CN108416251A - Efficient human motion recognition method based on quantum genetic algorithm optimization - Google Patents
Efficient human motion recognition method based on quantum genetic algorithm optimization Download PDFInfo
- Publication number
- CN108416251A CN108416251A CN201810014848.8A CN201810014848A CN108416251A CN 108416251 A CN108416251 A CN 108416251A CN 201810014848 A CN201810014848 A CN 201810014848A CN 108416251 A CN108416251 A CN 108416251A
- Authority
- CN
- China
- Prior art keywords
- angle
- human
- dry
- limb
- genetic algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; 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
- G06V40/23—Recognition of whole body movements, e.g. for sport training
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Physiology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Social Psychology (AREA)
- Psychiatry (AREA)
- Genetics & Genomics (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of efficient human motion recognition methods based on quantum genetic algorithm optimization, include the following steps：Within the duration of human action to be identified, M frame human body RGB D image datas are obtained, wherein M is the integer more than or equal to 2；Extraction obtains human joint points coordinate information per frame skeleton data, and the angle of multiple dry angles of limb in the frame skeleton data is extracted according to human joint points coordinate information；The description sample of human action to be identified is generated according to the angle of multiple dry angles of limb in M frame skeleton data；Support vector machines is optimized by quantum genetic algorithm with the supporting vector machine model after being optimized, and classified to description sample by the supporting vector machine model after optimization, human action to be identified to be identified.According to the method for the present invention, human action feature calculation complexity can be reduced, efficiency and the accuracy of human action identification are improved.
Description
Technical field
The present invention relates to mode identification technology, more particularly to a kind of efficient human body based on quantum genetic algorithm optimization
Action identification method.
Background technology
The research of Human bodys' response is in multiple fields such as computer video monitoring, senior health and fitness's monitoring and human-computer interactions
With important scientific meaning.The first purpose of Activity recognition is to provide user behavior relevant information, allow computer system actively
User is assisted to complete their task.Human bodys' response can be used for natural human-computer interaction, motion analysis, virtual reality,
The various fields such as augmented reality, education, video display cartoon making are widely used.
Human bodys' response starts from late nineteen nineties, and the Human bodys' response method of early stage is mainly based upon video sequence
It carries out.Research method can substantially be divided into two classes：The method of view-based access control model and sensor-based side.With Microsoft Kinect
The appearance of equal RGB-D sensor devices, the research of Human bodys' response have new opportunity.Such RGB-D body-sensings sensor can
To be tracked with whole body bone, it is possible to provide the information such as 15-20 human joint points posture, it is at low cost and with good portability,
Condition is provided a great convenience for the research of Human bodys' response.Research human body is carried out with RGB-D sensors both at home and abroad
The research of Human bodys' response.In research method, there are decision tree, bayes method, neighbour for Activity recognition main method
Sampling, fuzzy logic, neural network and support vector machines, Hidden Markov Model and condition random field etc..
Human bodys' response is an extremely challenging task, and human body behavior pattern has multi-parameter, is not easy to describe, and moves
Make the features such as identification is low.Microsoft's Kinect sensor can be used for the position tracking of skeleton artis, but these positions
The characteristics of data have dimensional space big, information content redundancy.These seriously affect the speed and accuracy rate of Human bodys' response.
Traditional support vector machines is being widely used in classification problem because it is simple and efficiently, but how to search and excellent
Change parameter, is still an open problem.According to classical Newtonian mechanics theory, people are only it is to be understood that original state and the drive of object
Power, so that it may to obtain object of which movement situation completely.But human body behavior act is complicated, the object studied is various, this side
Method can not be studied effectively.Therefore, it is the machine Learning Theory relied on that people, which have turned one's attention to data-driven,.Pass through
Statistical law inside data and design feature are recognized and are predicted to human body behavioural characteristic.Support vector machines passes through
Hyperplane is constructed, different classes of data point is kept apart, classification is achieved the purpose that with this.The penalty factor of support vector machines and
The two parameters of kernel functional parameter largely affect the accuracy of classification results, therefore how to search optimized parameter,
It is that support vector machines needs the major issue solved, however is reached at present with traditional support vector machines parameter optimization method
The effect arrived is ideal not enough.
Therefore, at present the efficiency and accuracy rate of the identification of human body behavior act are need to be improved.
Invention content
The present invention is directed to solve one of the technical problem in above-mentioned technology at least to a certain extent.For this purpose, the present invention
Purpose is to propose a kind of efficient human motion recognition method optimized based on quantum genetic algorithm, can improve human action and know
Other efficiency and accuracy.
In order to achieve the above objectives, the efficient human action identification side proposed by the present invention based on quantum genetic algorithm optimization
Method includes the following steps：Within the duration of human action to be identified, M frame human body RGB-D image datas are obtained, wherein M
For the integer more than or equal to 2；Extraction obtains human joint points coordinate information, and according to the human body per frame skeleton data
The angle of multiple dry angles of limb in the body joint point coordinate information extraction frame skeleton data；According in M frame skeleton data
The angle of multiple dry angles of limb generates the description sample of the human action to be identified；By quantum genetic algorithm to supporting vector
Machine is optimized with the supporting vector machine model after being optimized, and by the supporting vector machine model after optimization to the description
Sample is classified, the human action to be identified to be identified.
Efficient human motion recognition method according to the ... of the embodiment of the present invention based on quantum genetic algorithm optimization, passes through extraction
The angle of multiple dry angles of limb in skeleton data, and human action to be identified is generated according to the angle of the dry angle of multiple limbs
Sample is described, sample size can be greatly reduced, is calculated convenient for subsequent analysis, passes through the support optimized based on quantum genetic algorithm
Vector machine model classifies to description sample, can improve classification accuracy rate, so as to improve the effect of human action identification
Rate and accuracy.
In addition, being identified according to the efficient human action based on quantum genetic algorithm optimization that the above embodiment of the present invention proposes
Method can also have following additional technical characteristic：
Wherein, the dry angle of the limb is to do institute's shape as any two limb on vertex to connect the dry human joint points of multiple limbs
At angle.
Further, the dry human joint points of the multiple limbs of connection include buttocks artis, neck artis, backbone pass
Node, left shoulder joint node, left elbow joint point, left wrist joint point, right shoulder joint node, right elbow joint point, right wrist joint point, left buttocks
Artis, left knee joint point, left ankle artis, right hips artis, right knee joint point, right ankle artis.
Specifically, multiple dry angles of limb in the frame skeleton data are extracted according to the human joint points coordinate information
Angle includes：It is according to the coordinate of the dry human joint points of the multiple limbs of connection, to connect the dry human joint points of multiple limbs
The coordinate that two limbs on vertex do the human joint points of end calculates separately the vector that two limbs do place straight line；Calculate this two
Angle between the vector of straight line where a limb is dry, to obtain the angle of the dry dry angle of limb of two limbs.
Specifically, the human action to be identified is generated according to the angle of multiple dry angles of limb in M frame skeleton data
Description sample include：Extract angle of the dry angle of each limb in the M frames skeleton data respectively, with obtain with each
The corresponding M angle of the dry angle of limb；The variance for calculating the corresponding M angle of the dry angle of each limb, to obtain the people to be identified
The description sample of body action.
According to one embodiment of present invention, support vector machines is optimized by quantum genetic algorithm and is specifically included：
According to constraints yk(ωTφ(xk)+b)≥1-εkObtain the dual equation of Lagrange's equation：
Wherein, ψ is object function, and ω is the weighted vector of hyperplane, and b is bias function, εkFor k-th of slack variable,
And εk>=0, N are positive integer, λkAnd μkAll it is Lagrange multiplier, C is penalty factor, and n is natural number, corresponding n rank soft margins point
Class can use n=1, i.e., linear soft margin classification, xkFor the data of input, ykFor the classification of output.
Above-mentioned dual equation is converted according to optimal conditions, the supporting vector machine model after being optimized：
Wherein, xi、xjIt is inputted for two dimension, yi、yjIt is exported for two dimension, K (xi,xj)=φ (xi)φ(xj) it is kernel function, it is described
Supporting vector machine model after optimization meets constraints：
0≤λk≤ C k=1,2, L, N.
Further, described to use quantum genetic algorithm optimization support vector machines, execute following steps：Step 1：It is given first
Beginning algorithm parameter, and input training set data and test set data and respective labels；Step 2：Initialization population Q (t)；Step
Rapid three：Initialization population is measured, one group of fixed solution P (t) is obtained, for a string of binary system generations of initialization length
Code converts the fixed solution P (t) to penalty factor and kernel functional parameter, and trains supporting vector together with training sample
Machine calculates accuracy by test sample, evaluates current individual, retains optimum individual；Step 4：Judge accurate
Whether degree restrains or whether reaches maximum iteration, if so, algorithm terminates, otherwise executes step 5；Step 5：Utilization
Son rotation door rotation angle adjustable strategies Population Regeneration Q (t)；Step 6：It checks whether and meets catastrophe condition, if it is, retaining
Optimal value, initialization population, if not, executing step 7；Step 7：It enables iterations add one, is back to step 3 and continues to hold
Row；Step 8：Optimized parameter is exported, optimized parameter is used in combination to test test set.
Wherein, the initial algorithm parameter include maximum iteration, Population Size, variable binary length.
Wherein, the initialization population Q (t) includes：Equal rights processing is done to all genes, initializes all genesForIndicate that a chromosome all occurs with equiprobability situation in initial ranging.
Description of the drawings
Fig. 1 is the stream according to the efficient human motion recognition method based on quantum genetic algorithm optimization of the embodiment of the present invention
Cheng Tu；
Fig. 2 is the schematic diagram according to the human joint points of one embodiment of the invention；
Fig. 3 is the schematic diagram according to the dry angle of limb of one embodiment of the invention；
Fig. 4 is the schematic diagram according to the support vector machines of one embodiment of the invention；
Fig. 5 is to be identified according to the efficient human action based on quantum genetic algorithm optimization of one specific embodiment of the present invention
The flow chart of method；
Fig. 6 is the throwing and carry arm action schematic diagram according to one embodiment of the invention.
Specific implementation mode
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to for explaining the present invention, and is not considered as limiting the invention.
Know below in conjunction with the accompanying drawings to describe the efficient human action of the embodiment of the present invention optimized based on quantum genetic algorithm
Other method.
Fig. 1 is the stream according to the efficient human motion recognition method based on quantum genetic algorithm optimization of the embodiment of the present invention
Cheng Tu.
As shown in Figure 1, the efficient human motion recognition method based on quantum genetic algorithm optimization of the embodiment of the present invention, packet
Include following steps：
S1, within the duration of human action to be identified, obtain M frame human body RGB-D image datas, wherein M be more than
Integer equal to 2.
S2 is extracted per frame skeleton data, obtains human joint points coordinate information, and believe according to human synovial point coordinates
Breath extracts the angle of multiple dry angles of limb in the frame skeleton data.
In one embodiment of the invention, human body RGB-D image datas can be obtained by Kinect sensor, and extracted
Per the skeleton data in frame human body RGB-D image datas, and obtain the seat of the human joint points in skeleton data
Mark.Wherein, as shown in Fig. 2, human joint points may include buttocks artis, neck artis, joint of vertebral column point, left shoulder joint
Point, left elbow joint point, left wrist joint point, right shoulder joint node, right elbow joint point, right wrist joint point, left buttocks artis, left knee close
Node, left ankle artis, right hips artis, right knee joint point, right ankle artis, joint of head point, left hand joint point,
Right hand joint point, left foot artis and right foot joint point, above-mentioned human joint points are marked by 1~20 number respectively.
The embodiment of the present invention can obtain the coordinate information of above-mentioned 20 human joint points, be denoted as f=[f1,f2,…,fn,…,f20],
In, fnFor the coordinate (x of n-th of human joint pointsn, yn, zn), i.e., each human joint points include 3 dimension parameters.Therefore, above-mentioned 20
The coordinate information of a human joint points includes 60 dimension parameters.
Identification for human action does not need to the absolute spatial position for accurately acquiring each human joint points, therefore
By investigating the relative position of human joint points, have been able to provide enough information to effective identification that some are acted.
The dry angle of limb described in this step refers to connect the dry human joint points of multiple limbs as any two on vertex
Limb is dry to be formed by angle.In human joint points shown in Fig. 2, joint of head point, left hand joint point, right hand joint point, a left side
Foot joint point and right foot joint point only connect a limb and do, and other human joint points are all connected at least two limbs and do, that is, connect
The dry human joint points of multiple limbs include buttocks artis, neck artis, joint of vertebral column point, left shoulder joint node, left elbow joint
Point, left wrist joint point, right shoulder joint node, right elbow joint point, right wrist joint point, left buttocks artis, left knee joint point, left ankle
Artis, right hips artis, right knee joint point, right ankle artis.
The dry human joint points of multiple limbs are connected as vertex using any one, can form the dry angle of at least one limb.Even
Connecing the dry human joint points of two limbs has 13 (correspondence markings numbers 2,5,6,7,9,10,11,13,14,15,17,18,19),
1 angle can be formed respectively as vertex using it, totally 13 angles；The dry human joint points of three limbs of connection have 1 (correspondence markings
Number 1), according to permutation and combination knowledge, be using the angle number that it can be formed as vertexForm 3 angles；Connection four
The dry human joint points of limb have 1 (correspondence markings number 3), are using the angle number that it can be formed as vertexForm 6
Angle.Therefore, a total of 13+3+6=22 of the dry angle of limb.
For example, as shown in figure 3, to connect the dry buttocks artis of three limbs as vertex, θ can be formed1~θ3Three limbs
Dry angle can form θ to connect the dry neck artis of four limbs as vertex5~θ10Six dry angles of limb.
For 20 human joint points shown in Fig. 2, it can extract 22 limbs altogether and do angle theta1~θ22Angle, such as 1 institute of table
Show (human joint points are indicated with its number is marked in table 1).
Table 1
| Joint sequence number | Angle sequence number |
| 1 | θ1, θ2, θ3 |
| 2 | θ4 |
| 3 | θ5, θ6, θ7, θ8, θ9, θ10 |
| 4 | - |
| 5 | θ11 |
| 6 | θ12 |
| 7 | θ13 |
| 8 | - |
| 9 | θ14 |
| 10 | θ15 |
| 11 | θ16 |
| 12 | - |
| 13 | θ17 |
| 14 | θ18 |
| 15 | θ19 |
| 16 | - |
| 17 | θ20 |
| 18 | θ21 |
| 19 | θ22 |
| 20 | - |
Since the angle of the dry angle of each limb is 1 dimension parameter, thus the angle of the dry angle of above-mentioned 22 limbs may include 22
Tie up parameter.Therefore, it is possible to be 22 angle informations tieed up by the coordinate information dimensionality reduction of the human joint points of 60 dimensions, after substantially reducing
Continuous calculation amount can shorten human action identification and take.
In one embodiment of the invention, it is extracted according to human joint points coordinate information more in the frame skeleton data
The angle of a dry angle of limb specifically includes：According to connecting the coordinate of the dry human joint points of multiple limbs, dry to connect multiple limbs
Human joint points are that the coordinate of the human joint points of the dry end of two limbs on vertex calculates separately the dry place straight line of two limbs
Vector, and the angle between the vector of the dry place straight line of two limbs is calculated, to obtain the angle of the dry dry angle of limb of two limbs
Degree.
Wherein, limb do end refer to be formed the dry angle of limb a limb it is dry in the other end relative to vertex.
For example, it is assumed that n-th of human joint points is the dry human joint points of the multiple limbs of connection, two limbs connected
Dry end is respectively (n-1)th human joint points and (n+1)th human joint points, and the coordinate of three human joint points is：
The vector of straight line is where two limbs then connected are done：
A=fn-1-fn
B=fn+1-fn (2)
Two vectorial angles can be acquired, i.e. the angle of the dry dry angle of limb of two limbs is：
S3 generates the description sample of human action to be identified according to the angle of multiple dry angles of limb in M frame skeleton data
This.
Specifically, angle of the dry angle of each limb in M frame skeleton data can be extracted respectively, to obtain and each limb
The corresponding M angle of dry angle, and the variance of the corresponding M angle of the dry angle of each limb is calculated, it is dynamic to obtain human body to be identified
The description sample of work.
In the above-described embodiments, 22 angle groups are can extract, g=[g are denoted as1,g2,…,ga,…,g22], wherein gaIt is
The corresponding M angle of the dry angle of a limb, i.e. a-th of angle group, ga=[θa1,θa2,...,θaM]T。
Further, the variance of the corresponding M angle of the dry angle of each limb can be calculated according to formula once：
Wherein, jaFor the variance of the corresponding M angle of the dry angle of a-th of limb, μaFor the corresponding M angle of the dry angle of a-th of limb
The mean value of degree.
After the variance for calculating the corresponding M angle of the dry angle of each limb, it can will be calculated separately out according to 22 angles
Description sample of the variance as human action to be identified, you can by { j1,j2,…,ja,…,j22It is used as human action to be identified
Description sample.
S4 optimizes support vector machines by quantum genetic algorithm with the supporting vector machine model after being optimized,
And classified to description sample by the supporting vector machine model after optimization, human action to be identified to be identified.
Quantum genetic algorithm and traditional support vector machines are briefly described first below.
Quantum genetic algorithm is the optimization algorithm based on quantum calculation theory.By quantum theory, describe to lose with state vector
Coding is passed, and realizes the evolution of population using Quantum rotating gate.Due to the concurrency of quantum algorithm, quantum genetic algorithm is in population
In terms of size and search speed, it is better than traditional quantum algorithm.
For the coding of quantum genetic algorithm, common classics genetic algorithm encoding is compiled using binary system and the decimal system
Code.When using quantum bit, the mode of coding will will be different.It is that there is additivities and coherence between quantum state
, so it is different with classical bit, there is Entanglement between different quantum bits.For a binary bits, quantum ratio
Spy cannot simply be written as 0 or 1 state, but one kind between them is arbitrarily superimposed, therefore quantum bit can be written as：
| ψ ＞=α | 0 ＞+β | 1 ＞ (5)
Wherein | 0 ＞ and | 1 ＞ is a kind of vector, indicate system state in which.α and β is a pair of of plural number, corresponds to above-mentioned two
The probability amplitude of kind of state, α and β's square is the probability observed corresponding to above two state, and meets normalizing relationship：
|αi|2+|βi|2=1 (6)
It therefore meets a pair of (5), (6) two formula is plural to [α, β]TIt can be used for indicating a quantum bit, it is right
There is the chromosome of m bits that can be expressed as in one：
Each element of chromosome meets normalizing condition | αi|2+|βi|2=1, i=1,2, L, m.
For Quantum rotating gate, realized to quantum bit by quantum logic revolving door
Operation.The evolution of population may be implemented in the rotation of Quantum logic gates, and by the guiding of optimal condition, it can be in population
The gene of optimization is produced faster.It can accelerate entire algorithm in this way.The operation of Quantum logic gates can use the shape of matrix
Formula is expressed as：
Wherein,WithThe quantum bit of the chromosome in t generations and t+1 generations, G tables are indicated respectively
Show Quantum rotating gate：
θ is rotation angle, and the selection scheme of direction and size is as shown in table 2.
Table 2
In table 2, xiAnd biOptimal chromosome and current best chromosome in current population are indicated respectively.F (x) is to adapt to
Function is spent, Δ θ is rotation angle size, selects different rotation angles can be with control convergence speed.S(αi,βi) be rotation angle side
To.
From the point of view of the task type that machine learning execution part is reflected, it is broadly divided into classification and problem solving two parts.
In treatment classification problem, support vector machines is widely used since it is efficient and simple and direct by everybody.Support vector machines is a kind of
Supervised learning pattern describes the parameter space x of objectk∈Rn, a label space-like y is corresponded to respectivelyk∈R.This constitutes sample
This data space D=(xk,yk) | k=1,2, L, N, N are sample number.As shown in figure 4, support vector machines seeks to solve one group
Hyperplane, this group of hyperplane are two parallel hyperplane being made of two nearest points of distance in sorting group, and in order to reach
To best classification results, to make this two groups of hyperplane distances big as far as possible.
As seen from Figure 4, two solid line hyperplane are the hyperplane constituted to interface closest approach, supporting vector
The task of machine is sought to 2/ in solution figure | ω | maximum value.In order to facilitate solution, optimization problem is expressed as with mathematical expression：
Wherein, ψ is object function, and ω is the weighted vector of hyperplane, and b is bias function, is contained in restrictive condition yk(ωT
φ(xk)+b) >=1 in.This makes it possible to obtain Lagrange's equations：
To the Lagrange's equation carry out extreme value solution, you can obtain support vector machines parameter, for example, can obtain punishment because
Son and kernel functional parameter.
However, for practical problem, two classes may be spaced without so big, it is therefore desirable to introduce the general of soft margin classification
It reads.In this case, object function needs that a slack variable is added, and constraints may be modified such that yk(ωTφ(xk)+b)≥1-
εk, and εk≥0.The constraints is wider than above-mentioned constraints to be sent, and sample point appearance is also had between two hyperplane.Accordingly
The dual equation of Lagrange's equation is：
Wherein, ψ is object function, and ω is the weighted vector of hyperplane, and b is bias function, εkFor k-th of slack variable,
And εk>=0, N are positive integer, λkAnd μkAll it is Lagrange multiplier, C is penalty factor, and n is natural number, corresponding n rank soft margins point
Class, xkFor the data of input, ykFor the classification of output.
Desirable n=1, as linear soft margin classification.
According to optimal conditions：
It can obtain
Extremum conditions after derivation is taken back into Lagrangian, abbreviation optimized after supporting vector machine model：
Wherein, xi、xjIt is inputted for two dimension, yi、yjIt is exported for two dimension, K (xi,xj)=φ (xi)φ(xj) it is kernel function.
Relative to traditional support vector machines, the supporting vector machine model after optimization meets constraints：
It can be seen that Lagrange coefficient is not simply to take the real number not less than 0 can under the conditions of soft margin classification
With value is also less than penalty factor.
It is not directly to use x when describing sample point parameterk, but use its a mapping phi (xk).This be because
In many practical problems, cannot merely to be classified well to sample using linear interface, so needing more multiple
Miscellaneous interface keeps classifying quality more preferable, and this mapping is just played the role of this.For above-mentioned kernel function, common kernel function
Following a few classes can be divided into：
One, gaussian radial basis function：K(xi,xj)=exp (‖ xi-xj‖2/σ2)；
Two, Polynomial kernel function：K(xi,xj)=(xi·xj+c)d；
Three, Sigmoid kernel functions：K(xi,xj)=tan (k (xi·xj)+v)；
Four, linear kernel function：K(xi,xj)=xi·xj。
In one embodiment of the invention, with quantum genetic algorithm optimization support vector machines, following steps are executed：
Step 1：Initialization algorithm parameter, and input training set data and test set data and respective labels, label
Indicate the classification of data.Wherein, initial algorithm parameter include maximum iteration, Population Size, variable binary length
Deng.
Step 2：Initialization population Q (t) specifically can do equal rights processing to all genes, initialize all genesForIndicate that a chromosome all occurs with equiprobability situation in initial ranging.
Step 3：Initialization population is measured, one group of fixed solution P (t) is obtained, is a string of initialization length
Binary code converts fixed solution P (t) to penalty factor and kernel functional parameter, and training is supported together with training sample
Vector machine calculates accuracy by test sample, evaluates current individual, retains optimum individual.
Step 4：Judge whether accuracy restrains or whether reach maximum iteration, if so, algorithm terminates, otherwise
Execute step 5.
Step 5：Utilize quantum rotation door rotation angle adjustable strategies Population Regeneration Q (t) shown in table 2.
Step 6：It checks whether and meets catastrophe condition, if it is, retaining optimal value, initialization population, if not, holding
Row step 7.
Step 7：It enables iterations add one, is back to step 3 and continues to execute.
Step 8：Optimized parameter is exported, optimized parameter is used in combination to test test set.
By formula (15), (16) it is found that being to need to make support vector machines normal operation, penalty factor and kernel functional parameter σ
Two parameters to be determined, two parameters will directly affect the quality of classifying quality.How this two are fast and accurately determined
Parameter is that model construction of SVM is successfully crucial.It in an embodiment of the present invention, can be by quantum genetic algorithm to this two
Parameter is in optimized selection.
Specifically, as shown in figure 5, may include based on the efficient human motion recognition method that quantum genetic algorithm optimizes following
Step：
S101 obtains human joint points coordinate information by Kinect sensor.
S102, the angle of the dry angle of multiple limbs is calculated according to human synovial point coordinate data, and it is dynamic to generate human body to be identified
The description sample of work.
S103 carries out quantization coding to description sample.
S104 initializes iterations t=0 and maximum iteration T, and initializes global optimum variable Max_Global
And error floor and initialization algorithm parameter.
S105 measures quantum state Q (t), obtains input parameter.
S106 calculates fitness function, record local optimum Max_Curr.
S107 judges whether there is Max_Curr >=Max_Global.If so, thening follow the steps S108；If it is not, then holding
Row step S109.
S108 updates Max_Global.Initial global optimum's variable is carried out more according to local optimum Max_Curr
Newly.Step S109 is executed after having executed the step.
S109 rotates policy update quantum state Q (t) according to Quantum rotating gate.
S110, judges whether quantum genetic algorithm restrains.If so, thening follow the steps S113；If not, thening follow the steps
S111。
S111 judges whether there is t≤T.If so, thening follow the steps S112；If not, thening follow the steps S113.
S112, t=t+1.I.e. iterations add one, into searching process next time.
S113 exports optimal value.
In general, can first by above-mentioned two parameter carry out quantum coding, then operated by Quantum rotating gate, constantly to
Optimal solution adjusts.By initializing one group of parameter, the classification accuracy rate under this parameter is found out, using this accuracy as fitness
Function searches out the parameter with highest accuracy.
In one particular embodiment of the present invention, MSRC-12 data sets can be used, by 12 groups of Kinect system acquisitions
The data of action choose throwing and two kinds of actions of carry arm therein and verification are identified.It throws and what two kinds of carry arm acted divides
Solution is as shown in Figure 6.Kinect sensor acquires physical activity data with the frame per second of 30 frame per second.Such as above-described embodiment, acquisition
Mode can be to record the three-dimensional real-time coordinates of human joint points shown in Fig. 2, be calculated by the method for above-described embodiment more
The angle of a dry angle of limb.Herein can also angle theta be done to above-mentioned limb1~θ22Classify, wherein θ1~θ9Indicate human body center pillar
The dry relative angle of limb, θ10~θ16Indicate the dry relative angle of each limb in human upper limb part, θ17~θ22Indicate human body lower limbs part
The dry relative angle of each limb.
Then, respectively by grid search and quantum genetic algorithm come determine supporting vector machine model penalty factor and
Kernel functional parameter σ.In specific calculate, it is 80 that Population Size, which can be arranged, for quantum genetic calculation, and the length of quantum bit is 60
Position.The search range of the penalty factor of support vector machines is [2-2,24], the search range of kernel functional parameter σ is [2-4,24].It is right
In the algebraically of Genetic Algorithm Evolution, can calculate different algebraically as a result, being obtained respectively by grid search and quantum genetic algorithm
Support vector machines parameter, accuracy rate and take as shown in table 3.
Table 3
As can be seen from Table 3, quantum genetic algorithm develops by two generations can reach convergence.Quantum genetic algorithm passes through
Part-time complexity is sacrificed, accuracy rate is improved nearly 2.5 percentage points.
Confusion matrix can be as shown in table 4.
Table 4
(a) quantum genetic algorithm
(b) grid search
As it can be seen that the solution space of grid search receives certain limitation, obtained result with a distance from optimal solution more
Far.And quantum genetic algorithm passes through the characteristics of quantum parallelism, in the case of the time complexity slightly increased, greatly expands
The solution space of exhibition, so as to obtain more ideal result.On the other hand, people is identified by investigating the dry angle of joint limb
The method of body action, under suitable machine learning algorithm, accuracy rate can reach 95% or more.
In conclusion the efficient human action identification side according to the ... of the embodiment of the present invention based on quantum genetic algorithm optimization
Method by extracting the angle of multiple dry angles of limb in skeleton data, and generates according to the angle of the dry angle of multiple limbs and waits knowing
The description sample of other human action, can greatly reduce sample size, be calculated convenient for subsequent analysis, be calculated by being based on quantum genetic
The supporting vector machine model of method optimization classifies to description sample, classification accuracy rate can be improved, so as to improve human body
The efficiency of action recognition and accuracy.
In the description of the present invention, it is to be understood that, term "center", " longitudinal direction ", " transverse direction ", " length ", " width ",
" thickness ", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outside", " up time
The orientation or positional relationship of the instructions such as needle ", " counterclockwise ", " axial direction ", " radial direction ", " circumferential direction " be orientation based on ... shown in the drawings or
Position relationship is merely for convenience of description of the present invention and simplification of the description, and does not indicate or imply the indicated device or element must
There must be specific orientation, with specific azimuth configuration and operation, therefore be not considered as limiting the invention.
In addition, term " first ", " second " are used for description purposes only, it is not understood to indicate or imply relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include one or more this feature.In the description of the present invention, the meaning of " plurality " is two or more,
Unless otherwise specifically defined.
In the present invention unless specifically defined or limited otherwise, term " installation ", " connected ", " connection ", " fixation " etc.
Term shall be understood in a broad sense, for example, it may be being fixedly connected, may be a detachable connection, or integral；Can be that machinery connects
It connects, can also be electrical connection；It can be directly connected, can also can be indirectly connected through an intermediary in two elements
The interaction relationship of the connection in portion or two elements.It for the ordinary skill in the art, can be according to specific feelings
Condition understands the concrete meaning of above-mentioned term in the present invention.
In the present invention unless specifically defined or limited otherwise, fisrt feature can be with "above" or "below" second feature
It is that the first and second features are in direct contact or the first and second features pass through intermediary mediate contact.Moreover, fisrt feature exists
Second feature " on ", " top " and " above " but fisrt feature be directly above or diagonally above the second feature, or be merely representative of
Fisrt feature level height is higher than second feature.Fisrt feature second feature " under ", " lower section " and " below " can be
One feature is directly under or diagonally below the second feature, or is merely representative of fisrt feature level height and is less than second feature.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiments or example.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, changes, replacing and modification.
Claims (9)
1. a kind of efficient human motion recognition method based on quantum genetic algorithm optimization, which is characterized in that include the following steps：
Within the duration of human action to be identified, M frame human body RGB-D image datas are obtained, wherein M is more than or equal to 2
Integer；
Extraction obtains human joint points coordinate information, and according to the human joint points coordinate information per frame skeleton data
Extract the angle of multiple dry angles of limb in the frame skeleton data；
The description sample of the human action to be identified is generated according to the angle of multiple dry angles of limb in M frame skeleton data；
Support vector machines is optimized by quantum genetic algorithm with the supporting vector machine model after being optimized, and by excellent
Supporting vector machine model after change classifies to the description sample, the human action to be identified to be identified.
2. the efficient human motion recognition method according to claim 1 based on quantum genetic algorithm optimization, feature exist
It is to be formed by angle as any two limb on vertex is dry to connect the dry human joint points of multiple limbs in, the dry angle of limb.
3. the efficient human motion recognition method according to claim 2 based on quantum genetic algorithm optimization, feature exist
In the dry human joint points of the multiple limbs of connection include buttocks artis, neck artis, joint of vertebral column point, left shoulder joint
Point, left elbow joint point, left wrist joint point, right shoulder joint node, right elbow joint point, right wrist joint point, left buttocks artis, left knee close
Node, left ankle artis, right hips artis, right knee joint point, right ankle artis.
4. the efficient human motion recognition method according to claim 3 based on quantum genetic algorithm optimization, feature exist
In the angle for extracting multiple dry angles of limb in the frame skeleton data according to the human joint points coordinate information is specifically wrapped
It includes：
According to the coordinate of the dry human joint points of the multiple limbs of connection, to connect the dry human joint points of multiple limbs as vertex
The coordinate for the human joint points that two limbs do end calculates separately the vector of the dry place straight line of two limbs；
The angle between the vector of the dry place straight line of two limbs is calculated, to obtain the angle of the dry dry angle of limb of two limbs.
5. the efficient human motion recognition method according to claim 4 based on quantum genetic algorithm optimization, feature exist
In the description sample for generating the human action to be identified according to the angle of multiple dry angles of limb in M frame skeleton data has
Body includes：
Angle of the dry angle of each limb in the M frames skeleton data is extracted respectively, to obtain and the dry angle pair of each limb
The M angle answered；
The variance for calculating the corresponding M angle of the dry angle of each limb, to obtain the description sample of the human action to be identified.
6. the efficient human motion recognition method according to claim 1 based on quantum genetic algorithm optimization, feature exist
In being optimized and specifically included to support vector machines by quantum genetic algorithm：
According to constraints yk(ωTφ(xk)+b)≥1-εkObtain the dual equation of Lagrange's equation：
Wherein, ψ is object function, and ω is the weighted vector of hyperplane, and b is bias function, εkFor k-th of slack variable, and εk≥
0, N is positive integer, λkAnd μkAll it is Lagrange multiplier, C is penalty factor, and n is natural number, corresponding n rank soft margin classifications, can
Take n=1, i.e., linear soft margin classification, xkFor the data of input, ykFor the classification of output,
Above-mentioned dual equation is converted according to optimal conditions, the supporting vector machine model after being optimized：
Wherein, xi、xjIt is inputted for two dimension, yi、yjIt is exported for two dimension, K (xi,xj)=φ (xi)φ(xj) it is kernel function,
Supporting vector machine model after the optimization meets constraints：
7. the efficient human motion recognition method according to claim 6 based on quantum genetic algorithm optimization, feature exist
In, it is described to use quantum genetic algorithm optimization support vector machines, execute following steps：
Step 1：Initialization algorithm parameter, and input training set data and test set data and respective labels；
Step 2：Initialization population Q (t)；
Step 3：Initialization population is measured, one group of fixed solution P (t) is obtained, for initialize a string two of length into
Code processed converts the fixed solution P (t) to penalty factor and kernel functional parameter, and training is supported together with training sample
Vector machine calculates accuracy by test sample, evaluates current individual, retains optimum individual；
Step 4：Judge whether accuracy restrains or whether reach maximum iteration, if so, algorithm terminates, otherwise executes
Step 5；
Step 5：Utilize Quantum rotating gate rotation angle adjustable strategies Population Regeneration Q (t)；
Step 6：It checks whether and meets catastrophe condition, if it is, retaining optimal value, initialization population, if not, executing step
Rapid seven；
Step 7：It enables iterations add one, is back to step 3 and continues to execute；
Step 8：Optimized parameter is exported, optimized parameter is used in combination to test test set.
8. the efficient human motion recognition method according to claim 7 based on quantum genetic algorithm optimization, feature exist
In, the initial algorithm parameter include maximum iteration, Population Size, variable binary length.
9. the efficient human motion recognition method according to claim 8 based on quantum genetic algorithm optimization, feature exist
In the initialization population Q (t) includes：Equal rights processing is done to all genes, initializes all genesForIndicate that a chromosome all occurs with equiprobability situation in initial ranging.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810014848.8A CN108416251A (en) | 2018-01-08 | 2018-01-08 | Efficient human motion recognition method based on quantum genetic algorithm optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810014848.8A CN108416251A (en) | 2018-01-08 | 2018-01-08 | Efficient human motion recognition method based on quantum genetic algorithm optimization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108416251A true CN108416251A (en) | 2018-08-17 |
Family
ID=63125754
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810014848.8A Pending CN108416251A (en) | 2018-01-08 | 2018-01-08 | Efficient human motion recognition method based on quantum genetic algorithm optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108416251A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109840490A (en) * | 2019-01-25 | 2019-06-04 | 深圳大学 | Processing method, system, electronic device and the storage medium of human action characterization |
| CN110414839A (en) * | 2019-07-29 | 2019-11-05 | 四川长虹电器股份有限公司 | Load recognition methods and system based on quantum genetic algorithm and SVM model |
| CN112861696A (en) * | 2021-02-01 | 2021-05-28 | 电子科技大学中山学院 | Abnormal behavior identification method and device, electronic equipment and storage medium |
| CN112906438A (en) * | 2019-12-04 | 2021-06-04 | 内蒙古科技大学 | Human body action behavior prediction method and computer equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104537382A (en) * | 2015-01-12 | 2015-04-22 | 杭州电子科技大学 | Electromyographic signal gait recognition method for optimizing support vector machine based on genetic algorithm |
| CN106022213A (en) * | 2016-05-04 | 2016-10-12 | 北方工业大学 | Human body motion recognition method based on three-dimensional bone information |
| CN106053067A (en) * | 2016-05-24 | 2016-10-26 | 广东石油化工学院 | Bearing fault diagnosis method based on quantum genetic algorithm optimized support vector machine |
-
2018
- 2018-01-08 CN CN201810014848.8A patent/CN108416251A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104537382A (en) * | 2015-01-12 | 2015-04-22 | 杭州电子科技大学 | Electromyographic signal gait recognition method for optimizing support vector machine based on genetic algorithm |
| CN106022213A (en) * | 2016-05-04 | 2016-10-12 | 北方工业大学 | Human body motion recognition method based on three-dimensional bone information |
| CN106053067A (en) * | 2016-05-24 | 2016-10-26 | 广东石油化工学院 | Bearing fault diagnosis method based on quantum genetic algorithm optimized support vector machine |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109840490A (en) * | 2019-01-25 | 2019-06-04 | 深圳大学 | Processing method, system, electronic device and the storage medium of human action characterization |
| CN109840490B (en) * | 2019-01-25 | 2021-10-22 | 深圳大学 | Human motion representation processing method and system, electronic device and storage medium |
| CN110414839A (en) * | 2019-07-29 | 2019-11-05 | 四川长虹电器股份有限公司 | Load recognition methods and system based on quantum genetic algorithm and SVM model |
| CN112906438A (en) * | 2019-12-04 | 2021-06-04 | 内蒙古科技大学 | Human body action behavior prediction method and computer equipment |
| CN112906438B (en) * | 2019-12-04 | 2023-05-02 | 内蒙古科技大学 | Human body action behavior prediction method and computer equipment |
| CN112861696A (en) * | 2021-02-01 | 2021-05-28 | 电子科技大学中山学院 | Abnormal behavior identification method and device, electronic equipment and storage medium |
| CN112861696B (en) * | 2021-02-01 | 2023-08-18 | 电子科技大学中山学院 | Abnormal behavior identification method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Applications of deep learning to MRI images: A survey | |
| Balın et al. | Concrete autoencoders: Differentiable feature selection and reconstruction | |
| CN108416251A (en) | Efficient human motion recognition method based on quantum genetic algorithm optimization | |
| CN112101176B (en) | User identity recognition method and system combining user gait information | |
| Pishchulin et al. | Deepcut: Joint subset partition and labeling for multi person pose estimation | |
| WO2020258611A1 (en) | Lymph node ct detection system employing recurrent spatio-temporal attention mechanism | |
| Sanz et al. | A genetic tuning to improve the performance of fuzzy rule-based classification systems with interval-valued fuzzy sets: Degree of ignorance and lateral position | |
| Schuld et al. | Quantum computing for pattern classification | |
| Chen et al. | A novel convolutional neural network model based on beetle antennae search optimization algorithm for computerized tomography diagnosis | |
| EP3171297A1 (en) | Joint boundary detection image segmentation and object recognition using deep learning | |
| CN108921123A (en) | A kind of face identification method based on double data enhancing | |
| CN101241600B (en) | Chain-shaped bone matching method in movement capturing technology | |
| Wang et al. | Device-free human gesture recognition with generative adversarial networks | |
| Saadi et al. | Investigation of effectiveness of shuffled frog-leaping optimizer in training a convolution neural network | |
| Hu et al. | A novel feature incremental learning method for sensor-based activity recognition | |
| Ye et al. | Deep mixture generative autoencoders | |
| Ng et al. | Multi-localized sensitive autoencoder-attention-lstm for skeleton-based action recognition | |
| Ponce et al. | Stochastic parallel extreme artificial hydrocarbon networks: An implementation for fast and robust supervised machine learning in high-dimensional data | |
| Lui et al. | Enhanced decoupled active contour using structural and textural variation energy functionals | |
| Xu et al. | Diverse human motion prediction guided by multi-level spatial-temporal anchors | |
| CN109598219A (en) | A kind of adaptive electrode method for registering for robust myoelectric control | |
| CN110163131A (en) | Mix the human action classification method of convolutional neural networks and the optimization of microhabitat grey wolf | |
| Braga et al. | A semi-supervised self-organizing map for clustering and classification | |
| Fujii et al. | Decentralized policy learning with partial observation and mechanical constraints for multiperson modeling | |
| Wu et al. | An unsupervised real-time framework of human pose tracking from range image sequences |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180817 |
|
| RJ01 | Rejection of invention patent application after publication |