CN106096662B - Human motion state identification based on acceleration transducer - Google Patents
Human motion state identification based on acceleration transducer Download PDFInfo
- Publication number
- CN106096662B CN106096662B CN201610472604.5A CN201610472604A CN106096662B CN 106096662 B CN106096662 B CN 106096662B CN 201610472604 A CN201610472604 A CN 201610472604A CN 106096662 B CN106096662 B CN 106096662B
- Authority
- CN
- China
- Prior art keywords
- data
- cluster
- human motion
- window
- motion state
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The present invention provides a kind of human motion state recognition methods and system based on acceleration transducer, are divided into off-line phase and on-line stage；Wherein, off-line phase constructs human motion state identification model using K-Means clustering method, and the data based on existing tape label are trained research, proposes classification policy；Then on-line stage is based on Android phone and designs human body moving state identification real-time system, is designed respectively from 5 functions such as data acquisition, data processing, movement identification, model modification, data displayings；Finally by experiments have shown that clustering algorithm validity, the experimental results showed that, be feasible based on clustering method building human motion identification model, and the model has many advantages, such as that real-time is good, light weight easily adjusts.
Description
Technical field
The invention belongs to a kind of design of the clustering algorithm in data mining technology field more particularly to data mining classification
Method.
Background technique
As development of Mobile Internet technology rises the development with wireless sensor technology, the moment is all generating sensing data,
These data contain information abundant, have far-reaching research significance.Being widely used for pedometer is exactly that one of those grinds
Study carefully achievement.Movement identification is also a wherein more popular direction at present, by analyzing acceleration transducer data, finder
The fixed model of body movement.So far, which all mainly uses traditional sorting technique, rare to clustering method to be related to.
Due to the method take be sampling process, in advance it is to be understood that whole data sets, therefore be suitable only for static time sequence
Column data.Again since time series has mobility and unpredictability, data are all to generate with the time, and can not predict
The range of data variation, therefore, the method has significant limitation.
Human motion state Study of recognition mainly analyzes human body movement data, utilizes data mining and engineering
It practises technology mining and goes out the fixed mode in data, mainly there are two the purposes of aspect for the mode: a. moving state identification, i.e. basis
Available data judges that someone is in any motion state；B. identification, it can by similar between calculating mode
Degree may determine which people the motor pattern belongs to, in terms of being mainly used for criminal investigation.
It is divided into following 3 directions currently, can study moving state identification according to the data type difference of research:
(1) based on the moving state identification of sound: this method has an apparent defect: applicable scene it is very limited and
It is larger using the cost of this method.
(2) based on the moving state identification of image/video: this method mainly passes through the data that analysis mining camera acquires
To capture the sports category of human body.Since camera acquisition data are easy to the shadow by factors such as weather, light, distance, orientation
It rings, the scene used is also very limited, and can not come into operation for a long time since video image occupies memory space very much.
(3) based on the moving state identification of wearable device: this method mainly passes through in portable wearable device
Sensor acquires data and then analyzes and researches.Relative to above 2 kinds of methods, this method has following several advantages: a, it is at low cost and
Easy to carry: wearable device is cheap and small and exquisite can wear with oneself；B, strong interference immunity: acquisition data procedures are by extraneous ring
Border influences small；C, persistently obtain the ability of data: carrying can guarantee constantly to obtain data.
The moving situation that moving state identification can not only help people to monitor one day well, but also smart home
One important research field, it may bring a kind of new man-machine interaction mode, such as somatic sensation television game, to make people's lives
It is more intelligent.Thus, the Study of recognition of motion state has a very big significance the quality of life for improving the mankind.
Research foreign countries based on this respect are more early than domestic.Laboratory is done with regard to someone in foreign countries many years ago to pass through in human body body
Upper wearable sensors monitor the motion state of human body to identify, this mainly applies to above the rehabilitation of sportsman and postoperative patient,
It is extremely narrow using field.In smart home research field, it is thus proposed that sensor network system, by installing all kinds of sensings at home
Device, data are uniformly input to control platform, and then analysis identifies the specific movement that people stays at home；Somebody passes through in human lumbar
An acceleration transducer is fixed in portion, can very well identify on foot, running, stand etc. 9 kinds of sports category；Grinding also
In studying carefully, in order to guarantee the collecting efficiency of data and the reliability of transmission, acceleration transducer and storage are set by USB data line
It is standby connect and be worn on, although can finally reach 94% accuracy, wear it is very inconvenient, general user without
Method receives；Since the daily exercise of people is varied and very complicated, in order to extremely accurate identify more motor patterns,
Some researchers place acceleration biography by testing 5 human body parts such as the waist in people, buttocks, wrist, thigh, ankle
Sensor, last experimental result can accurately identify 20 multi-motion classifications, but since the sensor of wearing is too many, user's body
It tests very poor, hardly results in popularization, but this also illustrates multiple sensing datas can improve recognition accuracy.
The characteristics of only taking into account priori knowledge due to current most research methods, have ignored flow data dynamic change,
To cause that the disaggregated model of building accuracy rate under static data is high and the problems such as real experiences effect is poor.In addition, current
Research method does not account for user's difference, and trained disaggregated model is personalized serious.
Summary of the invention
In order to solve the problems, such as that in the prior art, the present invention provides a kind of human motion states based on acceleration transducer
Identification, the present invention are achieved through the following technical solutions:
A kind of human motion state recognition methods based on acceleration transducer, the method are divided into off-line phase and online
Stage；Wherein, the off-line phase constructs human motion state identification model using K-Means clustering method, is based on existing band
The data of label are trained research, propose classification policy；The classification policy is as follows: A. is provided by continuous window to pre-
The data of survey；B. each sample point in window according to being divided into corresponding cluster with a distance from the center from each cluster；C. basis
The category distribution situation of each cluster, calculate all sample points belong to each classification probability and, find out maximum probability value p and
Corresponding classification label；D. if p is more than or equal to first threshold, the sports category label of this window data is exported, it is no
It is then unpredictable, need user's hand labeled；The on-line stage is adopted in real time using the resulting model of off-line phase training
The acceleration information for collecting human body carries out the identification of human motion state.
As a further improvement of the present invention, the window size is not less than the period of studied human motion classification
Maximum value.
As a further improvement of the present invention, the window size is 2 seconds, and the step-length of window is 1 second.
As a further improvement of the present invention, the first threshold is 0.6.
As a further improvement of the present invention, the step B further include when some data point is divided into some cluster, if
More than N times of the cluster radius with a distance from the cluster cluster heart, then illustrates that this data point is noise spot, then carries out discard processing,
In, N is appropriately arranged with according to experiment effect, N >=1.2.
The human motion state identifying system based on acceleration transducer that the present invention also provides a kind of, the system pass through
Android program realizes, the system comprises: data acquisition module, data processing module, movement identification module, model are more
New module, data display module；Wherein, the acceleration information of the data collecting module collected human motion；The model is more
New module constructs human motion state identification model using K-Means clustering method, and the data based on existing tape label are instructed
Practice research, proposes classification policy；The classification policy is as follows: A. provides data to be predicted by continuous window；B. window
Each sample point in mouthful is according to being divided into corresponding cluster with a distance from the center from each cluster；C. according to the category distribution of each cluster
Situation, calculate all sample points belong to each classification probability and, find out maximum probability value p and corresponding classification label；
D. if p is more than or equal to first threshold, the sports category label of this window data is exported, it is otherwise unpredictable, it needs to use
Family hand labeled.
The beneficial effects of the present invention are: being feasible based on clustering method building human motion identification model, and the mould
Type has many advantages, such as that real-time is good, light weight easily adjusts.
Detailed description of the invention
Fig. 1 is the block diagram of human motion state recognition methods of the invention；
Acceleration information schematic diagram when Fig. 2 is on foot；
Fig. 3 is acceleration information schematic diagram when going upstairs；
Fig. 4 is acceleration information schematic diagram when going downstairs.
Specific embodiment
The present invention is further described for explanation and specific embodiment with reference to the accompanying drawing.
The present invention constructs human motion state identification model using K-Means clustering method, and is based on Android phone
Upper design human body movement recognition system.First in the method for off-line phase research and establishment disaggregated model, in order to solve presently, there are
The problem of, propose a kind of method based on clustering method building identification model；Then on-line stage is set based on Android phone
It counts human motion state and identifies real-time system, shown respectively from data acquisition, data processing, movement identification, model modification, data
It is designed Deng 5 functions；Finally by experiments have shown that clustering algorithm validity, the experimental results showed that, be based on clustering method
It is feasible for constructing human motion identification model, and the model has many advantages, such as that real-time is good, light weight easily adjusts.
Shown in attached drawing 1, human motion state Study of recognition is broadly divided into two stages: off-line phase and on-line stage.
Off-line phase is the building learning model stage.Data based on existing tape label are trained research, propose classification
Strategy verifies the quality of model.
Classification policy is as follows: firstly, providing data to be predicted by continuous window.As long as window size is not less than institute
The maximum value in the period of study movement classification, present invention uses 2 seconds window sizes, the step-length of window was 1 second, in this way
The data predicted every time have 50% Chong Die with last prediction data, guarantee the continuity of data.Then, every in window
A sample point is according to being divided into corresponding cluster with a distance from the center from each cluster；Then, according to the category distribution situation of each cluster,
Calculate all sample points belong to each classification probability and, find out maximum probability value p and corresponding classification label.Finally,
If p >=0.6, the sports category label of this window data is exported, it is otherwise unpredictable, need user's hand labeled.Point
Class is in order to identify noise data, when some data point is divided into some cluster, if with a distance from the cluster cluster heart being more than the cluster
1.5 times of radius then illustrate that this data point is noise spot, discard processing.Judge that the condition of noise can be according to experiment effect
Appropriate adjustment.
On-line stage is mainly based upon a human motion identification real-time system of Cell Phone Design.To use in this stage from
The resulting learning model of line stage-training.Movement recognition system will be designed using the model in Android intelligent,
The model is used under practical dynamic environment.
The present invention will realize the identifying system by Android program below.System mainly includes following module: data
Acquisition module, data processing module, movement identification module, model modification module, data are shown.It designs a model in system and updates mould
Block can guarantee that model changes with data variation, change as user changes, this design philosophy, which is more suitable for reality, to be made
With environment, this model has adaptive adjustment capability.When just having begun to use this model, accuracy may be less high, with
The number of the passage of time, adjustment increases, and model can increasingly be suitble to the user using it, and accuracy also can be higher and higher.
Data processing module is by denoising the data of acquisition, smoothly, after normalized, using clustering algorithm
It is modeled, and is deployed on mobile phone and is identified, for other models, there is the features such as speed is fast, model light weight,
Specific recognizer is as follows:
(1) gravitational acceleration component is sought using low-pass first order filter.
gj(i)=α gj(i-1)+(1-α)aj(i)
Wherein α is filtering parameter, and j indicates three coordinate axis components, gj(0)=aj(0), parameter is generally manually set
, and meet 0 < α < 1, filtering parameter can be adjusted according to experiment effect in experimentation, initial value is set as 0.8 and (comes from
Android document suggestion).It can also be calculated by following formula in addition to directly setting:
Wherein t refers to that the time constant of low-pass filter, dT refer to acquisition data frequency.
Acceleration of gravity has been acquired after the component on tri- axis of X, Y, Z using above-mentioned formula, directly uses respective coordinates
Original acceleration value on axis, which subtracts the corresponding component of acceleration of gravity, can be obtained the acceleration value of equipment real motion.
(2) influence of reference axis exchange is reduced by way of increasing dimension.
There are mainly three types of modes to reduce influence of the coordinate interchange to experimental result: (a) by way of seeking resultant acceleration
It is influenced to reduce.Three-dimensional data points p (x, y, z) is passed through into formulaIt is converted into one-dimensional data point p ', this
There are loss of data serious problems for kind mode, it has ignored the correlation of three between centers completely, so that last recognition accuracy is not
It can be too ideal；(b) X-axis and Z axis, Y-axis and Z axis are sought into resultant acceleration respectively, such data are just by three-dimensional drop at two dimension, so both
Part can be eliminated because coordinate interchange bring influences and can have good accuracy rate；(c) increase data dimension, pass through increase
Correspondence acceleration value after three weight components and elimination weight component is expanded data to nine dimension datas, thus by three-dimensional
Acceleration value and weight component are mapped, the influence of coordinate overturning is reduced.The present invention uses the third scheme.
(3) based on the cluster of sampled point
Each sampled point is directly treated as a sample by this mode, feature be exactly original acceleration value on three axis and
The acceleration value after gravity is eliminated, such as (" on foot ", x, y, z, x ', y ', z ') represents a sample.It is adopted by being then based on
The sample of the cluster of sampling point, cluster can be very big, and the process of cluster may take a long time.Clustering method is clustered using K-Means,
Choosing K-Means method cause is mainly that this method principle is simple, and setting parameter is easy, and can suitably be joined with cracking determination
Number.The target of K-Means cluster is so that cluster class Sample Similarity is high, and differences between samples are big between cluster.Following target letter can be passed through
Number is to realize:
N representative sample sum, i represent the id of cluster C, and C represents gathering, error (Ci) calculation formula is as follows:
Wherein ciRepresent cluster CiThe cluster heart,Give directions p and cluster heart ciThe distance between.Calculate distance
Formula it is varied, here using Euclidean distance.
In order to verify the validity of the method for the present invention, experimental situation is as follows: off-line phase mainly carries out on PC machine device real
It tests, constructs learning model.Off-line phase experimental situation is as follows: 64 Windows7 flagship edition systems, Python (2.7.8)；?
The stage has mainly used third party library.
The present invention is studied based on acceleration information, so having used the newest acceleration information of WISDM to construct mould
Type.Data set shown in table 1 is to collect in December, 2012, and it is as follows which collects process: under experimental situation, 36
Volunteer carries Android phone and walks, jogs, upstairs by using with a Android acquisition acceleration information APP acquisition
Ladder, the acceleration information going downstairs, sit quietly, standing under this six kinds of motion states.The process of acquisition is all to carry out certain as required
Movement, so data are all from tape label, i.e. data are all known class.After being uploaded to automatically after data acquisition
Platform.
Classification results after 1 data processing of table
As can be drawn from Table 1, which has " jogging ", " sitting quietly " and " standing " three kinds of categorical datas good
Classifying quality, accurate rate and recall rate all reach 85% or more, and classifier entirety recognition correct rate is 0.8195.Cause other classes
The reason of other effect data difference is: " going upstairs " and " going downstairs " data are unpredictable to be come out, and data " on foot " have all been categorized into,
Cause these three categorical data classification accuracies low.In order to solve this problem, the present invention studies under these three operating states
Accelerating curve (as shown in attached drawing 2 to 4), and individually having studied true classification is " going upstairs " or " going downstairs " and final classification
At the data of " on foot ".
The study found that the data that these mistakes are divided into " on foot " belong to probability " on foot " although maximum, with belong to " on
Away from little, the confidence level being categorized into " on foot " is not high for stair " or the probability difference of " going downstairs ".It can be sent out by observing attached drawing 2 to 4
Existing, " on foot " data and " going upstairs " data and " on foot " data and " going downstairs " data or difference are obvious, and " going upstairs "
Data and " going downstairs " data differences are smaller, are difficult to distinguish them only by this cluster mode, the present invention by this two
Kind data are merged into one kind, are named as " Stairs ".In order to distinguish " stair " data from " on foot " data, if data category
Maximum probability is only second in the probability of " stair " and is approached, then is predicted as " stair ".Experimental result is as shown in table 2.
Table 2 merges the data after stair activity
Observation table 2 as can be seen that above-mentioned solution " stair " data from distinguishing " on foot ".Whole classification
Accuracy reaches 0.8740, has preferable classifying quality to each sports category.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
Specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs, exist
Under the premise of not departing from present inventive concept, a number of simple deductions or replacements can also be made, all shall be regarded as belonging to of the invention
Protection scope.
Claims (10)
1. a kind of human motion state recognition methods based on acceleration transducer, which is characterized in that the method is divided into offline
Stage and on-line stage；Wherein, the off-line phase constructs human motion state identification model using K-Means clustering method,
Data based on existing tape label are trained research, propose classification policy；The classification policy is as follows: A. passes through continuous window
Mouth provides data to be predicted；B. each sample point in window is corresponding according to being divided into a distance from the center from each cluster
Cluster；C. according to the category distribution situation of each cluster, calculate all sample points belong to each classification probability and, find out maximum general
Rate value p and corresponding classification label；D. if p is more than or equal to first threshold, the sports category of this window data is exported
Label, it is otherwise unpredictable, need user's hand labeled；The on-line stage uses the resulting mould of off-line phase training
Type, the acceleration information for acquiring human body in real time carry out the identification of human motion state.
2. according to the method described in claim 1, it is characterized by: the window size is not less than studied human motion class
The maximum value in other period.
3. according to the method described in claim 2, the step-length of window is 1 second it is characterized by: the window size is 2 seconds.
4. according to the method described in claim 1, it is characterized by: the first threshold is 0.6.
5. according to the method described in claim 1, it is characterized by: the B further includes when some data point is divided into some cluster
When, if illustrating that this data point is noise spot more than N times of the cluster radius with a distance from the cluster cluster heart, then being abandoned
Processing, wherein N is appropriately arranged with according to experiment effect, N >=1.2.
6. a kind of human motion state identifying system based on acceleration transducer, the system is by Android program come real
It is existing, which is characterized in that the system comprises: data acquisition module, data processing module, movement identification module, model modification mould
Block, data display module；Wherein, the acceleration information of the data collecting module collected human motion；The model modification mould
Block constructs human motion state identification model using K-Means clustering method, and the data based on existing tape label, which are trained, grinds
Study carefully, proposes classification policy；The classification policy is as follows: A. provides data to be predicted by continuous window；B. in window
Each sample point according to from each cluster intentionally with a distance from be divided into corresponding cluster；C. according to the category distribution feelings of each cluster
Condition, calculate all sample points belong to each classification probability and, find out maximum probability value p and corresponding classification label；D.
If p is more than or equal to first threshold, the sports category label of this window data is exported, it is otherwise unpredictable, need user
Hand labeled.
7. system according to claim 6, it is characterised in that: the window size is not less than studied human motion classification
The maximum value in period.
8. system according to claim 7, it is characterised in that: the window size is 2 seconds, and the step-length of window is 1 second.
9. system according to claim 6, it is characterised in that: the first threshold is 0.6.
10. system according to claim 6, it is characterised in that: the B further include when some data point is divided into some cluster,
If illustrating that this data point is noise spot more than N times of the cluster radius with a distance from the cluster cluster heart, then carrying out at discarding
Reason, wherein N is appropriately arranged with according to experiment effect, N >=1.2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610472604.5A CN106096662B (en) | 2016-06-24 | 2016-06-24 | Human motion state identification based on acceleration transducer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610472604.5A CN106096662B (en) | 2016-06-24 | 2016-06-24 | Human motion state identification based on acceleration transducer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106096662A CN106096662A (en) | 2016-11-09 |
| CN106096662B true CN106096662B (en) | 2019-06-28 |
Family
ID=57252668
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610472604.5A Active CN106096662B (en) | 2016-06-24 | 2016-06-24 | Human motion state identification based on acceleration transducer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106096662B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110008998A (en) * | 2018-11-27 | 2019-07-12 | 美律电子(深圳)有限公司 | Label data generating system and method |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106598234B (en) * | 2016-11-28 | 2019-05-28 | 电子科技大学 | Gesture identification method based on inertia sensing |
| DE102016225648A1 (en) * | 2016-12-20 | 2018-06-21 | Bundesdruckerei Gmbh | Method and system for behavior-based authentication of a user |
| CN107016346A (en) * | 2017-03-09 | 2017-08-04 | 中国科学院计算技术研究所 | gait identification method and system |
| CN107273726B (en) * | 2017-06-02 | 2019-10-29 | 中国人民解放军信息工程大学 | Equipment owner's identity real-time identification method and its device based on acceleration cycle variation law |
| CN107396306A (en) * | 2017-06-30 | 2017-11-24 | 北京奇虎科技有限公司 | User Activity state identification method, device and mobile terminal based on mobile terminal |
| CN108008151A (en) * | 2017-11-09 | 2018-05-08 | 惠州市德赛工业研究院有限公司 | A kind of moving state identification method and system based on 3-axis acceleration sensor |
| CN108810272B (en) * | 2018-06-07 | 2020-10-13 | 中国人民解放军战略支援部队信息工程大学 | Behavior recognition model training method and device based on multiple sensors of mobile terminal |
| CN109063722B (en) * | 2018-06-08 | 2021-06-29 | 中国科学院计算技术研究所 | Behavior recognition method and system based on opportunity perception |
| CN109558841B (en) * | 2018-11-30 | 2023-06-02 | 歌尔科技有限公司 | Motion state identification method, motion state identification device and terminal |
| CN109886068B (en) * | 2018-12-20 | 2022-09-09 | 陆云波 | Motion data-based action behavior identification method |
| CN109978001B (en) * | 2019-02-21 | 2023-07-14 | 上海理工大学 | Empty hand channel motion state recognition device based on multilayer hybrid clustering algorithm |
| CN110349646B (en) * | 2019-07-15 | 2022-06-10 | 上海交通大学 | Priori knowledge clustering-based motion pattern recognition method and system |
| CN110926467B (en) * | 2019-11-11 | 2021-08-06 | 南京航空航天大学 | Self-adaptive pedestrian mobile phone attitude identification method based on mean value clustering algorithm |
| CN116449330B (en) * | 2023-06-20 | 2023-10-13 | 精华隆智慧感知科技（深圳）股份有限公司 | Indoor people number estimation method and device, computer equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103455472A (en) * | 2012-06-01 | 2013-12-18 | 索尼公司 | Information processing apparatus, information processing method and program |
| CN104020845A (en) * | 2014-03-27 | 2014-09-03 | 浙江大学 | Acceleration transducer placement-unrelated movement recognition method based on shapelet characteristic |
| CN105468713A (en) * | 2015-11-19 | 2016-04-06 | 西安交通大学 | Multi-model fused short text classification method |
-
2016
- 2016-06-24 CN CN201610472604.5A patent/CN106096662B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103455472A (en) * | 2012-06-01 | 2013-12-18 | 索尼公司 | Information processing apparatus, information processing method and program |
| CN104020845A (en) * | 2014-03-27 | 2014-09-03 | 浙江大学 | Acceleration transducer placement-unrelated movement recognition method based on shapelet characteristic |
| CN105468713A (en) * | 2015-11-19 | 2016-04-06 | 西安交通大学 | Multi-model fused short text classification method |
Non-Patent Citations (3)
| Title |
|---|
| Mining complex activities in the wild via a;Rai A et al;《Proceedings of the Sixth International》;20121231;43-51页 |
| 基于三轴加速度传感器的山羊行为特征分类与识别;郭东东 等;《家畜生态学报》;20140831;论文全文 |
| 基于加速度传感器的人体运动状态;彭际群;《中国优秀硕士学位论文全文数据库》;20160315;论文全文 |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110008998A (en) * | 2018-11-27 | 2019-07-12 | 美律电子(深圳)有限公司 | Label data generating system and method |
| CN110008998B (en) * | 2018-11-27 | 2021-07-13 | 美律电子(深圳)有限公司 | Label data generating system and method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106096662A (en) | 2016-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106096662B (en) | Human motion state identification based on acceleration transducer | |
| CN108171278B (en) | Motion pattern recognition method and system based on motion training data | |
| CN110287825B (en) | Tumble action detection method based on key skeleton point trajectory analysis | |
| US20160086023A1 (en) | Apparatus and method for controlling presentation of information toward human object | |
| US20110190008A1 (en) | Systems, methods, and apparatuses for providing context-based navigation services | |
| CN105868779B (en) | A kind of Activity recognition method based on feature enhancing and Decision fusion | |
| CN108288015A (en) | Human motion recognition method and system in video based on THE INVARIANCE OF THE SCALE OF TIME | |
| Jensen et al. | Classification of kinematic swimming data with emphasis on resource consumption | |
| CN108958482B (en) | Similarity action recognition device and method based on convolutional neural network | |
| CN110674875A (en) | Pedestrian motion mode identification method based on deep hybrid model | |
| CN106210269A (en) | A kind of human action identification system and method based on smart mobile phone | |
| CN112801000B (en) | Household old man falling detection method and system based on multi-feature fusion | |
| CN110084192A (en) | Quick dynamic hand gesture recognition system and method based on target detection | |
| Ding et al. | Energy efficient human activity recognition using wearable sensors | |
| CN112464738A (en) | Improved naive Bayes algorithm user behavior identification method based on mobile phone sensor | |
| CN114255508A (en) | OpenPose-based student posture detection analysis and efficiency evaluation method | |
| CN115346272A (en) | Real-time tumble detection method based on depth image sequence | |
| CN110163264A (en) | A kind of walking mode recognition methods based on machine learning | |
| CN111262637A (en) | Human body behavior identification method based on Wi-Fi channel state information CSI | |
| CN114550299A (en) | System and method for evaluating daily life activity ability of old people based on video | |
| CN107239147A (en) | A kind of human body context aware method based on wearable device, apparatus and system | |
| Tay et al. | Markerless gait estimation and tracking for postural assessment | |
| CN107688828A (en) | A kind of bus degree of crowding estimating and measuring method based on mobile phone sensor | |
| CN116758479B (en) | Coding deep learning-based intelligent agent activity recognition method and system | |
| CN109359543B (en) | Portrait retrieval method and device based on skeletonization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |