CN107862276A - A kind of Activity recognition method and terminal - Google Patents
A kind of Activity recognition method and terminal Download PDFInfo
- Publication number
- CN107862276A CN107862276A CN201711060185.5A CN201711060185A CN107862276A CN 107862276 A CN107862276 A CN 107862276A CN 201711060185 A CN201711060185 A CN 201711060185A CN 107862276 A CN107862276 A CN 107862276A
- Authority
- CN
- China
- Prior art keywords
- signal
- block signal
- segmentation
- window
- sensed 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Complex Calculations (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a kind of Activity recognition method and terminal, wherein, behavior recognition methods includes:The sensed data of behavior to be identified is gathered, the signal that sensed data generates is segmented, generates block signal, at least one section of block signal is chosen and carries out Activity recognition.Activity recognition method and terminal provided by the invention, realize the automatic segmentation of motor message in behavior to be identified, so that different types of behavior to be identified in sensed data to be separated, need to only its behavior type be identified to a certain section of block signal caused by segmentation, avoid processing frequently classification and Activity recognition computing, reduce frequently automatic identification computing overhead, recognition accuracy is costly and time consuming short.
Description
Technical field
The present invention relates to computer technology, espespecially a kind of Activity recognition method and terminal.
Background technology
Human behavior analysis based on smart motion sensor science and technology is applied to health medical treatment, social networks, game, religion
Educate, the field such as traffic.Pass through the assistance of intelligence computation, human behavioral mode, abnormal accident or disease hair, behavioural habits and motion letter
The behaviors such as breath statistics can identify.
At present, Activity recognition method is mainly by carrying out behavior classification to the behavioral data of identification, to same type of
Behavior repeats discriminant rules and reaches certain quantity, that is, is identified as differentiated behavior.It is however, accurate in order to ensure Activity recognition
Property, using current Activity recognition method, it is necessary to handle frequently classification and Activity recognition computing, take cost source.
The content of the invention
In order to solve the above-mentioned technical problem, the invention provides a kind of Activity recognition method and terminal, reduce frequently
Automatic identification computing overhead, recognition accuracy are costly and time consuming short.
On the one hand, the invention provides a kind of Activity recognition method, including:
Gather the sensed data of behavior to be identified;
The signal that the sensed data generates is segmented, generates block signal;
Choose at least one section of block signal and carry out Activity recognition.
On the other hand, the invention provides a kind of terminal, including:
Acquisition module, for gathering the sensed data of behavior to be identified;
Segmentation module, the signal for the sensed data to be generated are segmented, and generate block signal;
Identification module, Activity recognition is carried out for choosing at least one section of block signal.
Another further aspect, the invention provides a kind of terminal, including:Processor and memory, memory, which is used to store, to be performed
Instruction;Processor calls the execute instruction, for performing the operation of above-mentioned Activity recognition embodiment of the method.
Activity recognition method and terminal provided by the invention, by gathering the sensed data of behavior to be identified, will sense number
It is segmented according to the signal of generation, generates block signal, is chosen at least one section of block signal and carry out Activity recognition.The present invention is implemented
Example is segmented by the signal for generating sensed data, and each segment signal in block signal corresponds to a kind of motion row of user
For, it is achieved thereby that in behavior to be identified motor message automatic segmentation, by different types of row to be identified in sensed data
It is separated, need to only its behavior type be identified to a certain section of block signal caused by segmentation, avoids processing frequently
Classification and Activity recognition computing, reduce frequently automatic identification computing overhead, recognition accuracy is costly and time consuming short.
Other features and advantages of the present invention will be illustrated in the following description, also, partly becomes from specification
Obtain it is clear that or being understood by implementing the present invention.The purpose of the present invention and other advantages can be by specification, rights
Specifically noted structure is realized and obtained in claim and accompanying drawing.
Brief description of the drawings
Accompanying drawing is used for providing further understanding technical solution of the present invention, and a part for constitution instruction, with this
The embodiment of application is used to explain technical scheme together, does not form the limitation to technical solution of the present invention.
Fig. 1 is the flow chart for the Activity recognition method that the embodiment of the present invention one provides;
Fig. 2 is the normalized autocorrelation of reference window provided in an embodiment of the present invention and testing window first before the first zero crossing
The schematic diagram of quadrant;
Fig. 3 is the flow chart for the Activity recognition method that the embodiment of the present invention two provides;
Fig. 4 is the structural representation for the terminal that the embodiment of the present invention one provides;
Fig. 5 is the structural representation for the terminal that the embodiment of the present invention two provides;
Fig. 6 is the segmentation result that the actual measurement behavioral data collection 1 of the application present invention is walked, run, sitting;
Fig. 7 is the noise of switch step between segmentation algorithm provided in an embodiment of the present invention separation is driven away and run;
Fig. 8 is the segmentation result that the actual measurement behavioral data collection 2 of the application present invention is walked, sits, runs and stood.
Embodiment
For the object, technical solutions and advantages of the present invention are more clearly understood, below in conjunction with accompanying drawing to the present invention
Embodiment be described in detail.It should be noted that in the case where not conflicting, in the embodiment and embodiment in the application
Feature can mutually be combined.
Can be in the computer system of such as one group computer executable instructions the flow of accompanying drawing illustrates the step of
Perform.Also, although logical order is shown in flow charts, in some cases, can be with suitable different from herein
Sequence performs shown or described step.
Fig. 1 is the flow chart for the Activity recognition method that the embodiment of the present invention one provides, as shown in figure 1, Activity recognition method
Specifically include following steps:
S101:Gather the sensed data of behavior to be identified.
The executive agent of the embodiment of the present invention can be the intelligent terminal with sensor function, such as smart mobile phone;
Can be the wearable device with sensor function, such as Intelligent bracelet.
Specifically, intelligent terminal or wearable device gather the sensed data of behavior to be identified by sensor in real time.Its
In, behavior to be identified can refer to the active state of user, such as, walk, run, being seated by or stand;Use can also be referred to
Scene residing for family, such as outdoor, indoors or in public places etc.;Field residing for User Activity state and user can also be referred to
The combination of scape, for example stand in outdoor or be seated indoors.
S102:The signal that sensed data generates is segmented, generates block signal.
Specifically, each segment signal in block signal corresponds to a kind of motor behavior of user, such as, if by sensed data
The signal of generation is walked including user and runs two kinds of behaviors to be identified with user, then block signal includes two segment signals, one section of correspondence
The behavior that user walks, the behavior that one section of corresponding user runs.The embodiment of the present invention is divided by the signal for generating sensed data
Section, it is achieved thereby that in behavior to be identified motor message automatic segmentation, by different types of row to be identified in sensed data
It is separated, the sensed data of behavior to be identified is divided into different types of data subsequence.Wherein it is possible to by collection
After sensed data is filtered, the signal of sensed data generation is obtained.
It is specifically as follows specifically, the signal that the embodiment of the present invention generates sensed data carries out segmentation:Using based on height
The statistical analysis of this model is segmented, and produces the block signal changed by average and variance;Or calculated using sliding window auto-correlation
The sequential processing method of method is segmented, and produces the block signal by frequency and changes in amplitude;Or while using being based on Gauss
The statistical analysis of model and the sequential processing method of sliding window auto-correlation algorithm are segmented, and average, variance, frequency and width are pressed in generation
Spend the block signal of change.Wherein, the sequential processing method of the statistical analysis based on Gauss model and sliding window auto-correlation algorithm has
The segmentation method of body refers to the description of Examples hereinafter.
S103:Choose at least one section of block signal and carry out Activity recognition.
Specifically, the service that intelligent terminal of the embodiment of the present invention or wearable device provide the user according to their needs,
One or more block signals can be chosen to be identified.For example intelligent terminal or wearable device need to provide the user
Meter step service when user walks, then only need to choose block signal identification when corresponding user walks.That is, the embodiment of the present invention
Its type of sports is identified in corresponding type of sports data subsequence in block signal caused by need to being only segmented to S102.
Wherein, the embodiment of the present invention can apply feature extraction and sorting technique to carry out Activity recognition to each section of block signal, use
The realization principle that feature extraction and sorting technique carry out Activity recognition is same as the prior art, and the embodiment of the present invention is no longer superfluous herein
State.
Activity recognition method provided in an embodiment of the present invention, by gathering the sensed data of behavior to be identified, number will be sensed
It is segmented according to the signal of generation, generates block signal, is chosen at least one section of block signal and carry out Activity recognition.The present invention is implemented
Example is segmented by the signal for generating sensed data, and each segment signal in block signal corresponds to a kind of motion row of user
For, it is achieved thereby that in behavior to be identified motor message automatic segmentation, by different types of row to be identified in sensed data
It is separated, need to only its behavior type be identified to a certain section of block signal caused by segmentation, avoids processing frequently
Classification and Activity recognition computing, reduce frequently automatic identification computing overhead, recognition accuracy is costly and time consuming short.
Further, in the above-described embodiments, the signal that sensed data generates is segmented, generates block signal, bag
Include:Generate reference window and testing window corresponding to the signal of sensed data generation;Testing window slides into signal end from signal top,
When reference window and the signal autocorrelation difference of testing window are more than given threshold, produce segmentation and change point, generate and change in segmentation
First block signal of point segmentation, the first block signal are the block signals by frequency and changes in amplitude determination;By the first segmentation
Signal is as block signal.
Specifically, the embodiment of the present invention is segmented using the sequential processing method of sliding window auto-correlation algorithm, Fig. 2 is this
The reference window and the schematic diagram of normalized autocorrelation first quartile before the first zero crossing of testing window that inventive embodiments provide, such as
Shown in Fig. 2, sliding window auto-correlation algorithm is used for the reference window before measurement signal frequency and amplitude change and the testing window after change
Normalized autocorrelation difference.Wherein, reference window is fixed that testing window is to slide, and the window length of testing window answers long enough anti-
The most slow frequency of signal is reflected, while the expectation segmentation of minimum should be shorter than, wherein, minimum expectation segmentation refers to separated section
In most short segment length, when avoiding testing window length and being more than most short segment length, can't detect the block signal that most short section is grown, it is ensured that can
To detect the block signal of all segment lengths.Auto-correlation difference exceedes the threshold value of setting, new segmentation between reference window and testing window
Produce, reference window is moved to current section boundaries.Wherein, default threshold value be according to training data application sectionalization test,
What the accuracy of adjustment threshold value comparison result was drawn.Generally, the boundary point for being segmented the testing window of generation is not to change point
Perfect estimation, but it can be used for positioning to change initial position of point, in testing window, change determination a little by sliding testing window
Calculate the auto-correlation difference of reference window and testing window maximum.
Specifically, sliding window auto-correlation algorithm of the embodiment of the present invention is to use formulaCalculate
The signal autocorrelation difference of reference window and testing window.
Wherein, ATHR and FTHR is the Magnitude Difference and the respective threshold value of frequency-splitting of setting;Magnitude Difference ADIFF is by public affairs
FormulaIt is calculated, p (0)RIt is that testing window slides into signal top
When reference window autocorrelation value, p (0)TIt is the autocorrelation value of testing window when testing window slides into signal top;Frequency-splitting
FDIFF is calculated by formula F DIFF=B/C, and B is the difference section of reference window and testing window autocorrelation signal, and C is reference window
With the same section of testing window autocorrelation signal.
Wherein, the signal autocorrelation p (h) of reference window and testing window is by formulaCalculate, h is
Time interval in reference window and testing window auto-correlation computation, during h=0, testing window is located at signal top.I is reference window or survey
Try the index of the sequence of points in window window;xiIt is the sequence of points in reference window or testing window window;T is the window of reference window or testing window
It is long.
The embodiment of the present invention is segmented by using the sequential processing method of sliding window auto-correlation algorithm, in reference window and survey
When auto-correlation difference exceedes the threshold value of setting between examination window, new segmentation is produced, realizes and is segmented by the first of frequency and changes in amplitude
Signal.
Further, in the above-described embodiments, the signal that sensed data generates is segmented, generates block signal, bag
Include:Generate comparison function corresponding to sensed data;Minimize comparison function generate the second block signal, the second block signal be by
Average and the block signal of variance change.
Specifically, the embodiment of the present invention is segmented using the statistical analysis technique based on Gauss model, based on Gauss
The statistical analysis technique of model is the segments M according to calculating, is calculated by the comparison of the Gauss model generation of different mean variances
Function, this comparison function is minimized to produce the first block signal by average and variance change.
Specifically, comparison function corresponding to generation sensed data, including:
Using formula Xi=μi+σiεiGeneration observation data Xi;
Using formulaWith Calculating observation data XiWith default Gauss
Difference J (i, x) between model;
Using formula H (i)=J (i, x)+β M (i) andGenerate comparison function H (i);
Wherein, i is the index of observational variable, μiAnd σiIt is default Gauss model of the observational variable in default segmentation respectively
Average and standard deviation, εiIt is stochastic variable of the observational variable in default segmentation;M (i) is segmentation Number Sequence, and M (i) is by indexing i's
Dimension K (i) is determined;K (i) is that comparison function minimizes the change point sequence drawn;M is segmentation Number Sequence M (i) in standardization ratio
Compared with secondary deviation be more than 0.75 maximum;T is the total length for being segmented Number Sequence M (i);J be change point sequence index, j
=1,2 ... M;kjIt is to change j-th of change point, k in point sequencej-1It is to change -1 change point of jth in point sequence;kj-1+1:kj
It is from change point kj-1To next change point kjSequence of points,It is to index i from change point kj-1To next change point
kjThe average value of observation data corresponding to sequence of points.
The embodiment of the present invention is segmented by using the statistical analysis technique based on Gauss model, is calculated by different averages
The comparison function of the Gauss model generation of variance, minimizes this comparison function, to realize change by average and variance first
Block signal.
Fig. 3 is the flow chart for the Activity recognition method that the embodiment of the present invention two provides, as shown in figure 3, Activity recognition method
Specifically include following steps:
S301:Gather the sensed data of behavior to be identified.
It should be noted that S301 is identical with S101 implementation, S101 description is referred to, here is omitted.
S302:Generate reference window and testing window corresponding to the signal of sensed data generation.
S303:Testing window is slided into signal end from signal top, it is poor in the signal autocorrelation of reference window and testing window
When value is more than given threshold, produces segmentation and change point, generate and changing the first block signal of point segmentation.
Wherein, the first block signal is the block signal by frequency and changes in amplitude determination.
It should be noted that sequential processing methods of the S302 and S303 with using sliding window auto-correlation algorithm in above-described embodiment
The implementation being segmented is identical, and refer to that the above-mentioned sequential processing method using sliding window auto-correlation algorithm is segmented retouches
State, here is omitted.
S304:Comparison function corresponding to sensed data is generated, comparison function is minimized and generates the second block signal.
Wherein, the second block signal is the block signal determined by average and variance change.
It should be noted that S304 in above-described embodiment using the statistical analysis technique based on Gauss model with being segmented
Implementation it is identical, refer to the above-mentioned description being segmented using the statistical analysis technique based on Gauss model, herein no longer
Repeat.
It should be noted that S304 and S302 order can exchange, i.e., can also first carry out 304, then perform S302 and
S303, the embodiment of the present invention are simply illustrated exemplified by first carrying out S302, it is not limited to this.
S305:The first block signal and the second block signal are merged, generates final block signal.
Specifically, because the change that segmentation is concerned only with average and variance is insufficient, fusion of the embodiment of the present invention first
Block signal and the second block signal, generation fusion average, variance, amplitude and the final block signal of frequency change, using frequency
The sliding window auto-correlation algorithm of rate and changes in amplitude is segmented to correct caused by average and variance change, to determine rational segmentation knot
Fruit.The embodiment of the present invention can observe more effective data statistics by merging the first block signal and the second block signal
Characteristic, such as average, variance, amplitude, frequency, the change of these signal attributes has effectively divided different motion class in motor message
Type data subsequence, it is achieved thereby that the automatic segmentation of motor message.
Specifically, the first block signal of fusion and the second block signal can be:First block signal and second are segmented
Signal is compared with gold criterion, retain and be superimposed the first block signal and the second block signal in it is closest with gold criterion
Signal.Wherein, gold criterion is its main definitions different motion class well known to a person skilled in the art a formula of criteria
The standard type of type signal, the embodiment of the present invention is herein without repeating.
S306:Choose at least one section final block signal and carry out Activity recognition.
Specifically, S306 is identical with S101 implementation, S101 description is referred to, here is omitted.With S101 only
The final block signal of the embodiment of the present invention unlike one is to merge average, variance, amplitude and the block signal of frequency shift.
Activity recognition method provided in an embodiment of the present invention, by gathering the sensed data of behavior to be identified, using sliding window
Sequential processing method segmentation the first block signal of generation of auto-correlation algorithm, using the statistical analysis technique based on Gauss model point
The block signals of Duan Shengcheng second, the first block signal and the second block signal are merged, generate final block signal, choose at least one
The final block signal of section carries out Activity recognition.The embodiment of the present invention is segmented by the signal for generating sensed data, segmentation
Each segment signal in signal corresponds to a kind of motor behavior of user, it is achieved thereby that in behavior to be identified motor message it is automatic
Segmentation, different types of behavior to be identified in sensed data is separated, and only a certain section of block signal caused by segmentation need to be entered
Row identifies its behavior type, avoids processing frequently classification and Activity recognition computing, reduces frequently automatic identification
Computing overhead, recognition accuracy are costly and time consuming short.Meanwhile the automatic segmentation algorithm of design of the embodiment of the present invention mainly applies system
Count model and sequential processing method, by observation signal statistical property, without the guiding of additional label data, realize
Non-supervisory automatic segmentation.And the change of average, variance, amplitude and frequency is merged in final block signal so that raw
Into block signal it is more rationally and accurate, so as to improve the accuracy of Activity recognition.
Fig. 4 is the structural representation for the terminal that the embodiment of the present invention one provides, as shown in figure 4, the embodiment of the present invention provides
Terminal include:Acquisition module 41, segmentation module 42 and identification module 43.
Acquisition module 41, for gathering the sensed data of behavior to be identified;Segmentation module 42, for by the sensed data
The signal of generation is segmented, and generates block signal;Identification module 43, enter every trade for choosing at least one section of block signal
For identification.
The terminal provided in the present embodiment is used to performing the technical scheme of embodiment of the method shown in Fig. 1, its realization principle and
Realize that effect is similar, here is omitted.
Further, the segmentation module 42 is used to the signal that the sensed data generates being segmented, generation segmentation
Signal, including:
Generate reference window and testing window corresponding to the signal of the sensed data generation;By the testing window from signal top
Signal end is slided into, when the reference window and the signal autocorrelation difference of the testing window are more than given threshold, produces one
Individual segmentation changes point, generates the first block signal for changing point segmentation in the segmentation, first block signal is by frequency
The block signal determined with changes in amplitude;Using first block signal as the block signal;
Further, the segmentation module 42 is used for:
Generate reference window and testing window corresponding to the signal of the sensed data generation;By the testing window from signal top
Signal end is slided into, when the reference window and the signal autocorrelation difference of the testing window are more than given threshold, produce and divides
Section changes point, generates the first block signal for changing point segmentation in the segmentation, first block signal is by frequency and width
The block signal that degree change determines;
Generate comparison function corresponding to the sensed data;Minimize the comparison function and generate the second block signal, institute
It is the block signal by average and variance change to state the second block signal;
First block signal and second block signal are merged, generates final block signal;
The identification module 43 is used to choose at least one section of block signal and carries out Activity recognition, including:
Choose at least one section final block signal and carry out Activity recognition.
Further, the terminal also includes:Computing module.
Computing module, for using formulaCalculate the reference window and the testing window
Signal autocorrelation difference;
Wherein, ATHR and FTHR is the Magnitude Difference and the respective threshold value of frequency-splitting of setting;Magnitude Difference ADIFF is by public affairs
FormulaIt is calculated, p (0)RIt is that the testing window slides into signal
The autocorrelation value of reference window during top, p (0)TBe the testing window slide into testing window during signal top from
Correlation;Frequency-splitting FDIFF is calculated by formula F DIFF=B/C, and B is the reference window and the testing window auto-correlation
The difference section of signal, C are the same sections of the reference window and the testing window autocorrelation signal.
Further, the segmentation module 42 is used to generate comparison function corresponding to the sensed data, including:
Using formula Xi=μi+σiεiGeneration observation data Xi;
Using formulaWith Calculating observation data XiWith default Gauss
Difference J (i, x) between model;
Using formula H (i)=J (i, x)+β M (i) andGenerate comparison function H (i);
Wherein, i is the index of observational variable, μiAnd σiIt is default Gauss model of the observational variable in default segmentation respectively
Average and standard deviation, εiIt is stochastic variable of the observational variable in default segmentation;M (i) is segmentation Number Sequence, and M (i) is by indexing i's
Dimension K (i) is determined;K (i) is that comparison function minimizes the change point sequence drawn;M is segmentation Number Sequence M (i) in standardization ratio
Compared with secondary deviation be more than 0.75 maximum;T is the total length for being segmented Number Sequence M (i);J be change point sequence index, j
=1,2 ... M;kjIt is to change j-th of change point, k in point sequencej-1It is to change -1 change point of jth in point sequence;kj-1+1:kj
It is from change point kj-1To next change point kjSequence of points,It is to index i from change point kj-1To next change point
kjThe average value of observation data corresponding to sequence of points.
Fig. 5 is the structural representation for the terminal that the embodiment of the present invention two provides, as shown in figure 5, the embodiment of the present invention provides
Terminal include:Memory 51 and processor 52.
Memory 51 is used to store execute instruction, and processor 52 can be a central processing unit (Central
Processing Unit, CPU), or specific integrated circuit (Application Specific Integrated
Circuit, ASIC), or complete to implement one or more integrated circuits of the embodiment of the present invention.When base station is run, processing
Communicated between device 52 and memory 51, the call executive instruction of processor 52, for performing following operation:
Gather the sensed data of behavior to be identified;
The signal that the sensed data generates is segmented, generates block signal;
Choose at least one section of block signal and carry out Activity recognition.
Further, the signal that the sensed data generates is segmented by processor 52, generates block signal, including:
Generate reference window and testing window corresponding to the signal of the sensed data generation;
The testing window is slided into signal end from signal top, in the signal of the reference window and the testing window certainly
When associated differences are more than given threshold, produce segmentation and change point, generate the first block signal for changing point segmentation in the segmentation,
First block signal is the block signal by frequency and changes in amplitude determination;
Using first block signal as the block signal.
Further, processor 52 is additionally operable to:
Generate reference window and testing window corresponding to the signal of the sensed data generation;
The testing window is slided into signal end from signal top, in the signal of the reference window and the testing window certainly
When associated differences are more than given threshold, produce segmentation and change point, generate the first block signal for changing point segmentation in the segmentation,
First block signal is the block signal by frequency and changes in amplitude determination;
Generate comparison function corresponding to the sensed data;Minimize the comparison function and generate the second block signal, institute
It is the block signal by average and variance change to state the second block signal;
First block signal and second block signal are merged, generates final block signal;
Choose at least one section final block signal and carry out Activity recognition.
Further, processor 52 is additionally operable to:
Using formulaThe signal autocorrelation for calculating the reference window and the testing window is poor
Value;
Wherein, ATHR and FTHR is the Magnitude Difference and the respective threshold value of frequency-splitting of setting;Magnitude Difference ADIFF is by public affairs
FormulaIt is calculated, p (0)RIt is that the testing window slides into signal
The autocorrelation value of reference window during top, p (0)TBe the testing window slide into testing window during signal top from
Correlation;Frequency-splitting FDIFF is calculated by formula F DIFF=B/C, and B is the reference window and the testing window auto-correlation
The difference section of signal, C are the same sections of the reference window and the testing window autocorrelation signal.
Further, processor 52 generates comparison function corresponding to the sensed data, including:
Using formula Xi=μi+σiεiGeneration observation data Xi;
Using formulaWith Calculating observation data XiWith default Gauss
Difference J (i, x) between model;
Using formula H (i)=J (i, x)+β M (i) andGenerate comparison function H (i);
Wherein, i is the index of observational variable, μiAnd σiIt is default Gauss model of the observational variable in default segmentation respectively
Average and standard deviation, εiIt is stochastic variable of the observational variable in default segmentation;M (i) is segmentation Number Sequence, and M (i) is by indexing i's
Dimension K (i) is determined;K (i) is that comparison function minimizes the change point sequence drawn;M is segmentation Number Sequence M (i) in standardization ratio
Compared with secondary deviation be more than 0.75 maximum;T is the total length for being segmented Number Sequence M (i);J be change point sequence index, j
=1,2 ... M;kjIt is to change j-th of change point, k in point sequencej-1It is to change -1 change point of jth in point sequence;kj-1+1:kj
It is from change point kj-1To next change point kjSequence of points,It is to index i from change point kj-1To next change point
kjThe average value of observation data corresponding to sequence of points.
Experiment 1:
The smart mobile phone of embedded inertia motion sensor is attached to human body, gathers acceleration signal, generates human behavior
Data set 1:Walk, run, sit, Fig. 6 is the segmentation result that the actual measurement behavioral data collection 1 of the application present invention is walked, run, sitting, and Fig. 7 is this hair
The noise of switch step between the segmentation algorithm separation that bright embodiment provides is driven away and run.As shown in Figure 6 and Figure 7, it is not only different
Motion had successfully formed segmentation, measure the sensor drift of beginning, walk and run between switch step noise and measurement terminate
Noise all had successfully been isolated out.
Experiment 2:
Experiment obtains human behavior data set 2:Walk, sit, run and stand, Fig. 8 is the actual measurement behavioral data collection 2 of the application present invention
The segmentation result walked, sit, run and stood, as shown in figure 8, four kinds of daily behaviors are successfully separated.Experimental result based on segmentation,
Activity recognition is carried out using feature extraction and sorting technique, test result such as subordinate list 1, accuracy rate is more than 98%.
Table 1
Although disclosed herein embodiment as above, described content be only readily appreciate the present invention and use
Embodiment, it is not limited to the present invention.Technical staff in any art of the present invention, taken off not departing from the present invention
On the premise of the spirit and scope of dew, any modification and change, but the present invention can be carried out in the form and details of implementation
Scope of patent protection, still should be subject to the scope of the claims as defined in the appended claims.
Claims (10)
1. a kind of Activity recognition method, including:
Gather the sensed data of behavior to be identified;
The signal that the sensed data generates is segmented, generates block signal;
Choose at least one section of block signal and carry out Activity recognition.
2. Activity recognition method according to claim 1, it is characterised in that the signal for generating the sensed data
It is segmented, generates block signal, including:
Generate reference window and testing window corresponding to the signal of the sensed data generation;
The testing window is slided into signal end from signal top, in the reference window and the signal autocorrelation of the testing window
When difference is more than given threshold, produces segmentation and change point, generate the first block signal for changing point segmentation in the segmentation, it is described
First block signal is the block signal by frequency and changes in amplitude determination;
Using first block signal as the block signal.
3. Activity recognition method according to claim 1, it is characterised in that the signal for generating the sensed data
It is segmented, generates block signal, including:
Generate reference window and testing window corresponding to the signal of the sensed data generation;
The testing window is slided into signal end from signal top, in the reference window and the signal autocorrelation of the testing window
When difference is more than given threshold, produces segmentation and change point, generate the first block signal for changing point segmentation in the segmentation, it is described
First block signal is the block signal by frequency and changes in amplitude determination;
Generate comparison function corresponding to the sensed data;Minimize the comparison function and generate the second block signal, described the
Two-section signal is the block signal determined by average and variance change;
First block signal and second block signal are merged, generates final block signal;
The selection at least one section of block signal carries out Activity recognition, including:Choose at least one section final segmentation letter
Number carry out Activity recognition.
4. the Activity recognition method according to Claims 2 or 3, it is characterised in that the reference window and the testing window
Signal autocorrelation difference is obtained by following steps:
Using formulaCalculate the reference window and the signal autocorrelation difference of the testing window;
Wherein, ATHR and FTHR is the Magnitude Difference and the respective threshold value of frequency-splitting of setting;Magnitude Difference ADIFF is by formulaIt is calculated, p (0)RIt is that the testing window slides into the signal beginning
The autocorrelation value of reference window during end, p (0)TBe the testing window slide into testing window during signal top from phase
Pass is worth;Frequency-splitting FDIFF is calculated by formula F DIFF=B/C, and B is the reference window and testing window auto-correlation letter
Number difference section, C is the same section of the reference window and the testing window autocorrelation signal.
5. Activity recognition method according to claim 3, it is characterised in that compare corresponding to the generation sensed data
Compared with function, including:
Using formula Xi=μi+σiεiGeneration observation data Xi;
Using formulaWith Calculating observation data XiWith default Gauss
Difference J (i, x) between model;
Using formula H (i)=J (i, x)+β M (i) andGenerate comparison function H (i);
Wherein, i is the index of observational variable, μiAnd σiIt is average of the observational variable in the default Gauss model of default segmentation respectively
And standard deviation, εiIt is stochastic variable of the observational variable in default segmentation;M (i) be segmentation Number Sequence, M (i) by index i dimension K
(i) determine;K (i) is that comparison function minimizes the change point sequence drawn;M segmentation Number Sequence M (i) compare in standardization
Secondary deviation is more than 0.75 maximum;T is the total length for being segmented Number Sequence M (i);J be change point sequence index, j=1,
2 ... M;kjIt is to change j-th of change point, k in point sequencej-1It is to change -1 change point of jth in point sequence;kj-1+1:kjBe from
Change point kj-1To next change point kjSequence of points,It is to index i from change point kj-1To next change point kjSequence
The average value of observation data corresponding to row point.
A kind of 6. terminal, it is characterised in that including:
Acquisition module, for gathering the sensed data of behavior to be identified;
Segmentation module, the signal for the sensed data to be generated are segmented, and generate block signal;
Identification module, Activity recognition is carried out for choosing at least one section of block signal.
7. terminal according to claim 6, it is characterised in that the segmentation module is used for generate the sensed data
Signal is segmented, and generates block signal, including:
Generate reference window and testing window corresponding to the signal of the sensed data generation;
The testing window is slided into signal end from signal top, in the reference window and the signal autocorrelation of the testing window
When difference is more than given threshold, produces a segmentation and change point, generate the first block signal for changing point segmentation in the segmentation,
First block signal is the block signal by frequency and changes in amplitude determination;
Using first block signal as the block signal;
Or the segmentation module is used for:
Generate reference window and testing window corresponding to the signal of the sensed data generation;
The testing window is slided into signal end from signal top, in the reference window and the signal autocorrelation of the testing window
When difference is more than given threshold, produces segmentation and change point, generate the first block signal for changing point segmentation in the segmentation, it is described
First block signal is the block signal by frequency and changes in amplitude determination;
Generate comparison function corresponding to the sensed data;Minimize the comparison function and generate the second block signal, described the
Two-section signal is the block signal by average and variance change;
First block signal and second block signal are merged, generates final block signal;
The identification module is used to choose at least one section of block signal and carries out Activity recognition, including:
Choose at least one section final block signal and carry out Activity recognition.
8. terminal according to claim 7, it is characterised in that the terminal also includes:
Computing module, for using formulaCalculate the letter of the reference window and the testing window
Number auto-correlation difference;
Wherein, ATHR and FTHR is the Magnitude Difference and the respective threshold value of frequency-splitting of setting;Magnitude Difference ADIFF is by formulaIt is calculated, p (0)RIt is that the testing window slides into the signal beginning
The autocorrelation value of reference window during end, p (0)TBe the testing window slide into testing window during signal top from phase
Pass is worth;Frequency-splitting FDIFF is calculated by formula F DIFF=B/C, and B is the reference window and testing window auto-correlation letter
Number difference section, C is the same section of the reference window and the testing window autocorrelation signal.
9. terminal according to claim 7, it is characterised in that the segmentation module is corresponding for generating the sensed data
Comparison function, including:
Using formula Xi=μi+σiεiGeneration observation data Xi;
Using formulaWith Calculating observation data XiWith default Gauss
Difference J (i, x) between model;
Using formula H (i)=J (i, x)+β M (i) andGenerate comparison function H (i);
Wherein, i is the index of observational variable, μiAnd σiIt is average of the observational variable in the default Gauss model of default segmentation respectively
And standard deviation, εiIt is stochastic variable of the observational variable in default segmentation;M (i) be segmentation Number Sequence, M (i) by index i dimension K
(i) determine;K (i) is that comparison function minimizes the change point sequence drawn;M segmentation Number Sequence M (i) compare in standardization
Secondary deviation is more than 0.75 maximum;T is the total length for being segmented Number Sequence M (i);J be change point sequence index, j=1,
2 ... M;kjIt is to change j-th of change point, k in point sequencej-1It is to change -1 change point of jth in point sequence;kj-1+1:kjBe from
Change point kj-1To next change point kjSequence of points,It is to index i from change point kj-1To next change point kjSequence
The average value of observation data corresponding to row point.
A kind of 10. terminal, it is characterised in that including:Processor and memory, memory are used to store execute instruction;Processor
The execute instruction is called, for performing the Activity recognition method as described in claim 1-5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711060185.5A CN107862276B (en) | 2017-11-01 | 2017-11-01 | Behavior identification method and terminal |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711060185.5A CN107862276B (en) | 2017-11-01 | 2017-11-01 | Behavior identification method and terminal |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107862276A true CN107862276A (en) | 2018-03-30 |
CN107862276B CN107862276B (en) | 2021-05-18 |
Family
ID=61696620
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711060185.5A Active CN107862276B (en) | 2017-11-01 | 2017-11-01 | Behavior identification method and terminal |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107862276B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109151593A (en) * | 2018-09-30 | 2019-01-04 | 广州酷狗计算机科技有限公司 | Main broadcaster's recommended method, device storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104268577A (en) * | 2014-06-27 | 2015-01-07 | 大连理工大学 | Human body behavior identification method based on inertial sensor |
CN106339071A (en) * | 2015-07-08 | 2017-01-18 | 中兴通讯股份有限公司 | Method and device for identifying behaviors |
CN106503673A (en) * | 2016-11-03 | 2017-03-15 | 北京文安智能技术股份有限公司 | A kind of recognition methodss of traffic driving behavior, device and a kind of video acquisition device |
-
2017
- 2017-11-01 CN CN201711060185.5A patent/CN107862276B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104268577A (en) * | 2014-06-27 | 2015-01-07 | 大连理工大学 | Human body behavior identification method based on inertial sensor |
CN106339071A (en) * | 2015-07-08 | 2017-01-18 | 中兴通讯股份有限公司 | Method and device for identifying behaviors |
CN106503673A (en) * | 2016-11-03 | 2017-03-15 | 北京文安智能技术股份有限公司 | A kind of recognition methodss of traffic driving behavior, device and a kind of video acquisition device |
Non-Patent Citations (2)
Title |
---|
DANIELA SCHULZ ET AL.: "The sliding window correlation procedure for detecting hidden correlations: existence of behavioral subgroups illustrated with aged rats", 《JOURNAL OF NEUROSCIENCE METHODS》 * |
杨伟笃: "基于可穿戴设备的人体行为识别与状态监测方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109151593A (en) * | 2018-09-30 | 2019-01-04 | 广州酷狗计算机科技有限公司 | Main broadcaster's recommended method, device storage medium |
CN109151593B (en) * | 2018-09-30 | 2021-07-02 | 广州酷狗计算机科技有限公司 | Anchor recommendation method, device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN107862276B (en) | 2021-05-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Brodu et al. | Exploring two novel features for EEG-based brain–computer interfaces: Multifractal cumulants and predictive complexity | |
CN103717124B (en) | For obtaining and process the apparatus and method for of the survey measurements of biology | |
CN107886503A (en) | A kind of alimentary canal anatomical position recognition methods and device | |
US20210315486A1 (en) | System and Method for Automatic Evaluation of Gait Using Single or Multi-Camera Recordings | |
EP4071765A1 (en) | Method and system for drawing brain functional atlas | |
Yong et al. | Robust common spatial patterns for EEG signal preprocessing | |
Jensen et al. | Classification of kinematic swimming data with emphasis on resource consumption | |
CN110503082B (en) | Model training method based on deep learning and related device | |
CN106971130A (en) | A kind of gesture identification method using face as reference | |
CN109766796A (en) | A kind of depth pedestrian detection method towards dense population | |
Al-Ghannam et al. | Prayer activity monitoring and recognition using acceleration features with mobile phone | |
CN112115803A (en) | Mask state reminding method and device and mobile terminal | |
CN110210483A (en) | Medical image lesion region dividing method, model training method and equipment | |
KR101999271B1 (en) | Apparatus and method for determining of optimal eeg channel based on pso | |
Wissel et al. | Considerations on strategies to improve EOG signal analysis | |
Karas et al. | Adaptive empirical pattern transformation (ADEPT) with application to walking stride segmentation | |
CN111671420A (en) | Method for extracting features from resting electroencephalogram data and terminal equipment | |
CN114943924B (en) | Pain assessment method, system, equipment and medium based on facial expression video | |
Mehmood et al. | Hardensenet: A 1d densenet inspired convolutional neural network for human activity recognition with inertial sensors | |
US20150206064A1 (en) | Method for supervised machine learning | |
CN108985839A (en) | Shopping guide method and device in unmanned supermarket based on recognition of face | |
CN106570479B (en) | A kind of pet motions recognition methods of Embedded platform | |
CN107862276A (en) | A kind of Activity recognition method and terminal | |
Yang et al. | Automatic detection pipeline for accessing the motor severity of Parkinson’s disease in finger tapping and postural stability | |
CN113558602B (en) | Hypothesis-driven cognitive impairment brain network analysis method |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |