CN110532898A - A kind of physical activity recognition methods based on smart phone Multi-sensor Fusion - Google Patents
A kind of physical activity recognition methods based on smart phone Multi-sensor Fusion Download PDFInfo
- Publication number
- CN110532898A CN110532898A CN201910732567.0A CN201910732567A CN110532898A CN 110532898 A CN110532898 A CN 110532898A CN 201910732567 A CN201910732567 A CN 201910732567A CN 110532898 A CN110532898 A CN 110532898A
- Authority
- CN
- China
- Prior art keywords
- data
- physical activity
- axis
- sensor
- smart phone
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000037081 physical activity Effects 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 16
- 230000004927 fusion Effects 0.000 title claims abstract description 14
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims description 20
- 230000001133 acceleration Effects 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 5
- 230000008447 perception Effects 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 2
- 230000009467 reduction Effects 0.000 claims description 2
- 230000003595 spectral effect Effects 0.000 claims description 2
- 230000004899 motility Effects 0.000 claims 1
- 238000013461 design Methods 0.000 abstract description 3
- 230000003993 interaction Effects 0.000 abstract description 2
- 238000000605 extraction Methods 0.000 abstract 1
- 238000005516 engineering process Methods 0.000 description 11
- 230000000694 effects Effects 0.000 description 6
- 230000009471 action Effects 0.000 description 2
- 238000007635 classification algorithm Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000016776 visual perception Effects 0.000 description 2
- GCKMFJBGXUYNAG-HLXURNFRSA-N Methyltestosterone Chemical compound C1CC2=CC(=O)CC[C@]2(C)[C@@H]2[C@@H]1[C@@H]1CC[C@](C)(O)[C@@]1(C)CC2 GCKMFJBGXUYNAG-HLXURNFRSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Telephone Function (AREA)
Abstract
A kind of physical activity recognition methods based on smart phone Multi-sensor Fusion belongs to field of intelligent man-machine interaction, the multi-modal sensors such as inertial sensor, magnetometer and the barometer that the present invention is integrated using smart phone propose the physical activity recognition methods of the fusion multi-modal sensor of smart phone.Main contents of the invention are as follows: (1) establish the physical activity identification framework of the lay figure based on cartesian coordinate system and Multi-sensor Fusion；(2) human body carries out daily routines, and experimenter acquires multi-sensor data by carrying smart phone；(3) hot-tempered, feature extraction and selection are carried out to original sensor data by data prediction；(4) multi-modal data of the design based on Stacking merges integrated RSK-Stacking algorithm and is trained to obtain physical activity identification model to optimal physical activity data set, and then identifies to physical activity.
Description
Technical field
The invention belongs to field of intelligent man-machine interaction, it is a kind of based on smart phone Multi-sensor Fusion, can be applied to people
The algorithm of body activity recognition.
Background technique
Have physical activity identification technology according to the difference of used sensor technology, can generally be divided into following three
Class: wearable cognition technology (Wearable Sensors), context aware technology (Ambient Sensors) and visual perception skill
Art (Vision-based Sensors).Technology based on wearable perception may operate at indoor and outdoors environment, and provide good
Good action recognition.Further, since its is small in size, it is conveniently worn on the different parts (such as arm, waist and leg etc.) of body.
However, physically disposing the problems such as multiple sensors bring discomfort to wearer there are at high cost, meeting.Context aware technology and
Visual perception technology has the characteristics that accuracy rate is high, motion capture is intuitive, but sensor deployment and detection algorithm are complicated, and monitor
It is limited in scope, or even is easy exposure privacy of user.As smart phone computing capability is promoted and using universal, researcher's benefit
The research of physical activity identification technology is carried out with the sensor that smart phone integrates.Smart phone be not only integrated with accelerometer,
The multi-modal sensor such as magnetometer, gyroscope, thermometer, GPS and barometer, and it is strong there are also higher memory capacity and relatively
Computing capability, it is the excellent perception and computing platform for carrying out activity recognition.Single type sensor limits its identification activity
Type and accuracy.
The present invention is based on the physical activity data of intelligent mobile phone sensor acquisition, are based on intelligence with reference to Fig. 1 architecture design
Mobile phone multisensor and the physical activity of sorting algorithm (Random Forest, SVM, KNN) fusion identify Integrated Algorithm RSK-
Stacking。
Summary of the invention
The present invention is based on the physical activity data and traditional classification algorithm of smart phone acquisition, devise physical activity identification
Integrated Algorithm RSK-Stacking.The present invention relates to following 2 points:
(1) inertial sensor, magnetometer and the air pressure of intelligent mobile phone sensor acquisition are counted fusion one by the present invention
It rises, introduces sliding window technique and physical activity feature is extracted, and feature is selected by PCA principal component analysis.
(2) physical activity with reference to Fig. 1 architecture design based on smart phone Multi-sensor Fusion identifies Integrated Algorithm
RSK-Stacking sorts out physical activity feature, realizes physical activity recognizer.
Core algorithm of the present invention
(1) lay figure is established
During the motion, acceleration and angular speed can real-time change for human body.Forefathers are research shows that smart phone is placed on
It is to acquire its sensing data and identify everyday actions optimal site in the pocket of user's trousers.The present invention is comfortable from Portable device
Property and system reliability set out, by smart phone head downward, be vertically disposed in the pocket of user's trousers right front, establish
As shown in Figure 2 based on the lay figure of cartesian coordinate system.Wherein, ax、ayAnd azSmart phone is respectively represented along x-axis, y
The acceleration information that axis and z-axis obtain；ωx、ωyAnd ωzRespectively represent angle speed of the trunk around x-axis, y-axis and z-axis movement when
Degree evidence；mx、myAnd mzRespectively represent the geomagnetic data of x-axis, y-axis and z-axis.In addition, the gas under barometer perception current environment
Pressure.It is coordinate origin that the present invention, which selectes right pocket in front of the trousers of people, and the front-right of human body is the positive direction of x-axis, surface y
The positive direction of axis, front are the positive direction of z-axis.
On the basis of 3 dimension data that 3 axis accelerometers, gyroscope and magnetometer perceive respectively, the present invention is according to formula (1)
Calculate separately the conjunction value of acceleration, angular speed and magnetic force.Conjunction value only reflects the range value variation of acceleration, angular speed and magnetic force,
Independent of direction.
(2) physical activity data prediction
The purpose of data prediction is that the raw data set of sensor acquisition is converted to the feature vector for being conducive to classification.
Due to median filter can effectively smoothed data collection, remove the influence of peak signal in data, therefore the present invention uses intermediate value
Filtering algorithm filters multi-modal data and goes hot-tempered, and then data are normalized again, to improve training speed and identification
Accuracy.Accelerometer, gyroscope, magnetometer and the barometrical type and precision phase not to the utmost integrated due to different intelligent mobile phone
Together, the present invention is normalized to initial data between [- 1,1] using formula (2)；For magnetometer and barometer, using formula
(3) initial data that they are acquired is normalized between [- 1,1].In formula (2) (3), the initial data of x representative sensor,
xmeanThe average value of initial data is represented, range represents the range ability of acceleration or gyroscope, xvarRepresent initial data
Variance, xscaleRepresent the data after normalization.
Since physical activity data are consecutive data set, traditional classification algorithm is not directly applicable continuous data, therefore
The present invention is split the physical activity data after normalization using sliding window technique.Since physical activity frequency is usually small
In 20Hz, therefore the sample frequency of mobile phone sensor is set to 50Hz by the present invention.Lead in addition, human body completes a movable time
Often less than 2 seconds, sliding window size was set as 2s by the present invention, i.e., is divided with the sliding window that a capacity is 100 sample points
Sensor signal, adjacent sliding window carry out juxtaposition by 50%.The window that sample size is 100 can include one section of periodicity
Movable waveform.Mean value, variance, standard deviation, kurtosis, root mean square, maximum value, the minimum value of this patent selection physical activity data
Dimensionality reduction is carried out to feature vector with energy spectral density as the feature of physical activity and using PCA, to reduce calculation amount and raising
Nicety of grading.
(3) the physical activity recognizer RSK-Stacking based on Multi-sensor Fusion is devised
RSK-Stacking algorithm is different from common Ensemble Learning Algorithms Boosting and Bagging, carries out algorithm
Simple logical process is not done to the prediction result of base learner when fusion, then add one layer of learner, by
It practises device and is trained generation a new model, and then realize the fusion to base learner.Wherein, base learner is by different study
Algorithm generates.Given T primary learning algorithm S1,S2......STWith a first layer learning algorithm S.Stacking algorithm is main
Steps are as follows:
Training stage: primary learning layer is using K folding crosscheck method training base's learner.By initial training data set
D={ (x1,y1),(x2,y2).....(xh,yh) it is randomly divided into the similar Sub Data Set D of k size1,D2.....Dk,DjTable
Show the Sub Data Set (j=1,2.....k) of jth folding.It, will when each iteration followed by k iterationAs S
The input data set (t=1,2.....T) of algorithm, t-th of learning algorithm StInUpper training obtains primary learner Lt (j)。Lt (j)To DjIn each sample xiIt is predicted, prediction result is expressed as Lt (j)(xi).When T model carried out aforesaid operations it
Afterwards, T model is obtained to DjIn all samples prediction result.xiThe secondary training examples x'=(L of generation1 (j)(xi),L2 (j)
(xi),....,LT (j)(xi)), it is labeled as yi, after k iteration is completed, T model is obtained to the prediction result of all samplesFinally using D' as the input data of secondary learning algorithm S, secondary learner L'=S (D') is obtained.
Test phase: each learning algorithm StK primary learner L is generated on K folding training sett (j), j=1,2 ...,
K, this k learner is in test set DtestEach of k prediction result (L is generated on sample xt (1)(x),Lt (2)
(x),....,Lt (k)(x)), this k result is averaged according to formula (4):
Test sample x inputs T learning algorithm and obtains predicted valueIt is inputted secondary study
Device L', finally obtains prediction result
Pseudo-code of the algorithm is as shown in Figure 3.
Invention effect
Multi-sensor data by merging the acquisition of smart phone identifies physical activity, can overcome unitary class
Type sensor and the accuracy of identification of single sorting algorithm are insufficient, accurate to realize physical activity identification.RSK-Stacking model system
The accuracy rate of system has reached 99.0%, and susceptibility and specificity respectively reach 99.0% and 99.8%.Wherein, system to walking,
These three closely similar activities have carried out good Division identification with little Lou upstairs.
The core technology of this patent includes:
(1) innovative physical activity Data Preprocessing Technology: by the inertial sensor of intelligent mobile phone sensor acquisition, magnetic
Power meter and air pressure, which count, to be fused together, and is introduced sliding window technique and is extracted to physical activity feature, and passes through PCA master
Constituent analysis selects feature.
(2) data normalization technology is introduced simultaneously, inertial sensor, magnetometer and air pressure is counted and is converted into value and is
The real number value of [- 1,1] provides data basis to construct efficient physical activity identifying system；
(3) Integrated Algorithm (RSK-Stacking) towards physical activity identification is constructed, and is devised as shown in Figure 1
Physical activity identification framework realizes the quick and precisely monitoring of physical activity identification.
Detailed description of the invention
Fig. 1 is the lay figure based on cartesian coordinate system
Fig. 2 is physical activity identification framework.
Fig. 3 is RSK-Stacking pseudo-code of the algorithm.
Specific embodiment
(1) lay figure is referred to, the present invention is using the Samsung for running 5.0 operating system of Android
5 mobile phone of GALAXYS is used as the platform of data collection.Pass through self-developing and is mounted on the application program on mobile phone to acquire 4 kinds
Sensing data, sample frequency 50Hz.
(2) 10 students (4 female, 6 male) are as subject, and the age is between 24-30 years old.It is living that every subject completes 6 human bodies
It is dynamic (stands, walks, jogs, upstairs, downstairs and ride a bicycle), in order to eliminate at the beginning and end of acquiring data because of placement
The rejection of data of acquisition in first 2 seconds and last 2 seconds, so every kind of activity are had chosen 50050 numbers by error caused by mobile phone, experiment
According to.
(3) lay figure is referred to, the present invention is using physical activity data preprocessing method to sensor raw data
After being pre-processed, 3 groups of characteristic data sets are obtained.1st group contains acceleration and gyro data, names Data1；2nd group
Containing acceleration, gyroscope and magnetometer data, Data2 is named；3rd group contains acceleration, gyroscope, magnetometer and barometer
Data name Data3.1000 samples are chosen in every kind of activity.Data over-fitting when training in order to prevent, the present invention randomly select
90% data amount to 5400 samples as training dataset；The data of selection 10% amount to 600 as test data set
A sample；Ensure training dataset and test data set without juxtaposition, to improve the precision of identification.
(4) present invention instructs RSK-Stacking system using three group data sets (Data1, Data2, Data3)
Practice.Every group data set contains 6000 samples, and data set is divided into training set and verifying collection according to 9:1 ratio.Wherein just
Grade learning layer is using K folding crosscheck method training base's learner, and the output result of base's learner is as first layer learner
Input training data.Final Data3 verifying collection maintains 99.0% or so in the recognition accuracy of RSK-Stacking system,
Subsequent trained accuracy rate maintains stabilization no longer to rise.
(5) input of physical activity recognizer is the multi-sensor data that smart phone obtains, and is exported as physical activity
Determine result.
Claims (1)
1. the physical activity recognition methods based on smart phone Multi-sensor Fusion, which comprises the following steps:
(1) lay figure is established
By smart phone head downward, be vertically disposed in the pocket of user's trousers right front, the people based on cartesian coordinate system
Body motility model；Wherein, ax、ayAnd azRespectively represent the acceleration information that smart phone is obtained along x-axis, y-axis and z-axis；ωx、ωy
And ωzRespectively represent angular velocity data of the trunk around x-axis, y-axis and z-axis movement when；mx、myAnd mzRespectively represent x-axis, y-axis
With the geomagnetic data of z-axis；In addition, the air pressure under barometer perception current environment；Right pocket is coordinate in front of the trousers of selected people
Origin, the front-right of human body are the positive direction of x-axis, and surface is the positive direction of y-axis, and front is the positive direction of z-axis；
On the basis of 3 dimension data that 3 axis accelerometers, gyroscope and magnetometer perceive respectively, calculates separately and add according to formula (1)
The conjunction value of speed, angular speed and magnetic force；Conjunction value only reflects the range value variation of acceleration, angular speed and magnetic force, independent of direction；
(2) physical activity data prediction
Multi-modal data is filtered using median filtering algorithm and goes hot-tempered, then data are normalized again, to improve training
The accuracy of speed and identification；Initial data is normalized between [- 1,1] using formula (2)；For magnetometer and air pressure
Meter, is normalized to the initial data that they are acquired between [- 1,1] using formula (3)；In formula (2) (3), x representative sensor
Initial data, xmeanThe average value of initial data is represented, range represents the range ability of acceleration or gyroscope, xvarIt represents
The variance of initial data, xscaleRepresent the data after normalization；
The physical activity data after normalization are split using sliding window technique；The sample frequency of mobile phone sensor is determined
For 50Hz, sliding window size is set as 2s, i.e., divides sensor signal with the sliding window that a capacity is 100 sample points,
Adjacent sliding window carries out juxtaposition by 50%；Select the mean values of physical activity data, variance, standard deviation, kurtosis, root mean square,
Maximum value, minimum value and energy spectral density carry out dimensionality reduction to feature vector as the feature and use PCA of physical activity；
(3) the physical activity recognizer RSK-Stacking based on Multi-sensor Fusion is devised
Given T primary learning algorithm S1,S2......STIt is rapid as follows with a first layer learning algorithm S:
Training stage: primary learning layer is using K folding crosscheck method training base's learner；By initial training data set D=
{(x1,y1),(x2,y2).....(xh,yh) it is randomly divided into the similar Sub Data Set D of k size1,D2.....Dk,DjIt indicates
The Sub Data Set of jth folding, j=1,2.....k；It, will when each iteration followed by k iterationAs StIt calculates
The input data set of method, wherein t=1,2.....T, t-th learning algorithm StInUpper training obtains primary learner Lt (j)；Lt (j)To DjIn each sample xiIt is predicted, prediction result is expressed as Lt (j)(xi)；
After T model has carried out aforesaid operations, T model is obtained to DjIn all samples prediction result；xiTime generated
Grade training examples x'=(L1 (j)(xi),L2 (j)(xi),....,LT (j)(xi)), it is labeled as yi, after k iteration is completed, obtain
Prediction result of the T model to all samplesFinally using D' as the input data of secondary learning algorithm S,
Obtain secondary learner L'=S (D')；
Test phase: each learning algorithm StK primary learner L is generated on K folding training sett (j), j=1,2 ..., k, this
K learner is in test set DtestEach of k prediction result (L is generated on sample xt (1)(x),Lt (2)(x),....,Lt (k)(x)), this k result is averaged according to formula (4):
Test sample x inputs T learning algorithm and obtains predicted valueIt is inputted secondary learner L',
Finally obtain prediction result
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910732567.0A CN110532898A (en) | 2019-08-09 | 2019-08-09 | A kind of physical activity recognition methods based on smart phone Multi-sensor Fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910732567.0A CN110532898A (en) | 2019-08-09 | 2019-08-09 | A kind of physical activity recognition methods based on smart phone Multi-sensor Fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110532898A true CN110532898A (en) | 2019-12-03 |
Family
ID=68661569
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910732567.0A Pending CN110532898A (en) | 2019-08-09 | 2019-08-09 | A kind of physical activity recognition methods based on smart phone Multi-sensor Fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110532898A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021115064A1 (en) * | 2019-12-10 | 2021-06-17 | 中国科学院深圳先进技术研究院 | Fitness exercise recognition method based on wearable sensor |
| CN113065612A (en) * | 2021-05-18 | 2021-07-02 | 西京学院 | Human activity classification recognition method and system based on machine learning |
| WO2022052724A1 (en) * | 2020-09-09 | 2022-03-17 | 山东科技大学 | Intelligent skin based on small-size distributed optical fiber sensing array |
| CN114298277A (en) * | 2021-12-28 | 2022-04-08 | 四川大学 | Distributed deep learning training method and system based on layer sparsization |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106095101A (en) * | 2016-06-16 | 2016-11-09 | 福建师范大学 | Human bodys' response method based on power-saving mechanism and client |
| CN107402397A (en) * | 2017-06-30 | 2017-11-28 | 北京奇虎科技有限公司 | User Activity state based on mobile terminal determines method, device and mobile terminal |
| US20180264320A1 (en) * | 2017-03-14 | 2018-09-20 | Lumo BodyTech, Inc | System and method for automatic location detection for wearable sensors |
| CN109272365A (en) * | 2018-10-15 | 2019-01-25 | 广东工业大学 | A kind of flow of restaurant guests prediction technique, device and equipment based on data mining |
| CN109447099A (en) * | 2018-08-28 | 2019-03-08 | 西安理工大学 | A kind of Combining Multiple Classifiers based on PCA dimensionality reduction |
| US20200210561A1 (en) * | 2017-12-15 | 2020-07-02 | Alibaba Group Holding Limited | Biometric authentication, identification and detection method and device for mobile terminal and equipment |
-
2019
- 2019-08-09 CN CN201910732567.0A patent/CN110532898A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106095101A (en) * | 2016-06-16 | 2016-11-09 | 福建师范大学 | Human bodys' response method based on power-saving mechanism and client |
| US20180264320A1 (en) * | 2017-03-14 | 2018-09-20 | Lumo BodyTech, Inc | System and method for automatic location detection for wearable sensors |
| CN107402397A (en) * | 2017-06-30 | 2017-11-28 | 北京奇虎科技有限公司 | User Activity state based on mobile terminal determines method, device and mobile terminal |
| US20200210561A1 (en) * | 2017-12-15 | 2020-07-02 | Alibaba Group Holding Limited | Biometric authentication, identification and detection method and device for mobile terminal and equipment |
| CN109447099A (en) * | 2018-08-28 | 2019-03-08 | 西安理工大学 | A kind of Combining Multiple Classifiers based on PCA dimensionality reduction |
| CN109272365A (en) * | 2018-10-15 | 2019-01-25 | 广东工业大学 | A kind of flow of restaurant guests prediction technique, device and equipment based on data mining |
Non-Patent Citations (5)
| Title |
|---|
| ADDIS ABEBE ASSEFA等: "Bio-Mechanical Distracted Driver Recognition Based on Stacked Autoencoder and Convolutional Neural Network", 2019 IEEE 2ND INTERNATIONAL CONFERENCE ON INFORMATION COMMUNICATION AND SIGNAL PROCESSING (ICICSP) * |
| YUN-JUE CHEN等: "Unsupervised learning of space-time symmetric patterns in ＲGB-D videos for 4D human activity detection", 2017 17TH INTERNATIONAL SYMPOSIUM ON COMMUNICATIONS AND INFORMATION TECHNOLOGIES ( ISCIT） * |
| 何坚: "基于ZigBee 和CNN 的低功耗跌倒检测技术", 《天津大学学报(自然科学与工程技术版)》 * |
| 周义凯等: "基于 CNN 的人体姿态识别", 《计算机与现代化》 * |
| 曾禄杰: "基于多分类器的日常人体活动类型识别技术研究", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021115064A1 (en) * | 2019-12-10 | 2021-06-17 | 中国科学院深圳先进技术研究院 | Fitness exercise recognition method based on wearable sensor |
| WO2022052724A1 (en) * | 2020-09-09 | 2022-03-17 | 山东科技大学 | Intelligent skin based on small-size distributed optical fiber sensing array |
| US11754423B2 (en) | 2020-09-09 | 2023-09-12 | Shandong University Of Science And Technology | Intelligent skin based on small-size distributed optical fiber sensing array |
| CN113065612A (en) * | 2021-05-18 | 2021-07-02 | 西京学院 | Human activity classification recognition method and system based on machine learning |
| CN114298277A (en) * | 2021-12-28 | 2022-04-08 | 四川大学 | Distributed deep learning training method and system based on layer sparsization |
| CN114298277B (en) * | 2021-12-28 | 2023-09-12 | 四川大学 | Distributed deep learning training method and system based on layer sparsification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Khatun et al. | Deep CNN-LSTM with self-attention model for human activity recognition using wearable sensor | |
| CN105678222B (en) | A kind of mobile device-based Human bodys' response method | |
| CN110532898A (en) | A kind of physical activity recognition methods based on smart phone Multi-sensor Fusion | |
| CN106228200A (en) | A kind of action identification method not relying on action message collecting device | |
| CN110334573B (en) | Human motion state discrimination method based on dense connection convolutional neural network | |
| CN108245172B (en) | Human body posture recognition method free of position constraint | |
| CN106096662A (en) | Human motion state identification based on acceleration transducer | |
| Gil-Martín et al. | Human activity recognition adapted to the type of movement | |
| Su et al. | HDL: Hierarchical deep learning model based human activity recognition using smartphone sensors | |
| CN112464738B (en) | Improved naive Bayes algorithm user behavior identification method based on mobile phone sensor | |
| CN108958482B (en) | Similarity action recognition device and method based on convolutional neural network | |
| CN108985157A (en) | A kind of gesture identification method and device | |
| Hung et al. | Activity recognition with sensors on mobile devices | |
| CN106295675A (en) | A kind of Activity recognition method of based on smart mobile phone of high accuracy | |
| Büber et al. | Discriminative time-domain features for activity recognition on a mobile phone | |
| CN109765996A (en) | Insensitive gesture detection system and method are deviated to wearing position based on FMG armband | |
| Masnad et al. | Human activity recognition using smartphone sensors with context filtering | |
| Dehkordi et al. | Feature extraction and feature selection in smartphone-based activity recognition | |
| Yang et al. | Smartphones based online activity recognition for indoor localization using deep convolutional neural network | |
| CN106370180A (en) | Inertial sensor initial position recognition method based on dynamic time wrapping algorithm | |
| Aggelides et al. | A gesture recognition approach to classifying allergic rhinitis gestures using wrist-worn devices: a multidisciplinary case study | |
| Ghobadi et al. | A robust automatic gait monitoring approach using a single IMU for home-based applications | |
| CN112370058A (en) | Method for identifying and monitoring emotion of user based on mobile terminal | |
| Chakraborty et al. | An approach for designing low cost deep neural network based biometric authentication model for smartphone user | |
| Fu et al. | Ping pong motion recognition based on smart watch |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191203 |
|
| RJ01 | Rejection of invention patent application after publication |