CN109740761A - A kind of fall detection method based on multisensor Fusion Features - Google Patents
A kind of fall detection method based on multisensor Fusion Features Download PDFInfo
- Publication number
- CN109740761A CN109740761A CN201811452601.0A CN201811452601A CN109740761A CN 109740761 A CN109740761 A CN 109740761A CN 201811452601 A CN201811452601 A CN 201811452601A CN 109740761 A CN109740761 A CN 109740761A
- Authority
- CN
- China
- Prior art keywords
- data
- fall detection
- characteristic
- feature
- video image
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 230000004927 fusion Effects 0.000 title claims abstract description 11
- 230000001133 acceleration Effects 0.000 claims abstract description 37
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 18
- 239000000284 extract Substances 0.000 claims abstract description 12
- 238000013145 classification model Methods 0.000 claims abstract description 7
- 238000010276 construction Methods 0.000 claims abstract description 6
- 238000010801 machine learning Methods 0.000 claims abstract description 3
- 238000000034 method Methods 0.000 claims description 17
- 238000012706 support-vector machine Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000002123 temporal effect Effects 0.000 claims description 3
- 230000008901 benefit Effects 0.000 abstract description 2
- 238000012544 monitoring process Methods 0.000 description 6
- 230000006378 damage Effects 0.000 description 5
- 208000027418 Wounds and injury Diseases 0.000 description 3
- 208000014674 injury Diseases 0.000 description 3
- 206010000117 Abnormal behaviour Diseases 0.000 description 2
- 238000007635 classification algorithm Methods 0.000 description 2
- 230000004069 differentiation Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 208000003443 Unconsciousness Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000036760 body temperature Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000033764 rhythmic process Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of fall detection methods based on multisensor Fusion Features, comprising the following steps: S1: carrying out raw data acquisition, initial data includes acceleration information and video image data;S2: characteristic is carried out based on the collected initial data of step S1 and extracts simultaneously construction feature vector;S3: the characteristic extracted by machine learning algorithm training step S2 obtains svm algorithm classification model m;S4: real-time perfoming fall detection judgement.The present invention has many advantages, such as to cope with that complex situations, classification is accurate, Detection accuracy is high.
Description
Technical field
The present invention relates to the technical field of fall detection more particularly to a kind of tumbles based on multisensor Fusion Features
Detection method.
Background technique
The elderly is the disadvantaged group of society.With the increase at age, the body items function of old man all can slowly decline,
Phenomena such as action becomes slowly, and failure of memory and adaptability to changes are deteriorated, thus be easier with young man compared with to occur accident with
It comes to harm.The elderly is more fragile, once coming to harm, consequence is all often very serious.For example, gently can then be led after falling
The elderly is caused to remain unconscious, deformity or death heavy then that may cause the elderly.Therefore, when the elderly meets in daily life
When the similar unexpected injury fallen, it is badly in need of the timely relief of household or hospital.It is aged how the population that will face is coped with
Change problem improves the quality of life in the elderly's old age, it will is the important topic of country.
But today's society, along with the accelerating rhythm of life, the children for being in busy work are often difficult to take into account to old
The treatment of people, which results in the appearance of Empty nest elderly now.Up to the present, China's old solitary people enormous amount, these sky nests
What old man substantially enjoyed shows loving care for less than life.For society, solve the problems, such as that the monitoring of all the elderlys is unrealistic
's.Therefore, intelligentized real-time monitoring, which is necessary, to be realized to the elderly's daily life behavior, especially for such as falling
Etc. it be easy to cause the abnormal behaviour of the elderly's unexpected injury.By real-time monitoring, once the raw abnormal behaviour of old human hair, it can be with
It timely issues and alarms and send information to household and other caregivers, and then avoid the generation of great bodily injury.
Approximate solution: 2010, PHILIPS Co. was proposed Lifeline Emergency medical service system, can be quasi- in time
It really detects the tumble that old man occurs by accident or burst disease and requests to rescue.2010, Japanese WIN Human
Recorder company has developed a kind of wearable health detecting system for being fixed on chest, can pass through electrocardiogram, body temperature and body
The dynamic daily routines for waiting information monitorings old man.2012, Shenzhen Ai Fulai Science and Technology Ltd. was proposed " tumble automatic help hand
Machine " Ai Fulai A03 can be detected automatically when old man falls and be positioned and alarm, and ensure that old man is solitary and goes out period
It is healthy and safe.
Disadvantage: although having many fall detection schemes, there are still problems for the research of current fall detection scheme.
Traditional monitor and detection method is mainly video monitoring, also includes that acceleration transducer monitors, the methods of sonic sensor monitoring.
But traditional monitor and detection method is also only used only a certain sensor and is monitored based on recording.With collection mesh
The demand for marking the characteristic information of object is more and more, and single sensor collects the limitation of characteristic information due to it,
Through demand that cannot be current.
Summary of the invention
It can cope with that complex situations, classification be accurate, inspection it is an object of the invention to overcome the deficiencies of the prior art and provide a kind of
Survey the high fall detection method based on multisensor Fusion Features of accuracy rate
It is to achieve the above object, provided by the present invention that the technical scheme comprises the following steps:
S1: raw data acquisition is carried out, initial data includes acceleration information and video image data;
S2: characteristic is carried out based on the collected initial data of step S1 and extracts simultaneously construction feature vector;
S3: the characteristic extracted by machine learning algorithm training step S2 obtains svm algorithm classification model m;
S4: real-time perfoming fall detection judgement.
Further, acceleration information described in step S1 is acquired by being worn on the acceleration transducer of user's waist
It obtains;Video image data is collected by being mounted on the video image sensors in the place that user often occurs.
Further, the step S2 is based on the collected initial data progress characteristic extraction of step S1 and constructs spy
Levying vector, specific step is as follows:
For acceleration information:
A2-1: the timestamp of every acceleration information and every frame video image data is obtained, current frame image timestamp is taken
Acceleration information between previous frame image temporal stamp is as the corresponding acceleration information of current frame image;
A2-2: median filter process is carried out to this group of data in step S2-1;
A2-3: taking X, Y, Z axis acceleration amplitude component, and squared and evolution again obtains the acceleration amplitude of this group of data;
A2-4: take the maximum value in this group of amplitude as extraction acceleration amplitude feature 1a;
For video image data:
B2-1 takes current video frame data;
B2-2: video image binaryzation, morphology operations, motion smoothing, HSI are carried out to current video frame data and remove yin
After shadow, contours extract, the character contour of test object is obtained;
B2-3: external square aspect ratio features 2b, profile centroid feature 2c, external oblique ellipse declining angle are extracted according to character contour
Feature 2d;
Different characteristic finally based on the multiple sensors extracted in moment i constructs the characteristic vector α at current timei
=(1a, 2b, 2c, 2d).
Further, the characteristic that the step S3 is extracted by algorithm of support vector machine training step S2, specifically
Steps are as follows:
S3-1: according to video image data, judge the tumble state t_i of object to be detected in current frame image;
S3-2: judging every frame image, extracts all tumble label t=(t_1, t_2 ..., t_i);
S3-3: utilizing feature extracting method, obtains characteristic vector α _ i of present frame in training data;
S3-4: feature extraction is carried out to every frame image, extracts all characteristic vector α=(α _ 1, α _ 2 ..., α _ i);
S3-5: in conjunction with tumble label and characteristic vector, training sample set θ=(t, the α) of algorithm of support vector machine is obtained;
S3-6: being trained data using the training algorithm in the library libsvm, obtains the svm algorithm classification mould of sample set
Type m.
Further, specific step is as follows for the step S4 real-time perfoming fall detection judgement:
S4-1: each sensor raw data of current time t is obtained in real time;
S4-2: the characteristic 1a of each sensor of current time t is obtained using feature extracting methodt, 2bt, 2ct, 2dt, structure
Build characteristic vector α _ t;
S4-3: using the prediction technique in the library libsvm, the sample data svm algorithm classification model obtained based on step S3
M predicts α _ t;If prediction result is 1, judges that object to be detected is fallen, enable alarm processing immediately, remind detection
The physical condition of object;If prediction result is 0, judges that object to be detected is not fallen, and return step S4-1, carry out lower a period of time
The fall detection at quarter judges.
Compared with prior art, this programme principle and advantage is as follows:
Fall detection based on multisensor, comparison are used only the fall detection scheme of single-sensor, can cope with complexity
The case where.When there is the case where single-sensor is difficult differentiation, this programme passes through the data characteristics of comprehensive multiple sensors, right
Present case carries out Accurate classification.Such as acceleration transducer, traditional sentencing based on acceleration transducer dependent thresholds
Disconnected method changes similar jump to acceleration signature in practical situations and tumble acts, is difficult to distinguish the two, and
Gap of both scenes in video sensor feature is but very big.
Secondly, the accuracy rate for judgement of falling can be promoted.Traditional clustering threshold value is unsupervised segmentation algorithm, in Duo Te
The case where it is relatively low to levy accuracy rate in the case where judging, while being easy to appear over-fitting.And the support vector machines that this programme introduces is calculated
Rule is Supervised classification algorithm, and the accuracy rate of sorting algorithm, while support vector machines can be promoted by a large amount of training sample
It is suitble in algorithm principle to multiple features data classification, thus can be by increasing a greater variety of sensors and extracting their feature
As judgment basis, accuracy rate of the continuous lifting system to judgement of falling.
Detailed description of the invention
Fig. 1 is the tumble feature extraction of every frame image and the construction flow chart of characteristic vector in the present invention;
Fig. 2 is svm algorithm training sample set and to obtain the flow chart of disaggregated model in the present invention;
Fig. 3 is the flow chart of svm algorithm real-time perfoming fall detection judgement in the present invention.
Specific embodiment
The present invention is further explained in the light of specific embodiments:
Referring to figure 1, a kind of fall detection method based on multisensor Fusion Features described in the present embodiment, packet
Include following steps:
S1: raw data acquisition is carried out, initial data includes acceleration information and video image data;Wherein, accelerate
Degree evidence is collected by being worn on the acceleration transducer of user's waist;Video image data is frequent by being mounted on user
The video image sensors in the place of appearance collect.
S2: the specific steps that characteristic extracts simultaneously construction feature vector are carried out based on the collected initial data of step S1
It is as follows:
For acceleration information:
A2-1: the timestamp of every acceleration information and every frame video image data is obtained, current frame image timestamp is taken
Acceleration information between previous frame image temporal stamp is as the corresponding acceleration information of current frame image;
A2-2: median filter process is carried out to this group of data in step S2-1;
A2-3: taking X, Y, Z axis acceleration amplitude component, and squared and evolution again obtains the acceleration amplitude of this group of data;
A2-4: take the maximum value in this group of amplitude as extraction acceleration amplitude feature 1a;
For video image data:
B2-1 takes current video frame data;
B2-2: video image binaryzation, morphology operations, motion smoothing, HSI are carried out to current video frame data and remove yin
After shadow, contours extract, the character contour of test object is obtained;
B2-3: external square aspect ratio features 2b, profile centroid feature 2c, external oblique ellipse declining angle are extracted according to character contour
Feature 2d;
Different characteristic finally based on the multiple sensors extracted in moment i constructs the characteristic vector α at current timei
=(1a, 2b, 2c, 2d).
S3: the characteristic extracted by algorithm of support vector machine training step S2, the specific steps are as follows:
S3-1: according to video image data, judge the tumble state t_i of object to be detected in current frame image;
S3-2: judging every frame image, extracts all tumble label t=(t_1, t_2 ..., t_i);
S3-3: utilizing feature extracting method, obtains characteristic vector α _ i of present frame in training data;
S3-4: feature extraction is carried out to every frame image, extracts all characteristic vector α=(α _ 1, α _ 2 ..., α _ i);
S3-5: in conjunction with tumble label and characteristic vector, training sample set θ=(t, the α) of algorithm of support vector machine is obtained;
S3-6: being trained data using the training algorithm in the library libsvm, obtains the svm algorithm classification mould of sample set
Type m.
S4: real-time perfoming fall detection judgement, the specific steps are as follows:
S4-1: each sensor raw data of current time t is obtained in real time;
S4-2: the characteristic 1a of each sensor of current time t is obtained using feature extracting methodt, 2bt, 2ct, 2dt, structure
Build characteristic vector α _ t;
S4-3: using the prediction technique in the library libsvm, the sample data svm algorithm classification model obtained based on step S3
M predicts α _ t;If prediction result is 1, judges that object to be detected is fallen, enable alarm processing immediately, remind detection
The physical condition of object;If prediction result is 0, judges that object to be detected is not fallen, and return step S4-1, carry out lower a period of time
The fall detection at quarter judges.
The fall detection scheme of single-sensor, energy is used only in fall detection of the present embodiment based on multisensor, comparison
Cope with complicated situation.When there is the case where single-sensor is difficult differentiation, the present embodiment passes through comprehensive multiple sensors
Data characteristics carries out Accurate classification to present case.It is traditional based on acceleration transducer phase such as acceleration transducer
The judgment method of threshold value is closed, similar jump is changed to acceleration signature in practical situations and tumble acts, is difficult the two
It distinguishes, and gap of both scenes in video sensor feature is very big.
Secondly, the present embodiment can promote the accuracy rate for judgement of falling.Traditional clustering threshold value is that unsupervised segmentation is calculated
Method, in the case where multiple features judge, accuracy rate is relatively low, while the case where be easy to appear over-fitting.And the branch that the present embodiment introduces
Holding vector machine algorithm is Supervised classification algorithm, the accuracy rate of sorting algorithm can be promoted by a large amount of training sample, simultaneously
It is suitble in algorithm of support vector machine principle to multiple features data classification, thus can be by increasing a greater variety of sensors and extracting
Their feature is as judgment basis, accuracy rate of the continuous lifting system to judgement of falling.
The examples of implementation of the above are only the preferred embodiments of the invention, and implementation model of the invention is not limited with this
It encloses, therefore all shapes according to the present invention, changes made by principle, should all be included within the scope of protection of the present invention.
Claims (5)
1. a kind of fall detection method based on multisensor Fusion Features, which comprises the following steps:
S1: raw data acquisition is carried out, initial data includes acceleration information and video image data;
S2: characteristic is carried out based on the collected initial data of step S1 and extracts simultaneously construction feature vector;
S3: the characteristic extracted by machine learning algorithm training step S2 obtains svm algorithm classification model m;
S4: real-time perfoming fall detection judgement.
2. a kind of fall detection method based on multisensor Fusion Features according to claim 1, which is characterized in that step
Acceleration information described in rapid S1 is collected by being worn on the acceleration transducer of user's waist;Video image data is logical
It crosses and is mounted on the video image sensors in the place that user often occurs and collects.
3. a kind of fall detection method based on multisensor Fusion Features according to claim 1, which is characterized in that institute
It states step S2 and the specific steps of simultaneously construction feature vector is extracted such as based on the collected initial data progress characteristic of step S1
Under:
For acceleration information:
A2-1: obtain every acceleration information and every frame video image data timestamp, take current frame image timestamp with it is upper
Acceleration information between one frame image temporal stamp is as the corresponding acceleration information of current frame image;
A2-2: median filter process is carried out to this group of data in step S2-1;
A2-3: taking X, Y, Z axis acceleration amplitude component, and squared and evolution again obtains the acceleration amplitude of this group of data;
A2-4: take the maximum value in this group of amplitude as extraction acceleration amplitude feature 1a;
For video image data:
B2-1 takes current video frame data;
B2-2: video image binaryzation, morphology operations, motion smoothing, HSI are carried out to current video frame data and remove shade, wheel
After exterior feature extracts, the character contour of test object is obtained;
B2-3: external square aspect ratio features 2b, profile centroid feature 2c, external oblique ellipse declining angle feature are extracted according to character contour
2d;
Different characteristic finally based on the multiple sensors extracted in moment i constructs the characteristic vector α at current timei=(1a,
2b,2c,2d)。
4. a kind of fall detection method based on multisensor Fusion Features according to claim 1, which is characterized in that institute
State the characteristic that step S3 is extracted by algorithm of support vector machine training step S2, the specific steps are as follows:
S3-1: according to video image data, judge the tumble state t_i of object to be detected in current frame image;
S3-2: judging every frame image, obtains all tumble label t=(t_1, t_2 ..., t_i);
S3-3: utilizing feature extracting method, obtains characteristic vector α _ i of present frame in training data;
S3-4: feature extraction is carried out to every frame image, extracts all characteristic vector α=(α _ 1, α _ 2 ..., α _ i);
S3-5: in conjunction with tumble label and characteristic vector, training sample set θ=(t, the α) of algorithm of support vector machine is obtained;
S3-6: being trained data using the training algorithm in the library libsvm, obtains the svm algorithm classification model m of sample set.
5. a kind of fall detection method based on multisensor Fusion Features according to claim 1, which is characterized in that institute
Stating the judgement of step S4 real-time perfoming fall detection, specific step is as follows:
S4-1: each sensor raw data of current time t is obtained in real time;
S4-2: the characteristic 1a of each sensor of current time t is obtained using feature extracting methodt, 2bt, 2ct, 2dt, building spy
Levy vector α _ t;
S4-3: using the prediction technique in the library libsvm, the sample data svm algorithm classification model m obtained based on step S3 is right
α _ t is predicted;If prediction result is 1, judges that object to be detected is fallen, enable alarm processing immediately, remind test object
Physical condition;If prediction result is 0, judges that object to be detected is not fallen, and return step S4-1, carry out subsequent time
Fall detection judgement.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811452601.0A CN109740761A (en) | 2018-11-30 | 2018-11-30 | A kind of fall detection method based on multisensor Fusion Features |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811452601.0A CN109740761A (en) | 2018-11-30 | 2018-11-30 | A kind of fall detection method based on multisensor Fusion Features |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109740761A true CN109740761A (en) | 2019-05-10 |
Family
ID=66358329
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811452601.0A Pending CN109740761A (en) | 2018-11-30 | 2018-11-30 | A kind of fall detection method based on multisensor Fusion Features |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109740761A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110211334A (en) * | 2019-06-25 | 2019-09-06 | 启迪数华科技有限公司 | Campus Security alarming method for power and device based on big data neural network |
CN111008227A (en) * | 2019-12-27 | 2020-04-14 | 广西民族师范学院 | Data analysis processing platform |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011177234A (en) * | 2010-02-26 | 2011-09-15 | Tama Tlo Ltd | Fall detection system, fall detection device, fall detection method and program |
CN106991790A (en) * | 2017-05-27 | 2017-07-28 | 重庆大学 | Old man based on multimode signature analysis falls down method of real-time and system |
WO2018029193A1 (en) * | 2016-08-08 | 2018-02-15 | Koninklijke Philips N.V. | Device, system and method for fall detection |
CN108615050A (en) * | 2018-04-11 | 2018-10-02 | 南京邮电大学 | A kind of human body attitude method of discrimination based on support vector machines |
-
2018
- 2018-11-30 CN CN201811452601.0A patent/CN109740761A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011177234A (en) * | 2010-02-26 | 2011-09-15 | Tama Tlo Ltd | Fall detection system, fall detection device, fall detection method and program |
WO2018029193A1 (en) * | 2016-08-08 | 2018-02-15 | Koninklijke Philips N.V. | Device, system and method for fall detection |
CN106991790A (en) * | 2017-05-27 | 2017-07-28 | 重庆大学 | Old man based on multimode signature analysis falls down method of real-time and system |
CN108615050A (en) * | 2018-04-11 | 2018-10-02 | 南京邮电大学 | A kind of human body attitude method of discrimination based on support vector machines |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110211334A (en) * | 2019-06-25 | 2019-09-06 | 启迪数华科技有限公司 | Campus Security alarming method for power and device based on big data neural network |
CN111008227A (en) * | 2019-12-27 | 2020-04-14 | 广西民族师范学院 | Data analysis processing platform |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105448041B (en) | A kind of falling over of human body intelligence control system and its method | |
CN108629300B (en) | Fall detection method | |
CN110477925A (en) | A kind of fall detection for home for the aged old man and method for early warning and system | |
CN109166275B (en) | Human body falling detection method based on acceleration sensor | |
CN105760861B (en) | Epileptic seizure monitoring method and system based on depth data | |
CN103581852B (en) | The method, device and mobile terminal system of falling over of human body detection | |
CN103577836B (en) | Falling over of human body detection model method for building up and model system | |
CN108509897A (en) | A kind of human posture recognition method and system | |
CN106241584A (en) | A kind of intelligent video monitoring system based on staircase safety and method | |
CN112489368A (en) | Intelligent falling identification and detection alarm method and system | |
CN109389041A (en) | A kind of fall detection method based on joint point feature | |
CN111242004A (en) | Automatic alarm method and system based on elevator monitoring data processing | |
JP2008536208A (en) | Space-time self-organizing map | |
CN104361321A (en) | Methods of judging fall behaviors and body balance for old people | |
CN109740761A (en) | A kind of fall detection method based on multisensor Fusion Features | |
CN102488512A (en) | Automatic alarm system for electrocatdiogram monitoring and alarm method thereof | |
CN106407996A (en) | Machine learning based detection method and detection system for the fall of the old | |
CN105303183A (en) | Children's attitude identification and analysis system and method based on wearable equipment | |
Ren et al. | Low-power fall detection in home-based environments | |
CN110415152A (en) | A kind of safety monitoring system | |
CN115736902A (en) | Constructor management system based on intelligent wearable equipment | |
CN108898108A (en) | A kind of user's abnormal behaviour monitoring system and method based on sweeping robot | |
CN113679302A (en) | Monitoring method, device, equipment and storage medium based on sweeping robot | |
CN111626273A (en) | Fall behavior recognition system and method based on atomic action time sequence characteristics | |
Xie et al. | ART: adaptive and real-time fall detection using COTS smart watch |
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: 20190510 |
|
RJ01 | Rejection of invention patent application after publication |