
Figure FDA0002374597670000032
Figure FDA0002374597670000033
in the above formula, U (t), V (t), W (t) are weight matrices for updating errors at time t, U (t +1), V (t +1), and W (t +1) are weight matrices for updating errors at time t +1, z is a time mark, eh(t-z)，xt-zH (t-z-1) represents a hidden layer error at the time t-z, an input unit at the time t-z and a hidden layer matrix at the time t-z, α is a learning rate, β is a normalization parameter;
s2015: will output layer error eoAs an indicator of the effectiveness of the training algorithm, when eoIs less than a threshold TeWhen so, the training is finished.
6. The edge calculation-based fall detection method according to claim 1, wherein the alarm information includes a name, a sex, an age, a medical record number, a current location, and the like of an elderly person who is currently falling.
7. An edge-computing-based fall detection system applying the fall detection method according to any one of claims 1 to 6, comprising an intelligent terminal, an edge detection computing terminal and a cloud platform; the intelligent terminal comprises a three-axis acceleration sensor and is used for acquiring three-axis acceleration data of the action of the wearer and sending the three-axis acceleration data to the edge detection computing end; the edge detection computing end computes the received triaxial acceleration data by using a falling judgment model and judges whether the intelligent terminal wearer falls or not, and if the intelligent terminal wearer falls, alarm information is sent to the cloud platform; the cloud platform is used for receiving the falling alarm information sent by the edge detection computing end and maintaining the basic data of the intelligent terminal wearer.
8. The system for detecting falls based on edge calculation as claimed in claim 7, wherein the edge detection calculation end comprises an intelligent gateway, a fall detection power unit and a rule engine; the intelligent gateway is used for receiving the triaxial acceleration data sent by the intelligent terminal and sending the triaxial acceleration data to the fall detection power measuring unit; the fall detection power unit processes the triaxial acceleration data according to an algorithm selected by the rule engine, judges whether the intelligent terminal wearer falls or not, and sends a judgment result to the intelligent gateway; and if the judgment result is that the intelligent gateway falls down, the intelligent gateway sends alarm information to the cloud platform.
9. The edge-computing-based fall detection system of claim 8, wherein the cloud platform comprises a cloud management server, which communicates with the edge detection computing terminal and maintains basic information, medical information and fall data of all monitored intelligent terminal wearers in the responsible area.
10. An edge-computing-based fall detection system as claimed in claim 9, wherein the cloud management server is further configured to configure a rules engine to select a fall detection method.
CN202010061335.XA 2020-01-19 2020-01-19 Fall detection method and system based on edge calculation Pending CN111325946A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010061335.XA CN111325946A (en) 2020-01-19 2020-01-19 Fall detection method and system based on edge calculation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010061335.XA CN111325946A (en) 2020-01-19 2020-01-19 Fall detection method and system based on edge calculation

Publications (1)

Publication Number Publication Date
CN111325946A true CN111325946A (en) 2020-06-23

Family

ID=71167250

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010061335.XA Pending CN111325946A (en) 2020-01-19 2020-01-19 Fall detection method and system based on edge calculation

Country Status (1)

Country Link
CN (1) CN111325946A (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111818173A (en) * 2020-07-21 2020-10-23 郑州大学 Medication reminding system and method based on active big data perception
CN112183228A (en) * 2020-09-09 2021-01-05 青岛联合创智科技有限公司 Intelligent community endowment service system and method
CN112883392A (en) * 2021-02-24 2021-06-01 山东万博科技股份有限公司 Edge calculation method and device based on artificial intelligence
CN112927475A (en) * 2021-01-27 2021-06-08 浙江理工大学 Fall detection system based on deep learning
CN113205440A (en) * 2021-04-28 2021-08-03 华奥系科技(汕头)有限公司 Smart community endowment service Internet of things system
CN113287805A (en) * 2021-05-27 2021-08-24 郑州大学 Fall protection method and wearable fall protection device
CN113989570A (en) * 2021-11-01 2022-01-28 南通大学 Fall detection alarm method based on embedded edge artificial intelligence
CN115530774A (en) * 2021-06-30 2022-12-30 荣耀终端有限公司 Epilepsy detection method and device

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030096590A1 (en) * 2001-11-21 2003-05-22 Kazuhiko Satoh Interconnecting device
CN104036147A (en) * 2014-06-26 2014-09-10 杜波 Home endowment service information reporting and processing method, client and processing system
CN106599914A (en) * 2016-12-07 2017-04-26 广东工业大学 Multi-sensor fused wristed falling detection method and device
CN110390565A (en) * 2019-07-23 2019-10-29 厦门市佳音在线股份有限公司 The method and system of intelligent gateway adaptive management are realized by AI edge calculations

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030096590A1 (en) * 2001-11-21 2003-05-22 Kazuhiko Satoh Interconnecting device
CN104036147A (en) * 2014-06-26 2014-09-10 杜波 Home endowment service information reporting and processing method, client and processing system
CN106599914A (en) * 2016-12-07 2017-04-26 广东工业大学 Multi-sensor fused wristed falling detection method and device
CN110390565A (en) * 2019-07-23 2019-10-29 厦门市佳音在线股份有限公司 The method and system of intelligent gateway adaptive management are realized by AI edge calculations

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
牛德姣 等: "基于递归神经网络的跌倒检测系统", 《智能系统学报》 *

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111818173A (en) * 2020-07-21 2020-10-23 郑州大学 Medication reminding system and method based on active big data perception
CN112183228A (en) * 2020-09-09 2021-01-05 青岛联合创智科技有限公司 Intelligent community endowment service system and method
CN112927475A (en) * 2021-01-27 2021-06-08 浙江理工大学 Fall detection system based on deep learning
CN112927475B (en) * 2021-01-27 2022-06-10 浙江理工大学 Fall detection system based on deep learning
CN112883392A (en) * 2021-02-24 2021-06-01 山东万博科技股份有限公司 Edge calculation method and device based on artificial intelligence
CN113205440A (en) * 2021-04-28 2021-08-03 华奥系科技(汕头)有限公司 Smart community endowment service Internet of things system
CN113205440B (en) * 2021-04-28 2023-02-28 华奥系科技(汕头)有限公司 Smart community endowment service Internet of things system
CN113287805A (en) * 2021-05-27 2021-08-24 郑州大学 Fall protection method and wearable fall protection device
CN115530774A (en) * 2021-06-30 2022-12-30 荣耀终端有限公司 Epilepsy detection method and device
CN115530774B (en) * 2021-06-30 2024-03-26 荣耀终端有限公司 Epilepsy detection method and device
CN113989570A (en) * 2021-11-01 2022-01-28 南通大学 Fall detection alarm method based on embedded edge artificial intelligence

Similar Documents

Publication Publication Date Title
CN111325946A (en) Fall detection method and system based on edge calculation
CN105118236B (en) Paralysis falls to monitor and preventing mean and its processing method
Oniga et al. Human activity recognition using neural networks
CN107092806A (en) It is a kind of towards the intelligentized information fusion of old man&#39;s household and method for early warning
CN103211599A (en) Method and device for monitoring tumble
Fortino et al. Fall-MobileGuard: A smart real-time fall detection system
CN205050303U (en) Human paralysis of intelligence is monitoring devices
CN109726672A (en) It is a kind of to fall down detection method based on human skeleton sequence and convolutional neural networks
CN109887238A (en) A kind of fall detection system and detection alarm method of view-based access control model and artificial intelligence
CN110659677A (en) Human body falling detection method based on movable sensor combination equipment
CN104269025A (en) Wearable type single-node feature and position selecting method for monitoring outdoor tumble
Lubina et al. Artificial neural networks in accelerometer-based human activity recognition
CN111461042A (en) Fall detection method and system
CN107019501B (en) Remote tumble detection method and system based on genetic algorithm and probabilistic neural network
Farooq et al. Real time monitoring and recognition of eating and physical activity with a wearable device connected to the eyeglass
CN115662631A (en) AI intelligence discrimination-based nursing home management system
CN113096792B (en) Intelligent health monitoring and early warning method and system based on machine learning
Amir et al. Real-time threshold-based fall detection system using wearable IoT
Xie et al. ART: adaptive and real-time fall detection using COTS smart watch
CN110414590A (en) Physical activity recognition methods based on Intelligent mobile equipment and convolutional neural networks
CN113869179A (en) Multi-mode feature attention fused home-based and aged-care intelligent monitoring method and system
Pereira et al. Fall detection on ambient assisted living using a wireless sensor network
CN112927475B (en) Fall detection system based on deep learning
Wu et al. Fall detection monitoring system based on mems sensor
Sánchez et al. A new adaptive algorithm for detecting falls through mobile devices

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20200623

RJ01 Rejection of invention patent application after publication