CN110047247A - A kind of smart home device accurately identifying Falls in Old People - Google Patents
A kind of smart home device accurately identifying Falls in Old People Download PDFInfo
- Publication number
- CN110047247A CN110047247A CN201910426282.4A CN201910426282A CN110047247A CN 110047247 A CN110047247 A CN 110047247A CN 201910426282 A CN201910426282 A CN 201910426282A CN 110047247 A CN110047247 A CN 110047247A
- Authority
- CN
- China
- Prior art keywords
- algorithm
- formula
- output
- smart home
- vector
- 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
Links
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
- G08B21/0407—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
-
- 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
Landscapes
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Gerontology & Geriatric Medicine (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Psychiatry (AREA)
- Psychology (AREA)
- Social Psychology (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention belongs to smart home device technical fields, a kind of smart home device accurately identifying Falls in Old People is provided, including an acceleration transducer, a gyroscope and central processing unit, input of the data that acceleration transducer, gyroscope are collected after a miniature denoiser processing as central processing unit, the corresponding movement of this batch data can be exported after the processing of the built-in algorithm of central processing unit to judge whether old man falls, real time information is then sent to guardian by communication module.Present device is optimized on detection algorithm, and present device does not need and other smart home devices are used in combination, using simple, detection accuracy is high.
Description
Technical field
The invention belongs to smart home device technical fields, and in particular to a kind of intelligent family for accurately identifying Falls in Old People
Occupy equipment.
Background technique
Smart home is always an important directions of Internet of Things industry development, and people utilize sensor and smart machine group
At the certain functions of network implementations to improve the sense of security and comfort level of family, and the family that kinsfolk includes old man is come
Say, the nurse to old man how is realized using the smart machine in family, and this seems and is even more important, studies show that, old man because
Injury caused by tumble is the master for leading to old man's personal safety because of the secondary injury that burst disease causes tumble and is formed
It threatens.
The smart home device of problems is solved on the market at present often through to one built-in acceleration of elders wear
The smart machine of sensor, and collect the X of acceleration transducer, the acceleration a in tri- directions Y, Zx, ay, az, and utilize and add
Speed formulaResultant acceleration SVM is found out, and judges whether it is less than given threshold value, if it is less than
It is taken as state of weightlessness, such judgment method is often less accurate.Then another solution is exactly again
The smart machine is combined and is used with smart home behavioral value equipment network in other families, although compensating for single equipment in this way
Judge Falls Among Old People inaccurate problem, but also brings along other problem.First, will certainly band using so many equipment
Carry out this excessively high problem of cost.Second, a devices in system is excessive, then the stability of this system must be declined, though
The stability of system so can be maintained at by higher level with existing technology, but the personal safety for being related to old man still needs
To consider this disadvantage.
Summary of the invention
It is an object of the invention to overcoming the deficiencies in the prior art described above, a kind of Falls in Old People that accurately identifies is provided
Smart home device, which does not need and other smart home devices are used in combination, using simple, detection accuracy is high.
The technical solution adopted in the present invention is as follows.
The present invention provides a kind of smart home device for accurately identifying Falls in Old People, including an acceleration transducer,
One gyroscope and central processing unit, the data that acceleration transducer, gyroscope are collected are after a miniature denoiser processing
As the input of central processing unit, it is corresponding dynamic that this batch data can be exported after the processing of the built-in algorithm of central processing unit
Make to judge whether old man falls, real time information is then sent to guardian by communication module.
In the above-mentioned technical solutions, the communication module includes WIFI communication module, ZigBee communication module, mobile communication
Module.
In the above-mentioned technical solutions, the built-in algorithm of the central processing unit includes that can accurately identify the behavior of user
The fall detection algorithm of movement, the generation of algorithm the following steps are included:
(1) data acquire: collection process require subject wears' smart home device, Walk Simulation, running, above go downstairs,
Repose and fall six kinds of states, records from the accelerometer sensor (linear acceleration) and gyroscope built in smart home device
Data on the x, y, z axis of (angular speed), frequency acquisition are 50 hertz (50 data points i.e. per second), are carried out to initial data pre-
Processing: denoiser pretreatment acceleration transducer and gyroscope signal are used；Data are divided into 2.56 seconds (128 data points)
Fixation window, have 50% overlapping between window；Acceleration transducer data are decomposed into acceleration of gravity data and body adds
Speed data two parts.
(2) dataset representation: the reliability of training algorithm and verification algorithm needs training set and test set, the two set
It is all by sample Qi={ X1 (i), Y1 (i), Z1 (i), X2 (i), Y2 (i), Z2 (i), X3 (i), Y3 (i), Z3 (i), D } and composition, herein, what i referred to
It is sample serial number in set, X1, X2, X3Refer respectively to body acceleration, gyroscope angular speed and total acceleration in the X-axis direction
The value of degree, Y, Z axis expression are identical with this, and D is a M n dimensional vector n, for indicating six kinds of states.
(3) algorithm frame is chosen: using one-dimensional convolutional neural networks as algorithm frame；As identification Falls Among Old People
Smart home device, the accurate rate of identification have to be maintained at very high level.This requires have to collect a large amount of number
According to training algorithm is carried out to improve its accuracy and generalization ability.Moreover, selected algorithm frame must also have fastly
Speed processing mass data ability with realize quickly identification tumble behavior, just can guarantee so quickly by recognition result substantially without
Delay passes.In view of the above two o'clock, present invention employs one-dimensional convolutional neural networks as algorithm frame.
(4) training and tuning of algorithm
(4-1) is as follows by the adjustment of one-dimensional convolutional neural networks algorithm model overall structure, can preferably handle data.
It includes 21 dimension convolutional layers, followed by one dropout layers, then plus a pond layer, in order to make algorithm model preferably from
Learning characteristic in input data defines two convolutional layers as one group, and quickly, dropout layers can slow down to CNN pace of learning
Habit process simultaneously keeps final algorithm model effect more preferable, and pond layer reduces the feature learnt, keeps its reservation most important
After element, CNN and pond layer, the feature acquired is launched into a long vector, using a full articulamentum, then arrives
Up to output layer, predicted；(4-2) is standardized pretreatment to data.
After (4-3) has put up algorithm model, be trained using training dataset to algorithm model: (4-3-1) is selected
Training group, randomly chooses 9600 samples as training group respectively from sample set, and herein, the unit number of middle layer can be with
It is adjusted according to trained effect, for ease of description, is set to L, the vector of input is Qi={ X1 (i), Y1 (i),
Z1 (i), X2 (i), Y2 (i), Z2 (i), X2 (i), Y3 (i), Z3 (i), the input vector of middle layer is H={ h0, h1..., hL, network is practical
Output vector is Y={ y0, y1..., yM, and with D={ d0, d1..., dMCome indicate target export, output unit i to hidden unit
The weight of j is Vij, and the weight of hidden unit j to output unit k are Wjk, in addition use θkAnd φjCome respectively indicate output unit and
The threshold value of implicit unit；
The output of middle layer each unit are as follows:
And the output of output layer each unit are as follows:
Wherein, f (x) is excitation function, using S type function formula are as follows:
(4-3-2) is by each weight Vij, WjkWith threshold value φj, θkIt is set to the random value close to 0, and initializes precision controlling
Parameter ε and learning rate α；
(4-3-3) takes an input vector Q from training groupi, and give its target output vector Di；
(4-3-4) calculates a middle layer output vector H using formula one, then the reality of network is calculated with formula two
Output arrow Y；
(4-3-5) is by k-th of element y in output vectorkWith the kth element d in target vectorkIt is compared, calculates
M output error item:
δk=(dk-yk)yk(1-yk) formula four
L error term is also calculated to the hidden unit of middle layer:
(4-3-6) successively calculates the adjustment amount of each weight:
ΔWjk(n)=(a/ (1+L)) * (Δ Wjk(n-1)+1)*δk*δjFormula six
ΔVij(n)=(a/ (1+N)) * (Δ Vij(n-1)+1)*δk*hjFormula seven
With the adjustment amount of threshold value:
Δθk(n)=(a/ (1+L)) * (Δ θk(n-1)+1)*δkFormula eight
Δφj(n)=(a/ (1+L)) * (Δ φj(n-1)+1)*δjFormula nine
(4-3-7) adjusts weight:
Wjk(n+1)=Wjk(n)+ΔWjk(n) formula ten
Vij(n+1)=Vij(n)+ΔVij(n) formula 11
Adjust threshold value:
θk(n+1)=θk(n)+Δθk(n) formula 12
φj(n+1)=φj(n)+Δφj(n) formula ten
(4-3-8) after k every experience 1 to M, whether judge index meets required precision: E≤ε, and wherein E is overall error letter
Number, andIf conditions are not met, being returned to step (4-3-3), continue iteration；If satisfaction enters
In next step；
(4-3-9) training terminates, and weight and threshold value is saved hereof, classifier algorithm is formed.
The invention proposes a kind of smart home device for accurately identifying Falls in Old People, one is based on one built in the equipment
The algorithm model of convolutional neural networks is tieed up, the data which has passed through acquisition are finished as training set training, can be accurately
Identify the motion state of user.The present invention is algorithmically optimized, in the number of sensors of human body behavior for identification
In conditional situation, accurate recognition effect same as multisensor Human bodys' response system is reached, relative to entire
Human bodys' response system, present device cost is lower, convenient to wear, the more flexible multiplicity of usage mode, relative to other
Non-systemic Human bodys' response smart home device, have higher precision.
Detailed description of the invention
Fig. 1 is the composition schematic diagram of smart home device of the present invention.
Fig. 2 is the flow chart of fall detection algorithm in the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, with reference to the accompanying drawings and embodiments, to this
Invention is further elaborated.
As shown in Figure 1, the present embodiment provides a kind of smart home device for accurately identifying Falls in Old People, including one adds
Velocity sensor, a gyroscope and central processing unit, the data that acceleration transducer, gyroscope are collected pass through a miniature drop
Input after device of making an uproar processing as central processing unit can export this lot number after the processing of the built-in algorithm of central processing unit
Judge whether old man falls according to corresponding movement, real time information is then sent to guardian by communication module.
In the above-described embodiments, the communication module includes WIFI communication module, ZigBee communication module, mobile communication mould
Block.
In the above-described embodiments, the built-in algorithm of the central processing unit include can accurately identify user behavior it is dynamic
The fall detection algorithm of work, the generation of algorithm the following steps are included:
(1) data acquire: collection process require subject wears' smart home device, Walk Simulation, running, above go downstairs,
Repose and fall six kinds of states, records from the accelerometer sensor (linear acceleration) and gyroscope built in smart home device
Data on the x, y, z axis of (angular speed), recording frequency are 50 hertz, pre-process to initial data: using miniature noise reduction
Device pre-processes acceleration transducer and gyroscope signal；Data are divided into 2.56 seconds fixation windows, have 50% between window
Overlapping；Acceleration transducer data are decomposed into acceleration of gravity data and body acceleration data two parts.
(2) dataset representation: the reliability of training algorithm and verification algorithm needs training set and test set, the two set
It is all by sample Qi={X1 (i), Y1 (i), Z1 (i), X2 (i), Y2 (i), Z2 (i), X3 (i), Y3 (i), Z3 (i), D } and composition, i is referred to herein
Sample serial number in set, X1, X2, X3Refer respectively to body acceleration, gyroscope angular speed and total acceleration in the X-axis direction
Value, Y, Z axis expression be identical with this, D is a M n dimensional vector n, for indicating six kinds of states.
(3) algorithm frame is chosen: using one-dimensional convolutional neural networks as algorithm frame.
(4) training and tuning of algorithm
(4-1) is as follows by the adjustment of one-dimensional convolutional neural networks algorithm model overall structure, it includes 21 dimension convolutional layers, so
After be one dropout layers, then plus a pond layer, it is fixed in order to make algorithm model learning characteristic preferably from input data
Adopted two convolutional layers are as one group, and dropout layers can slow down learning process and keep final algorithm model effect more preferable, and pond
Change layer to reduce feature learn, make its most important element of reservation, after CNN and pond layer, the feature acquired is unfolded
Output layer is then reached, is predicted using a full articulamentum at a long vector；
(4-2) is standardized pretreatment to data；
After (4-3) has put up algorithm model, be trained using training dataset to algorithm model: (4-3-1) is selected
Training group, randomly chooses 9600 samples as training group respectively from sample set, and herein, the unit number of middle layer can be with
It is adjusted according to trained effect, is set to L, the vector of input is Qi={ X1 (i), Y1 (i), Z1 (i), X2 (i), Y2 (i),
Z2 (i), X3 (i), Y3 (i), Z3 (i), the input vector of middle layer is H={ h0, h1..., hL, network reality output vector is Y=
{y0, y1..., yM, and with D={ d0, d1..., dMCome indicate target export, the weight of output unit i to hidden unit j are Vij,
And the weight of hidden unit j to output unit k are Wjk, in addition use θkAnd φjTo respectively indicate the threshold of output unit and implicit unit
Value；
The output of middle layer each unit are as follows:
And the output of output layer each unit are as follows:
Wherein, f (x) is excitation function, using S type function formula are as follows:
(4-3-2) is by each weight Vij, WjkWith threshold value φj, θkIt is set to the random value close to 0, and initializes Accuracy Controlling Parameter
ε and learning rate α；
(4-3-3) takes an input vector Q from training groupi, and give its target output vector Di；
(4-3-4) calculates a middle layer output vector H using formula one, then the reality of network is calculated with formula two
Output arrow Y；
(4-3-5) is by k-th of element y in output vectorkWith the kth element d in target vectorkIt is compared, calculates
M output error item:
δk=(dk-yk)yk(1-yk) formula four
L error term is also calculated to the hidden unit of middle layer:
(4-3-6) successively calculates the adjustment amount of each weight:
ΔWjk(n)=(a/ (1+L)) * (Δ Wjk(n-1)+1)*δk*δjFormula six
ΔVij(n)=(a/ (1+N)) * (Δ Vij(n-1)+1)*δk*hjFormula seven
With the adjustment amount of threshold value:
Δθk(n)=(a/ (1+L)) * (Δ θk(n-1)+1)*δkFormula eight
Δφj(n)=(a/ (1+L)) * (Δ φj(n-1)+1)*δjFormula nine
(4-3-7) adjusts weight:
Wjk(n+1)=Wjk(n)+ΔWjk(n) formula ten
Vij(n+1)=Vij(n)+ΔVij(n) formula 11
Adjust threshold value:
θk(n+1)=θk(n)+Δθk(n) formula 12
φj(n+1)=φj(n)+Δφj(n) formula 13
(4-3-8) after k every experience 1 to M, whether judge index meets required precision: E≤ε, and wherein E is overall error letter
Number, andIf conditions are not met, being returned to step (4-3-3), continue iteration；If satisfaction enters
In next step；
(4-3-9) training terminates, and weight and threshold value is saved hereof, classifier algorithm is formed.
The content being not described in detail in this specification belongs to the prior art well known to those skilled in the art.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent replacements, and improvements done within the spirit and principles of the present invention should be included in
Within protection scope of the present invention.
Claims (3)
1. a kind of smart home device for accurately identifying Falls in Old People, it is characterised in that: the smart home device includes one
Acceleration transducer, a gyroscope and central processing unit, the data that acceleration transducer, gyroscope are collected pass through a noise reduction
Input after device processing as central processing unit passes through communication module to monitoring after the processing of the built-in algorithm of central processing unit
Human hair send real time information.
2. the smart home device according to claim 1 for accurately identifying Falls in Old People, it is characterised in that: the communication
Module includes WIFI communication module, ZigBee communication module, mobile communication module.
3. the smart home device according to claim 1 for accurately identifying Falls in Old People, it is characterised in that the center
The built-in algorithm of processor includes the fall detection algorithm that can accurately identify the behavior act of user, and the generation of algorithm includes
Following steps:
(1) data acquire: collection process requires subject wears' smart home device, and Walk Simulation, above goes downstairs, reposes at running
With six kinds of states of falling, the number on the x, y, z axis from accelerometer sensor and gyroscope built in smart home device is recorded
According to recording frequency is 50 hertz, pre-processes to initial data: being believed using denoiser pretreatment acceleration sensor and gyroscope
Number；Data are divided into 2.56 seconds fixation windows, there is 50% overlapping between window；Acceleration sensor data are decomposed into gravity
Acceleration information and body acceleration data two parts；
(2) dataset representation: the reliability of training algorithm and verification algorithm needs training set and test set, the two set are all
By sample Qi={ X1 (i),Y1 (i),Z1 (i),X2 (i),Y2 (i),Z2 (i),X3 (i),Y3 (i),Z3 (i), D } and composition, i refers to gathering herein
Middle sample serial number, X1, X2, X3The value of body acceleration, gyroscope angular speed and total acceleration in the X-axis direction is referred respectively to,
Y, Z axis expression is identical with this, and D is a M n dimensional vector n, for indicating six kinds of states；
(3) algorithm frame is chosen: using one-dimensional convolutional neural networks as algorithm frame；
(4) training and tuning of algorithm
(4-1) is as follows by the adjustment of one-dimensional convolutional neural networks algorithm model overall structure, it includes 21 dimension convolutional layers, followed by
One dropout layers, then plus a pond layer in order to make algorithm model learning characteristic preferably from input data define two
A convolutional layer is as one group, and dropout layers can slow down learning process and keep final algorithm model effect more preferable, and pond layer
The feature learnt is reduced, its is made to retain most important element, after CNN and pond layer, the feature acquired is launched into one
Then a long vector reaches output layer, is predicted using a full articulamentum；
(4-2) is standardized pretreatment to data；
After (4-3) has put up algorithm model, algorithm model is trained using training dataset:
(4-3-1) selectes training group, and randomly 9600 samples of selection are as training group respectively from sample set, herein, in
The unit number of interbed can be adjusted according to trained effect, be set to L, and the vector of input is Qi={ X1 (i),Y1 (i),Z1 (i),X2 (i),Y2 (i),Z2 (i),X3 (i),Y3 (i),Z3 (i), the input vector of middle layer is H={ h0, h1..., hL, network is real
Border output vector is Y={ y0, y1..., yM, and with D={ d0, d1..., dMCome indicate target export, output unit i to hidden list
The weight of first j is Vij, and the weight of hidden unit j to output unit k are Wjk, in addition use θkAnd φjTo respectively indicate output unit
With the threshold value of implicit unit；
The output of middle layer each unit are as follows:
And the output of output layer each unit are as follows:
Wherein, f (x) is excitation function, using S type function formula are as follows:
(4-3-2) is by each weight Vij, WjkWith threshold value φj, θkIt is arranged to close to 0 random value, and initializes precision controlling ginseng
Number ε and learning rate α；
(4-3-3) takes an input vector Q from training groupi, and give its target output vector Di；
(4-3-4) calculates a middle layer output vector H using formula one, then the reality output of network is calculated with formula two
Swear Y；
(4-3-5) is by k-th of element y in output vectorkWith the kth element d in target vectorkIt is compared, calculates M
Output error item:
δk=(dk-yk)yk(1-yk) formula four
L error term is also calculated to the hidden unit of middle layer:
(4-3-6) successively calculates the adjustment amount of each weight:
ΔWjk(n)=(a/ (1+L)) * (Δ Wjk(n-1)+1)*δk*δjFormula six
ΔVij(n)=(a/ (1+N)) * (Δ Vij(n-1)+1)*δk*hjFormula seven
With the adjustment amount of threshold value:
Δθk(n)=(a/ (1+L)) * (Δ θk(n-1)+1)*δkFormula eight
Δφj(n)=(a/ (1+L)) * (Δ φj(n-1)+1)*δjFormula nine
(4-3-7) adjusts weight:
Wjk(n+1)=Wjk(n)+ΔWjk(n) formula ten
Vij(n+1)=Vij(n)+ΔVij(n) formula 11
Adjust threshold value:
θk(n+1)=θk(n)+Δθk(n) formula 12
φj(n+1)=φj(n)+Δφj(n) formula 13
(4-3-8) after k every experience 1 to M, whether judge index meets required precision: E≤ε, and wherein E is overall error function, andIf conditions are not met, being returned to step (4-3-3), continue iteration；If satisfaction enters next
Step；
(4-3-9) training terminates, and weight and threshold value is saved hereof, classifier algorithm is formed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910426282.4A CN110047247A (en) | 2019-05-21 | 2019-05-21 | A kind of smart home device accurately identifying Falls in Old People |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910426282.4A CN110047247A (en) | 2019-05-21 | 2019-05-21 | A kind of smart home device accurately identifying Falls in Old People |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110047247A true CN110047247A (en) | 2019-07-23 |
Family
ID=67282938
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910426282.4A Pending CN110047247A (en) | 2019-05-21 | 2019-05-21 | A kind of smart home device accurately identifying Falls in Old People |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110047247A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110928298A (en) * | 2019-10-31 | 2020-03-27 | 山东大学 | Automatic cruise electric sickbed and elevator interaction method and system |
| CN111710129A (en) * | 2020-06-12 | 2020-09-25 | 电子科技大学 | Real-time pre-collision falling detection method for old people |
| CN112150766A (en) * | 2020-08-28 | 2020-12-29 | 永安行科技股份有限公司 | Early warning method and device for remote safety prevention and control |
| CN112933579A (en) * | 2019-12-11 | 2021-06-11 | 中移(苏州)软件技术有限公司 | Motion quality evaluation method and device and storage medium |
| CN113706827A (en) * | 2021-09-03 | 2021-11-26 | 浙江远图互联科技股份有限公司 | Wireless acquisition system for vital signs of household old people |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103680063A (en) * | 2013-12-16 | 2014-03-26 | 嘉兴学院 | Old person abnormal condition judging method based on mobile phone accelerometer and gyroscope |
| CN103745271A (en) * | 2014-01-02 | 2014-04-23 | 上海大学 | Forecasting method for induction thermal deposition calcium-phosphate coating process on basis of neural network |
| CN106846729A (en) * | 2017-01-12 | 2017-06-13 | 山东大学 | A kind of fall detection method and system based on convolutional neural networks |
| US20180070889A1 (en) * | 2016-09-09 | 2018-03-15 | Qualcomm Incorporated | Devices and methods for fall detection based on phase segmentation |
| CN108416979A (en) * | 2018-03-08 | 2018-08-17 | 辽宁工程技术大学 | A kind of intelligence the elderly's tumbling alarm system |
-
2019
- 2019-05-21 CN CN201910426282.4A patent/CN110047247A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103680063A (en) * | 2013-12-16 | 2014-03-26 | 嘉兴学院 | Old person abnormal condition judging method based on mobile phone accelerometer and gyroscope |
| CN103745271A (en) * | 2014-01-02 | 2014-04-23 | 上海大学 | Forecasting method for induction thermal deposition calcium-phosphate coating process on basis of neural network |
| US20180070889A1 (en) * | 2016-09-09 | 2018-03-15 | Qualcomm Incorporated | Devices and methods for fall detection based on phase segmentation |
| CN106846729A (en) * | 2017-01-12 | 2017-06-13 | 山东大学 | A kind of fall detection method and system based on convolutional neural networks |
| CN108416979A (en) * | 2018-03-08 | 2018-08-17 | 辽宁工程技术大学 | A kind of intelligence the elderly's tumbling alarm system |
Non-Patent Citations (1)
| Title |
|---|
| 陈向震: "基于深度学习的人脸表情识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110928298A (en) * | 2019-10-31 | 2020-03-27 | 山东大学 | Automatic cruise electric sickbed and elevator interaction method and system |
| CN110928298B (en) * | 2019-10-31 | 2021-08-06 | 山东大学 | Automatic cruise electric sickbed and elevator interaction method and system |
| CN112933579A (en) * | 2019-12-11 | 2021-06-11 | 中移(苏州)软件技术有限公司 | Motion quality evaluation method and device and storage medium |
| CN112933579B (en) * | 2019-12-11 | 2023-03-31 | 中移(苏州)软件技术有限公司 | Motion quality evaluation method and device and storage medium |
| CN111710129A (en) * | 2020-06-12 | 2020-09-25 | 电子科技大学 | Real-time pre-collision falling detection method for old people |
| CN111710129B (en) * | 2020-06-12 | 2021-06-01 | 电子科技大学 | Real-time pre-collision falling detection method for old people |
| CN112150766A (en) * | 2020-08-28 | 2020-12-29 | 永安行科技股份有限公司 | Early warning method and device for remote safety prevention and control |
| CN113706827A (en) * | 2021-09-03 | 2021-11-26 | 浙江远图互联科技股份有限公司 | Wireless acquisition system for vital signs of household old people |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110047247A (en) | A kind of smart home device accurately identifying Falls in Old People | |
| CN111027487B (en) | Behavior recognition system, method, medium and equipment based on multi-convolution kernel residual error network | |
| Chen et al. | Intelligent fall detection method based on accelerometer data from a wrist-worn smart watch | |
| Ashry et al. | CHARM-deep: Continuous human activity recognition model based on deep neural network using IMU sensors of smartwatch | |
| Xu et al. | Human activity recognition based on random forests | |
| Vu et al. | Transportation mode detection on mobile devices using recurrent nets | |
| Wisesa et al. | Fall detection algorithm based on accelerometer and gyroscope sensor data using Recurrent Neural Networks | |
| CN106580282A (en) | Human body health monitoring device, system and method | |
| Shiranthika et al. | Human activity recognition using CNN & LSTM | |
| WO2010083562A1 (en) | Activity detection | |
| CN111009321A (en) | Application method of machine learning classification model in juvenile autism auxiliary diagnosis | |
| CN109009145A (en) | A kind of tumble judgment method based on wearable device | |
| CN110084286A (en) | A kind of human motion recognition method of sensor-based ECOC technology | |
| CN109976526A (en) | A kind of sign Language Recognition Method based on surface myoelectric sensor and nine axle sensors | |
| Yang et al. | PD-ResNet for classification of Parkinson’s disease from gait | |
| CN106030246A (en) | Device, method and system for counting the number of cycles of a periodic movement of a subject | |
| CN107657277A (en) | A kind of human body unusual checking based on big data and decision method and system | |
| CN107945534A (en) | A kind of special bus method for predicting based on GMDH neutral nets | |
| Khokhlov et al. | Design of activity recognition systems with wearable sensors | |
| Wang et al. | Motion recognition for smart sports based on wearable inertial sensors | |
| Ding et al. | Cascaded convolutional neural network with attention mechanism for mobile EEG-based driver drowsiness detection system | |
| CN104586402A (en) | Feature extracting method for body activities | |
| Shi et al. | Design of motor skill recognition and hierarchical evaluation system for table tennis players | |
| CN109119155A (en) | ICU mortality prediction assessment system based on deep learning | |
| Cao et al. | Exploration of new community fitness mode using intelligent life technology and AIoT |
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: 20190723 |
|
| RJ01 | Rejection of invention patent application after publication |