CN110110674A - A kind of gesture identification method based on acceleration MEMS and basic stroke - Google Patents
A kind of gesture identification method based on acceleration MEMS and basic stroke Download PDFInfo
- Publication number
- CN110110674A CN110110674A CN201910391987.7A CN201910391987A CN110110674A CN 110110674 A CN110110674 A CN 110110674A CN 201910391987 A CN201910391987 A CN 201910391987A CN 110110674 A CN110110674 A CN 110110674A
- Authority
- CN
- China
- Prior art keywords
- gesture
- code
- moment
- information
- feature
- 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
- 238000000034 method Methods 0.000 title claims abstract description 52
- 230000001133 acceleration Effects 0.000 title claims abstract description 40
- 238000007514 turning Methods 0.000 claims abstract description 59
- 230000033001 locomotion Effects 0.000 claims abstract description 14
- 230000011218 segmentation Effects 0.000 claims abstract description 10
- 230000003068 static effect Effects 0.000 claims description 20
- 238000011156 evaluation Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 13
- 230000009467 reduction Effects 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 239000012634 fragment Substances 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 7
- 238000013519 translation Methods 0.000 description 6
- 238000011088 calibration curve Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000036544 posture Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/28—Recognition of hand or arm movements, e.g. recognition of deaf sign language
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/30—Writer recognition; Reading and verifying signatures
- G06V40/37—Writer recognition; Reading and verifying signatures based only on signature signals such as velocity or pressure, e.g. dynamic signature recognition
- G06V40/382—Preprocessing; Feature extraction
- G06V40/388—Sampling; Contour coding; Stroke extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/30—Writer recognition; Reading and verifying signatures
- G06V40/37—Writer recognition; Reading and verifying signatures based only on signature signals such as velocity or pressure, e.g. dynamic signature recognition
- G06V40/394—Matching; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Health & Medical Sciences (AREA)
- Character Discrimination (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of gesture identification method based on acceleration MEMS and basic stroke, comprising the following steps: according to the motion profile of gesture, convert gesture to the sequence formed with label, establish the code identification model based on code feature；The target gesture of established standards and its corresponding first code；Acquire hand signal, by hand signal be sent into code identification model in, obtain sequence label, using it is more when decision making algorithm, obtain the basic stroke of gesture；It presorts to the meaning of gesture, obtains the similarity between gesture and the first code for possessing identical turning code number of features, take result of the corresponding gesture of first code for possessing maximum similarity as gesture identification.The present invention describes the relationship between the stroke of gesture, first concepts such as code and code feature, and a whole set of complete, gesture identification method based on code feature is formd, by turning code feature, realize the automatic segmentation of gesture during gesture identification, algorithm is simple, and accuracy of identification is high.
Description
Technical field
It is the present invention relates to a kind of gesture identification method, in particular to a kind of based on acceleration MEMS and basic stroke
Gesture identification method.
Background technique
Gesture identification is the process identified to human body gesture, in fields such as medical monitoring, the Web-based instruction, electronic equipments
It is with a wide range of applications.Moreover, gesture is as a kind of natural, intuitive, convenient man-machine interaction mode, also greatly letter
Change human-computer interaction process, meanwhile, the diversification of man-machine interaction mode also promotes the development of gesture information collector.Currently,
Gesture information collector for gesture identification research is broadly divided into: 1) picture pick-up device；Gesture identification system based on picture pick-up device
System carries out Image Acquisition to hand and ambient enviroment, identifies gesture wherein included.2) wearable device, it by built-in or
The mode that person integrates by collector fitting, is fixed on hand, acquires the movement or biological information of hand, complete the knowledge of gesture
Not.3) novel information acquisition mode, the information of such acquisition include WIFI, radar etc., are current state-of-the art gesture identification sides
Method.
The application background of gesture identification directly determines the selection of gesture collector.For example, the placement of picture pick-up device can be serious
Influence user experience, while being also difficult to meet the requirement of real-time tracking；And the gesture identification based on novel information acquisition mode
Method is not yet mature, and the gesture of identification is also universal simple；When carrying out single object tracking gesture identification, they are unable to satisfy
It is required that.
Summary of the invention
That in order to solve the above technical problem, the present invention provides a kind of algorithms is simple, accuracy of identification is high micro- based on acceleration
The gesture identification method of Mechatronic Systems and basic stroke.
Technical proposal that the invention solves the above-mentioned problems is: a kind of hand based on acceleration MEMS and basic stroke
Gesture recognition methods, comprising the following steps:
Step 1: the identification model of training standard code realizes the one-to-one correspondence of gesture path and standard code；And gesture is turned
Turn to standard code sequence；
Step 2: sample gesture and its corresponding first code are determined；
Step 3: decision making algorithm when utilizing more obtains the basic stroke of gesture；
Step 4: firstly, presorting according to basic stroke number opponent gesture meaning；Then, calculate gesture to be measured with
Possess the similarity between same basic stroke number purpose sample code, takes the corresponding sample hand of the first code for possessing maximum similarity
Result of the gesture as gesture identification.
The above-mentioned gesture identification method based on acceleration MEMS and basic stroke, the step 1 specific steps
Are as follows:
1-1) according to the actual path of gesture, gesture is divided into 4 classes and label are as follows: turning, static state, arc, straight line；Turning table
Show that gesture path tour, theoretical duration are 0, is denoted as 4, embodies the gesture of beginning and the knot segmentation and to(for) gesture
Beam；Static representations gesture remains static, and the forward and backward hand of corresponding gesture is in the motionless period, is denoted as 3, is convenient for hand
The beginning and end of gesture identifying system identification gesture；Arc indicates that gesture path is the period of arcuation, is denoted as 2, is gesture identification
The foundation of system resolution gesture；Straight line indicates that gesture path is the linear period, is denoted as 1, is that gesture recognition system is differentiated
The foundation of gesture；It is divided into the above four classes label in each moment gesture, then each gesture acquired is converted into above four class
The sequence of label composition；Theoretically, the label of gesture at different times, obtains according to motion profile, to complete from gesture
For sequence to the transformation of sequence label, this mode is one-to-one as coding, therefore tag along sort is denoted as standard code,
Correspondingly, by the gesture information at each moment classify and label process, be denoted as standard code identification；
The raw information at each moment 1-2) is called time information, the corresponding standard code information at each moment is called
Standard code information；In standard code identification process, standard code information, uses for reference speed pass corresponding with curve movement in order to obtain
Each moment and information at the time of N number of moment below are carried out a variety of conversions, have obtained transitional information, transitional information includes by system
Angle (Angle) and variance (Var), in which:
In formula,Angle information between at the time of indicating moment t and be spaced kT, k value are integer；T adopts for information
The period of collection, T=0.02s；AtIt is the acceleration of t moment；At+kTIt is the acceleration at t+kT moment；π is constant；
AngletThe set of the angle information of expression moment t, element representation moment t therein and interval kT (k=1,2 ...
N the angle information between at the time of)；
In formula,Covariance information between at the time of indicating moment t and be spaced kT；
VartIndicate the set of the covariance information of moment t, element representation moment t therein and interval kT (k=1,2 ... N)
At the time of between covariance information；
When being in the stationary state of standard at certain moment, without velocity and acceleration, therefore, the value of angle and covariance information
It is stable near 0, when be in standard turning state when, the variation of speed is very big, thus acceleration information variation it is obvious that
So the mean value and variance yields for angle and covariance information are very big；This theoretically, by be arranged threshold value, identify static
With two class code feature of turning；
In moment t, N number of time interval is considered, obtain the information aggregate about feature FIt is right
In feature F, the mean value and variance of set measure acceleration and velocity variations in moment t, consider in different features, L, N, threshold
In the case where value, L is turning range, and the feature with highest evaluation of estimate, L, N and threshold value is selected to construct filter function:
The filter function of " turning " standard code are as follows:
VAR(Ft)/E(Ft)≥ε+λv
In formula, E (Ft) it is set (Ft) mean value, VAR (Ft) it is set (Ft) variance, ε is constant；λ is speed sensitive
Coefficient；V is gesture speed；
The filter function of " static state " standard code are as follows:
In formula, α, β are constant；
In order to identify " turning ", " straight line " standard code using filter function, selectAndAs evaluation parameter, choosing
Select feature, L, N, threshold value with highest evaluation of estimate；
For the evaluation parameter of gesture initial time, calculated result indicates that the reliability of algorithm is better closer to 1；Wherein
t1The initial time of ' the gesture strokes identified for algorithm；t1For the theoretical initial time of gesture strokes；t0When being originated for gesture
It carves；
For the evaluation parameter of gesture strokes length, calculated result indicates that the reliability of algorithm is better closer to 1；Its
Middle leniFor the length of i-th of stroke of gesture, Leni' it is LeniThe middle correct occasion length of identification, i are gesture strokes serial number；
Information at the time of for that cannot be identified using angle and variance is carried out two classification using more SVM, obtains straight line
Or the degree of membership of arc code feature；
When carrying out two classification using SVM, key is to map in sample, is obtained by the method training of statistical learning
Two classification of one complete paired data of optimal hyperlane B, in training optimal hyperlane, constraint condition and objective function such as formula:
In formula, w is hyperplane method vector；B is that hyperplane is displaced item；C is the penalty factor for controlling marginal classification mistake, C
The sample points of more big then misclassification are fewer, δjFor non-negative slack variable, Nu is the number of data point to be sorted；Work as utilization
When SVM carries out standard code identification, xjIndicate j-th point of time, yiValue indicate j-th point of standard code, value n, n=
1,2；
When carrying out two classification using SVM, the set expression of the element with identical standard code n is Ln；SVM classifier classification
The element set that obtained standard code is n is Sn, classified using SVM, classification results are the reliability δ of standard code nn:
δnFor measuring the reliability of svm classifier, at the time of it indicates that the standard code result for classification is " n ", there are δn
Probabilistic standard code be n, while having 1- δnProbability be 3-n.
The above-mentioned gesture identification method based on acceleration MEMS and basic stroke, in the step 2, standard
Target gesture is number from 0 to 9, corresponding member code are as follows: 0 corresponding yuan of code is (3,4,1,4,1,4,1,4,1,4,3), and 1 is corresponding
First code be (3,4,1,4,3), 2 corresponding yuan of codes are (3,4,2,4,1,4,1,4,3), 3 corresponding yuan of codes be (3,4,2,4,2,
4,3), 4 corresponding yuan of codes are (3,4,1,4,1,4,1,4,3), and 5 corresponding yuan of codes are (3,4,1,4,1,4,2,4,3), and 6 is corresponding
First code be (3,4,2,4,2,4,3), 7 corresponding yuan of codes are (3,4,1,4,1,4,3), 8 corresponding yuan of codes be (3,4,2,4,1,
4,2,4,1,4,3), 9 corresponding yuan of codes are (3,4,2,4,1,4,3).
The above-mentioned gesture identification method based on acceleration MEMS and basic stroke, in the step 3, gesture
The identification of basic stroke the following steps are included:
I) it deletes in sequence label, at the time of all yards of features are static；
II) by the code feature at turning code feature l moment of two sides, it is assigned a value of turning, prevents the single turning period from being known
Not Wei multiple segmentations turning；
III) turning code feature reduction, one turning in all continuous turnings is substituted, reduction sequence is obtained；
IV) to the sequence between adjacent corners code feature, decision when carrying out more: reference template is expressed asWhat Re was indicated is two turnings code features and its intermediate length M standard code sequence fragment,
Wherein,When indicating to be identified by code feature, the degree of membership that moment t belongs to straight line is obtained using SVM, while moment t belongs to arc
Degree of membership isBy it is more when decision, obtain the period stroke classification be straight line degree of membership:
In formula, i indicates serial number of the stroke in gesture sequence,Indicate that i-th of stroke of gesture belongs to being subordinate to for straight line
Degree,Indicate that moment t carves the degree of membership for belonging to straight line, the value of t is 1-M.
The above-mentioned gesture identification method based on acceleration MEMS and basic stroke, in the step 4, when more
After decision, the data in gesture sequence are made of turning and straight line degree of membership, wherein straight line degree of membership is higher, the period
A possibility that belonging to straight line stroke is higher, and otherwise, a possibility that arc stroke is higher；In this case, turning code feature is utilized
Number carry out decision, to presort to the meaning of gesture,For gesture sequence
Reduction, whereinIndicate i-th1A stroke belongs to the degree of membership of " straight line " code feature；Indicate sample
This gesture member code, whereinIndicate i-th2The standard code of a stroke, its value include: 1 (straight line), 2 (arcs)；i1、i2For hand
The number of stroke in gesture, if meeting equation i1=i2, then the similarity between two sequences is as follows:
In formula,Indicate the similarity between gesture R and first code G,Indicate that i-th of stroke belongs to straight line
Degree of membership；Consider several sample gesture member codes, forms setIf sample g meets equation:
Then, the corresponding gesture of sample g is the gesture identification result of gesture to be measured.
The beneficial effects of the present invention are: the present invention according to the motion profile of gesture, converts gesture to and uses label first
The sequence of composition establishes the code identification model based on code feature；Then the target gesture of established standards and its corresponding first code；
Acquire hand signal again, by hand signal be sent into step 1 code identification model in, obtain sequence label, using it is more when decision
Algorithm obtains the basic stroke of gesture；It finally presorts to the meaning of gesture, obtain gesture and possesses identical turning Ma Te
The similarity between first code of number is levied, result of the corresponding gesture of first code for possessing maximum similarity as gesture identification is taken.
Entire method describes the relationship between the stroke of gesture, first concepts such as code and code feature, and form a whole set of is complete,
The automatic segmentation of gesture during gesture identification is realized, is had by turning code feature based on the gesture identification method of code feature
Have the advantages that algorithm is simple, accuracy of identification is high.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Fig. 2 is the flow chart of code feature identification in code identification model of the invention.
Fig. 3 is the flow chart of svm classifier of the invention.
Fig. 4 is the decision diagram of the number of the invention using turning code feature.
Fig. 5 is the structural block diagram of gesture acquisition system in the embodiment of the present invention.
Fig. 6 is gesture path figure in the embodiment of the present invention.
Fig. 7 is original acceleration waveform diagram in the embodiment of the present invention.
Fig. 8 is normalized Acceleration pulse figure in the embodiment of the present invention.
Fig. 9 is code signature waveform figure in the embodiment of the present invention.
Specific embodiment
The present invention is further illustrated with reference to the accompanying drawings and examples.
As shown in Figure 1, a kind of gesture identification method based on acceleration MEMS and basic stroke, including following step
It is rapid:
Step 1: it according to the motion profile of gesture, converts gesture to the sequence formed with label, establishes and be based on code feature
Code identification model.Specific steps are as follows:
1-1) dimensionless is the concept in calculus, i.e., with number 0 be the limit and be infinitely close to 0 variable.In micro- product
During point, the area of curve, is usually all that area is carried out to segmentation summation in order to obtain, and last section length approach is infinitely small
When result be exactly curve area.But in calculus calculating process, as long as the sector boss angle value of variable is small to certain in fact
Degree, the result integrated just have very high precision.This thought is used for reference, speed is that accelerating curve and time shaft are constituted
Area under the curve, during gesture identification, it is ensured that information collection frequency is sufficiently high, so that it may straight using the acceleration of any time
It is reversed should moment velocity information.Certainly, the velocity information of single point in time not can be carried out the identification of gesture, when only combining this
The velocity information with n (n > 1) moment below is carved, the gesture at the moment can be classified, obtain the classification of the moment gesture.
Therefore, according to the actual path of gesture, gesture is divided into 4 classes and label are as follows: turning, static state, arc, straight line；Turning
It indicates that gesture path tour, theoretical duration are 0, is denoted as 4, embodies the gesture of beginning and the knot segmentation and to(for) gesture
Beam；Static representations gesture remains static, and the forward and backward hand of corresponding gesture is in the motionless period, is denoted as 3, is convenient for hand
The beginning and end of gesture identifying system identification gesture；Arc indicates that gesture path is the period of arcuation, is denoted as 2, is gesture identification
The foundation of system resolution gesture；Straight line indicates that gesture path is the linear period, is denoted as 1, is that gesture recognition system is differentiated
The foundation of gesture；It is divided into the above four classes label in each moment gesture, then each gesture acquired is converted into above four class
The sequence of label composition；Theoretically, the label of gesture at different times, obtains according to motion profile, to complete from gesture
For sequence to the transformation of sequence label, this mode is one-to-one as coding, therefore tag along sort is denoted as a yard feature,
Correspondingly, the gesture information at each moment classify and the process of label, it is denoted as the identification of yard feature；
Table 1
1-2) moment is a bit on time shaft, refers to certain in a flash, not only without size, but also does not have direction.Code feature is known
Other purpose exactly obtains corresponding code feature of each moment.It, for convenience of description, will be every in code feature identification process
The raw information at a moment is called time information, and the corresponding code characteristic information at each moment is called a yard characteristic information.Such as Fig. 2
Shown, in code feature identification process, code characteristic information, uses for reference the corresponding relationship of speed and curve movement in order to obtain, will be every
A moment carries out a variety of conversions with information at the time of N number of moment below, has obtained transitional information, transitional information includes angle
(Angle) and variance (Var), in which:
In formula,Angle information between at the time of indicating moment t and be spaced kT, k value are integer；T adopts for information
The period of collection, T=0.02s；AtIt is the acceleration of t moment；At+kTIt is the acceleration at t+kT moment；π is constant；
AngletThe set of the angle information of expression moment t, element representation moment t therein and interval kT (k=1,2 ...
N the angle information between at the time of)；
In formula,Covariance information between at the time of indicating moment t and be spaced kT；
VartIndicate the set of the covariance information of moment t, element representation moment t therein and interval kT (k=1,2 ... N)
At the time of between covariance information.
Theoretically, when being in the stationary state of standard at certain moment, without velocity and acceleration, therefore, angle and side
The value of poor information is stable near 0, and when the turning state for the standard of being in, the variation of speed is very big, therefore acceleration information
Variation is it is obvious that so mean value and variance yields for angle and covariance information are very big；This theoretically, pass through setting threshold
Value identifies static and two class code feature of turning.
In moment t, N number of time interval is considered, obtain the information aggregate about feature FIt is right
In feature F, the mean value and variance of set measure acceleration and velocity variations in moment t, consider in different features, L, N, threshold
In the case where value, L is turning range, and the feature with highest evaluation of estimate, L, N and threshold value is selected to construct filter function:
The filter function of " turning " standard code are as follows:
VAR(Ft)/E(Ft)≥ε+λv
In formula, E (Ft) it is set (Ft) mean value, VAR (Ft) it is set (Ft) variance, ε is constant；λ is speed sensitive
Coefficient；V is gesture speed；
The filter function of " static state " standard code are as follows:
In formula, α, β are constant.
In order to identify " turning ", " straight line " standard code using filter function, selectAndAs evaluation parameter, choosing
Select feature, L, N, threshold value with highest evaluation of estimate；
For the evaluation parameter of gesture initial time, calculated result indicates that the reliability of algorithm is better closer to 1；Wherein
t1The initial time of ' the gesture strokes identified for algorithm；t1For the theoretical initial time of gesture strokes；t0When being originated for gesture
It carves；
For the evaluation parameter of gesture strokes length, calculated result indicates that the reliability of algorithm is better closer to 1；Its
Middle leniFor the length of i-th of stroke of gesture, Leni' it is LeniThe middle correct occasion length of identification, i are gesture strokes serial number.
Information at the time of for that cannot be identified using angle and variance carries out two classification using more SVM, such as Fig. 3 institute
Show, obtains the degree of membership of straight line or arc code feature；
When carrying out two classification using SVM, key is to map in sample, is obtained by the method training of statistical learning
Two classification of one complete paired data of optimal hyperlane B, in training optimal hyperlane, constraint condition and objective function such as formula:
In formula, w is hyperplane method vector；B is that hyperplane is displaced item；C is the penalty factor for controlling marginal classification mistake, C
The sample points of more big then misclassification are fewer, δjFor non-negative slack variable, Nu is the number of data point to be sorted；Work as utilization
When SVM carries out standard code identification, xjIndicate j-th point of time, yiValue indicate j-th point of standard code, value n, n=
1,2；
When carrying out two classification using SVM, the set expression of the element with identical standard code n is Ln；SVM classifier classification
The element set that obtained standard code is n is Sn, classified using SVM, classification results are the reliability δ of standard code nn:
δnFor measuring the reliability of svm classifier, at the time of it indicates that the standard code result for classification is " n ", there are δn
Probabilistic standard code be n, while having 1- δnProbability be 3-n.
Step 2: the target gesture of established standards and its corresponding first code.
First code of gesture refers to the classification according to the standard trajectory of gesture in the geometric locus at each moment, obtained length
Spend minimum, most simplified code characteristic sequence.It will be apparent that first code of gesture, also complies with people's recognizing for " stroke " this concept
Know.So first code of gesture, it may also be said to be the basic stroke of standard gesture, it will not follow the personal habits of writer and
Variation.Gesture member code is obtained by the standard trajectory of gesture, includes translation and two processes of reduction.
The translation of gesture member code refers to, according to gesture in the classification of each moment gesture, the gesture path of standard is translated
For the process of code characteristic sequence.The translation of gesture member code is similar to conventional code feature identification.But the translation of gesture member code is not
Specific algorithm or program is needed to identify sequence, it is to obtain according to people for the cognition of standard gesture path
's.
The purpose of the reduction of gesture member code is the influence rejected gesture and write length, pays close attention to the change of the curve category of gesture
Change.For example, standard trajectory is the gesture of straight line, no matter its length no matter how long, the code feature obtained after translation is " straight
How many is a at the time of line ", its gesture member code is all one " straight line ".Theoretically, for identical gesture, writer's
Under different writing styles, they can regard the gesture path for the standard of being different as.But as long as the personal habits of writer not
Will affect the correct of yard feature identify-will not namely obscure straightway, segmental arc, round corner section, then they can possess phase
Same gesture member code.Meanwhile standard gesture path will not influence gesture member after the operation such as rotation, translation and scaling
The result of code.Therefore, the gesture identification method based on first code feature, can effectively avoid personal writing style to gesture identification
Influence.
As shown in table 2, the target gesture of established standards is number from 0 to 9, corresponding member code are as follows: 0 corresponding yuan of code is
(3,4,1,4,1,4,1,4,1,4,3), 1 corresponding yuan of code are (3,4,1,4,3), 2 corresponding yuan of codes be (3,4,2,4,1,4,1,
4,3), 3 corresponding yuan of codes are (3,4,2,4,2,4,3), and 4 corresponding yuan of codes are (3,4,1,4,1,4,1,4,3), 5 corresponding yuan
Code is (3,4,1,4,1,4,2,4,3), and 6 corresponding yuan of codes are (3,4,2,4,2,4,3), 7 corresponding yuan of codes be (3,4,1,4,1,
4,3), 8 corresponding yuan of codes are (3,4,2,4,1,4,2,4,1,4,3), and 9 corresponding yuan of codes are (3,4,2,4,1,4,3).Different hands
The possible first code having the same of gesture, such as gesture 3 and 6；At this point, additional consider that first code duration is distinguished, such as gesture 3
All for tool there are two " arc " standard code, two sections of arc duration of gesture 3 are approximately equal with 6, the previous arc of gesture 6 it is lasting when
Between compared with the latter arc length.
Table 2
| 0 | P、C、L、C、L、C、L、C、L、C、P |
| 1 | P、C、L、C、P |
| 2 | P、C、A、C、L、C、L、C、P |
| 3 | P、C、A、C、A、C、P |
| 4 | P、C、L、C、L、C、L、C、P |
| 5 | P、C、L、C、L、C、A、C、P |
| 6 | P、C、A、C、A、C、P |
| 7 | P、C、L、C、L、C、P |
| 8 | P、C、A、C、L、C、A、C、L、C、P |
| 9 | P、C、A、C、L、C、P |
Step 3: hand signal is sent into the code identification model of step 1, obtains label sequence by acquisition hand signal
Column, using it is more when decision making algorithm, obtain the basic stroke of gesture.
The meaning of gesture, can be under certain background condition, and according to the stroke of gesture, the identification for obtaining basic stroke includes
Following steps:
I it) deletes in sequence label, at the time of all yards of features are static；During gesture identification, static moment section is right
Answer gesture garbage section, practical significance no for the identification of gesture；But gesture can detecte by quiet code feature
The beginning and end moment.
II) by the code feature at turning code feature 1 moment of two sides, it is assigned a value of turning, prevents the single turning period from being known
Not Wei multiple segmentations turning；Turning code feature, corresponds to current time, and gesture classification is just changing or useful section of gesture
Beginning and end；The segmentation to gesture is embodied, gesture sequence is divided into different two classes of front and back.The value of L, will affect
Global code feature accuracy of identification, " turning " discrimination.
III) turning code feature reduction, one turning in all continuous turnings is substituted, reduction sequence is obtained；
IV) to the sequence between adjacent corners code feature, decision when carrying out more: reference template sequence is expressed asWhat F was indicated is two turnings code features and its intermediate length M standard code sequence fragment,
In,When indicating to be identified by code feature, the degree of membership that moment t belongs to straight line is obtained using SVM, while moment t belongs to being subordinate to for arc
Degree isBy it is more when decision, obtain the period stroke classification be straight line degree of membership:
In formula, i is serial number of the stroke in gesture sequence,Indicate that i-th of stroke belongs to the degree of membership of straight line,
Indicate that moment t carves the degree of membership for belonging to straight line, the value of t is 1-M.
Step 4: presorting to the meaning of gesture, obtains gesture and possesses first code of identical turning code number of features
Between similarity, take result of the corresponding gesture of first code for possessing maximum similarity as gesture identification.
When more after decision, the data in gesture sequence are made of turning and straight line degree of membership, wherein straight line degree of membership
Higher, a possibility which belongs to straight line stroke, is higher, and otherwise, a possibility that arc stroke is higher；In this case, sharp
Decision is carried out with the number of turning code feature, to presort to the meaning of gesture, decision diagram is as shown in Figure 4.For the reduction of gesture sequence, whereinIndicate i-th1A stroke belongs to " straight line " code feature
Degree of membership；Indicate sample gesture member code, whereinIndicate i-th2The standard code of a stroke,
Its value includes: 1 (straight line), 2 (arcs)；i1、i2For the number of stroke in gesture, if meeting equation i1=i2, then two sequence
Between similarity it is as follows:
In formula,Indicate the similarity between gesture R and first code G,Indicate that i-th of stroke belongs to straight line
Degree of membership；Consider several sample gesture member codes, forms setIf sample g meets equation:
Then, the corresponding gesture of sample g is the gesture identification result of gesture to be measured.
Embodiment
The information acquisition system that the present invention uses is divided into wired and wireless two class of information collection, they adequately meet
Requirement to portability, their hardware connection are as shown in Figure 5.During the experiment, cable information acquisition system is selected herein
Verify the accuracy of gesture identification method.
The sensor that the present invention uses is nine axis gyroscope postures of Shenzhen Wei Te intelligent technology limited research and development
Sensor module JY901.In gesture acquisition, sensor and ring shape fixture group are integral, are worn on finger.Its ruler
Very little is 15.24mm X 15.24mm X 2mm, this effectively guarantees user experience.JY901 provides four kinds of different measuring range accelerations
± 2g, ± 4g, ± 8g and ± 16g, operating voltage 3.3V-5V, acceleration stability are 0.01g, and attitude measurement stability is
0.01°.Output frequency adjustable range is 0.1hz~200Hz, and serial ports rate class selects 2400,4800,9600,19200,
38400,57600,115200,230400,460800,921600HZ。
In the present invention, the three-dimensional acceleration data output of acquisition is passed into computer.Output frequency selects 50HZ, also
Be collection period be 0.02s.It moves the acceleration measured according to us to be most worth, it is determined that it is ± 2g that acceleration, which measures,.
The meaning and label of gesture are the emphasis of traditional gesture identification method research.Gesture identification side proposed by the present invention
Method can not only obtain the meaning and label of gesture, moreover it is possible to obtain the gesture calibration curve information of gesture different moments.In order to verify simultaneously
The accuracy and gesture meaning of gesture calibration curve information identification, gesture path movement are executed by robot.Also, to every kind of gesture
Motion profile and length are provided, the reference value of gesture calibration curve information is conveniently obtained.When machine executes gesture motion, movement
Speed is 8m/min, and the track of gesture is all reduced to the most common straight line and arc, the track of corresponding gesture and rail
The parameter of mark is as shown in Figure 6.
In gesture data collection process, the beginning and stopping of the acquisition of each gesture, all by sensor bundled software
Switch control.Each gesture is required with following: 1) during gesture acquires, collector uses gesture after opening to execution
To closing between collector after the preceding and execution that uses gesture, hand information all remain stationary state, and the time at 4 seconds or more,
Ensure beginning and the finish time of algorithm identification gesture.2) it in the every section of gesture data acquired, only exists one and uses gesture.
In this experiment, data set includes 10 kinds of gestures (0-9) and two Chinese character gestures, and every kind of gesture is by machine
People executes 50 times according to fixed track, altogether 600 gesture samples.Part of test results of this method on above-mentioned data set
Fig. 7 as Figure 7-9 is original acceleration waveform diagram, and the standard code in each period is by artificially demarcating.Fig. 8 is original adds
Speed is into the waveform diagram crossed after normalizing, it can be seen that having certain separability between various criterion code in figure.Fig. 9 is that code is special
Waveform diagram is levied, wherein at the time of standard code value κ=4, corresponding standard code is " turning "；The corresponding standard code of κ=3 be " static state " when
It carves；At the time of standard 1 < code value κ, indicate that the degree of membership at the moment " straight line " is κ, the degree of membership of corresponding " arc " is 1- κ.Its
Middle table 3 is the confusion matrix of gesture identification accuracy.
Table 3
| Matrix | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |
| 0 | ||||||||||
| 1 | ||||||||||
| 2 | ||||||||||
| 3 | ||||||||||
| 4 | ||||||||||
| 5 | ||||||||||
| 6 | ||||||||||
| 7 | ||||||||||
| 8 | ||||||||||
| 9 |
When carrying out gesture identification using context of methods to verify, the reliability of obtained gesture calibration curve information, the present invention
Following data is proposed to be evaluated.The gesture total duration Theoretically Total Duration (TTD) of theory,
The total gesture duration Calculated Total Duration (CTD) calculated, the code characteristic duration calculated
Calculated Duration for X, X are four class code features (CDX, X can be L, A, C, P), curve different in track
Duration T heoretically Duration for X (TDX).Wherein, the value of CTD/TTD illustrates the party closer to 1
Total gesture duration that method obtains and theoretical gesture duration are closer, and method is more reliable.It is bent for straight line and camber line
Line, CDL/TDL and CDA/TDA illustrate that this method is more reliable closer to 1.For corner features, CDC illustrates the party closer to 0
Method is more reliable.For quiet code feature, it is in static since it actually represents gesture, it is quiet due to the regulation that gesture requires
Only all in the dead section of gesture in code characteristic theory, therefore it is not evaluated.
Claims (5)
1. a kind of gesture identification method based on acceleration MEMS and basic stroke, comprising the following steps:
Step 1: the identification model of training standard code realizes the one-to-one correspondence of gesture path and standard code；And it converts gesture to
Standard code sequence；
Step 2: sample gesture and its corresponding first code are determined；
Step 3: decision making algorithm when utilizing more obtains the basic stroke of gesture；
Step 4: firstly, presorting according to basic stroke number opponent gesture meaning；Then, it calculates gesture to be measured and possesses
Similarity between same basic stroke number purpose sample code takes the corresponding sample gesture of the first code for possessing maximum similarity to make
For the result of gesture identification.
2. the gesture identification method according to claim 1 based on acceleration MEMS and basic stroke, feature
It is, the step 1 specific steps are as follows:
1-1) according to the actual path of gesture, gesture is divided into 4 classes and label are as follows: turning, static state, arc, straight line；Turning indicates hand
Gesture track tour, theoretical duration are 0, are denoted as 4, embody the gesture of beginning and the end segmentation and to(for) gesture；It is quiet
State indicates that gesture remains static, and the forward and backward hand of corresponding gesture is in the motionless period, is denoted as 3, is convenient for gesture identification
The beginning and end of system identification gesture；Arc indicates that gesture path is the period of arcuation, is denoted as 2, is gesture recognition system point
Distinguish the foundation of gesture；Straight line indicates that gesture path is the linear period, is denoted as 1, is that gesture recognition system differentiates gesture
Foundation；It is divided into the above four classes label in each moment gesture, then each gesture acquired is converted into the above four classes set of tags
At sequence；Theoretically, the label of gesture at different times, obtains according to motion profile, thus complete from gesture sequence to
The transformation of sequence label, this mode is one-to-one as coding, therefore tag along sort is denoted as standard code, accordingly
, by the gesture information at each moment classify and label process, be denoted as standard code identification；
The raw information at each moment 1-2) is called time information, the corresponding standard code information at each moment is called standard
Code information；In standard code identification process, standard code information, uses for reference the corresponding relationship of speed and curve movement in order to obtain, will
Each moment carries out a variety of conversions with information at the time of N number of moment below, has obtained transitional information, transitional information includes angle
(Angle) and variance (Var), in which:
In formula,Angle information between at the time of indicating moment t and be spaced kT, k value are integer；T is information collection
Period, T=0.02s；AtIt is the acceleration of t moment；At+kTIt is the acceleration at t+kT moment；π is constant；
AngletIndicate moment t angle information set, element representation moment t therein and interval kT (k=1,2 ... N) when
Angle information between quarter；
In formula,Covariance information between at the time of indicating moment t and be spaced kT；
VartAt the time of indicating the set of the covariance information of moment t, element representation moment t therein and interval kT (k=1,2 ... N)
Between covariance information；
When being in the stationary state of standard at certain moment, without velocity and acceleration, therefore, the value of angle and covariance information is steady
Be scheduled near 0, when be in standard turning state when, the variation of speed is very big, thus acceleration information variation it is obvious that so
It is very big for the mean value and variance yields of angle and covariance information；This theoretically, by the way that threshold value is arranged, identify static and turn
Two class code feature of angle；
In moment t, N number of time interval is considered, obtain the information aggregate F about feature Ft=(Ft 1, Ft 2..., Ft N), for feature
F, the mean value and variance of set measure the acceleration and velocity variations in moment t, consider different features, L, N, threshold value feelings
Under condition, L is turning range, and the feature with highest evaluation of estimate, L, N and threshold value is selected to construct filter function:
The filter function of " turning " standard code are as follows:
VAR(Ft)/E(Ft)≥ε+λv
In formula, E (Ft) it is set (Ft) mean value, VAR (Ft) it is set (Ft) variance, ε is constant；λ is speed sensitive coefficient；
V is gesture speed；
The filter function of " static state " standard code are as follows:
In formula, α, β are constant；
In order to identify " turning ", " straight line " standard code using filter function, selectAndAs evaluation parameter, selection tool
There are feature, L, N, the threshold value of highest evaluation of estimate；
For the evaluation parameter of gesture initial time, calculated result indicates that the reliability of algorithm is better closer to 1；Wherein t1' be
The initial time for the gesture strokes that algorithm identifies；t1For the theoretical initial time of gesture strokes；t0For gesture initial time；
For the evaluation parameter of gesture strokes length, calculated result indicates that the reliability of algorithm is better closer to 1；Wherein leni
For the length of i-th of stroke of gesture, Leni' it is LeniThe middle correct occasion length of identification, i are gesture strokes serial number；
Information at the time of for that cannot be identified using angle and variance, using more SVM carry out two classification, obtain straight line or
The degree of membership of arc code feature；
When carrying out two classification using SVM, key is to map in sample, obtains one by the method training of statistical learning
Two classification of the complete paired data of optimal hyperlane B, in training optimal hyperlane, constraint condition and objective function such as formula:
In formula, w is hyperplane method vector；B is that hyperplane is displaced item；C is the penalty factor for controlling marginal classification mistake, and C is bigger
Then the sample points of misclassification are fewer, δjFor non-negative slack variable, Nu is the number of data point to be sorted；When utilize SVM
When carrying out standard code identification, xjIndicate j-th point of time, yiValue indicate j-th point of standard code, value n, n=1,2；
When carrying out two classification using SVM, the set expression of the element with identical standard code n is Ln；SVM classifier classifies to obtain
Standard code be n element set be Sn, classified using SVM, classification results are the reliability δ of standard code nn:
δnFor measuring the reliability of svm classifier, at the time of it indicates that the standard code result for classification is " n ", there are δnIt is general
Rate standard code is n, while having 1- δnProbability be 3-n.
3. the gesture identification method according to claim 2 based on acceleration MEMS and basic stroke, feature
Be, in the step 2, the target gesture of standard is number from 0 to 9, corresponding member code are as follows: 0 corresponding yuan of code be (3,4,
1,4,1,4,1,4,1,4,3), 1 corresponding yuan of code is (3,4,1,4,3), and 2 corresponding yuan of codes are (3,4,2,4,1,4,1,4,3),
3 corresponding yuan of codes are (3,4,2,4,2,4,3), and 4 corresponding yuan of codes are (3,4,1,4,1,4,1,4,3), and 5 corresponding yuan of codes are
(3,4,1,4,1,4,2,4,3), 6 corresponding yuan of codes are (3,4,2,4,2,4,3), 7 corresponding yuan of codes be (3,4,1,4,1,4,
3), 8 corresponding yuan of codes are (3,4,2,4,1,4,2,4,1,4,3), and 9 corresponding yuan of codes are (3,4,2,4,1,4,3).
4. the gesture identification method according to claim 3 based on acceleration MEMS and basic stroke, feature
Be, in the step 3, the identification of the basic stroke of gesture the following steps are included:
I) it deletes in sequence label, at the time of all yards of features are static；
II) by the code feature at turning code feature l moment of two sides, it is assigned a value of turning, the single turning period is prevented to be identified as
The turning of multiple segmentations；
III) turning code feature reduction, one turning in all continuous turnings is substituted, reduction sequence is obtained；
IV) to the sequence between adjacent corners code feature, decision when carrying out more: reference template is expressed asWhat Re was indicated is two turnings code features and its intermediate length M standard code sequence fragment,
Wherein,When indicating to be identified by code feature, the degree of membership that moment t belongs to straight line is obtained using SVM, while moment t belongs to arc
Degree of membership isBy it is more when decision, obtain the period stroke classification be straight line degree of membership:
In formula, i indicates serial number of the stroke in gesture sequence,Indicate that i-th of stroke of gesture belongs to the degree of membership of straight line,Table
Show that moment t carves the degree of membership for belonging to straight line, the value of t is 1-M.
5. the gesture identification method according to claim 3 based on acceleration MEMS and basic stroke, feature
It is, in the step 4, when more after decision, the data in gesture sequence are made of turning and straight line degree of membership, wherein
A possibility that straight line degree of membership is higher, which belongs to straight line stroke is higher, and otherwise, a possibility that arc stroke is higher；At this
Under kind situation, decision is carried out using the number of turning code feature, to presort to the meaning of gesture,For the reduction of gesture sequence, whereinIndicate i-th1A stroke belongs to " straight line " code feature
Degree of membership；Indicate sample gesture member code, whereinIndicate i-th2The standard code of a stroke,
Its value includes: 1 (straight line), 2 (arcs)；i1、i2For the number of stroke in gesture, if meeting equation i1=i2, then two sequence
Between similarity it is as follows:
In formula,Indicate the similarity between gesture R and first code G,Indicate that i-th of stroke belongs to being subordinate to for straight line
Degree；Consider several sample gesture member codes, forms setIf sample g meets equation:
Then, the corresponding gesture of sample g is the gesture identification result of gesture to be measured.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910391987.7A CN110110674B (en) | 2019-05-13 | 2019-05-13 | Gesture recognition method based on acceleration micro-electromechanical system and basic strokes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910391987.7A CN110110674B (en) | 2019-05-13 | 2019-05-13 | Gesture recognition method based on acceleration micro-electromechanical system and basic strokes |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110110674A true CN110110674A (en) | 2019-08-09 |
| CN110110674B CN110110674B (en) | 2022-12-13 |
Family
ID=67489581
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910391987.7A Active CN110110674B (en) | 2019-05-13 | 2019-05-13 | Gesture recognition method based on acceleration micro-electromechanical system and basic strokes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110110674B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103257711A (en) * | 2013-05-24 | 2013-08-21 | 河南科技大学 | Space gesture input method |
| CN103679213A (en) * | 2013-12-13 | 2014-03-26 | 电子科技大学 | 3D gesture recognition method |
| US20190087654A1 (en) * | 2017-09-15 | 2019-03-21 | Huazhong University Of Science And Technology | Method and system for csi-based fine-grained gesture recognition |
-
2019
- 2019-05-13 CN CN201910391987.7A patent/CN110110674B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103257711A (en) * | 2013-05-24 | 2013-08-21 | 河南科技大学 | Space gesture input method |
| CN103679213A (en) * | 2013-12-13 | 2014-03-26 | 电子科技大学 | 3D gesture recognition method |
| US20190087654A1 (en) * | 2017-09-15 | 2019-03-21 | Huazhong University Of Science And Technology | Method and system for csi-based fine-grained gesture recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110110674B (en) | 2022-12-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Frolova et al. | Most probable longest common subsequence for recognition of gesture character input | |
| Pisharady et al. | Recent methods and databases in vision-based hand gesture recognition: A review | |
| Davis et al. | Recognizing hand gestures | |
| Várkonyi-Kóczy et al. | Human–computer interaction for smart environment applications using fuzzy hand posture and gesture models | |
| US7095401B2 (en) | System and method for gesture interface | |
| Wilson et al. | Learning visual behavior for gesture analysis | |
| Nickel et al. | Visual recognition of pointing gestures for human–robot interaction | |
| CN106354252B (en) | A kind of continuation character gesture track recognition method based on STDW | |
| Xia et al. | Vision-based hand gesture recognition for human-robot collaboration: a survey | |
| Zhang et al. | Single object tracking with fuzzy least squares support vector machine | |
| US20070124703A1 (en) | Command input method using motion recognition device | |
| Yin et al. | Real-time continuous gesture recognition for natural human-computer interaction | |
| CN103488294A (en) | Non-contact gesture control mapping adjustment method based on user interactive habits | |
| CN104616028A (en) | Method for recognizing posture and action of human limbs based on space division study | |
| CN105335711B (en) | Fingertip Detection under a kind of complex environment | |
| CN104038799A (en) | Three-dimensional television-oriented gesture manipulation method | |
| CN111368762A (en) | Robot gesture recognition method based on improved K-means clustering algorithm | |
| CN105426882A (en) | Method for rapidly positioning human eyes in human face image | |
| Zhi et al. | Teaching a robot sign language using vision-based hand gesture recognition | |
| CN110717385A (en) | Dynamic gesture recognition method | |
| Zhang et al. | Robotic control of dynamic and static gesture recognition | |
| Choudhury et al. | A CNN-LSTM based ensemble framework for in-air handwritten Assamese character recognition | |
| Schumacher et al. | Active learning of ensemble classifiers for gesture recognition | |
| KR20120089948A (en) | Real-time gesture recognition using mhi shape information | |
| CN110110674A (en) | A kind of gesture identification method based on acceleration MEMS and basic stroke |
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 | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20190809 Assignee: SUZHOU MEDCOIL HEALTHCARE Co.,Ltd. Assignor: XIANGTAN University Contract record no.: X2024980006092 Denomination of invention: A gesture recognition method based on acceleration microelectromechanical system and basic strokes Granted publication date: 20221213 License type: Common License Record date: 20240523 |