CN108564100A - The method of mobile terminal and its generation classification of motion model, storage device - Google Patents
The method of mobile terminal and its generation classification of motion model, storage device Download PDFInfo
- Publication number
- CN108564100A CN108564100A CN201711337023.1A CN201711337023A CN108564100A CN 108564100 A CN108564100 A CN 108564100A CN 201711337023 A CN201711337023 A CN 201711337023A CN 108564100 A CN108564100 A CN 108564100A
- Authority
- CN
- China
- Prior art keywords
- action
- data
- value
- disaggregated model
- action data
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/24765—Rule-based classification
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application discloses a kind of method generating classification of motion model, this method includes：The first disaggregated model of acquisition for mobile terminal；Mobile terminal detect and record user predetermined period action data；Temporal signatures value and frequency domain character value are obtained according to action data；The second disaggregated model is obtained according to temporal signatures value, frequency domain character value and the first disaggregated model；Wherein, temporal signatures value includes acceleration signature value and environmental characteristic value.Disclosed herein as well is a kind of mobile terminals and a kind of storage device.By the above-mentioned means, the application can make the classification of motion model of generation more accurate when classifying to the action of user.
Description
Technical field
This application involves electronic device fields, more particularly to the side of a kind of mobile terminal and its generation classification of motion model
Method, storage device.
Background technology
Optimize with the continuous upgrading of mobile Internet and Intelligent hardware, people’s lives idea and demand are also constantly sent out
Changing increasingly pays close attention to own health, also increasingly pays close attention to daily motion conditions.
The classification of motion method of human body is that the action data recorded by terminal is matched with classification of motion model at present,
To classify to the action of user.And general classification of motion model is only the combination of acceleration rate threshold, and by this
Classification of motion model can not accurately classify to the action of user when classifying to the action of user.
Invention content
The application mainly solving the technical problems that provide a kind of method of mobile terminal and its generation classification of motion model,
Storage device can make the classification of motion model of generation more accurate when classifying to the action of user.
In order to solve the above technical problems, the technical solution that the application uses is：A kind of generation classification of motion mould is provided
The method of type, this method include：The first disaggregated model of acquisition for mobile terminal；Mobile terminal detects and records user in predetermined period
Action data；Temporal signatures value and frequency domain character value are obtained according to action data；According to temporal signatures value, frequency domain character value with
And first disaggregated model obtain the second disaggregated model；Wherein, temporal signatures value includes acceleration signature value and environmental characteristic value.
In order to solve the above technical problems, another technical solution that the application uses is：A kind of classification of motion method, the party
Method includes：The first disaggregated model of acquisition for mobile terminal；Mobile terminal detects and records the action data of user, the action data packet
Include the second action data in the first action data and the second time period in first time period；According to the first action data and
One disaggregated model obtains the second disaggregated model；Second action data and the second disaggregated model are matched with to the second action number
According to classification.
In order to solve the above technical problems, another technical solution that the application uses is：A kind of mobile terminal is provided, the shifting
Dynamic terminal includes processor and the memory that is connect with the processor, and for storing computer program, processor is used for memory
Call computer program to execute the above method.
In order to solve the above technical problems, another technical solution that the application uses is：A kind of storage device is provided, this is deposited
Storage device can store computer program, which can be performed to realize the above method.
The advantageous effect of the application is：The case where being different from the prior art, the classification mould of the application acquisition for mobile terminal first
Type；Mobile terminal detect and record user predetermined period action data；Temporal signatures value and frequency are obtained according to action data
Characteristic of field value；The second disaggregated model is obtained according to temporal signatures value, frequency domain character value and the first disaggregated model；Wherein, time domain
Characteristic value includes acceleration signature value and environmental characteristic value.By the above-mentioned means, the method that the application generates classification of motion model
Due to obtaining temporal signatures value and frequency domain character value according to action data, and utilize temporal signatures value and frequency domain character value and first
Disaggregated model obtain the second disaggregated model, enable the terminals to synthetic user movement when time domain data and frequency domain data to user
Action classify, to keep the classification of motion model of generation more accurate when classifying to the action of user.
Description of the drawings
Fig. 1 is the flow diagram for the method that the embodiment of the present application mobile terminal generates classification of motion model；
Fig. 2 is the flow diagram of the embodiment of the present application mobile terminal classification of motion method；
Fig. 3 is the hardware architecture diagram of the embodiment of the present application mobile terminal；
Fig. 4 is the schematic diagram of the embodiment of the present application storage device.
Specific implementation mode
It is understandable to enable the above objects, features, and advantages of the application to become apparent, below in conjunction with the accompanying drawings, to the application
Specific implementation mode be described in detail.It is understood that specific embodiment described herein is only used for explaining this Shen
Please, rather than the restriction to the application.It also should be noted that illustrating only for ease of description, in attached drawing and the application
Relevant part rather than entire infrastructure.Based on the embodiment in the application, those of ordinary skill in the art are not making creation
Property labour under the premise of all other embodiment for being obtained, shall fall in the protection scope of this application.
Term " first ", " second " in the application etc. be for distinguishing different objects, rather than it is specific suitable for describing
Sequence.In addition, term " comprising " and " having " and their any deformations, it is intended that cover and non-exclusive include.Such as comprising
The step of process of series of steps or unit, method, system, product or equipment are not limited to list or unit, and
It further includes the steps that optionally not listing or unit to be, or further includes optionally for these processes, method, product or equipment
Intrinsic other steps or unit.
Referenced herein " embodiment " is it is meant that a particular feature, structure, or characteristic described can wrap in conjunction with the embodiments
It is contained at least one embodiment of the application.Each position in the description occur the phrase might not each mean it is identical
Embodiment, nor the independent or alternative embodiment with other embodiments mutual exclusion.Those skilled in the art explicitly and
Implicitly understand, embodiment described herein can be combined with other embodiments.
Referring to Fig. 1, Fig. 1 is the flow signal for the method that the embodiment of the present application mobile terminal generates classification of motion model
Figure.
In the present embodiment, the method which generates classification of motion model may comprise steps of：
Step S11：The first disaggregated model of acquisition for mobile terminal.
Optionally, which can be smart mobile phone, wearable smart machine or tablet computer, wherein can
Wearable intelligent equipment can be smartwatch or intelligent glasses etc., and in other embodiments, mobile terminal may be it
Its moveable portable intelligent terminal device, the application are not restricted this.
Optionally, the first disaggregated model may include the disaggregated model of multiple actions, for example, walking, running, cycling, trip
It the disaggregated models of actions such as swims, go upstairs, going downstairs, jumping.In the present embodiment, the first disaggregated model include it is static action,
The disaggregated model of action, running action, cycling action and other five kinds of actions of action on foot, in other embodiments, first
Disaggregated model can also include the disaggregated model of other actions, and the application is not also restricted this.
The acquisition of first disaggregated model can be there are many mode.For example, in one embodiment, user sends per a kind of action
The characteristic and model structure of model, wherein characteristic includes that temporal signatures data, frequency domain character data and environment are special
Data etc. are levied, processor obtains these characteristics, and then gets the first disaggregated model, and optionally, processor can also incite somebody to action
First disaggregated model is stored into memory, to call the first disaggregated model in memory when carrying out the classification of motion.Another
In one embodiment, user can send the action data in predetermined period per a kind of action to mobile terminal, for example, user sends
The action that each in static action, action of walking, running action, cycling action and other actions within one hour acts
To processor, processor receives these action datas and stores that data in memory data, and processor utilizes these
Action data establishes the model per a kind of action, to get the first disaggregated model.
Step S12：Mobile terminal detect and record user predetermined period action data.
In the present embodiment, mobile terminal is carried on the body of user, is moved with the movement of user.For example, when movement
When terminal is smartwatch, mobile terminal can be worn in wrist by user, alternatively, when mobile terminal is smart mobile phone, be used
Mobile terminal can be placed in pocket or packet by family, in this way, user is during exercise, drive mobile terminal to generate movement, in turn
So that mobile terminal is detected and the action data of user is recorded.
Specifically, detecting and the action data for recording user may include：The processor of mobile terminal controls sensor
The action data of user is detected, processor reads the action data of the user of sensor detection and stores the action data of user
In memory.After the action data for reading the user of sensor detection, the action data of user can be cached easy to power down
The property lost memory can also be to store to non-power-failure volatile memory, and the embodiment of the present application does not limit this.Processor can
To read the data of a period of time inner sensor detection of caching, such as read the action number of the user cached in a period of time
According to for example, in the present embodiment, processor reads the action data for caching user on a sensor in one minute.
In the present embodiment, sensor may include environmental sensor, acceleration transducer and gravity sensor etc.,
In, environmental sensor can also include light sensor and range sensor etc., and acceleration transducer may include that three axis accelerate
Spend sensor or linear 3-axis acceleration sensor.Wherein, when acceleration transducer is 3-axis acceleration sensor, three axis add
Velocity sensor is used to detect and record the acceleration information of three axis (X-axis, Y-axis, Z axis) of user during exercise, and adds
Speed data can be influenced by mobile terminal acceleration of gravity；It is linear 3-axis acceleration sensor in acceleration transducer
When, 3-axis acceleration sensor is also used for detecting and recording the acceleration of three axis (X-axis, Y-axis, Z axis) of user during exercise
Data, but acceleration information can or can not be influenced by mobile terminal acceleration of gravity；Gravity sensor is for detecting and remembering
Employ the gravity 3-axis acceleration data of family during exercise；Light sensor is mobile whole during exercise for detecting and recording user
The environment bright angle value of environment where end；And range sensor for detect and record user during exercise mobile terminal with
The distance value of surrounding objects.
It, can be with it is to be appreciated that when the sensor of mobile terminal includes 3-axis acceleration sensor and gravity sensor
There is no linear 3-axis acceleration sensor, similarly, when the sensor of mobile terminal includes linear 3-axis acceleration sensor,
Can there is no 3-axis acceleration sensor and gravity sensor.
In other embodiments, sensor may include not only the sensor, can also include angular transducer, gyro
At least one of sensors such as instrument, height sensor and position sensor.Wherein, angular transducer is for detecting and recording
The angle-data of mobile terminal；Gyroscope is used to detect and record the angle and bearing data of mobile terminal；Height sensor is used
In detection altitude data；And position sensor is used for detection position data.
Optionally, in one embodiment, detection user can be the processing of mobile terminal in the action data of predetermined period
Device control sensor with predetermined sampling frequency detection user predetermined period action data.The predetermined sampling frequency can be
10Hz is to any value between 50Hz.For example, the predetermined sampling frequency can be 10Hz, 30Hz or 50Hz, in the present embodiment
In, which is 30Hz.
Optionally, in other embodiments, detection user may include in the action data of predetermined period：Processor according to
The action data of user determines sample frequency；Processor controls sensor with sample frequency detection user in the dynamic of predetermined period
Make data.
For example, the user action data acquisition user that processor is detected according to first time period sensor is in first time period
Interior action data, processor analyze the type of action of first time period user according to the user action data of first time period,
Then, processor obtains the sample frequency in predetermined period according to the type of action of first time period user, wherein at the first time
Section can be a period before predetermined period.
By the above-mentioned means, processor is acute in the movement of the user for the type of action characterization for determining first time period user
When strong, control sensor detects the action data of user in predetermined period with larger sample frequency；Processor is determining
When the movement of the user of the type of action characterization of one period user is gentle, control sensor is in predetermined period with relatively low sampling frequency
Rate detects the action data of user.
In other embodiments, processor can obtain current time, then according to current time from prestoring the time
Sample frequency corresponding with the acquisition of sample frequency mapping table, then processor is according to the sample frequency of acquisition control sensor
The action data of user is detected with the sample frequency.
For example, pre-stored time and sample frequency mapping table are processor according to the different time detected in the past
The different sample frequencys that are set of action data of the user of section, according to this one-to-one relationship, generated time with adopt
Sample frequency mapping table is simultaneously stored, wherein more violent, the sampling set of action for the user that action data is characterized
Frequency is faster.
In the present embodiment, predetermined period can be chosen according to actual conditions, for example, predetermined period can choose 5
Any duration between minute to 5 hours, for example, predetermined period can be 5 minutes, 1 hour or 5 hours.The application is to this
It is not construed as limiting.
Step S13：Temporal signatures value and frequency domain character value are obtained according to action data.
Wherein, temporal signatures value includes acceleration signature value and environmental characteristic value.
Optionally, before step S13, the method that mobile terminal generates classification of motion model can also include that will act number
According to adding window, to be divided into more parts of action datas.Wherein, the data volume of every part of action data of more parts of action datas is identical, and wantonly two
There is 20%~80% identical data in the adjacent action data of part.In the present embodiment, every part of action of more parts of action datas
The data volume of data is identical, and has 50% identical data in wantonly two parts of adjacent action datas.
Action data is divided into more parts of action datas by above-mentioned, it not only can be by the action number of a long period
According to the action data for being divided into several short periods, convenient for being analyzed action data and being calculated, and since adjacent is appointed
The identical data for having 50% in two parts of action datas, can more efficiently use the action data being recorded, to make point
Class is more accurate.
Optionally, by action data adding window, after being divided into more parts of action datas, mobile terminal generates classification of motion mould
The method of type can also include being filtered to every part of action data in more parts of action datas.
Optionally, it may include 3-axis acceleration data in every part of action data to be filtered to every part of action data
It is filtered.
Optionally, it can also includes environment bright degrees of data in every part of action data to be filtered to every part of action
And/or the distance value of mobile terminal and surrounding objects is filtered.
Optionally, it is filtered selected filter to every part of action data can be the same or different, filter
High-pass filter, low-pass filter, bandpass filter or bandstop filter, the application can also be selected without limitation.
In the present embodiment, it is low-pass Gaussian filter to be filtered selected filter to every part of action data, and is utilized
The method of low pass gaussian filtering is filtered every part of action data.Specifically, action data is carried out low pass gaussian filtering
It is exactly that the action data and a Gaussian kernel are subjected to convolution, i.e.,：
Iσ=I*Gσ (I)
Wherein, GσThe one-dimensional gaussian kernel function for being σ for standard deviation, wherein the formula of one-dimensional gaussian kernel function is as follows：
In the present embodiment, low pass is carried out to the every part of action data for detecting and being recorded using formula (1) and formula (2)
Gaussian filtering.
In other embodiments, to every part of action data be filtered can according to user action data carry out it is adaptive
It should filter.For example, changing the phase relation of filter type and/or filter according to the action data of the user in the first period
Number, so as to carry out adaptive-filtering according to the action data of user, the application is not restricted the method for adaptive-filtering.
It optionally, can be to the collected gravity of gravity sensor when mobile terminal includes gravity accelerometer
3-axis acceleration data are filtered, and specific filtering mode please refers to described above, and details are not described herein again.In this reality
It applies in example, the mode that be filtered to the 3-axis acceleration data that 3-axis acceleration sensor detects and to gravity sensitive
The mode that the gravity 3-axis acceleration data that device detects are filtered is identical, for example, the two is all to utilize low pass Gauss
Filter is filtered every part of action data in more parts of action datas.
Optionally, after every part of action data is filtered in more parts of action datas, mobile terminal generation action point
The method of class model can also include modifying the 3-axis acceleration data for detecting and being recorded to remove acceleration of gravity
Influence, obtain linear 3-axis acceleration data.
Specifically, when acceleration transducer in the terminal is 3-axis acceleration sensor, since three axis accelerate
Degree sensor detects and the 3-axis acceleration data being recorded can be influenced by gravity 3-axis acceleration, it is therefore desirable to detection
And the 3-axis acceleration data being recorded are modified to remove the influence of acceleration of gravity.Concrete modification mode can be filtering
3-axis acceleration data, filtered gravity 3-axis acceleration data and difference equation afterwards obtains no acceleration of gravity number
According to the linear 3-axis acceleration data of influence.In the present embodiment, the expression formula of difference equation is：
Y (n)=A*y (n-1)+(1-A) * x (n) (3)
Wherein, y (n-1) represents filtered 3-axis acceleration data, and x (n) represents filtered gravity 3-axis acceleration
Data, y (n) represent the linear 3-axis acceleration data that no gravity influences, and A values can be set according to actual conditions, this Shen
Please this is not restricted.
Optionally, after modifying to 3-axis acceleration data, the method that mobile terminal generates classification of motion model may be used also
To include carrying out vector sum addition to each 3-axis acceleration data of filtered every part of action data.Specifically, filtering
And each linear 3-axis acceleration data modified are respectively
When, the vector sum of the 3-axis acceleration data is By above-mentioned
Mode, for example, when being placed on pocket, can weaken mobile terminal with respect to pocket when mobile terminal is unlocked
The influence for generating movement and classification results being generated.
In the present embodiment, environmental characteristic value includes：Environment bright angle value every part of action data in more parts of action datas
Mean value and/or the distance value of mobile terminal and surrounding objects every part of action data in more parts of action datas mean value.
It is to be appreciated that in the present embodiment, environment bright angle value in more parts of action datas every part of action data it is equal
Value can be：The mean value of the environmental data of every part of action data after filtered.In other embodiments, environment bright angle value exists
The mean value of every part of action data can be in more parts of action datas：Without filtered every part of action data environmental data it is equal
Value.Similarly, in the present embodiment, the distance value of mobile terminal and surrounding objects every part of action data in more parts of action datas
Mean value can be：The mean value of the mobile terminal of every part of action data after filtered and the distance value of surrounding objects.Other
In embodiment, the mean value of every part of action data can also in more parts of action datas for the distance value of mobile terminal and surrounding objects
It is：Without the mean value of the distance value of the mobile terminal and surrounding objects of filtered every part of action data.
In other embodiments, environmental characteristic value can also include：Humidity data every part of action in more parts of action datas
The mean value of data, the mean value of temperature data every part of action data in more parts of action datas.
In another embodiment, environmental characteristic value can also include：Environment bright angle value is every part in more parts of action datas
The side of the variance and/or mobile terminal of action data and distance value every part of action data in more parts of action datas of surrounding objects
Difference.
In another embodiment, environmental characteristic value can also include：Environment bright angle value is every part in more parts of action datas
The standard deviation and/or mobile terminal of action data and the distance value of surrounding objects every part of action data in more parts of action datas
Standard deviation.
Acceleration signature value can be that every part of action data is filtered in more parts of action datas, change and vector sum is added
Acceleration information.
Optionally, frequency domain character value may include：The 3-axis acceleration data of filtered every part of action data are carried out
Fast Fourier Transform (FFT), and the Nth power of acquisition preceding 2 maintains number, wherein N is natural number.Optionally, filtered every part is acted
It can be to accelerate to three axis of filtered every part of action data that the 3-axis acceleration data of data, which carry out Fast Fourier Transform (FFT),
Each axle acceleration data of degrees of data carries out Fast Fourier Transform (FFT) respectively, and, it obtains each axle acceleration data and passes through
Preceding 2 Nth power after Fast Fourier Transform (FFT) maintains number, for example, obtaining preceding 4 maintains number, obtains preceding 8 and maintains number, obtain preceding 16
It maintains number or obtains preceding 32 and maintain number etc., the application is not construed as limiting this.
In the present embodiment, frequency domain character value may include：To the 3-axis acceleration number of filtered every part of action data
According to each axle acceleration data carry out Fast Fourier Transform (FFT) respectively, and obtain preceding 16 and maintain number.
Wherein, the Java code for obtaining the coefficient of Fast Fourier Transform (FFT) is as follows：
By the above-mentioned means, each axle acceleration data for getting the 3-axis acceleration data of every part of action data carries out
The coefficient value of Fast Fourier Transform (FFT), and take preceding 16 data as frequency domain character value.
Step S14：The second disaggregated model is obtained according to temporal signatures value, frequency domain character value and the first disaggregated model.
Optionally, obtaining the second disaggregated model according to temporal signatures value, frequency domain character value and the first disaggregated model can be with
Including：Data sample is obtained according to temporal signatures value and frequency domain character value, and obtains grader；According to data sample, grader
And first disaggregated model obtain the second disaggregated model.
Optionally, data sample includes training sample and test sample.
Specifically, data sample is divided at least two parts of samples, using any part of sample at least two parts of samples as
Test sample at least will remove test sample as training sample by two parts of samples.
In the present embodiment, data sample is divided into three parts of samples, wherein wantonly two parts of samples are as training sample, it is remaining
A sample is as test sample.
Optionally, it can be that data sample is equally divided into three parts data sample to be divided into three parts, specifically, by more parts
Temporal signatures value and frequency domain character value in action data are divided into three parts.
Optionally, grader can there are many selections, for example, grader can select Bayes classifier, neural network
One or more combinations of grader, Logistic graders, support vector machine classifier, the application are not restricted this.
In the present embodiment, grader is support vector machine classifier.
After data sample is divided into training sample and test sample, support vector machine classifier and training sample are utilized
First disaggregated model is updated.
Since the first disaggregated model in the application is static action, action of walking, running action, cycling acts and it
It acts the disaggregated model of this five kinds actions, and support vector machine classifier is proposed to solve two class classification problems, therefore,
The first disaggregated model of the application creates 5 support vector machine classifiers, respectively by the way of multiple classifiers combinations
For：The models that static action model and four kinds of actions for removing static action combine, action model and removing, which are walked, on foot acts
The models that combine of four kinds of actions, running action model and the models combined except four kinds of the action actions that go jogging, cycling action
The models that model is combined with four kinds of actions for removing cycling action, and, other action models with remove four kinds of other actions
Act the model combined.
It is to be appreciated that in other embodiments, being selected according to the difference of the action of classification, other quantity can be created
Support vector machine classifier, concrete implementation method is identical with the creation method of the application, and the application is not restricted this.
In the present embodiment, it is updated using grader and first the first disaggregated model of training sample pair and may include：
5 support vector machine classifiers in the characteristic value and disaggregated model of every part of action data in training sample are obtained, according to training
5 support vector machine classifiers in sample in the characteristic value and disaggregated model of every part of action data obtain in training sample every part
Type of action representated by action data, according to every part of action number in more parts of action datas in the first disaggregated model and training sample
The first disaggregated model is updated according to representative type of action, and then obtains the second disaggregated model.
Specifically, according to 5 supporting vectors in the characteristic value and disaggregated model of every part of action data in training sample
One embodiment that machine grader obtains the type of action representated by every part of action data in training sample can be：Processor can
First to detect the action of the running of user, i.e., the corresponding vector of running action is positive collection in support vector machine classifier, is removed
The vector that four kinds of actions of running action are corresponding is negative collection, walks in processor detection user, cycles, static and other dynamic
When making, method is similar, and details are not described herein again.It is understood that the sequence of detection user action can change.Optionally, exist
In training process, training is tied in order to avoid a kind of data volume of action data and the data volume of four kinds of action datas imbalance
Fruit has an impact, and can choose the data volume of a quarter of above-mentioned negative concentration as negative collection.Trained sample is got in processor
It, can be according to every part of action number in more parts of action datas in training sample after type of action in this representated by every part of action data
It is trained according to representative the first disaggregated model of type of action pair, and then updates the first disaggregated model, obtain the second classification mould
Type.
In the present embodiment, in order to make full use of the user data for detecting and being recorded, by the action number in predetermined period
According to being divided into three parts, respectively L, M and N.It, can be according to the first disaggregated model and training sample when updating the first disaggregated model
In the first disaggregated model of type of action pair in more parts of action datas representated by every part of action data updated three times, and every time
Examine whether updated model meets preset requirement by test sample after update.
For example, the L and M of action data is used to be trained as the first disaggregated model of training sample pair first, test is utilized
Sample N tests the first disaggregated model after the L of action data and M training.And meet preset requirement in test result
When, use the L and N of action data to continue to train as the first disaggregated model of training sample pair, using test sample M to through dynamic
Make the first disaggregated model after L and the N training of data to be tested, and when test result meets preset requirement, with action number
According to M and N be further continued for being trained as the first disaggregated model of training sample pair, obtain updated first disaggregated model, profit
The first disaggregated model after the M of action data and N training is tested with test sample L, and is met in test result pre-
If it is required that when, using updated first disaggregated model as the second disaggregated model.
Optionally, in one embodiment, it can be to classify using updated first that test result, which meets preset requirement,
The first test sample of category of model, if update before the first disaggregated model classify the first training sample classification results and update after
The first disaggregated model classify the first test sample classification results meet scheduled error, it may be considered that test result meets
It is required that；In another embodiment, it can be the characteristic value in updated first disaggregated model that test result, which meets preset requirement,
Meet scheduled functional relation with the characteristic value in test sample, it may be considered that test result meets the requirements, the application is to this
It is not restricted.
Optionally, in above-mentioned implementation process, if the first disaggregated model of any test sample pair or trained sample instruction
The first disaggregated model after white silk tested after test result when not meeting preset value, can data sample be divided into again
Two training samples and the second test sample, and be updated using second the first disaggregated model of training sample pair, until test is tied
Fruit meets the requirements, wherein the data that the data that the second training sample includes and aforementioned training sample include are different；Second test specimens
Originally the data that the data for including and aforementioned test sample include are different.
In other embodiments, the first action data can also be divided into other parts, and training sample and test sample can also
It is divided according to actual conditions, for example, can the first action data be divided into four parts, training sample is used as by arbitrary three parts,
Remaining a as test sample, alternatively, being used as training sample by arbitrary two parts, remaining two parts are used as test sample.This
Application is not restricted this.
Referring to Fig. 2, Fig. 2 is the flow diagram of the embodiment of the present application classification of motion method.
In the present embodiment, the classification of motion method of the mobile terminal may comprise steps of：
Step S21：The first disaggregated model of acquisition for mobile terminal.
Wherein, the specific execution method of the first disaggregated model of acquisition for mobile terminal is similar with the method for abovementioned steps S11, this
Place repeats no more.
Step S22：Mobile terminal detects and records the action data of user, and action data includes in first time period
The second action data in one action data and second time period.
The specific execution method for detecting and recording the action data of user is similar with the method for abovementioned steps S12, herein not
It repeats again.
Step S23：The second disaggregated model is obtained according to the first action data and the first disaggregated model.
The method that the specific method of the second disaggregated model sees above-mentioned steps S13 and step S14 is obtained, it is no longer superfluous herein
It states.
Step S24：Second action data and the second disaggregated model are matched to classify to the second action data.
Specifically, matching the second action data and the second disaggregated model with can be with to the classification of the second action data
Including：By the second action data in the second disaggregated model temporal signatures value and frequency domain character value match, to will use
The action at family is divided into five kinds of static action, action of walking, running action, cycling action and other actions actions.
Optionally, after step S24, the exercise data with the user for detecting and being recorded increases, can also basis
Second action data and the second disaggregated model obtain third disaggregated model, by third action data and the progress of third disaggregated model
It is equipped with and classifies to third action data ...
By the above-mentioned means, disaggregated model increases with the action data for the user for detecting and being recorded, continuous basis
The action data of user is updated disaggregated model, and disaggregated model is made to be more in line with the motor habit of user, to improve
The accuracy of sorting technique.
Referring to Fig. 3, Fig. 3 is the hardware architecture diagram of the embodiment of the present application mobile terminal.In the present embodiment, mobile
Terminal 20 includes processor 21, bus 22, the memory 23 being connect by bus 22 with processor 21 and sensor 24, memory
23 for storing computer program, and processor 21 is for calling computer program to execute the mobile end of above-mentioned any one embodiment
End generates the method and classification of motion method of classification of motion model.
Sensor 24 may include environmental sensor, acceleration transducer and gravity sensor etc., wherein environmental sensor
Further include light sensor and range sensor etc., acceleration transducer can be that 3-axis acceleration sensor or linear three axis add
Velocity sensor, sensor 24 can also include at least one of gyroscope, height sensor and position sensor, sensing
The explanation of 24 concrete function of device refers to the description in any one embodiment, and details are not described herein again.
Referring to Fig. 4, Fig. 4 is the schematic diagram of the embodiment of the present application storage device.In the present embodiment, storage device 30 is deposited
Computer program is contained, which can be performed to realize the mobile terminal generation action of above-mentioned any one embodiment
The method of disaggregated model.
Optionally, storage device 30 can be USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory),
Random access memory (RAM, Random Access Memory), magnetic disc, CD or server etc. are various can to store journey
The medium of sequence code.
Optionally, which can also be the memory 23 in above-described embodiment.
The case where being different from the prior art, the first disaggregated model of the application acquisition for mobile terminal；Mobile terminal detects and remembers
Employ family predetermined period action data；Temporal signatures value and frequency domain character value are obtained according to action data；According to time domain spy
Value indicative, frequency domain character value and the first disaggregated model obtain the second disaggregated model；Wherein, temporal signatures value includes acceleration signature
Value and environmental characteristic value.By the above-mentioned means, the application generates the method for classification of motion model due to being obtained according to action data
Temporal signatures value and frequency domain character value, and obtain the second classification using temporal signatures value and frequency domain character value and the first disaggregated model
Model, time domain data and frequency domain data when enabling the terminals to synthetic user movement classify to the action of user, to
Keep the classification of motion model of generation more accurate when classifying to the action of user.
It these are only presently filed embodiment, be not intended to limit the scope of the claims of the application, it is every to utilize the application
Equivalent structure or equivalent flow shift made by specification and accompanying drawing content is applied directly or indirectly in other relevant technologies
Field includes similarly in the scope of patent protection of the application.
Claims (10)
1. a kind of method generating classification of motion model, which is characterized in that the method includes：
The first disaggregated model of acquisition for mobile terminal；
The mobile terminal detect and record user predetermined period action data；
Temporal signatures value and frequency domain character value are obtained according to the action data；
The second disaggregated model is obtained according to the temporal signatures value, the frequency domain character value and first disaggregated model；
Wherein, the temporal signatures value includes acceleration signature value and environmental characteristic value.
2. the method according to claim 1 for generating classification of motion model, which is characterized in that described according to the action
Before data acquisition temporal signatures value and frequency domain character value, the method further includes：
By the action data adding window, to be divided into more parts of action datas；
To every part of action data filtering in the more parts of action datas；
Wherein, the data volume of every part of action data is identical, and has 20%~80% in wantonly two parts of adjacent action datas
Identical data.
3. the method according to claim 2 for generating classification of motion model, which is characterized in that the action data includes three
Axle acceleration data, it is described temporal signatures value and frequency domain character value are obtained according to the action data before, the method is also
Including：
Vector sum addition is carried out to each 3-axis acceleration data of filtered every part of action data.
4. the method according to claim 2 for generating classification of motion model, which is characterized in that described according to the action number
Include according to frequency domain character value is obtained：Fast Fourier is carried out to the 3-axis acceleration data of filtered every part of action data
Transformation, and the Nth power of acquisition preceding 2 maintains number, wherein N is natural number.
5. the method according to claim 2 for generating classification of motion model, which is characterized in that the action data includes ring
The distance value of border brightness values and/or terminal and surrounding objects, the environmental characteristic value include：The environment bright angle value is in institute
The mean value of every part of action data and/or the distance value of the terminal and the surrounding objects are stated in more parts of action datas described more
The mean value of every part of action data in part action data.
6. the method according to claim 1 for generating classification of motion model, which is characterized in that described according to time domain spy
Value indicative, the frequency domain character value and first disaggregated model obtain the second disaggregated model：
Data sample is obtained according to the temporal signatures value and the frequency domain character value, and obtains grader；
The second disaggregated model is obtained according to the data sample, the grader and first disaggregated model.
7. the method according to claim 6 for generating classification of motion model, which is characterized in that
The data sample includes training sample and test sample；
The grader includes support vector machine classifier；
It is described to include according to the data sample, the grader and first disaggregated model the second disaggregated model of acquisition：
First disaggregated model is updated according to the training sample, the support vector machine classifier；
Updated first disaggregated model is tested according to the test sample and generates test result；
If the test result meets preset requirement, using updated first disaggregated model as described second point
Class model.
8. a kind of classification of motion method, which is characterized in that the method includes：
The first disaggregated model of acquisition for mobile terminal；
The mobile terminal detects and records the action data of user, and the action data includes first dynamic in first time period
Make the second action data in data and second time period；
The second disaggregated model is obtained according to first action data and first disaggregated model；
Second action data and second disaggregated model are matched to classify to second action data.
9. a kind of mobile terminal, which is characterized in that the mobile terminal includes processor and the storage that is connected to the processor
Device, the memory is for storing computer program, and the processor is for calling the computer program to be wanted with perform claim
Seek the method described in 1-8 any one.
10. a kind of storage device, which is characterized in that the storage device is stored with computer program, the computer program energy
It is enough performed to realize the method described in claim 1-8 any one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711337023.1A CN108564100A (en) | 2017-12-12 | 2017-12-12 | The method of mobile terminal and its generation classification of motion model, storage device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711337023.1A CN108564100A (en) | 2017-12-12 | 2017-12-12 | The method of mobile terminal and its generation classification of motion model, storage device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108564100A true CN108564100A (en) | 2018-09-21 |
Family
ID=63530276
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711337023.1A Pending CN108564100A (en) | 2017-12-12 | 2017-12-12 | The method of mobile terminal and its generation classification of motion model, storage device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108564100A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110205359A1 (en) * | 2010-02-19 | 2011-08-25 | Panasonic Corporation | Video surveillance system |
| CN103076619A (en) * | 2012-12-27 | 2013-05-01 | 山东大学 | System and method for performing indoor and outdoor 3D (Three-Dimensional) seamless positioning and gesture measuring on fire man |
| CN105528613A (en) * | 2015-11-30 | 2016-04-27 | 南京邮电大学 | Behavior identification method based on GPS speed and acceleration data of smart phone |
| CN105956558A (en) * | 2016-04-26 | 2016-09-21 | 陶大鹏 | Human movement identification method based on three-axis acceleration sensor |
| CN106210269A (en) * | 2016-06-22 | 2016-12-07 | 南京航空航天大学 | A kind of human action identification system and method based on smart mobile phone |
| CN106237604A (en) * | 2016-08-31 | 2016-12-21 | 歌尔股份有限公司 | Wearable device and the method utilizing its monitoring kinestate |
| CN107103297A (en) * | 2017-04-20 | 2017-08-29 | 武汉理工大学 | Gait identification method and system based on mobile phone acceleration sensor |
-
2017
- 2017-12-12 CN CN201711337023.1A patent/CN108564100A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110205359A1 (en) * | 2010-02-19 | 2011-08-25 | Panasonic Corporation | Video surveillance system |
| CN103076619A (en) * | 2012-12-27 | 2013-05-01 | 山东大学 | System and method for performing indoor and outdoor 3D (Three-Dimensional) seamless positioning and gesture measuring on fire man |
| CN105528613A (en) * | 2015-11-30 | 2016-04-27 | 南京邮电大学 | Behavior identification method based on GPS speed and acceleration data of smart phone |
| CN105956558A (en) * | 2016-04-26 | 2016-09-21 | 陶大鹏 | Human movement identification method based on three-axis acceleration sensor |
| CN106210269A (en) * | 2016-06-22 | 2016-12-07 | 南京航空航天大学 | A kind of human action identification system and method based on smart mobile phone |
| CN106237604A (en) * | 2016-08-31 | 2016-12-21 | 歌尔股份有限公司 | Wearable device and the method utilizing its monitoring kinestate |
| CN107103297A (en) * | 2017-04-20 | 2017-08-29 | 武汉理工大学 | Gait identification method and system based on mobile phone acceleration sensor |
Non-Patent Citations (1)
| Title |
|---|
| 徐川龙: "基于三维加速度传感器的人体行为识别", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Human activity recognition based on wearable sensor using hierarchical deep LSTM networks | |
| Quaid et al. | Wearable sensors based human behavioral pattern recognition using statistical features and reweighted genetic algorithm | |
| CN110263870B (en) | Motion state identification method and device, intelligent terminal and storage medium | |
| Arif et al. | Better physical activity classification using smartphone acceleration sensor | |
| Otebolaku et al. | User context recognition using smartphone sensors and classification models | |
| JP7544379B2 (en) | Information processing device, information processing method, trained model generation method and program | |
| US10588517B2 (en) | Method for generating a personalized classifier for human motion activities of a mobile or wearable device user with unsupervised learning | |
| CN108703760A (en) | Human motion gesture recognition system and method based on nine axle sensors | |
| KR20140116481A (en) | Activity classification in a multi-axis activity monitor device | |
| CN106999748B (en) | Systems, devices and methods relating to athletic data | |
| WO2017003593A1 (en) | Customized network traffic models to detect application anomalies | |
| RU2606880C2 (en) | Method, device and software for activity sensor data processing | |
| CN110163264B (en) | A walking pattern recognition method based on machine learning | |
| WO2012134797A1 (en) | Methods, devices, and apparatuses for activity classification using temporal scaling of time-referenced features | |
| Ranakoti et al. | Human fall detection system over IMU sensors using triaxial accelerometer | |
| Lu et al. | Mobile online activity recognition system based on smartphone sensors | |
| CN108827290A (en) | A kind of human motion state inverting device and method | |
| CN108564100A (en) | The method of mobile terminal and its generation classification of motion model, storage device | |
| WO2014191803A1 (en) | Acceleration-based step activity detection and classification on mobile devices | |
| Hashim et al. | Machine learning-based human activity recognition using neighbourhood component analysis | |
| Krishnaprabha et al. | Predicting human activity from mobile sensor data using CNN architecture | |
| Othmen et al. | A novel on-wrist fall detection system using Supervised Dictionary Learning technique | |
| JP2012108836A (en) | Interpersonal property estimation device, estimation method and estimation program based on daily measurement data | |
| Liu et al. | Recognition of aerobics movement posture based on multisensor movement monitoring | |
| Skoglund et al. | Activity tracking using ear-level accelerometers |
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: 20180921 |
|
| RJ01 | Rejection of invention patent application after publication |