CN106502398A - A kind of semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration - Google Patents
A kind of semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration Download PDFInfo
- Publication number
- CN106502398A CN106502398A CN201610918275.2A CN201610918275A CN106502398A CN 106502398 A CN106502398 A CN 106502398A CN 201610918275 A CN201610918275 A CN 201610918275A CN 106502398 A CN106502398 A CN 106502398A
- Authority
- CN
- China
- Prior art keywords
- body movement
- semantization
- simple body
- sequence
- activity
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/011—Arrangements for interaction with the human body, e.g. for user immersion in virtual reality
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Human Computer Interaction (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
A kind of semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration, is comprised the steps:(1) based on simple body movement descriptive semantics activity, simple body movement characteristic view is built;(2) based on potential theme distribution descriptive semantics activity, potential theme distribution characteristic view is built;(3) multiple views are carried out with Cooperative Study based on semi-supervised technology, and carry out learning outcome fusion obtaining semantization activity recognition model.The present invention improves generalization ability and the adaptability of identification model based on multi views descriptive semantics activity;Identification model is trained using unlabeled data based on Cooperative Study technology, the not enough problem of mark sample is overcome.
Description
Technical field
The present invention relates to machine learning and human-computer interaction technology, and in particular to one kind is based on acceleration transducer and multi views
The semantization activity recognition method of integrated study.
Background technology
The activity of user is to understand one of user context and the most important information of demand, and acceleration transducer have sensitive
Degree is high, the low advantage of power consumption.Therefore, based on the activity recognition of acceleration transducer be general fit calculation and field of human-computer interaction most
One of important research contents.Current has focused largely on simple body work based on the activity recognition research of acceleration transducer
In dynamic (such as walk, run, standing) identification.Compared with simple body movement, semantization activity refers to has a meal, works, doing shopping
Complicated activities of daily living.Semantization activity can provide more rich user context information, while recognizing that difficulty is bigger.
Existing semantization activity recognition method mainly has following a few classes:
(1) method similar with the identification of simple body movement is adopted in model layer, introduce more rich feature in characteristic layer.
For example, B.V.Mirchevska, V.Janko et al. are in " Recognition of high-level
Activities with a smartphone " (international conference UbiComp 2015:In 1453-1461) from GPS, mike,
The feature for extracting complexity in the multiple sensors such as acceleration transducer, biosensor is used for training semantization activity recognition mould
Type.
(2) semantization activity is regarded as a series of combination of simple body movemenies, semantic using hierarchical model identification
Change activity.For example, L.Liu, Y.Peng, M.Liu et al. are in " Sensor-based human activity recognition
system with a multilayered model using time series shapelets”(Knowledge-Based
Systems 90(2015):Conform to the principle of simplicity to identify semanteme in unmarried body active sequenceses based on Time Series Matching algorithm in 138-152)
Change activity.
However, existing semantization recognition methodss there are the following problems:
(1) single view descriptive semantics activity is based on:Single view is difficult in adapt to different semantization activities not on the same day
The normal complexity under living environment, therefore existing easily is affected by noise data, is difficult to cover all semantization activity changes and advises
The problems such as rule.
(2) need have mark sample training model in a large number:Semantization identification model needs have mark sample to be instructed in a large number
Practice.However, due to the semantization activity complexity of itself, user is difficult to provide in daily life enough has mark sample.
Content of the invention
Have to overcome that the identification model generalization ability and adaptability of existing semantization recognition methodss are poor, needing in a large number
The deficiency of mark sample training model, the invention provides a kind of identification model generalization ability and adaptability, available of improving
The limited semantization activity recognition side learnt based on acceleration transducer and Multi-view Integration for having mark sample training model
Method.
The technical solution adopted for the present invention to solve the technical problems is:
A kind of semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration, the semantization are lived
Dynamic recognition methodss are comprised the following steps:
(1) based on simple body movement descriptive semantics activity, simple body movement characteristic view is built, step is as follows:
(1-1) simple body movement identification model training:A simple body movement training set is given, i.e., is labelled with a large number
Simple body movement type, length are the acceleration information sequence of w, first, extract all kinds of from each acceleration information sequence
Temporal signatures and frequency domain character, form motion feature vector;Then, based on the simple body movement type mark of motion feature vector sum
Note, training obtain simple body movement identification model;
(1-2) simple body movement sequence is generated:To the acceleration that each length of semantization active samples, i.e., one is W
Degrees of data sequence, wherein W>W, first, is divided into the data window that multiple sizes are w, forms data window sequence;So
Afterwards, above-mentioned motion feature vector is extracted from each data window, and is inputted the simple body movement identification that training is obtained
Model, obtains simple body movement recognition result;Finally, data window sequence is converted into simple body movement sequence;
(1-3) simple body movement characteristic view builds:First, simple body is extracted from each simple body movement sequence
Body active characteristics, including following three types:
Set feature:Calculate the ratio of every kind of simple body movement type occurrence number and simple body movement sequence length
Value;
Sequence signature:First, multiple for the same type of all continuous appearance in simple body movement sequence simple bodies are lived
Dynamic pressure is condensed to 1, obtains compressing simple body movement sequence;Then, length is excavated from compression simple body movement sequence
For 2 to the length all sequences patterns for being M;Finally, each sequence pattern is calculated in the pressure corresponding to simple body movement sequence
The number of times occurred in the unmarried body active sequenceses of breviaty;
Temporal characteristics:First, all single persistent period of every kind of simple body movement type are calculated;Then, calculate every
Plant average, intermediate value and the standard deviation of simple body movement type single persistent period;
Then, based on above-mentioned simple body movement feature construction characteristic vector, and as descriptive semantics activity
Simple body movement characteristic view;
(2) based on potential theme distribution descriptive semantics activity, potential theme distribution characteristic view is built, step is as follows:
(2-1) acceleration information window sequence:To each semantization active samples, being divided into multiple sizes is
The data window of w, forms data window sequence;Then, above-mentioned motion feature vector is extracted from each data window, and to fortune
Dynamic characteristic vector is normalized;
(2-2) data window cluster sequence is generated:First, based on the Euclidean distance metric data window between motion feature vector
Distance between mouthful, clusters to data window so that the corresponding data window cluster of each data window;Then, by data
Series of windows is converted into data window cluster sequence;
(2-3) potential theme distribution characteristic view builds:First, data window cluster is regarded as " word ", by data window
Cluster sequence regards " document " as, excavates potential theme based on LDA algorithm, and obtains " theme " distribution of " document ";Then, it is based on
" theme " distribution of " document " obtains probability vector of the data window sequence comprising different potential themes, and as description language
The potential theme distribution characteristic view of adoptedization activity;
(3) two kinds of characteristic views are carried out with Cooperative Study based on semi-supervised technology, and carry out fusion to learning outcome obtaining
Semantization activity recognition model.
Further, in step (3), given have mark semantization active samples collection L and without mark semantization activity sample
The step of this collection U, training semantization activity recognition model, is as follows:
(3-1) Training:First, simple body is built for all samples in L based on simple body movement characteristic view
Body active eigenvector, and identification model SM is trained based on semantization Activity Type mark and simple body movement characteristic vector;
Then, potential theme distribution characteristic vector is built for all samples in L based on potential theme distribution characteristic view, and based on semanteme
Change Activity Type mark and potential theme distribution characteristic vector trains identification model TM;
(3-2) semi-supervised training:First, all samples in U are identified based on identification model SM, are every class semantization
Recognition confidence highest n sample is picked out in activity, and recognition result is marked as which, is obtained pseudo- mark sample set and is put into
L;Then, all samples in U are identified based on identification model TM, are that every class semantization activity picks out recognition confidence most
N high sample, recognition result is marked as which, is obtained pseudo- mark sample set and is put into L;
(3-3) algorithm iteration:If sample size is not enough or iterationses exceed specified threshold in U, SM and TM is exported, instead
It, then turn to step (3-1);
(3-4) Model Fusion:To there is each sample in mark semantization active samples collection L, respectively using SM and TM pair
Which is identified, and obtains the probability that SM and TM recognizes which is every class semantization activity, and then obtains 2 probability vectors;Then, will
This 2 probability vectors and semantization Activity Type mark build new sample set NL as new sample;Finally, based on NL, adopt
Final semantization activity recognition model FM is obtained with Logistic Regression Algorithm for Training.
Further, in step (1-1), simple body movement identification model is obtained using C4.5 Algorithm for Training.
Further, in step (1-3), during extracting sequence signature, simple from compression based on Apriori algorithm
It is the 2 all sequences patterns for arriving that length is M to excavate length in body movement sequence.
In step (2-2), data window is clustered based on K-Medoids algorithms.
Beneficial effects of the present invention are mainly manifested in:1st, based on multi views descriptive semantics activity, identification model is improve
Generalization ability and adaptability.2nd, identification model is trained using unlabeled data based on Cooperative Study technology, has overcome mark
The problem that note sample is not enough.
Description of the drawings
Fig. 1 is the flow chart of the semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration;
Fig. 2 is the flow chart that simple body movement characteristic view builds;
Fig. 3 is the flow chart that potential theme distribution characteristic view builds;
Flow charts of the Fig. 4 for semantization activity recognition model training.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.
With reference to Fig. 1~Fig. 4, a kind of semantization activity recognition side learnt based on acceleration transducer and Multi-view Integration
Method, the semantization recognition methodss are comprised the following steps:
(1) based on simple body movement descriptive semantics activity, simple body movement characteristic view is built.
(2) based on potential theme distribution descriptive semantics activity, potential theme distribution characteristic view is built.
(3) two kinds of characteristic views are carried out with Cooperative Study based on semi-supervised technology, and carry out fusion to learning outcome obtaining
Semantization activity recognition model.
With reference to Fig. 2, in step (1), the detailed step for building simple body movement characteristic view is as follows:
(1-1) simple body movement identification model training:Give a simple body movement training set (to be labelled with a large number
The acceleration information sequence of simple body movement type, length for w), first, extract from each acceleration information sequence all kinds of
Temporal signatures (include:Average, standard deviation, quartile deviation, energy) and frequency domain character (include:Frequency amplitude, frequency domain entropy), formed
Motion feature vector.Then, based on the simple body movement type mark of motion feature vector sum, obtained using C4.5 Algorithm for Training
Simple body movement identification model.
(1-2) simple body movement sequence is generated:To (acceleration of i.e. one length for W of each semantization active samples
Degrees of data sequence, wherein W>W), first, the data window that multiple sizes are w is divided into, data window sequence is formed.So
Afterwards, above-mentioned motion feature vector is extracted from each data window, and is inputted the simple body movement identification that training is obtained
Model, obtains simple body movement recognition result.Finally, data window sequence is converted into simple body movement sequence.
(1-3) simple body movement characteristic view builds:First, simple body is extracted from each simple body movement sequence
Body active characteristics, including following three types:
Set feature:Calculate the ratio of every kind of simple body movement type occurrence number and simple body movement sequence length
Value.
Sequence signature:First, multiple for the same type of all continuous appearance in simple body movement sequence simple bodies are lived
Dynamic pressure is condensed to 1, obtains compressing simple body movement sequence;Then, based on Apriori algorithm from the simple body movement sequence of compression
It is the 2 all sequences patterns for arriving that length is M to excavate length in row;Finally, each sequence pattern is calculated in simple body movement
The number of times for compressing appearance in simple body movement sequence corresponding to sequence.
Temporal characteristics:First, all single persistent period of every kind of simple body movement type are calculated;Then, calculate every
The average of kind simple body movement type single persistent period, intermediate value, standard deviation.
Then, based on above-mentioned simple body movement feature construction characteristic vector, and as descriptive semantics activity
Simple body movement characteristic view.
With reference to Fig. 3, in step (2), the detailed step for building potential theme distribution characteristic view is as follows:
(2-1) acceleration information window sequence:To (acceleration of i.e. one length for W of each semantization active samples
Degrees of data sequence), the data window that multiple sizes are w is divided into, data window sequence is formed.Then, from each data
Above-mentioned motion feature vector is extracted in window, and motion feature vector is normalized.
(2-2) data window cluster sequence is generated:First, based on the Euclidean distance metric data window between motion feature vector
Distance between mouthful, is clustered to data window based on K-Medoids algorithms so that the corresponding data window of each data window
Cluster.Then, data window sequence is converted into data window cluster sequence.
(2-3) potential theme distribution characteristic view builds:First, data window cluster is regarded as " word ", by data window
Cluster sequence regards " document " as, excavates potential theme based on LDA algorithm, and obtains " theme " distribution of " document ".Then, it is based on
" theme " distribution of " document " obtains probability vector of the data window sequence comprising different potential themes, and as description language
The potential theme distribution characteristic view of adoptedization activity.
With reference to Fig. 4, in step (3), given have mark semantization active samples collection L and without mark semantization activity sample
This collection U, trains the detailed step of semantization activity recognition model as follows:
(3-1) Training:First, simple body is built for all samples in L based on simple body movement characteristic view
Body active eigenvector, and identification model SM is trained based on semantization Activity Type mark and simple body movement characteristic vector.
Then, potential theme distribution characteristic vector is built for all samples in L based on potential theme distribution characteristic view, and based on semanteme
Change Activity Type mark and potential theme distribution characteristic vector trains identification model TM.
(3-2) semi-supervised training:First, all samples in U are identified based on identification model SM, are every class semantization
Recognition confidence highest n sample is picked out in activity, and recognition result is marked as which, obtains pseudo- mark sample set USM,n×S
And it is put into L (wherein, quantity of the S for semantization Activity Type).Then, all samples in U are known based on identification model TM
Not, it is that recognition confidence highest n sample is picked out in every class semantization activity, recognition result is marked as which, puppet is obtained
Mark sample set UTM,n×SAnd it is put into L.
(3-3) algorithm iteration:If sample size is not enough or iterationses exceed specified threshold in U, SM and TM is exported.Instead
It, then turn to step (3-1).
(3-4) Model Fusion:To there is each sample in mark semantization active samples collection L, respectively using SM and TM pair
Which is identified, and obtains the probability that SM and TM recognizes which is every class semantization activity, and then obtains 2 probability vectors (wherein,
PSM,ikProbability of the sample i for semantization Activity Type k, P are recognized for SMTM,ikIt is semantization Activity Type k for TM identification sample i
Probability, 1≤k≤S).Then, this 2 probability vectors and semantization Activity Type mark are built new as new sample
Sample set NL.Finally, based on NL, obtain final semantization activity recognition using Logistic Regression Algorithm for Training
Model FM.
Claims (5)
1. a kind of semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration, it is characterised in that:Institute
State semantization activity recognition method to comprise the following steps:
(1) based on simple body movement descriptive semantics activity, simple body movement characteristic view is built, step is as follows:
(1-1) simple body movement identification model training:A simple body movement training set is given, i.e., is labelled with a large number simple
Body movement type, length are the acceleration information sequence of w, first, extract all kinds of time domains from each acceleration information sequence
Feature and frequency domain character, form motion feature vector;Then, based on the simple body movement type mark of motion feature vector sum,
Training obtains simple body movement identification model;
(1-2) simple body movement sequence is generated:To the acceleration number of degrees that each length of semantization active samples, i.e., one is W
According to sequence, wherein W>W, first, is divided into the data window that multiple sizes are w, forms data window sequence;Then, from
Above-mentioned motion feature vector is extracted in each data window, and is inputted the simple body movement identification model that training is obtained,
Obtain simple body movement recognition result;Finally, data window sequence is converted into simple body movement sequence;
(1-3) simple body movement characteristic view builds:First, extract simple body from each simple body movement sequence to live
Dynamic feature, including following three types:
Set feature:Calculate the ratio of every kind of simple body movement type occurrence number and simple body movement sequence length;
Sequence signature:First, by multiple for the same type of all continuous appearance in simple body movement sequence simple body movement pressures
1 is condensed to, obtains compressing simple body movement sequence;Then, from compress excavate in simple body movement sequence length for 2 to
All sequences pattern of the length for M;Finally, it is simple in the compression corresponding to simple body movement sequence that each sequence pattern is calculated
The number of times occurred in body movement sequence;
Temporal characteristics:First, all single persistent period of every kind of simple body movement type are calculated;Then, every kind of letter is calculated
The average of unmarried body Activity Type single persistent period, intermediate value and standard deviation;
Then, based on above-mentioned simple body movement feature construction characteristic vector, and as the simple of descriptive semantics activity
Body movement characteristic view;
(2) based on potential theme distribution descriptive semantics activity, potential theme distribution characteristic view is built, step is as follows:
(2-1) acceleration information window sequence:To each semantization active samples, multiple sizes are divided into for w's
Data window, forms data window sequence;Then, above-mentioned motion feature vector is extracted from each data window, and to motion
Characteristic vector is normalized;
(2-2) data window cluster sequence is generated:First, based between the Euclidean distance metric data window between motion feature vector
Distance, clusters to data window so that the corresponding data window cluster of each data window;Then, by data window
Sequence Transformed for data window cluster sequence;
(2-3) potential theme distribution characteristic view builds:First, regard data window cluster as " word ", data window is clustered
Sequence regards " document " as, excavates potential theme based on LDA algorithm, and obtains " theme " distribution of " document ";Then, based on " text
" theme " distribution of shelves " obtains probability vector of the data window sequence comprising different potential themes, and as descriptive semantics
The potential theme distribution characteristic view of change activity;
(3) two kinds of characteristic views are carried out with Cooperative Study based on semi-supervised technology, and carry out fusion to learning outcome obtaining semanteme
Change activity recognition model.
2. the semantization activity recognition method for being learnt based on acceleration transducer and Multi-view Integration as claimed in claim 1,
It is characterized in that:In step (3), given have mark semantization active samples collection L and without mark semantization active samples collection
The step of U, training semantization activity recognition model, is as follows:
(3-1) Training:First, build simple body based on simple body movement characteristic view for all samples in L to live
Dynamic characteristic vector, and identification model SM is trained based on semantization Activity Type mark and simple body movement characteristic vector;Then,
It is that all samples build potential theme distribution characteristic vector in L based on potential theme distribution characteristic view, and is lived based on semantization
Dynamic type mark and potential theme distribution characteristic vector train identification model TM;
(3-2) semi-supervised training:First, all samples in U are identified based on identification model SM, are every class semantization activity
Recognition confidence highest n sample is picked out, recognition result is marked as which, obtained pseudo- mark sample set and be put into L;So
Afterwards, all samples in U are identified based on identification model TM, are that recognition confidence highest is picked out in every class semantization activity
N sample, recognition result is marked as which, is obtained pseudo- mark sample set and is put into L;
(3-3) algorithm iteration:If sample size is not enough or iterationses exceed specified threshold in U, SM and TM is exported, conversely,
Step (3-1) is then turned to;
(3-4) Model Fusion:To there is each sample in mark semantization active samples collection L, which is entered using SM and TM respectively
Row identification, obtains the probability that SM and TM recognizes which is every class semantization activity, and then obtains 2 probability vectors;Then, by this 2
Individual probability vector and semantization Activity Type mark build new sample set NL as new sample;Finally, based on NL, employing
Logistic Regression Algorithm for Training obtains final semantization activity recognition model FM.
3. the semantization activity recognition side for being learnt based on acceleration transducer and Multi-view Integration as claimed in claim 1 or 2
Method, it is characterised in that:In step (1-1), simple body movement identification model is obtained using C4.5 Algorithm for Training.
4. the semantization activity recognition side for being learnt based on acceleration transducer and Multi-view Integration as claimed in claim 1 or 2
Method, it is characterised in that:In step (1-3), during extracting sequence signature, based on Apriori algorithm from the simple body of compression
It is the 2 all sequences patterns for arriving that length is M to excavate length in body active sequenceses.
5. the semantization activity recognition side for being learnt based on acceleration transducer and Multi-view Integration as claimed in claim 1 or 2
Method, it is characterised in that:In step (2-2), data window is clustered based on K-Medoids algorithms.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610918275.2A CN106502398B (en) | 2016-10-21 | 2016-10-21 | A kind of semantization activity recognition method based on Multi-view Integration study |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610918275.2A CN106502398B (en) | 2016-10-21 | 2016-10-21 | A kind of semantization activity recognition method based on Multi-view Integration study |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106502398A true CN106502398A (en) | 2017-03-15 |
CN106502398B CN106502398B (en) | 2019-01-29 |
Family
ID=58319386
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610918275.2A Active CN106502398B (en) | 2016-10-21 | 2016-10-21 | A kind of semantization activity recognition method based on Multi-view Integration study |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106502398B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109492678A (en) * | 2018-10-24 | 2019-03-19 | 浙江工业大学 | A kind of App classification method of integrated shallow-layer and deep learning |
CN110008998A (en) * | 2018-11-27 | 2019-07-12 | 美律电子(深圳)有限公司 | Label data generating system and method |
CN112988124A (en) * | 2021-05-10 | 2021-06-18 | 湖南高至科技有限公司 | Multi-view platform-independent model system |
US20220382845A1 (en) * | 2019-04-30 | 2022-12-01 | TruU, Inc. | Supervised and Unsupervised Techniques for Motion Classification |
CN117251770A (en) * | 2023-11-17 | 2023-12-19 | 北京新兴科遥信息技术有限公司 | Method for identifying low-utility land |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101389004A (en) * | 2007-09-13 | 2009-03-18 | 中国科学院自动化研究所 | Moving target classification method based on on-line study |
CN104331431A (en) * | 2014-10-22 | 2015-02-04 | 浙江中烟工业有限责任公司 | Mobile application ordering method of situational awareness |
CN105956614A (en) * | 2016-04-25 | 2016-09-21 | 中国科学院上海高等研究院 | Time series semantization predicting method and time series semantization predicting system |
-
2016
- 2016-10-21 CN CN201610918275.2A patent/CN106502398B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101389004A (en) * | 2007-09-13 | 2009-03-18 | 中国科学院自动化研究所 | Moving target classification method based on on-line study |
CN104331431A (en) * | 2014-10-22 | 2015-02-04 | 浙江中烟工业有限责任公司 | Mobile application ordering method of situational awareness |
CN105956614A (en) * | 2016-04-25 | 2016-09-21 | 中国科学院上海高等研究院 | Time series semantization predicting method and time series semantization predicting system |
Non-Patent Citations (2)
Title |
---|
徐常有: "《人体运动序列数据的语义化分析方法研究》", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
陈秋迪: "《语义化多角色可变形运动模型研究》", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109492678A (en) * | 2018-10-24 | 2019-03-19 | 浙江工业大学 | A kind of App classification method of integrated shallow-layer and deep learning |
CN109492678B (en) * | 2018-10-24 | 2021-11-23 | 浙江工业大学 | App classification method integrating shallow layer learning and deep learning |
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 |
US20220382845A1 (en) * | 2019-04-30 | 2022-12-01 | TruU, Inc. | Supervised and Unsupervised Techniques for Motion Classification |
US12008092B2 (en) * | 2019-04-30 | 2024-06-11 | TruU, Inc. | Supervised and unsupervised techniques for motion classification |
CN112988124A (en) * | 2021-05-10 | 2021-06-18 | 湖南高至科技有限公司 | Multi-view platform-independent model system |
CN112988124B (en) * | 2021-05-10 | 2021-07-30 | 湖南高至科技有限公司 | Multi-view platform-independent model system |
CN117251770A (en) * | 2023-11-17 | 2023-12-19 | 北京新兴科遥信息技术有限公司 | Method for identifying low-utility land |
CN117251770B (en) * | 2023-11-17 | 2024-02-13 | 北京新兴科遥信息技术有限公司 | Method for identifying low-utility land |
Also Published As
Publication number | Publication date |
---|---|
CN106502398B (en) | 2019-01-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Su et al. | Activity recognition with smartphone sensors | |
CN106502398A (en) | A kind of semantization activity recognition method learnt based on acceleration transducer and Multi-view Integration | |
Zhu et al. | Efficient human activity recognition solving the confusing activities via deep ensemble learning | |
Shuvo et al. | A hybrid approach for human activity recognition with support vector machine and 1D convolutional neural network | |
US10061389B2 (en) | Gesture recognition system and gesture recognition method | |
Asim et al. | Context-aware human activity recognition (CAHAR) in-the-Wild using smartphone accelerometer | |
Zhang et al. | Human activity recognition with HMM-DNN model | |
CN109688990A (en) | For providing a user the method and system of attached sensory information | |
Tran et al. | Data augmentation for inertial sensor-based gait deep neural network | |
Wang et al. | Sequential weakly labeled multiactivity localization and recognition on wearable sensors using recurrent attention networks | |
Nawaratne et al. | Hierarchical two-stream growing self-organizing maps with transience for human activity recognition | |
Chetty et al. | Body sensor networks for human activity recognition | |
Wen et al. | Adaptive activity learning with dynamically available context | |
CN106529504A (en) | Dual-mode video emotion recognition method with composite spatial-temporal characteristic | |
Garcia-Ceja et al. | An improved three-stage classifier for activity recognition | |
Huan et al. | Human complex activity recognition with sensor data using multiple features | |
Zhang et al. | CSI-based location-independent human activity recognition using feature fusion | |
Khosla et al. | Assistive robot enabled service architecture to support home-based dementia care | |
Wang | Three-dimensional convolutional restricted Boltzmann machine for human behavior recognition from RGB-D video | |
Dungkaew et al. | Impersonal smartphone-based activity recognition using the accelerometer sensory data | |
Wang et al. | Human activity recognition using smart phone embedded sensors: A linear dynamical systems method | |
Huan et al. | A hybrid cnn and blstm network for human complex activity recognition with multi-feature fusion | |
Zhang et al. | A dropconnect deep computation model for highly heterogeneous data feature learning in mobile sensing networks | |
Yao et al. | Revisiting Large-Kernel CNN Design via Structural Re-Parameterization for Sensor-Based Human Activity Recognition | |
Ismael et al. | A study on human activity recognition using smartphone |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20191126 Address after: Room 506-2, Block E, building 1, 1378 Wenyi West Road, Cangqian street, Yuhang District, Hangzhou City, Zhejiang Province Patentee after: Hangzhou smart strategy Technology Co., Ltd Address before: 310014, Zhejiang City, No. 18 Chao Wang Road, Zhejiang University of Technology Patentee before: Zhejiang University of Technology |
|
TR01 | Transfer of patent right |