CN102592150B - Gait identification method of bidirectional two-dimensional principal component analysis based on fuzzy decision theory - Google Patents

Gait identification method of bidirectional two-dimensional principal component analysis based on fuzzy decision theory Download PDF

Info

Publication number
CN102592150B
CN102592150B CN 201210011983 CN201210011983A CN102592150B CN 102592150 B CN102592150 B CN 102592150B CN 201210011983 CN201210011983 CN 201210011983 CN 201210011983 A CN201210011983 A CN 201210011983A CN 102592150 B CN102592150 B CN 102592150B
Authority
CN
China
Prior art keywords
gait
image
training sample
average
energygram
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.)
Expired - Fee Related
Application number
CN 201210011983
Other languages
Chinese (zh)
Other versions
CN102592150A (en
Inventor
张前进
卜文绍
徐素莉
郑国强
陈祥涛
李劲伟
张松灿
孙炎增
李佩佩
王桂泉
祁志娟
王雯霞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Henan University of Science and Technology
Original Assignee
Henan University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Henan University of Science and Technology filed Critical Henan University of Science and Technology
Priority to CN 201210011983 priority Critical patent/CN102592150B/en
Publication of CN102592150A publication Critical patent/CN102592150A/en
Application granted granted Critical
Publication of CN102592150B publication Critical patent/CN102592150B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Abstract

The invention provides a gait identification method of a bidirectional two-dimensional principal component analysis based on a fuzzy decision theory, The method comprises the following steps of: firstly, pre-processing an image of a gait sequence and extracting a human body movement outline; determining a gait period and calculating an average image of a gait energy image with a whole period; constructing a training sample of an average gait energy image and dividing the training sample into N secondary image sets; calculating the best projection matrix in a row direction and an array direction of each secondary image set, and calculating a characteristic matrix of secondary images of each training sample; calculating the characteristic matrix of each secondary image by the average gait energy image to be indentified; calculating a membership degree of each training sample to the image to be indentified; and determining a classification result according to the maximum membership degree principle. According to the gait identification method disclosed by the invention, the average gait energy image is divided into a plurality of the secondary images and a coefficient matrix dimensional quantity of the average gait energy image is analyzed and reduced by the bidirectional two-dimensional principal component analysis, so that the problem that the coefficient matrix dimensional quantity of the average gait energy image in the gait identification is too high is solved, the identification efficiency is improved and the identification speed is accelerated.

Description

Gait recognition method based on the bidirectional two-dimensional principal component analysis (PCA) of fuzzy theory decision-making
Technical field
The present invention relates to a kind of image processing techniques, specifically a kind of bidirectional two-dimensional principal component analysis (PCA) gait recognition method based on the fuzzy theory decision-making.
Background technology
Gait Recognition is a kind of novel biometrics identification technology.Gait Recognition is as a kind of living things feature recognition method, and the posture of walking according to the people is exactly carried out people's authentication and identification.Gait feature has certain rhythm and cycle characteristics, and is periodic in essence.As behavioural characteristic habitually, a people's gait is can great changes will take place within considerable time, has stronger stability.Be different from the recognition technologies such as recognition of face, fingerprint recognition, iris recognition, Gait Recognition has following characteristics as a kind of biological characteristic: be easy to observe, be difficult to camouflage,, remote identification low to the systemic resolution requirement etc.Therefore Gait Recognition is a kind of desirable non-infringement biometrics identification technology.The method of existing Gait Recognition is a lot, be divided into generally model-based methods and based on the method for statistics: model-based methods is intended to build the 2D of a human body or the motion structure model of 3D, by extracting characteristics of image, they is mapped to the gait pattern that characterizes human body on the constituent of model.Come the characterization gait motion based on the statistical property that the method for adding up is the spatiotemporal mode that profile produces in image by the pedestrian.Existing gait recognition method is all generally the framework that adopts training video sequence under identical walking states condition and test video sequence to set up Gait Recognition.In actual application, it is very greatly that the video sequence of test changes, the difference of for example wearing clothes, and difference of belongings etc. state all can have a huge impact the recognition performance of reality.In Gait Recognition, the application of gait energygram is also more extensive in addition, but the average gait energygram in gait feature identification is too much as the matrix of coefficients element, and the problem that dimension is too high is larger on the recognition speed impact.In this case, reasonably select characteristic quantity can characterize real gait feature, select rational gait recognition method, discrimination and the recognition speed that can improve gait become an important problem.
Find through the retrieval to prior art, " decomposing Gait Recognition with the motion excursion characteristic based on energygram " that the people such as Ma Qinyong delivered on to the 549th page the 545th page of " photoelectron laser " (the 20th volume fourth phase) proposed to utilize average neighbour figure to rebuild abnormal image with mean profile figure, then the gait energygram with object is decomposed into two parts, and be respectively a series of expanded images of every part generation, thereby construct Energy Decomposition figure.Use at last Energy Decomposition figure and motion excursion figure jointly to classify.The character representation mode of this gait recognition method has avoided traditional gait energygram to be subject to the shortcoming of gait shape width impact to a great extent.For the existing problem that has still kept a large amount of static in shape information based on gait deflection graph algorithm, this feature representation mode is paid attention to the contour edge zone more, and can lose some inner Global Information.Simultaneously, this gait profile diagram abnormality detection and correcting algorithm calculated amount are larger, also have the space that promotes on recognition speed." a kind of identification algorithm based on average gait energygram " by people such as retrieval Zhang Qianjin etc. delivered on to 44 pages the 39th page of " Journal of Engineering Graphics " (the 32nd volume first phase) proposed a kind of built-in type hidden Markov model personal identification method based on the gait energygram.At first extract the side profile of movement human, count gait cycle according to the swinging distance of gait lower limb, obtain average gait energygram.Each zone to energygram is analyzed, and utilizes two-dimension discrete cosine transform to observe piece be converted into observation vector energygram, realizes training and the identification of built-in type hidden Markov model.On USF and CASIA gait data storehouse, the algorithm that proposes is tested at last.This algorithm is based on built-in type hidden Markov model, and this model can expend certain calculated amount in the process of training and identification; Utilize simultaneously two-dimension discrete cosine transform to observe piece be converted into observation vector energygram, the information of gait has partial loss, and recognition result is had certain impact.This just impels the district to seek a kind of new gait recognition method, promotes the speed of identification when having higher discrimination.
Summary of the invention
Technical matters to be solved by this invention is to provide a kind of gait recognition method of the bidirectional two-dimensional principal component analysis (PCA) (DTPCA) based on the fuzzy theory decision-making.Obtain average gait energygram and be a plurality of subimages with the image segmentation that obtains by preconditioning technique, utilize the bidirectional two-dimensional principal component analysis (PCA) to reduce the matrix of coefficients dimension of average gait quantum of energy image, solve in the Gait Recognition average gait energygram as the too high problem of matrix of coefficients dimension, accelerated recognition speed.
The present invention solves the problems of the technologies described above the technical scheme that adopts to be: the gait recognition method based on the bidirectional two-dimensional principal component analysis (PCA) of fuzzy theory decision-making comprises the steps:
Step 1, the image in a gait sequence is carried out pre-service: the median method that adopts the nonlinear smoothing technology, the gray-scale value of each pixel is set to the intermediate value of all the pixel gray-scale values adjacent with this point, recover background image from image sequence, with the intermediate value pixel value of image as a setting of the N continuous two field picture pixel value of input;
The profile that difference operation is extracted human motion is carried out in step 2, use indirectly;
Step 3, simply connected movement destination image is carried out the quasi periodic analysis, by analyzing the human body contour shape over time, minimum determine gait cycle to the cyclical variation process of maximum again according to the lower limb profile width of people's body side surface from being up to;
The gait cycle that step 4, basis are determined builds the energygram of each gait cycle, and then calculates the average image of the gait energygram that has complete cycle in gait sequence, obtains an average gait energygram as training sample;
Step 5, according to the method for step 1 to four, build
Figure 201210011983X100002DEST_PATH_IMAGE002
Individual average gait energygram is as training sample
Figure 201210011983X100002DEST_PATH_IMAGE004
, each sample All be divided into NNumber of sub images , the subimage of all training image sample correspondence positions forms the subgraph image set
Figure 201210011983X100002DEST_PATH_IMAGE010
Step 6, establish ,
Figure 201210011983X100002DEST_PATH_IMAGE014
, wherein
Figure 201210011983X100002DEST_PATH_IMAGE016
With
Figure 201210011983X100002DEST_PATH_IMAGE018
Represent respectively training sample
Figure 201210011983X100002DEST_PATH_IMAGE020
With
Figure 201210011983X100002DEST_PATH_IMAGE022
iIndividual row vector,
Figure 201210011983X100002DEST_PATH_IMAGE024
The average image that represents all training samples; Utilize covariance matrix
Figure 201210011983X100002DEST_PATH_IMAGE026
To every number of sub images collection
Figure 243572DEST_PATH_IMAGE010
Obtain the optimum projection matrix on line direction
Step 7, establish
Figure 201210011983X100002DEST_PATH_IMAGE030
,
Figure 201210011983X100002DEST_PATH_IMAGE032
, here With
Figure 201210011983X100002DEST_PATH_IMAGE036
Expression respectively
Figure 201210011983X100002DEST_PATH_IMAGE038
With
Figure 201210011983X100002DEST_PATH_IMAGE040
Figure 201210011983X100002DEST_PATH_IMAGE042
Individual column vector is utilized formula To every number of sub images collection Ask the optimum projection matrix on column direction
Figure 201210011983X100002DEST_PATH_IMAGE048
Step 8, utilize formula
Figure 201210011983X100002DEST_PATH_IMAGE050
Ask each training image sample
Figure 201210011983X100002DEST_PATH_IMAGE052
All subimages
Figure DEST_PATH_IMAGE054
Eigenmatrix
Figure DEST_PATH_IMAGE056
, wherein
Figure 970832DEST_PATH_IMAGE054
Expression the iOf individual training sample jNumber of sub images,
Figure DEST_PATH_IMAGE058
, MBe the number of training sample,
Figure DEST_PATH_IMAGE060
, NBe the number of subgraph image set,
Figure DEST_PATH_IMAGE062
For
Figure DEST_PATH_IMAGE064
Correspondence is not less than the diagonal matrix that 1 eigenwert forms,
Figure DEST_PATH_IMAGE066
For less than 1 eigenwert characteristic of correspondence vector,
Figure DEST_PATH_IMAGE068
Be an index between 0-1;
Step 9, to an average gait energygram picture to be identified B, at first be partitioned into each number of sub images
Figure DEST_PATH_IMAGE070
, to each self-corresponding projector space projection, try to achieve the eigenmatrix of every number of sub images with every number of sub images
Figure DEST_PATH_IMAGE072
Step 10, the distance between each training sample eigenmatrix in the eigenmatrix that obtains of calculation procedure nine and step 8 then
Figure DEST_PATH_IMAGE074
, according to formula
Figure DEST_PATH_IMAGE076
Obtain the subimage of sample to be identified
Figure DEST_PATH_IMAGE078
Degree of membership to training sample
Figure DEST_PATH_IMAGE080
, wherein,
Figure DEST_PATH_IMAGE082
, Be fuzzy factor;
Step 11, to the degree of membership addition of average each number of sub images of gait energygram to be identified and each training sample subimage summation, obtain image to be identified for the degree of membership of each training sample
Figure DEST_PATH_IMAGE086
, according to maximum membership grade principle, last classification results is , i.e. facial image to be identified and SIndividual training sample belongs to a class together.
The invention has the beneficial effects as follows: consider that average gait energygram is too much as the matrix of coefficients element, the problem that dimension is too high is directly calculated, and calculated amount is too large.The average gait energygram that the present invention will obtain looks like to be divided into a plurality of subimages, utilize the bidirectional two-dimensional principal component analysis (PCA) to reduce the matrix of coefficients dimension of average gait quantum of energy image, solved in the Gait Recognition average gait energygram as the too high problem of matrix of coefficients dimension, improve discrimination, accelerated recognition speed.For the more difficult characteristics of classification in identification, introduced the better effects if that fuzzy theory makes classification in the process of categorised decision.The present invention tests the method that proposes on CASIA gait data storehouse, and result shows that the method has recognition performance preferably, and discrimination and recognition speed thereof are better than existing method.
Description of drawings
Fig. 1 the inventive method is extracted the gait sequence profile.
The analysis of Fig. 2 the inventive method gait cycle.
Fig. 3 the inventive method gait frame and gait energygram.
The identifying of the average gait energygram of Fig. 4 the inventive method.
The training of Fig. 5 the inventive method bidirectional two-dimensional principal component analysis (PCA) and identification block diagram.
90 ° of visual angle experimental results of Fig. 6 the inventive method.
Embodiment
The below elaborates to embodiments of the invention, and the present embodiment is implemented under take technical solution of the present invention as prerequisite, provided detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to following embodiment.
Embodiment
The preconditioning technique of step 1, image.At first obtain background image.Adopt the median method of nonlinear smoothing technology, the gray-scale value of each pixel is set to the intermediate value of all the pixel gray-scale values adjacent with this point, recover background image from image sequence, with the intermediate value pixel value of image as a setting of the N continuous two field picture pixel value of input.Order Represent the sequence that comprises the N two field picture, a background image
Figure DEST_PATH_IMAGE092
Can be expressed as:
Figure DEST_PATH_IMAGE094
, wherein
Figure DEST_PATH_IMAGE096
Be
Figure DEST_PATH_IMAGE098
The gray-scale value at place,
Figure DEST_PATH_IMAGE100
Be background image.
The extraction of step 2, human motion profile: use and indirectly carry out difference operation, In formula
Figure DEST_PATH_IMAGE104
With
Figure DEST_PATH_IMAGE106
Respectively that present image and background image are in pixel
Figure DEST_PATH_IMAGE108
The brightness value at place.This function can detect according to the brightness of each pixel in background image its Susceptible change, the brightness of present image and background image near the time, can detect better the profile of human motion, as shown in Figure 1.
Step 3, to the quasi periodic analysis of simply connected movement destination image: to obtaining the profile of human body image, use plural form
Figure DEST_PATH_IMAGE110
Be expressed as:
Figure DEST_PATH_IMAGE112
, in formula
Figure DEST_PATH_IMAGE114
With
Figure DEST_PATH_IMAGE116
Coordinate figure for this pixel.An important clue determining the inherent motion of pedestrian be the human body contour outline shape over time.For the complexity that reduces to calculate, two-dimensional contour shape is changed the change in time and space pattern that the one-dimensional distance signal is expressed gait motion that is converted to.The motion change that stresses the following human body lower limbs of waist selects lowest edge point as a reference point, from bottom to top the profile border is expanded into the most left and the rightest 2 points of identical ordinate
Figure DEST_PATH_IMAGE118
With
Figure DEST_PATH_IMAGE120
Distance
Figure DEST_PATH_IMAGE122
Figure DEST_PATH_IMAGE124
, can find out from human body side profile image sequence, gait data is the time-variable data of quasi periodic.The lower limb profile width of people's body side surface can be through the variation of one-period, namely have one minimum again to the change procedure of maximum from being up to, utilize the profile width of two legs to change to carry out periodicity analysis herein, determine gait cycle to the cyclical variation process of maximum according to the lower limb profile width of people's body side surface again from being up to minimum.During another employing profile analysis gait, can't distinguish the left and right leg from profile, as shown in Figure 2.
Step 4, according to the gait cycle of determining, build the energygram of each gait cycle.Suppose the moment in a sequence tThe time, corresponding pretreated scale-of-two gait contour images is
Figure DEST_PATH_IMAGE126
, the gait energygram of gray level is defined as follows:
Figure DEST_PATH_IMAGE128
, N is the total number of frame in complete gait cycle here, xWith yThe coordinate figure that corresponding is in the 2D image, tRefer to the frame number in gait cycle.After obtaining the energygram of each gait cycle, then have the average image of the gait energygram of complete cycle in the sequence of calculation, obtain an average gait energygram as training sample.Have in the gait sequence that suppose object is nIndividual complete gait cycle, the kThe gait energygram of individual gait cycle is expressed as
Figure DEST_PATH_IMAGE130
, the average gait energygram of this gait sequence is: , as shown in Figure 3.
Step 5, according to the method for step 1 to four, build
Figure 518880DEST_PATH_IMAGE002
Individual average gait energygram is as training sample
Figure 347378DEST_PATH_IMAGE004
, each sample
Figure 99433DEST_PATH_IMAGE006
All be divided into the N number of sub images
Figure 859579DEST_PATH_IMAGE008
, the subimage of all training image sample correspondence positions forms the subgraph image set
Figure 64295DEST_PATH_IMAGE010
Step 6, establish
Figure 637097DEST_PATH_IMAGE012
,
Figure 243659DEST_PATH_IMAGE014
, wherein
Figure 174706DEST_PATH_IMAGE016
With
Figure 368183DEST_PATH_IMAGE018
Represent respectively training sample
Figure 42878DEST_PATH_IMAGE020
With iIndividual row vector.Suppose that training sample is total MIndividual, the kIndividual training sample is with one
Figure DEST_PATH_IMAGE134
Matrix
Figure DEST_PATH_IMAGE136
Represent, the average image of all training samples is used
Figure 183058DEST_PATH_IMAGE024
Represent, can obtain the covariance matrix (scatter matrix) of image
Figure DEST_PATH_IMAGE138
For:
Figure DEST_PATH_IMAGE140
, utilize covariance matrix
Figure DEST_PATH_IMAGE142
To every number of sub images collection
Figure 671022DEST_PATH_IMAGE010
Obtain the optimum projection matrix on line direction
Figure 883828DEST_PATH_IMAGE028
Specifically ask method to be: to adopt criterion
Figure DEST_PATH_IMAGE144
, wherein, The covariance matrix of the projection properties vector of expression training sample,
Figure DEST_PATH_IMAGE148
Expression
Figure DEST_PATH_IMAGE150
Mark, the meaning that maximizes criterion finds a projecting direction exactly
Figure DEST_PATH_IMAGE152
, make all samples project to
Figure 91081DEST_PATH_IMAGE152
After upper, the total population scatter matrix of projection sample is maximum,
Figure DEST_PATH_IMAGE154
Can in the hope of
Figure 737832DEST_PATH_IMAGE028
Step 7, establish
Figure DEST_PATH_IMAGE156
,
Figure 342120DEST_PATH_IMAGE032
, here
Figure 854223DEST_PATH_IMAGE034
With
Figure 289883DEST_PATH_IMAGE036
Expression respectively
Figure DEST_PATH_IMAGE158
With
Figure 373114DEST_PATH_IMAGE040
Figure 261436DEST_PATH_IMAGE042
Individual column vector, image covariance matrix so
Figure 81625DEST_PATH_IMAGE138
Another kind be defined as
Figure DEST_PATH_IMAGE160
, utilize formula
Figure 748623DEST_PATH_IMAGE044
To every number of sub images collection
Figure 628854DEST_PATH_IMAGE046
Ask the optimum projection matrix on column direction
Figure 4472DEST_PATH_IMAGE048
, the same step 6 of method.
Step 8, utilize formula
Figure 861307DEST_PATH_IMAGE050
Ask each training image sample
Figure 271560DEST_PATH_IMAGE052
All subimages Eigenmatrix
Figure 185606DEST_PATH_IMAGE056
, wherein
Figure 804323DEST_PATH_IMAGE054
Expression the iOf individual training sample jNumber of sub images,
Figure 803503DEST_PATH_IMAGE058
, MBe the number of training sample,
Figure DEST_PATH_IMAGE162
, NBe the number of subgraph image set,
Figure 461755DEST_PATH_IMAGE062
For
Figure 811965DEST_PATH_IMAGE064
Correspondence is not less than the diagonal matrix that 1 eigenwert forms,
Figure 777647DEST_PATH_IMAGE066
For less than 1 eigenwert characteristic of correspondence vector,
Figure 896913DEST_PATH_IMAGE068
Be an index between 0-1,
Figure DEST_PATH_IMAGE164
Be exactly that this diagonal matrix has added an exponential quantity,
Figure 401100DEST_PATH_IMAGE068
Be the number between 0-1, result is each the value fetching number that is equivalent to diagonal matrix, for example
Figure DEST_PATH_IMAGE166
,
Figure DEST_PATH_IMAGE168
The time,
Figure 612507DEST_PATH_IMAGE164
Be just
Figure DEST_PATH_IMAGE170
Utilize formula
Figure DEST_PATH_IMAGE172
Ask each training image sample
Figure 487273DEST_PATH_IMAGE052
All subimages
Figure DEST_PATH_IMAGE174
Eigenmatrix
Figure 664307DEST_PATH_IMAGE056
Step 9, to an average gait energygram picture to be identified
Figure DEST_PATH_IMAGE176
, at first be partitioned into each number of sub images
Figure DEST_PATH_IMAGE178
, to each self-corresponding projector space projection, obtain the projection matrix on each number of sub images row, column direction, according to formula with every number of sub images
Figure DEST_PATH_IMAGE180
Try to achieve the eigenmatrix of every number of sub images
Step 10, the then eigenmatrix that obtains of calculation procedure nine and the distance between training sample, , the subimage of sample to be identified
Figure DEST_PATH_IMAGE184
Degree of membership to training sample can be according to formula
Figure DEST_PATH_IMAGE186
Obtain, wherein,
Figure DEST_PATH_IMAGE188
The expression subimage To the mean value between training sample, Being fuzzy factor, is the constant that can control fog-level, this paper fuzzy factor tValue get 0.5.
Step 11, to the degree of membership addition of average each number of sub images of gait energygram to be identified and each training sample subimage summation, obtain image to be identified for the degree of membership of each training sample
Figure 470217DEST_PATH_IMAGE086
, according to maximum membership grade principle, last classification results is
Figure 794102DEST_PATH_IMAGE088
, i.e. facial image to be identified and SIndividual training sample belongs to a class together, the identifying of average gait energygram as shown in Figure 4, the training of DTPCA and identification block diagram are as shown in Figure 5.
In the present invention, owing to directly matrix being calculated, calculated amount is too large, dimension is too high, generally all that the variable of getting certain characteristic properties that can represent matrix is expressed in the processing of reality, the present invention comes the character of expression matrix with the optimum projection matrix of matrix, can reduce the dimension of matrix, is convenient to calculate.Establish in the present invention that average gait energygram looks like to be divided into NThe number of sub images collection, the subgraph image set obtains through the bidirectional two-dimensional principal component analysis (PCA) pOn individual line direction and qOptimum projection matrix on individual column direction With , the present invention is divided into a large matrix minor matrix of several p*q, is convenient to calculate.
Implementation result
According to above-mentioned steps, storehouse Dataset B carries out Gait Recognition to Institute of Automation Research of CAS's CASIA gait data.This database is comprised of 124 people, everyone sequence be from 0 ° to 180 °, with 18 ° of 11 different visual angles that increase progressively.Everyone in database has 10 walking sequences, 6 normal walking sequences (Set A) wherein arranged, the sequence of 2 satchels (Set B), 2 sequences (Set C) of wearing overcoat.As training subset (Set A1), subset (Set A2) is tested in remaining conduct front 3 sequences in everyone Set A.The previous sequence of Set B and Set C is as training subset ((Set B1) and (Set C1)), surplus next sequence is as test subset ((Set B2) and (Set C2)), all experiments all realize on the PC computing machine, this machines configurations is: AMD Turion 64 * 2 Mobile Technology TL-50 dual core processors, 2GB internal memory.
At first the present invention selects 25 people in the storehouse to test, and the visual angle of sequence is the visual angle, positive side of 90 °.A check addition test is stayed in employing, and experimental result as shown in Figure 6.Can find out from experimental result, the experiment effect of experiment A is better than experiment B and C.This is that the average gait energygram of relative Set B and Set C set is less as the actual information dimension when walking due to people in Set A set, and the eigenmatrix after the DTPCA conversion more can characterize its characteristics of image effectively.Set B Set A set relative to Set C set, image is comparatively complicated, and after the DTPCA conversion, eigenmatrix has represented the most information of image, compares and has also lost some average gait energy image feature informations when Set A is integrated into conversion.
The below discusses the comparison of the inventive method and other several methods.The database use CASIA gait data storehouse of experiment, the classification of database selects 40 people in the storehouse to test with top method.The sample object direction of motion of test becomes 90 ° of visual angles with the camera light direction of principal axis, sorter is the nearest neighbor classifier of fuzzy theory decision-making technique, and the discrimination of gained is respectively the mean value of three subset.Algorithm of the present invention obviously is better than other several algorithms, and algorithm is as follows:
MBPSS method: Baofeng Guo, Mark S. Nixon. Gait feature Subset by mutual information[J]. IEEE Trans on Systens, Man, and Cybernetics-Part A:Systems and Humans, 2009,39 (1): 36-46;
ACPE method: based on the gait authentication [J] of acceleration signature point extraction. king's rush, Yuan Tao, Liang Can. Tsing-Hua University's journal (natural science edition), 2009,49 (10): 25-28;
Fanbean method: Wang Kejun, beautifully adorned Xian is firelight or sunlight. based on the Algorithm for gait recognition [J] of linear interpolation. and Central China University of Science and Technology's journal (natural science edition), 2010,38 (2): 41-44;
E-HMM method: Zhang Qianjin, Chen Xiangtao, Bu Wenshao. a kind of identification algorithm [J] based on average gait energygram. Journal of Engineering Graphics, 2011,32 (1): 39-44;
Many Fusion Features method: Wang Kejun, beautifully adorned Xian is firelight or sunlight, Liu Lili etc. based on the information fusion Gait Recognition [J] of energy. and Central China University of Science and Technology's journal (natural science edition), 2009,37 (5): 14-17.
The comparative result of algorithms of different discrimination is:
Figure 201210011983X100002DEST_PATH_IMAGE001
Algorithms of different comparative result consuming time is: (unit is ms)
Experiment shows, the inventive method as can be seen from the table, the recognition effect of this paper and performance consuming time obviously are better than additive method.MBPSS method and ACPE algorithm are all more classical gait recognition methods, and the complexity of these two kinds of algorithms is apparently higher than this paper algorithm, but its discrimination is not so good as this paper method, and consuming time higher than this paper.The Fanbean algorithm is to use the Fanbean of linear interpolation to hint obliquely in conjunction with the general matrix low-rank to estimate feature extraction algorithm, and the complexity of this algorithm is higher than this paper algorithm, and pedestrian's clothing situation is too relied on.The e-HMM algorithm utilizes the space characteristics information of the average gait energygram of two-dimension discrete cosine transform acquisition, has introduced the method for built-in type hidden Markov model, and calculated amount is greater than this paper algorithm, and its discrimination is also high not as this paper algorithm, and consuming time the longest.Many Feature Fusion Algorithms carry out dimensionality reduction to original sequence preferably in the process of identification, it is medium that computation complexity belongs to, but its discrimination is consuming time also longer not as this paper algorithm.Can be found out by table, with other current several Algorithm for gait recognitions comparisons, this paper algorithm is better than other algorithms on recognition performance, and aspect consuming time all is better than other several algorithms.Obtained discrimination preferably during normal gait, knapsack changes on the identification impact greatly than overcoat.

Claims (1)

1. based on the gait recognition method of the bidirectional two-dimensional principal component analysis (PCA) of fuzzy theory decision-making, it is characterized in that: comprise the steps:
Step 1, the image in a gait sequence is carried out pre-service: the median method that adopts the nonlinear smoothing technology, the gray-scale value of each pixel is set to the intermediate value of all the pixel gray-scale values adjacent with this point, recover background image from image sequence, with the intermediate value pixel value of image as a setting of the N continuous two field picture pixel value of input;
The profile that difference operation is extracted human motion is carried out in step 2, use indirectly;
Step 3, simply connected movement destination image is carried out the quasi periodic analysis, by analyzing the human body contour shape over time, minimum determine gait cycle to the cyclical variation process of maximum again according to the lower limb profile width of people's body side surface from being up to;
The gait cycle that step 4, basis are determined builds the energygram of each gait cycle, and then calculates the average image of the gait energygram that has complete cycle in gait sequence, obtains an average gait energygram as training sample;
Step 5, according to the method for step 1 to four, build
Figure 201210011983X100001DEST_PATH_IMAGE001
Individual average gait energygram is as training sample
Figure 201210011983X100001DEST_PATH_IMAGE002
, each sample
Figure 201210011983X100001DEST_PATH_IMAGE003
All be divided into the N number of sub images
Figure 201210011983X100001DEST_PATH_IMAGE004
, the subimage of all training image sample correspondence positions forms the subgraph image set
Figure 201210011983X100001DEST_PATH_IMAGE005
Step 6, establish ,
Figure DEST_PATH_IMAGE007
, wherein
Figure 201210011983X100001DEST_PATH_IMAGE008
With
Figure DEST_PATH_IMAGE009
Represent respectively training sample
Figure 201210011983X100001DEST_PATH_IMAGE010
With
Figure DEST_PATH_IMAGE011
iIndividual row vector, The average image that represents all training samples; Utilize covariance matrix
Figure 201210011983X100001DEST_PATH_IMAGE012
To every number of sub images collection
Figure 314901DEST_PATH_IMAGE005
Obtain the optimum projection matrix on line direction
Step 7, establish
Figure 201210011983X100001DEST_PATH_IMAGE014
,
Figure DEST_PATH_IMAGE015
, here
Figure 201210011983X100001DEST_PATH_IMAGE016
With
Figure DEST_PATH_IMAGE017
Expression respectively
Figure 201210011983X100001DEST_PATH_IMAGE018
With
Figure DEST_PATH_IMAGE019
Figure 201210011983X100001DEST_PATH_IMAGE020
Individual column vector is utilized formula
Figure DEST_PATH_IMAGE021
To every number of sub images collection
Figure 201210011983X100001DEST_PATH_IMAGE022
Ask the optimum projection matrix on column direction
Figure DEST_PATH_IMAGE023
Step 8, utilize formula
Figure 201210011983X100001DEST_PATH_IMAGE024
Ask each training image sample
Figure DEST_PATH_IMAGE025
All subimages
Figure 201210011983X100001DEST_PATH_IMAGE026
Eigenmatrix , wherein Expression the iOf individual training sample jNumber of sub images,
Figure 201210011983X100001DEST_PATH_IMAGE028
, MBe the number of training sample,
Figure DEST_PATH_IMAGE029
, NBe the number of subgraph image set,
Figure 201210011983X100001DEST_PATH_IMAGE030
For
Figure DEST_PATH_IMAGE031
Correspondence is not less than the diagonal matrix that 1 eigenwert forms,
Figure 201210011983X100001DEST_PATH_IMAGE032
For less than 1 eigenwert characteristic of correspondence vector,
Figure DEST_PATH_IMAGE033
Be an index between 0-1;
Step 9, to an average gait energygram picture to be identified B, at first be partitioned into each number of sub images , to each self-corresponding projector space projection, try to achieve the eigenmatrix of every number of sub images with every number of sub images
Figure DEST_PATH_IMAGE035
Step 10, the distance between each training sample eigenmatrix in the eigenmatrix that obtains of calculation procedure nine and step 8 then , according to formula Obtain the subimage of sample to be identified
Figure 201210011983X100001DEST_PATH_IMAGE038
Degree of membership to training sample
Figure DEST_PATH_IMAGE039
, wherein,
Figure 201210011983X100001DEST_PATH_IMAGE040
,
Figure DEST_PATH_IMAGE041
Be fuzzy factor;
Step 11, to the degree of membership addition of average each number of sub images of gait energygram to be identified and each training sample subimage summation, obtain image to be identified for the degree of membership of each training sample , according to maximum membership grade principle, last classification results is
Figure DEST_PATH_IMAGE043
, i.e. gait image to be identified and SIndividual training sample belongs to a class together.
CN 201210011983 2012-01-16 2012-01-16 Gait identification method of bidirectional two-dimensional principal component analysis based on fuzzy decision theory Expired - Fee Related CN102592150B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 201210011983 CN102592150B (en) 2012-01-16 2012-01-16 Gait identification method of bidirectional two-dimensional principal component analysis based on fuzzy decision theory

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201210011983 CN102592150B (en) 2012-01-16 2012-01-16 Gait identification method of bidirectional two-dimensional principal component analysis based on fuzzy decision theory

Publications (2)

Publication Number Publication Date
CN102592150A CN102592150A (en) 2012-07-18
CN102592150B true CN102592150B (en) 2013-06-19

Family

ID=46480760

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 201210011983 Expired - Fee Related CN102592150B (en) 2012-01-16 2012-01-16 Gait identification method of bidirectional two-dimensional principal component analysis based on fuzzy decision theory

Country Status (1)

Country Link
CN (1) CN102592150B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102855321B (en) * 2012-09-07 2015-11-18 陕西师范大学 Human bodys' response method
CN103593651B (en) * 2013-10-28 2016-10-05 西京学院 Based on gait and the coal mine down-hole personnel authentication identifying method of two dimension discriminant analysis
CN103577805A (en) * 2013-10-28 2014-02-12 汕头大学 Gender identification method based on continuous gait images
CN107092865A (en) * 2017-03-29 2017-08-25 西北工业大学 A kind of new Gait Recognition system based on Kinect
US10552671B2 (en) 2017-11-22 2020-02-04 King Fahd University Of Petroleum And Minerals Multi-kernel fuzzy local Gabor feature extraction method for automatic gait recognition
CN110569711B (en) * 2019-07-19 2022-07-15 沈阳工业大学 Human body action oriented recognition method
CN113642623B (en) * 2021-08-05 2023-08-18 深圳大学 Complex support vector machine classification method based on unitary space multi-feature fusion

Non-Patent Citations (10)

* Cited by examiner, † Cited by third party
Title
《一种基于平均步态能量图的身份识别算法》;张前进等;《工程图学学报》;20111231(第1期);39-44 *
《基于Radon 变换特征提取的步态识别》;张前进等;《数据采集与处理》;20100331;第25卷(第2期);268-272 *
《基于模糊主成分分析的步态识别》;徐素莉等;《计算机工程》;20110228;第37卷(第3期);192-194 *
《基于步态能量图的隐马尔可夫模型身份识别》;徐素莉等;《微电子学与计算机》;20101231;第27卷(第12期);117-120 *
张前进,徐素莉.《基于嵌入式隐马尔可夫模型的步态识别》.《信息与控制》.2010,第39卷(第1期),25-29. *
张前进等.《一种基于平均步态能量图的身份识别算法》.《工程图学学报》.2011,(第1期),39-44.
张前进等.《基于Radon 变换特征提取的步态识别》.《数据采集与处理》.2010,第25卷(第2期),268-272.
徐素莉等.《基于模糊主成分分析的步态识别》.《计算机工程》.2011,第37卷(第3期),192-194.
徐素莉等.《基于步态能量图的隐马尔可夫模型身份识别》.《微电子学与计算机》.2010,第27卷(第12期),117-120.
陈祥涛，张前进.《基于核主成分分析的步态识别方法》.《计算机应用》.2011,第31卷(第5期),1237-1241. *

Also Published As

Publication number Publication date
CN102592150A (en) 2012-07-18

Similar Documents

Publication Publication Date Title
CN102592150B (en) Gait identification method of bidirectional two-dimensional principal component analysis based on fuzzy decision theory
Keskin et al. Hand pose estimation and hand shape classification using multi-layered randomized decision forests
Williams et al. Advanced image classification using wavelets and convolutional neural networks
Ali et al. Human action recognition in videos using kinematic features and multiple instance learning
Liu et al. Human action recognition using boosted EigenActions
Pazhoumand-Dar et al. Joint movement similarities for robust 3D action recognition using skeletal data
Luo et al. Robust arbitrary view gait recognition based on parametric 3D human body reconstruction and virtual posture synthesis
CN102122354B (en) Adaptive characteristic block selection-based gait identification method
CN104616316B (en) Personage's Activity recognition method based on threshold matrix and Fusion Features vision word
CN107085716A (en) Across the visual angle gait recognition method of confrontation network is generated based on multitask
CN103942577A (en) Identity identification method based on self-established sample library and composite characters in video monitoring
CN101630364A (en) Method for gait information processing and identity identification based on fusion feature
CN104933414A (en) Living body face detection method based on WLD-TOP (Weber Local Descriptor-Three Orthogonal Planes)
CN106778474A (en) 3D human body recognition methods and equipment
CN109784206A (en) A kind of gait recognition method based on the learning machine that transfinites
Gürel et al. Design of a face recognition system
Fan et al. Human gait recognition based on discrete cosine transform and linear discriminant analysis
CN102142083A (en) Face recognition method based on LDA (Linear Discriminant Analysis) subspace learning
Chen et al. Multi-pose face ensemble classification aided by Gabor features and deep belief nets
Li et al. Spatial and temporal information fusion for human action recognition via Center Boundary Balancing Multimodal Classifier
CN113887509B (en) Rapid multi-modal video face recognition method based on image set
Tafazzoli et al. Improving human gait recognition using feature selection
CN107871110A (en) A kind of pedestrian target recognition methods based on Multi-scale model correlation
Rashwan et al. Gait representation and recognition from temporal co-occurrence of flow fields
Guan et al. Research on algorithm of human gait recognition based on sparse representation

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
C17 Cessation of patent right
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20130619

Termination date: 20140116