CN102302370B - Method and device for detecting tumbling - Google Patents
Method and device for detecting tumbling Download PDFInfo
- Publication number
- CN102302370B CN102302370B CN 201110180341 CN201110180341A CN102302370B CN 102302370 B CN102302370 B CN 102302370B CN 201110180341 CN201110180341 CN 201110180341 CN 201110180341 A CN201110180341 A CN 201110180341A CN 102302370 B CN102302370 B CN 102302370B
- Authority
- CN
- China
- Prior art keywords
- behavior
- data
- model
- regular
- fall detection
- 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.)
- Active
Links
Images
Landscapes
- Alarm Systems (AREA)
Abstract
The invention provides a method and device for detecting tumbling. The method provided by the invention comprises the steps of: firstly, collecting behavioral data of a user; secondly, identifying behavior of the user according to the behavioral data of the user; thirdly, segmenting out behavioral switching data from the collected data according to a behavioral identification result, and warping the behavioral switching data into equilong characteristic vectors; and fourthly, performing tumbling detection according to the warped characteristic vectors. According to the method and device provided by the invention, tumbling detection is performed on the basis of behavioral switching, a great amount of normal behavioral data can be filtered, the complexity of a feature space is reduced, the separating capacity of models is enhanced, and the model detection rate is improved.
Description
Technical field
The present invention relates to general fit calculation and health supervision field, be specifically related to a kind of fall detection method and apparatus.
Background technology
According to world's disease control and pre-preventing tissue statistics, have approximately 1/3rd over-65s at home old man every year above falling once, nearly 10% can cause grievous injury or disease in disposable the falling.Falling becomes one of key factor of harm old people and other special populations.Timely fall detection and relief can be won the quality time for treatment and rescue, to the independent living ability that improves the user, ensure that it is healthy and improve the medical monitoring level and all have very important effect.Current, China has entered aging society and has presented accelerated development situation, and old solitary people and " family not living home " increase gradually, and the urgent need research and development are easy to carry, detection is accurate, judge real-time falling detection device, with satisfied widely social need.
Existing fall detection method can roughly be divided into two classes: based on the detection method of threshold value with based on the detection method of model.
Detection method based on threshold value is the relevant parameter setting threshold in advance, if related parameter values does not satisfy threshold condition in the user behavior process, then is judged as and falls.For example, patent application 200910021227.3 is according to the acceleration transducer data, utilize first cut-off frequency to carry out filtering for the high pass filter of 0.5Hz, data flow after filtering is cut into the data segment that a plurality of units were 1 second, then calculate variance and the intermediate value of every section window, assessed value adopted average for ten seconds, fell by assessed value is compared to detect with predefined threshold value; Patent application 200880012293.8 is by calculating the amplitude of acceleration signal X, Y, Z
And judge that when surpassing setting threshold the user falls.For the alert rate of the mistake that reduces to detect, this invention by the whether movement of falling object of checkout equipment, whether equipment rotates and next-door neighbour's degree of equipment and user's body is come the filtered noise data; Patent application 200910145045.7 gathers the three-dimensional acceleration information of metastomium on the user, and information carried out fusion treatment, comprehensive human body be hit and impact before and after go up angle of inclination of body variation judge whether to fall down, when other impact generation of level is fallen down in judgement, use acceleration 〉=threshold value 3.5g to be standard; Patent application US2011025493-A1 adopts the composite value of acceleration information, and the passing threshold method detects the weightlessness of the moment of falling and the overweight state of the moment of contacting to earth and identifies and fall; Patent application US2009048540-A1 adopts threshold method to come instantaneous variation and the change of user's body from the vertical direction to the horizontal direction of sense acceleration; The acceleration that patent application US2010121603-A1 arranges thigh direction threshold value, thigh and waist concerns the change threshold of threshold value and thigh direction, if testing result surpasses these three threshold values in real time, thinks that then the user falls.In addition, adopt that the related application of threshold method also has 200810204106.8,201010265588.5, US2006279426-A1, EP1870037-A1, AU2009247584-A1, US2009292227-A1, WO2010126878-A1 etc.
Detection method general using machine learning algorithm based on model is distinguished normal behaviour and Deviant Behavior (falling) from training data learning classification model.The mode that patent application 201010285585.8 passing threshold methods and model method combine, after threshold decision, utilize a class support vector machines (Support Vector Machine, SVM) carry out pattern recognition, carrying out secondary judges, test vector and sample set are compared, when test vector is not within sample set, is judged as and falls; Document [J.Yin, Q.Yang, J.J.Pan, Sensor-Based Abnormal Human-ActivityDetection, IEEE Transactions on Knowledge and Data Engineering, v.20 (8), pp.1082-1090,2008.] from walk, stair activity, the normal behaviour data such as run, sit down extract feature, utilize first a class SVM to fall the normal sample of high probability as the abnormality detection model filter, recycling nuclear nonlinear regression model (NLRM) is with the model of cognition of unsupervised mode build exception sample.
Existing detection method is not filtered efficiently to detecting data, but to user's Continuous behavior data analysis.The transients of (as running, go downstairs etc.) is higher with the similarity of falling because indivedual normal behaviours, and the impact of additive noise behavioral data is so reduced the accuracy rate of detection; In addition, existing detection method is difficult to realize the balance between verification and measurement ratio and the alert rate of mistake: the artificial setting threshold of threshold method needs is big or small, the defined differentiation of model method border also is subjected to the impact of relevant parameter, is difficult to find the optimum between normal behaviour and the Deviant Behavior (falling) to distinguish the border.So although fall detection has several different methods now, existing method all can not satisfy fall detection simultaneously to the requirement of high detection rate and low mistake police's rate.Yet fall detection is the problem of a cost-sensitive, on the one hand, the undetected survey of falling is compared the undetected survey consequence of normal behaviour and is wanted much serious, requires model that high verification and measurement ratio is arranged; On the other hand, false alarm can cause user's dislike frequently, reduces it to the degree of belief of checkout gear, is unfavorable for practical application and the popularization of device.So, be badly in need of at present high and low detection method and the device of the alert rate of mistake of verification and measurement ratio.
Summary of the invention
The technical problem to be solved in the present invention is to realize satisfying simultaneously the fall detection method of high detection rate and the alert rate of low mistake.
According to one aspect of the invention, a kind of fall detection method is provided, comprising:
1) gathers user behavior data;
2) according to user behavior data identification user behavior;
3) from the data that gather, be partitioned into the behavior switch data according to the behavior recognition result, and with the isometric characteristic vector of the regular one-tenth of behavior switch data;
4) carry out fall detection according to the characteristic vector after regular.
In said method, described step 2) further comprise:
21) the user behavior data intercepting is window data;
22) the window data extraction feature of intercepting formed test sample book;
23) utilize the behavior model of cognition that test sample book is carried out user behavior identification.
In said method, described step 21) also comprises after: window data is carried out pretreatment.
In said method, described pretreatment comprises that the vacancy value fills up and data filtering.
In said method, described feature comprises average, standard variance, zero-crossing rate, percentile, coefficient of association, power spectral density, frequency domain entropy and/or spectrum peak position.
In said method, described behavior model of cognition obtains by off-line training.
In said method, described behavior model of cognition is Decision-Tree Classifier Model, supporting vector machine model, multilayer perceptron neutral net or hidden Markov model.
In said method, described step 3) regular described in is to utilize to observe the HMM of density finish based on Gauss.
In said method, described step 3) regular described in is the feature of directly extracting described behavior switch data, and the Feature Conversion of extracting is become isometric vector.
In said method, described step 4) fall detection is to utilize a class support vector machines model to carry out described in.
In said method, utilize a class support vector machines model to carry out fall detection after, for the test sample book that does not belong to the behavior of falling, recycling weighting k nearest neighbor algorithm further determines whether really not belong to the behavior of falling.
According to a further aspect of the invention, also provide a kind of falling detection device, having comprised:
Data acquisition module is used for gathering user behavior data;
The behavior recognition device is used for according to user behavior data identification user behavior;
Data processing module is used for being partitioned into the behavior switch data according to the behavior recognition result from the data that gather, and with the isometric characteristic vector of the regular one-tenth of behavior switch data;
The fall detection module is used for carrying out fall detection according to the characteristic vector after regular.
In said apparatus, described fall detection module comprises first detection module, is used for utilizing a class support vector machines model to carry out fall detection according to the characteristic vector after regular.
In said apparatus, described fall detection module also comprises the second detection module, be used for after described first detection module carries out fall detection, for the test sample book that does not belong to the behavior of falling, recycling weighting k nearest neighbor algorithm further determines whether really not belong to the behavior of falling.
In said apparatus, described user behavior data is from accelerometer and/or gyroscope.
In said apparatus, it is to utilize to observe the HMM of density finish based on Gauss that described data processing module carries out regular to the behavior switch data.
Than prior art, detection method provided by the invention and the device with the behavior switch data as detected object, with fall detection from transferring to the behavior switch data as feature space as feature space with the normal behaviour data, can effectively filter normal behaviour data and noise data, reduce the complexity of feature space, improve power of test and the accuracy rate of model.In addition, the detection method of the preferred embodiments of the present invention and the related anomaly association detection model of device have merged the advantage of model detection algorithm and distance detection algorithm, can improve the separating capacity of model, thus better implementation model verification and measurement ratio and the by mistake balance of police's rate.
Description of drawings
Fig. 1 (a) and Fig. 1 (b) are respectively the sketch maps of switching place that occurs in different behaviors and sudden change place that occurs in identical behavior of falling of falling;
Fig. 2 is the main functional modules sketch map of fall detection method in accordance with a preferred embodiment of the present invention;
Fig. 3 is the flow chart of fall detection method in accordance with a preferred embodiment of the present invention;
Fig. 4 is the flow chart of behavior identification in accordance with a preferred embodiment of the present invention;
Fig. 5 is the classification boundaries sketch map of anomaly association behavior detection model in accordance with a preferred embodiment of the present invention.
The specific embodiment
In order to make purpose of the present invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing, fall detection method and apparatus is according to an embodiment of the invention further described.Should be appreciated that specific embodiment described herein only in order to explain the present invention, is not intended to limit the present invention.
In actual life, switching place or sudden change place of falling and generally occurring in behavior are shown in Fig. 1 (a) and Fig. 1 (b).Wherein Fig. 1 (a) shows to fall and occurs in switching place of different behaviors, and particularly, the people has occured to fall when running, and then becomes static (lying low).Fig. 1 (b) shows to fall and occurs in sudden change place of identical behavior, and particularly, the people falls in a swoon suddenly when standing (static), and (lying low) still remains static after falling in a swoon.
In order effectively to filter normal behaviour data and noise data, reduce the complexity of feature space, strengthen the separating capacity of detection method, realize simultaneously high detection rate and the alert rate of low mistake, the present invention is based on above-mentioned objective fact a kind of fall detection method and device that switches based on behavior is provided.As shown in Figure 2, this fall detection method comprises three steps: behavior identification, behavior switch data are cut apart and regular, abnormality detection.Particularly, at first, utilize the behavior recognition methods that user's Continuous behavior data are identified according to the behavioral data of sensor acquisition; Then, according to the result of behavior identification user's Continuous behavior data are cut apart, the switch data between extracting adjacent lines and being filters out a large amount of normal behaviour data and noise data, and with the isometric characteristic vector of the regular one-tenth of this switch data; At last, utilize method for detecting abnormality, the behavior switch data is carried out abnormality detection, identify unusual behavior switch data, think the event of falling has occured and to report to the police.
Fig. 3 shows the flow chart of fall detection method in accordance with a preferred embodiment of the present invention, and the below is described in detail fall detection method of the present invention according to this flow chart.
1, behavior identification
The behavior model of cognition that behavior identification can utilize training to obtain carries out.Show model training and two parts of ONLINE RECOGNITION with solid line part and dotted portion respectively among Fig. 4.Shown in solid line part among Fig. 4, model of cognition training part specifically comprises the following steps:
Utilize the sensing data of every kind of typical behaviors of equipment collection user such as accelerometer, gyroscope, utilize length for the window of l sensing data to be intercepted, wherein should typical case's behavior such as static, walk, running, stair activity etc.;
Each window data to intercepting utilizes preprocessing algorithms that data are carried out pretreatment, and Preprocessing Algorithm includes but not limited to that the vacancy value is filled up, data filtering etc.The vacancy value is filled up for additional missing data, makes it to become continuous time series data; Data filtering is used for filtering out isolated point data or high-frequency noise data etc.;
Window data after pretreatment extracts feature, with the feature composition characteristic vector that extracts, and add the behavior category label, generate training sample, typical feature includes but not limited to average, standard variance, zero-crossing rate, percentile, coefficient of association, power spectral density, frequency domain entropy, spectrum peak position etc.;
Training set according to being comprised of all training samples utilizes decision tree (Decision Tree, DT) sorting algorithm, training DT multicategory classification model, i.e. behavior model of cognition.
One of ordinary skill in the art will appreciate that, adopted simple among this embodiment of the present invention and effective DT disaggregated model, but one of ordinary skill in the art will appreciate that the multicategory classification model that also can adopt other, for example: SVM, multilayer perceptron neutral net (Multi-Layer Perceptron neural network, MLP) and HMM (Hidden Markov Model, HMM) etc.
It is similar to utilize above-mentioned behavior model of cognition to carry out process and the training process of behavior identification, as the solid line of Fig. 4 partly shown in, specifically comprise the following steps:
10) by the sensor acquisition user behavior data, utilize length for the window of l sensing data to be intercepted;
20) to each window data of intercepting, utilize preprocessing algorithms that data are carried out pretreatment;
30) window data after pretreatment extracts feature, and with the feature composition test sample book of this extraction, the difference of this test sample book and above-mentioned training sample is that it does not comprise the behavior category label;
40) utilize the behavior model of cognition that test sample book is carried out behavior identification.
One of ordinary skill in the art will appreciate that the training of above-mentioned behavior model of cognition can off-line execution, above-mentioned behavior ONLINE RECOGNITION process normally repeats, continuous.
2, the behavior switch data is cut apart and is regular
10) according to the behavior recognition result, judge whether generation behavior switching, if the generation behavior is switched, then the Resurvey data if the behavior switching has occured, then carry out steps 20).
20) part of switching for the generation behavior, the behavior switch data between from the raw sensor sampled data, being partitioned into adjacent lines and being.General, the persistent period of behavior handoff procedure is shorter, often less than the length of a behavior identification window.Suppose with A, B, C, D ... represent the behavior classification, then the subordinate act A process that switches to behavior B may present following three kinds of recognition results: 1) AA...AB...BB; 2) AA...ACB...BB; 3) AA...ACDB...BB, wherein C can be identical with D, be that the behavior switch data may be included in the cut-off window and initial window of adjacent behavior, in one or two transition windows between also may being included in adjacent lines and being, and the behavioral data of transition windows may be identified as other behavior classification.The corresponding primitive behavior data of above-mentioned three kinds of situations are cut apart, extracted respectively AB, ACB, ACDB as the behavior switch data.
30) because the length of behavior switch data is incomplete same, need to carries out regular its length that makes to it and could utilize the abnormality detection model to identify after identical.Suppose to have N section normal behaviour switch data Y
i, 1≤i≤N is divided into the M class.Use the Baum-Welch algorithm to observe the HMM model of density, the corresponding parameter lambda of each model for one of each class data training based on Gauss
j, 1≤j≤M.Logarithm similarity between every one piece of data and each HMM model can be expressed as:
L(Y
i；λ
j)＝log P(Y
i|λ
j)，1≤i≤N，1≤j≤M (1)
Accordingly, the behavior switch data Y that is uneven in length
iCan the isometric characteristic vector x of regular one-tenth
i=＜L (Y
iλ
1) ..., L (Y
iλ
M).Thus, any one section behavior switch data can be according to the isometric characteristic vector of the regular one-tenth of said method.
Certainly, except the regular method of above-mentioned data, also can directly extract feature to the behavior switch data, convert thereof into isometric vector, the feature that wherein can be used for extracting includes but not limited to average, standard variance, zero-crossing rate etc.
3, abnormality detection
The abnormality detection process comprises the following steps:
10) detect the event of whether falling, if do not fall event, then the Resurvey data if the event of falling has occured, then carry out steps 20);
At first, by training to obtain the model for detection of the event of falling.
The vector that obtains after the normal behaviour switch data is regular is as training sample (normal sample), utilize a class support vector machines (Support Vector Machine, SVM) Algorithm for Training one class svm classifier model, and by regulating " reject rate " parameter δ (0≤δ≤1) the normal sample of preset proportion is got rid of outside the border to realize " the compacting " on border, make a distinction first with the sample that confidence level is large, as shown in Figure 5.
At online detection-phase, the behavior switch data that needs are detected carries out regular, utilizes vector after regular as test sample book, inputs a class SVM model.If test sample book drops within the classifying face, then be judged to be normal.
Preferably, if test sample book does not drop within the classifying face, then utilize weighting k nearest neighbor (KNearest Neighbor, KNN) algorithm is judged again, wherein weighting KNN algorithm is a kind of " laziness " learning algorithm, do not need model training, only need to keep all normal samples so that the distance between online detection-phase calculating test sample book and training sample.Suppose test sample book x and its k arest neighbors training sample x
1..., x
kDistance and distance weighting be respectively d
1..., d
kAnd ω
1..., ω
k, then the weighted average position of k arest neighbors training sample is
Calculate sample x with
Between Euclidean distance
If
Then x belongs to normal sample, otherwise is judged as unusual.Distance threshold τ (0＜τ＜+∞) be used for regulating detection sensitivity, also can solve a traditional class KNN algorithm too loose problem of detection boundaries in the sparse situation of sample.By utilizing this weighting KNN algorithm, the Binding distance threshold value is set classification boundaries, and the normal sample nearer apart from training sample originally made a distinction, and has avoided the too loose problem of classification boundaries under the sparse condition of sample.
Combinations thereof abnormality detection model can further improve the separating capacity of model in conjunction with the advantage of a class SVM and weighting k nearest neighbor, effectively the balance of implementation model verification and measurement ratio and the alert rate of mistake.
One of ordinary skill in the art will appreciate that, except an above-mentioned class SVM and weighting k nearest neighbor, can also adopt any other a class sorting algorithm or make up a class sorting algorithm, for example: a class K mean algorithm, a class principal component analysis (Principal Component Analysis, PCA) algorithm, a class Self-organizing Maps (Self-Organizing Map, SOM) algorithm etc.
20) affair alarm of falling;
30) determine whether continue to detect, if so, Resurvey data then, otherwise, finish.
According to a further aspect of the invention, also provide a kind of falling detection device, having comprised:
Data acquisition module is used for gathering user behavior data;
The behavior recognition device is used for according to user behavior data identification user behavior;
Data processing module is used for being partitioned into the behavior switch data according to the behavior recognition result from the data that gather, and with the isometric characteristic vector of the regular one-tenth of behavior switch data;
The fall detection module is used for carrying out fall detection according to the characteristic vector after regular.
This fall detection module comprises first detection module, is used for utilizing a class support vector machines model to carry out fall detection according to the characteristic vector after regular.
This fall detection module also comprises the second detection module, is used for after described first detection module carries out fall detection, and for the test sample book that does not belong to the behavior of falling, recycling weighting k nearest neighbor algorithm further determines whether really not belong to the behavior of falling.
In the said apparatus, described data acquisition module gathers user behavior data from accelerometer and/or gyroscope.
In the said apparatus, it is to utilize to observe the HMM of density finish based on Gauss that described data processing module carries out regular to the behavior switch data.
In sum, the behavior that the present invention is based on is switched and is carried out fall detection, extract feature construction fall detection model from the normal switch data between adjacent behavior, namely with behavior switch data construction feature space, can filter out a large amount of normal behaviour data, reduce the complexity of feature space, strengthen the separating capacity of model, improve the verification and measurement ratio of model.
Should be noted that and understand, in the situation that does not break away from the desired the spirit and scope of the present invention of accompanying claim, can make to the present invention of foregoing detailed description various modifications and improvement.Therefore, the scope of claimed technical scheme is not subjected to the restriction of given any specific exemplary teachings.
Claims (16)
1. fall detection method comprises:
1) gathers user behavior data;
2) according to user behavior data identification user behavior;
3) from the data that gather, be partitioned into the behavior switch data according to the behavior recognition result, and with the isometric characteristic vector of the regular one-tenth of behavior switch data;
4) carry out fall detection according to the characteristic vector after regular.
2. method according to claim 1 is characterized in that, described step 2) further comprise:
21) the user behavior data intercepting is window data;
22) the window data extraction feature of intercepting formed test sample book;
23) utilize the behavior model of cognition that test sample book is carried out user behavior identification.
3. method according to claim 2 is characterized in that, described step 21) after also comprise: window data is carried out pretreatment.
4. method according to claim 3 is characterized in that, described pretreatment comprises that the vacancy value fills up and data filtering.
5. according to claim 2 to 4 each described methods, it is characterized in that described feature comprises average, standard variance, zero-crossing rate, percentile, coefficient of association, power spectral density, frequency domain entropy and/or spectrum peak position.
6. according to claim 2 to 4 each described methods, it is characterized in that described behavior model of cognition obtains by off-line training.
7. according to claim 2 to 4 each described methods, it is characterized in that described behavior model of cognition is Decision-Tree Classifier Model, supporting vector machine model, multilayer perceptron neutral net or hidden Markov model.
8. according to claim 1 to 4 each described methods, it is characterized in that regular described in the described step 3) is to utilize to observe the HMM of density finish based on Gauss.
9. according to claim 1 to 4 each described methods, it is characterized in that regular described in the described step 3) is the feature of directly extracting described behavior switch data, and the Feature Conversion of extracting is become isometric characteristic vector.
10. according to claim 1 to 4 each described methods, it is characterized in that fall detection described in the described step 4) is to utilize a class support vector machines model to carry out.
11. method according to claim 10 is characterized in that, utilize a class support vector machines model to carry out fall detection after, for the test sample book that does not belong to the behavior of falling, recycling weighting k nearest neighbor algorithm further determines whether really not belong to the behavior of falling.
12. a falling detection device comprises:
Data acquisition module is used for gathering user behavior data;
The behavior recognition device is used for according to user behavior data identification user behavior;
Data processing module is used for being partitioned into the behavior switch data according to the behavior recognition result from the data that gather, and with the isometric characteristic vector of the regular one-tenth of behavior switch data;
The fall detection module is used for carrying out fall detection according to the characteristic vector after regular.
13. device according to claim 12 is characterized in that, described fall detection module comprises first detection module, is used for utilizing a class support vector machines model to carry out fall detection according to the characteristic vector after regular.
14. device according to claim 13, it is characterized in that, described fall detection module also comprises the second detection module, be used for after described first detection module carries out fall detection, for the test sample book that does not belong to the behavior of falling, recycling weighting k nearest neighbor algorithm further determines whether really not belong to the behavior of falling.
15. to 14 each described devices, it is characterized in that described user behavior data is from accelerometer and/or gyroscope according to claim 12.
16. to 14 each described devices, it is characterized in that according to claim 12 it is to utilize to observe the HMM of density finish based on Gauss that described data processing module carries out regular to the behavior switch data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110180341 CN102302370B (en) | 2011-06-30 | 2011-06-30 | Method and device for detecting tumbling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110180341 CN102302370B (en) | 2011-06-30 | 2011-06-30 | Method and device for detecting tumbling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102302370A CN102302370A (en) | 2012-01-04 |
| CN102302370B true CN102302370B (en) | 2013-01-30 |
Family
ID=45376365
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201110180341 Active CN102302370B (en) | 2011-06-30 | 2011-06-30 | Method and device for detecting tumbling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102302370B (en) |
Families Citing this family (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102610056B (en) * | 2012-03-16 | 2013-09-04 | 清华大学 | Mobile phone wearing mode-oriented falling event detection system and method |
| CN103070690B (en) * | 2012-12-04 | 2016-03-16 | 东蓝数码有限公司 | The control method of intelligent terminal's monitor system |
| CN103251411A (en) * | 2013-04-08 | 2013-08-21 | 杭州电子科技大学 | Complexity based pressure center nonlinear feature extraction method |
| CN103248955B (en) * | 2013-04-22 | 2017-07-28 | 深圳Tcl新技术有限公司 | Personal identification method and device based on intelligent remote control system |
| CN103295364B (en) * | 2013-05-15 | 2016-06-01 | 上海交通大学 | A kind of fall events detection system based on two mirror |
| CN103417219A (en) * | 2013-09-11 | 2013-12-04 | 重庆大学 | Wearable human body falling detection device |
| CN104586398A (en) * | 2013-10-30 | 2015-05-06 | 广州华久信息科技有限公司 | Old man falling detecting method and system based on multi-sensor fusion |
| CN103970271B (en) * | 2014-04-04 | 2017-06-20 | 浙江大学 | The daily routines recognition methods of fusional movement and physiology sensing data |
| CN103927851B (en) * | 2014-04-04 | 2016-06-29 | 余志峰 | A kind of individualized multi thresholds fall detection method and system |
| CN104392583A (en) * | 2014-11-27 | 2015-03-04 | 北京工业大学 | Fall detection and alarm system and method based on KNN algorithm |
| CN104510475A (en) * | 2014-12-15 | 2015-04-15 | 中国科学院计算技术研究所 | Human body falling-down detection method and system |
| CN104598880A (en) * | 2015-03-06 | 2015-05-06 | 中山大学 | Behavior identification method based on fuzzy support vector machine |
| CN105101092A (en) * | 2015-09-01 | 2015-11-25 | 上海美慧软件有限公司 | Mobile phone user travel mode recognition method based on C4.5 decision tree |
| CN105125220A (en) * | 2015-10-20 | 2015-12-09 | 重庆软汇科技股份有限公司 | Falling-down detection method |
| EP3185205A1 (en) * | 2015-12-21 | 2017-06-28 | Thomson Licensing | Method and device for detecting behavioral patterns of a user |
| CN105708470B (en) * | 2016-01-21 | 2018-11-13 | 山东大学 | Based on doppler detector and the united fall detection system of sensor and method |
| CN105869354B (en) * | 2016-04-29 | 2017-12-01 | 华南理工大学 | A kind of Falls Among Old People detection method based on attractor propagation algorithm |
| CN105997094B (en) * | 2016-05-09 | 2019-03-29 | 北京科技大学 | A kind of gesture recognition device and method |
| CN105873110B (en) * | 2016-05-27 | 2020-01-10 | 集道成科技(北京)有限公司 | User behavior correction method and device |
| CN107886103B (en) * | 2016-09-29 | 2023-12-08 | 日本电气株式会社 | Method, device and system for identifying behavior patterns |
| CN107995011B (en) * | 2016-10-26 | 2022-07-22 | 通用电气公司 | Alarm control method and device for patient monitor and self-adaptive alarm method for patient monitor |
| CN106618499B (en) * | 2016-12-14 | 2020-11-10 | 深圳先进技术研究院 | Fall detection device, fall detection method and device |
| CN106709471A (en) * | 2017-01-05 | 2017-05-24 | 宇龙计算机通信科技(深圳)有限公司 | Fall detection method and device |
| CN106725495A (en) * | 2017-01-13 | 2017-05-31 | 深圳先进技术研究院 | A kind of fall detection method, apparatus and system |
| CN106875630B (en) * | 2017-03-13 | 2018-12-04 | 中国科学院计算技术研究所 | A kind of wearable fall detection method and system based on hierarchical classification |
| CN108932217A (en) * | 2017-05-27 | 2018-12-04 | 深圳市中电电力技术股份有限公司 | The method and device of energy consumption statistic |
| CN108399712A (en) * | 2018-03-01 | 2018-08-14 | 深圳明创自控技术有限公司 | A kind of monitoring and warning system for capableing of early warning in time |
| CN109685111B (en) * | 2018-11-26 | 2023-04-07 | 深圳先进技术研究院 | Motion recognition method, computing system, intelligent device and storage medium |
| CN111403039A (en) * | 2020-03-19 | 2020-07-10 | 中国科学院深圳先进技术研究院 | Dynamic balance evaluation method, device, equipment and medium |
| CN111540168A (en) * | 2020-04-20 | 2020-08-14 | 金科龙软件科技(深圳)有限公司 | Tumble detection method and equipment and storage medium |
| CN112861696B (en) * | 2021-02-01 | 2023-08-18 | 电子科技大学中山学院 | Abnormal behavior identification method and device, electronic equipment and storage medium |
| CN113989570A (en) * | 2021-11-01 | 2022-01-28 | 南通大学 | Fall detection alarm method based on embedded edge artificial intelligence |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101950464A (en) * | 2010-09-17 | 2011-01-19 | 中国科学院深圳先进技术研究院 | Method and system for fall monitoring and warning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5120795B2 (en) * | 2005-11-15 | 2013-01-16 | 学校法人日本大学 | Human posture motion discrimination device and energy consumption calculation device |
-
2011
- 2011-06-30 CN CN 201110180341 patent/CN102302370B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101950464A (en) * | 2010-09-17 | 2011-01-19 | 中国科学院深圳先进技术研究院 | Method and system for fall monitoring and warning |
Non-Patent Citations (5)
| Title |
|---|
| Jie Yin etc..Sensor-based abnormal human-activity detection.《IEEE Transactions on Knowledge and Data Engineering》.2008,第20卷(第8期),1082-1090. |
| JP特开2007-160076A 2007.06.28 |
| Sensor-based abnormal human-activity detection;Jie Yin etc.;《IEEE Transactions on Knowledge and Data Engineering》;20080831;第20卷(第8期);第1082-1090页 * |
| 王双全.申请上海交通大学工学博士学位论文基于无线传感器网络的行为识别与目标定位研究.《中国博士学位论文全文数据库信息科学辑》.2011,(第4期),I140-39. |
| 申请上海交通大学工学博士学位论文基于无线传感器网络的行为识别与目标定位研究;王双全;《中国博士学位论文全文数据库信息科学辑》;20110415(第4期);正文第13-83页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102302370A (en) | 2012-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102302370B (en) | Method and device for detecting tumbling | |
| CN111209848B (en) | Real-time falling detection method based on deep learning | |
| CN104665849B (en) | A kind of high ferro dispatcher based on the interaction of many physiological signals multi-model stress detection method | |
| CN107422643A (en) | Smart home monitoring method and system based on vibration detection | |
| CN103323749B (en) | The partial discharge diagnostic method of multi-classifier information fusion | |
| CN111436944B (en) | Falling detection method based on intelligent mobile terminal | |
| CN109166275B (en) | Human body falling detection method based on acceleration sensor | |
| CN106714220A (en) | WSN (Wireless Sensor Network) anomaly detection method based on MEA-BP neural network | |
| CN108509897A (en) | A kind of human posture recognition method and system | |
| CN108511055B (en) | Ventricular premature beat recognition system and method based on classifier fusion and diagnosis rules | |
| CN105139029B (en) | A kind of Activity recognition method and device of prison prisoner | |
| CN106910314B (en) | A kind of personalized fall detection method based on the bodily form | |
| CN107609477A (en) | It is a kind of that detection method is fallen down with what Intelligent bracelet was combined based on deep learning | |
| CN106228200A (en) | A kind of action identification method not relying on action message collecting device | |
| CN108710822A (en) | Personnel falling detection system based on infrared array sensor | |
| CN109243604A (en) | A kind of construction method and building system of the Kawasaki disease risk evaluation model based on neural network algorithm | |
| CN111931568A (en) | Human body falling detection method and system based on enhanced learning | |
| Ning et al. | Fall detection algorithm based on gradient boosting decision tree | |
| CN107019501B (en) | Remote tumble detection method and system based on genetic algorithm and probabilistic neural network | |
| CN109567832A (en) | A kind of method and system of the angry driving condition of detection based on Intelligent bracelet | |
| CN114881079A (en) | Human body movement intention abnormity detection method and system for wearable sensor | |
| CN105869354A (en) | Method for old man falling-down detection based on attractor propagation algorithm | |
| CN113095379A (en) | Human motion state identification method based on wearable six-axis sensing data | |
| CN107007285A (en) | Fall detection method based on pressure and acceleration transducer | |
| CN111646332A (en) | Method and system for identifying abnormal operation of elevator |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |