CN108268854A - A kind of tutor auxiliary platform big data intelligent analysis method of feature based identification - Google Patents
A kind of tutor auxiliary platform big data intelligent analysis method of feature based identification Download PDFInfo
- Publication number
- CN108268854A CN108268854A CN201810106613.1A CN201810106613A CN108268854A CN 108268854 A CN108268854 A CN 108268854A CN 201810106613 A CN201810106613 A CN 201810106613A CN 108268854 A CN108268854 A CN 108268854A
- Authority
- CN
- China
- Prior art keywords
- image
- gradient
- pixel
- feature
- height
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- 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
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of tutor auxiliary platform big data intelligent analysis methods of feature based identification, include the following steps：S1, student's still image is obtained；S2, the direction gradient feature for extracting image；S3, machine learning；S4, sliding window；S5, choose the classification of motion under maximum probability action, identification maneuver；S6, people's progress face recognition is sent out to action in S5.The present invention can extract students ' behavior feature, so as to identify that the behavior of student, action and the earth facilitate the raising managed and be conducive to quality of instruction to student.
Description
Technical field
The present invention relates to a kind of big data intelligent analysis method, the teaching more particularly to a kind of identification of feature based is auxiliary
Help big data intelligent analysis method.
Background technology
Growing with big data analysis technology, people, which start to put forth effort on, to be applied to education by big data technology and leads
Have in domain, such as Publication No. CN107316261A, CN106023013A, CN104573071A Chinese invention patent application
Relevant record.
And on classroom, generally at least tens students, teacher can not observe the dynamic of each student in teaching constantly
Make, consequently, it is possible to reducing the interaction between student and teacher or making teacher that can not find do not listen to the teacher conscientiously in time
It is raw.
Therefore, just seem particularly significant to the behavior of student, motion analysis, but domestic ad hoc analysis not yet at present
Students ' behavior, the technology of action.Therefore applicant proposes a kind of tutor auxiliary platform big data intellectual analysis side of feature based identification
Method can extract students ' behavior feature, so as to identify the action of student, to carry out teaching supervision, improve the quality of teaching.
Invention content
In view of the drawbacks described above of the prior art, the technical problems to be solved by the invention are to provide a kind of feature based
The tutor auxiliary platform big data intelligent analysis method of identification.
To achieve the above object, the present invention provides a kind of tutor auxiliary platform big data intellectual analysis of feature based identification
Method includes the following steps：
S1, student's still image is obtained；
S2, the direction gradient feature for extracting image；The model that the machine learning of S3 trains is judged whether by calculating
Belong to some classification；
S3, machine learning；
S4, sliding window；
S5, choose the classification of motion under maximum probability action, identification maneuver；
S6, people's progress face recognition is sent out to action in S5.
Preferably, the direction gradient feature of S2, taking-up image, includes the following steps：
S21, coloured image are converted into corresponding gray level image, and gray-scale map is regarded as a 3-D view, on image
Each pixel represent that wherein x represents the x coordinate of the pixel in the picture with (x, y, gray), y represents the picture
The y-coordinate of vegetarian refreshments in the picture, gray represent the gray value of the pixel, and conversion formula is as follows：
Gray=0.3*R+0.59*G+0.11*B；
S22, adjust image color contrast, reduce because local light according to uneven generation shade to Gradient Features
Caused by influence, using Gamma correction methods, specific formula for calculation is as follows：
I (x, y)=I (x, y)gamma
Gamma ＞ 1 are taken, then gray scale can be allowed to improve contrast than dark part；
S23, its Gradient Features are calculated to each pixel on image, can extract out here between pixel
Gradient relation, while can also further inhibit influence of the illumination to image, specific formula for calculation is as follows：
The gray value of gray level image respective pixel by being obtained in S22 is represented in table；Since the direction of gray level image has
X directions and y directions, therefore gradient direction needs to calculate the gradient of both direction respectively and gradient is obtained according to amplitude formula
Size：
The gradient magnitude and gradient direction of point (x, y) are obtained according to amplitude formula：
S24, image is divided into several " cell sections " by pixel, to all pixels point in each " cell section "
Gray scale direction counted；
S25, the Gradient Features extracted in " cell factory " are standardized, by several adjacent " cell factories "
The Gradient Features extracted are spliced into the Gradient Features of a large area and then utilize the mode that two norms standardize by gradient
Feature normalization, it is specific as follows：
Assuming that v is the feature vector before standardization, | | v | |2Two norms of feature vector v are represented, asking for formula is：
Feature vector after standardization is：
Wherein, ε is the numerical value of a very little, prevents denominator in formula above from occurring being similar to 0 situation, and ε can be
0.01
S26, it after the gradient eigenvector in all sections being made of " cell section " is standardized, needs institute
There is the gradient eigenvector in section to be spliced, the corresponding gradient eigenvector of composition whole image, the Gradient Features of image
Vector dimension calculation formula is as follows：
Wherein, size is the dimension of the Gradient Features for the image being finally obtained, BlockSize.width and
BlockSize.height is the corresponding width in section and height, and CellSize.width and CellSize.height are " cells
The corresponding width in section " and height, WinSize.width and WinSize.height it is corresponding be the corresponding width of detection image
Degree and height, BlockStride.width and BlockStride.height it is corresponding be section stepping width and height,
nbin
Preferably, S3, machine learning, include the following steps：Sought out in S2 the Gradient Features of each sample image with
Afterwards, the Gradient Features of student's one kind behavior act are trained using SVM, corresponding disaggregated model；Disaggregated model can be lifted
Hand, read, write constant gradient characteristic model；
S31, user mark some relevant retrieval results as positive sample and some incoherent negative samples of retrieval first
This, the result that then image indexing system is retrieved based on the update of these feedback samples；Alternately above-mentioned two step alternating into
Row can allow computer gradually to learn the demand of user, enhance search function；
S311, image gradient features space is divided into multiple regions, wherein, the specific farthest positive sample of plane of classifying is made
For most correlated samples, sample of the nearest sample of distance classification plane as most information content；
The association feedback method of S312, image indexing system based on SVM：SVM be a kind of effective two classification, more points
Class algorithm, below by taking two classification problems as an example (And yi={+1, -1 }), more classification problems can be with analogy；
Wherein, xiIt is a n dimensional feature vector, yiIt is the classification information belonging to feature vector,
SVM is divided two class sample points by a hyperplane, i.e.,：
wTX+b=0
Wherein, w is an input vector, and x is an adaptive weighting vector, and b is a biasing, and SVM passes through maximization
Geometrical boundaryThe parameter w and b of optimal hyperlane are obtained, and is met：
yi(wTxi+b)≥+1
By introducing Lagrange coefficient, solving dual problem can obtain：
And at the same time meet：
αi≥0
In antithesis pattern, only there is inner product form in data point, can be by these in order to obtain the more preferable expression of data
Data point is mapped to by replacement operation in the Xi Baite inner product spaces：
xi·xj→φ(xi)·φ(xj)=K (xi,xj)
Wherein, K () is one and function；
Can obtain dual problem and function representation form：
For any given and function, SVM classifier can be expressed as：
F (x)=sgn (f (x))
Wherein, f (x) is the decision function of svm classifier hyperplane：
For given image sample, when | f (x) | it is corresponding to predict that certainty factor is higher when higher, work as on the contrary | f (x) | more
It is corresponding to predict that certainty factor is lower when low.
Preferably, S4, sliding window, include the following steps：
S41, selected digital image region, area-of-interest is chosen to be by image；
S42, selected area-of-interest equal proportion is zoomed into feature detection window size；
S43, the image for meeting size is asked for into HOG features；
S44, by the HOG sought out with being compared previously by the trained models of SVM, to region to be measured into action
Work is classified；
S45, selected image-region move to right a step-length, until sliding window is moved to the right end of image one's own profession；
S46, selected image-region move down a step-length until sliding window is moved to last column of image；
S47, the size in amplification selected digital image region；
S48, S41 to S47 is repeated, you can the region of traversal all sizes of current image.
Preferably, S5, choose the action of maximum probability under the classification of motion, identification maneuver includes the following steps：
S51, traverse sliding window all detection images and establish image pyramid；
S52, current all target windows containing current action classification are collected, and counts and obtain the corresponding window of highest
W；
S53, all windows for having obvious overlapping with window W are eliminated；
S54, next window for having highest response certainty factor is moved to, S51-S53 is repeated under current scale；
S55, after the completion of this process, the pyramidal next scale of mobile image, and repeat S51-S55 process.
Preferably, S6, in S5 action send out people carry out face recognition, include the following steps：
S61, the LBP feature extraction modes using equivalent formulations, by jump of the LBP coding modes " equivalence " between 0 and 1
Change problem；
S62, in the design process of system, transition times in binary coding less than 2 times LBP coding uniformly return
Class is an equivalent formulations class, and improved procedure in this way, binary-coded type can greatly reduce, and will not lose
It breaks one's promise breath, binary group/cording quantity is also from original 2p+ 2 kinds of p (p-1) is reduced to, wherein, p represents the sampling in neighborhood
The number of pixel.
The beneficial effects of the invention are as follows：The present invention can recognize that the behavior, action and the earth of student is facilitated to learning
The raw raising managed and be conducive to quality of instruction.
Description of the drawings
Fig. 1 is cell interval division schematic diagram of the present invention.
Fig. 2 is LBP conversion regime schematic diagrames.
Specific embodiment
The invention will be further described with reference to the accompanying drawings and examples：
Referring to Fig. 1-Fig. 2, a kind of tutor auxiliary platform big data intelligent analysis method of feature based identification, including walking as follows
Suddenly：
S1, student's still image is obtained, student's still image is obtained using image acquisition equipment, such as camera；
S2, the direction gradient feature for extracting image；
S3, machine learning；
S4, sliding window；
S5, choose the classification of motion under maximum probability action, identification maneuver；
S6, people's progress face recognition is sent out to action in S5.
Further, the direction gradient feature of S2, taking-up image, includes the following steps：
S21, because color is little to the contribution degree of Gradient Features, for the efficiency subsequently calculated, by colored triple channel
(RGB) it is converted into corresponding gray level image, and gray-scale map is regarded as a 3-D view, each pixel on image
(x, y, gray) can be used to represent, wherein x represents the x coordinate of the pixel in the picture, and y represents the pixel in image
In y-coordinate, gray represents the gray value (0 to 255) of the pixel, and conversion formula is as follows：
Gray=0.3*R+0.59*G+0.11*B；
S22, adjust image color contrast, reduce because local light according to uneven generation shade to Gradient Features
Caused by influence, the step can inhibit influence of the high-frequency noise to image gradient features simultaneously, and the method mainly used is
Gamma correction methods, specific formula for calculation are as follows：
I (x, y)=I (x, y)gamma
Gamma ＞ 1 are taken, then gray scale can be allowed to improve contrast than dark part；
S23, its Gradient Features are calculated (including gradient magnitude and gradient side to each pixel on image
To), the step can extract the gradient relation between pixel, while can further inhibit illumination to image
It influences, specific formula for calculation is as follows：
| H(x-1,y-1) | H(x,y-1) | H(x+1,y-1) |
| H(x-1,y) | H(x,y) | H(x+1,y) |
| H(x-1,y+1) | H(x,y+1) | H(x+1,y+1) |
The gray value of gray level image respective pixel by being obtained in S22 is represented in table, such as：H (x, y) is represented in gray scale
Gray value at (x, y) of figure is H (x, y)；Since the direction of gray level image is there are two (x directions and y directions), gradient
Direction needs to calculate the gradient of both direction respectively and gradient magnitude is obtained according to amplitude formula：
The gradient magnitude and gradient direction of point (x, y) are obtained according to amplitude formula：
S24, image is divided into several " cell sections " by pixel, to all pixels point in each " cell section "
Gray scale direction counted.
In the present invention, the size in each " cell section " is 8 pixel *, 8 pixels, to gradient direction using every 20 degree works
For interval division, weights when gradient magnitude is as statistics；
Assuming that gradient magnitude G (x, y)=2, α (x, y)=137.6 ° of a certain pixel (x, y) in " cell factory ", then
This pixel region z6 should be increased into 2 statistics in statistics；
S25, the Gradient Features extracted in " cell factory " are standardized, because in image difference " cell factory "
Intensity of illumination may be influenced to be different by intensity of illumination, sometimes graded can be very big, in the present invention,
The Gradient Features of a large area are spliced into using the Gradient Features for extracting several adjacent " cell factories "
Then Gradient Features are standardized using the mode of two norms standardization, concrete principle is as follows：
Assuming that v is the feature vector before standardization, | | v | |2Two norms of feature vector v are represented, asking for formula is：
Feature vector after standardization is：
Wherein, ε is the numerical value of a very little, prevents denominator in formula above from occurring being similar to 0 situation；
S26, it after the gradient eigenvector in all sections being made of " cell section " is standardized, needs institute
There is the gradient eigenvector in section to be spliced, the corresponding gradient eigenvector of composition whole image, the Gradient Features of image
Vector dimension calculation formula is as follows：
Wherein, size is the dimension of the Gradient Features for the image being finally obtained, BlockSize.width and
BlockSize.height is the corresponding width in section and height, and CellSize.width and CellSize.height are " cells
The corresponding width in section " and height, WinSize.width and WinSize.height it is corresponding be the corresponding width of detection image
Degree and height, BlockStride.width and BlockStride.height it is corresponding be section stepping width and height,
What nbin was represented is the regional channel number used in gradient angle statistics in Step4；
In the present invention, CellSize.width=CellSize.height=8 is taken,
BlockSize.width=BlockSize.height=16,
BlockStride.width=BlockStride.height=8,
WinSize.width=64, BlockStride.height=128,
Nbin=9.
S3, machine learning after seeking out the Gradient Features of each sample image, need the side using machine learning
The Gradient Features of student's one kind behavior act are trained by method, corresponding disaggregated model；
The training method that the present invention uses is SVM (Support Vector Machine, support vector machines), this is one
Kind has the statistical learning algorithm of supervision, can minimize experience error and maximize Geometry edge.
The present invention is based on the image detecting systems of content, mainly include the following steps：
User marks some relevant retrieval results and retrieves incoherent negative samples as positive sample and some first,
Then result of the image indexing system based on the update retrieval of these feedback samples；Alternately above-mentioned two step alternately, can
Computer to be allowed gradually to learn the demand of user, enhance search function；It is specific as follows：
Image gradient features space is divided into multiple regions, wherein, the specific farthest positive sample of plane of classifying is used as most phase
Close sample, sample of the nearest sample of distance classification plane as most information content；
The association feedback method of image indexing system based on SVM：SVM is a kind of effective two classification, more classification calculations
Method, below by taking two classification problems as an example (And yi={+1, -1 }), more classification problems can be with analogy；
Wherein, xiIt is a n dimensional feature vector, yiIt is the classification information belonging to feature vector,
SVM is divided two class sample points by a hyperplane, i.e.,：
wTX+b=0
Wherein, w is an input vector, and x is an adaptive weighting vector, and b is a biasing.SVM passes through maximization
Geometrical boundaryThe parameter w and b of optimal hyperlane are obtained, and is met：
yi(wTxi+b)≥+1
By introducing Lagrange coefficient, solving dual problem can obtain：
And at the same time meet：
αi≥0
In antithesis pattern, only there is inner product form in data point, can be by these in order to obtain the more preferable expression of data
Data point is mapped to by replacement operation in the Xi Baite inner product spaces：
xi·xj→φ(xi)·φ(xj)=K (xi,xj)
Wherein, K () is one and function；
Can obtain dual problem and function representation form：
For any given and function, SVM classifier can be expressed as：
F (x)=sgn (f (x))
Wherein, f (x) is the decision function of svm classifier hyperplane：
For given image sample, when | f (x) | it is corresponding to predict that certainty factor is higher when higher, work as on the contrary | f (x) | more
It is corresponding to predict that certainty factor is lower when low.
The present invention can detect the behavior act of fixed area student, moreover it is possible to implement function such as：It is same in detection image
Multiple targets of kind action determine to detect position of these targets of these actions in present image.Accomplish this two
Point needs the method using sliding window.Sliding window is a kind of technology for computer vision, it includes will in image
The inspection of movable part (sliding window) and each section is detected using image pyramid.This is in order to multiple dimensioned
Lower detection object.
Sliding window solves positioning texts and pictures by scanning the smaller area of larger image, and then in the difference of same image
Multiple scanning under scale.This technology needs each image resolving into multiple portions, and then losing those is less likely to include
The part of object, and classify to remainder.
The core thinking of sliding window is mainly examined for an area-of-interest and scale to H features every time
It surveys window size and then performs corresponding feature extraction and classification；
S4, sliding window, include the following steps：
S41, selected digital image region, area-of-interest is chosen to be by image；
S42, selected area-of-interest equal proportion is zoomed into feature detection window size；
S43, the image for meeting size is asked for into HOG features；
S44, by the HOG sought out with being compared previously by the trained models of SVM, to region to be measured into action
Work is classified；
S45, selected image-region move to right a step-length, until sliding window is moved to the right end of image one's own profession；
S46, selected image-region move down a step-length until sliding window is moved to last column of image；
S47, the size in amplification selected digital image region.
S48, S41 to S47 is repeated, you can the region of traversal all sizes of current image.
The problem of method of sliding window can bring region to be overlapped, region overlapping are referred to image execution action point
The adjacent sliding window of same action has the phenomenon that intersection when class judges.Each window can lose several pixels every time,
This means that a sliding window can be matched with the different location of the same action of same student, therefore, it is desirable to be made
It is screened with non-maximum suppression algorithm to overlapping image, non-maximum suppression algorithm can use one group of given overlapping region
Maximum scores or maximum region are filtered out the region that can most represent out student's operating position.
Non-maxima suppression (NMS) non-maxima suppression as the term suggests be exactly inhibit be not maximum element, search office
The maximum in portion.What this was locally represented is a neighborhood, and neighborhood is there are two changeable parameters, when the dimension of neighborhood, second is that adjacent
The size in domain.General NMS algorithms are not discussed here, but for extracting the highest window of score in target detection
's.Such as in pedestrian detection, the extracted feature of sliding window, after categorized device Classification and Identification, each window can obtain one
A score.But sliding window can cause many windows to there is a situation where comprising with other windows or largely intersect.At this moment
It just needs to use NMS to choose score highest in those neighborhoods (being the maximum probability under the classification of motion), and inhibit those
The low window of score.
S5, choose the classification of motion under maximum probability action, specific steps are as follows：
S51, traverse sliding window all detection images and establish image pyramid；
S52, current all target windows (certainty factor is more than certain threshold value) containing current action classification are collected, and
Statistics obtains the corresponding window W of highest；
S53, all windows for having obvious overlapping (overlapping region area percentage is more than certain threshold value) with window W are eliminated；
S54, next window for having highest response certainty factor is moved to, is repeated the above process under current scale.
S55, after the completion of this process, the pyramidal next scale of mobile image, and repeat S51-S55 process.
Preferably, S6, face recognition include the following steps：
After student's action is determined, needing through the method judgement action of recognition of face is made by which student
, human face detection and tracing technology also needs the method for using feature extraction and machine learning, in the face that the present invention uses
It is characterized in LBP (Local Binary Pattern, Chinese are local binary patterns) features and Adaboost training methods.
Original LBP character description methods and computational methods, include the following steps：
For any one pixel (x, y) selected on image, its gray value is denoted as H (x, y), is with H (x, y)
The gray value of other 8 pixels in threshold value and its neighbouring 3*3 region is compared, if the gray value of adjacent pixel
More than H (x, y), then labeled as 1；If the gray value of adjacent pixel is less than H (x, y), labeled as 0, finally by adjacent pixels
Binary value be encoded to decimal number, formula is as follows：
Wherein, LBP (xc,yc) that represent is pixel (xc,yc) at LBP characteristic values, icThat represent is pixel (xc,
yc) at gray value, ipThat represent is pixel (xp,yp) at gray value, pixel (xp,yp) in pixel (xc,yc)
Adjacent domain, function s () is comparison function, specific as follows；
In this way, LBP (xc,yc) it can reflect pixel (xc,yc) at LBP features.
Uniform Pattern LBP features (also referred to as equivalent formulations or uniform pattern), it is special for original LBP
Sign extraction is to increase in the size exponentially grade with LBP zonings to the time of calculation amount and memory space, such as：3*3's
The binary coding range in region is 0-28, the binary coding range in the region of 4*4 is 0-215, the region of 32*32 two into
Coding range processed is 0-21023.Binary coding mode huge in this way is unfavorable for the extraction, storage and subsequent machine of feature
Device learns.
In the present invention, the LBP feature extraction modes of equivalent formulations have been used so that original LBP features are dropped
Dimension does not disable in the case of making data volume reduction while and best goes back the LBP features of original image.It is described above in order to solve
The problem of binary coded patterns are more, the present invention use the LBP feature extraction modes of equivalent formulations, include the following steps：
By jump problem of the LBP coding modes " equivalence " between 0 and 1, such as：00000000 (0 saltus step),
11111111 (0 saltus step), 10010111 (4 saltus steps).
In the design process of system, LBP coding of the transition times in binary coding less than 2 times is uniformly classified as
One equivalent formulations class, improved procedure in this way, binary-coded type can greatly reduce, and will not lose letter
Breath, binary group/cording quantity is also from original 2p+ 2 kinds of p (p-1) is reduced to, wherein, p represents the sampled pixel in neighborhood
The number of point.
Such as：The More General Form binary coding range in the region of 3*3 is 0-58, the binary coding range in the region of 4*4
It is 0-212, the binary coding range in the region of 32*32 is 0-932.
Part is not described in detail by the present invention, is the known technology of those skilled in the art.
The preferred embodiment of the present invention described in detail above.It should be appreciated that those of ordinary skill in the art without
Creative work is needed according to the present invention can to conceive and makes many modifications and variations.Therefore, all technologies in the art
Personnel can be obtained under this invention's idea by logical analysis, reasoning, or a limited experiment on the basis of existing technology
Technical solution, all should be in the protection domain being defined in the patent claims.
Claims (8)
1. a kind of tutor auxiliary platform big data intelligent analysis method of feature based identification, which is characterized in that include the following steps：
S1, student's still image is obtained；
S2, the direction gradient feature for extracting image；
S3, machine learning；
S4, sliding window；
S5, choose the classification of motion under maximum probability action, identification maneuver；
S6, people's progress face recognition is sent out to action in S5.
2. the tutor auxiliary platform big data intelligent analysis method of feature based identification as described in claim 1, which is characterized in that
S2, the direction gradient feature for taking out image, include the following steps：
S21, coloured image are converted into corresponding gray level image, and gray-scale map is regarded as a 3-D view, every on image
One pixel all represents that wherein x represents the x coordinate of the pixel in the picture, and y represents the pixel and exists with (x, y, gray)
Y-coordinate in image, gray represent the gray value of the pixel, and conversion formula is as follows：
Gray=0.3*R+0.59*G+0.11*B；
S22, adjust image color contrast, reduce because local light according to uneven generation shade caused by Gradient Features
It influences, using Gamma correction methods, specific formula for calculation is as follows：
I (x, y)=I (x, y)gamma
Gamma ＞ 1 are taken, then gray scale can be allowed to improve contrast than dark part；
S23, its Gradient Features are calculated to each pixel on image, the gradient that can extract out here between pixel is closed
System, while can also further inhibit influence of the illumination to image, specific formula for calculation is as follows：
The gray value of gray level image respective pixel by being obtained in S22 is represented in table；Since there are x directions in the direction of gray level image
With y directions, therefore gradient direction needs to calculate the gradient of both direction respectively and gradient magnitude is obtained according to amplitude formula：
The gradient magnitude and gradient direction of point (x, y) are obtained according to amplitude formula：
S24, image is divided into several " cell sections " by pixel, to the gray scale of all pixels point in each " cell section "
Direction is counted；
S25, the Gradient Features extracted in " cell factory " are standardized, several adjacent " cell factories " is extracted
Gradient Features be spliced into a large area Gradient Features then using the standardization of two norms mode by Gradient Features mark
Standardization, it is specific as follows：
Assuming that v is the feature vector before standardization, | | v | |2Two norms of feature vector v are represented, asking for formula is：
Feature vector after standardization is：
Wherein, ε is the numerical value of a very little, prevents denominator in formula above from occurring being similar to 0 situation；
S26, it after the gradient eigenvector in all sections being made of " cell section " is standardized, needs all sections
In gradient eigenvector spliced, the corresponding gradient eigenvector of composition whole image, the gradient eigenvector dimension of image
It is as follows to spend calculation formula：
Wherein, size is the dimension of the Gradient Features for the image being finally obtained, BlockSize.width and
BlockSize.height is the corresponding width in section and height, and CellSize.width and CellSize.height are " cells
The corresponding width in section " and height, WinSize.width and WinSize.height it is corresponding be the corresponding width of detection image
And height, BlockStride.width and BlockStride.height it is corresponding be section stepping width and height, nbin
What is represented is the regional channel number used in gradient angle statistics in S24.
3. the tutor auxiliary platform big data intelligent analysis method of feature based identification as claimed in claim 2, which is characterized in that
In S24, CellSize.width=CellSize.height=8 is taken；
BlockSize.width=BlockSize.height=16；
BlockStride.width=BlockStride.height=8
Win S.ize w=i 6dt4, h BlockStride.height=128, nbin=9.
4. the tutor auxiliary platform big data intelligent analysis method of feature based identification as claimed in claim 2, which is characterized in that
In S26, the size in each " cell section " is 8 pixel *, 8 pixels, to gradient direction using every 20 degree as interval division, gradient
Weights when size is as statistics；
Assuming that gradient magnitude G (x, y)=2, α (x, y)=137.6 ° of a certain pixel (x, y) in " cell factory ", then should
This pixel region z6 is increased into 2 statistics in statistics.
5. the tutor auxiliary platform big data intelligent analysis method of feature based identification as described in claim 1, which is characterized in that
S3, machine learning, include the following steps：It, will using SVM algorithm after seeking out the Gradient Features of each sample image in S2
The Gradient Features of student's one kind behavior act are trained, corresponding disaggregated model；
S31, user mark first some relevant retrieval results as positive sample and some retrieve incoherent negative samples, so
Result of the image indexing system based on the update retrieval of these feedback samples afterwards；Alternately above-mentioned two step alternately, can allow
Computer gradually learns the demand of user, enhances search function；
S311, image gradient features space is divided into multiple regions, wherein, the specific farthest positive sample of plane of classifying is used as most phase
Close sample, sample of the nearest sample of distance classification plane as most information content；
The association feedback method of S312, image indexing system based on SVM：SVM is a kind of effective two classification, more classification calculations
Method, below by taking two classification problems as an example (And yi={+1, -1 }), more classification problems can be with analogy；
Wherein, xiIt is a n dimensional feature vector, yiIt is the classification information belonging to feature vector,
SVM is divided two class sample points by a hyperplane, i.e.,：
wTX+b=0
Wherein, w is an input vector, and x is an adaptive weighting vector, and b is a biasing, and SVM is by maximizing geometry
BoundaryThe parameter w and b of optimal hyperlane are obtained, and is met：
yi(wTxi+b)≥+1
By introducing Lagrange coefficient, solving dual problem can obtain：
And at the same time meet：
αi≥0
In antithesis pattern, only there is inner product form in data point, can be by these data points in order to obtain the more preferable expression of data
It is mapped in the Xi Baite inner product spaces by replacement operation：
xi·xj→φ(xi)·φ(xj)=K (xi,xj)
Wherein, K () is one and function；
Can obtain dual problem and function representation form：
For any given and function, SVM classifier can be expressed as：
F (x)=sgn (f (x))
Wherein, f (x) is the decision function of svm classifier hyperplane：
For given image sample, as | f (x) | it is corresponding to predict that certainty factor is higher when higher, opposite as | f (x) | when lower,
Corresponding prediction certainty factor is lower.
6. the tutor auxiliary platform big data intelligent analysis method of feature based identification as described in claim 1, which is characterized in that
S4, sliding window, include the following steps：
S41, selected digital image region, area-of-interest is chosen to be by image；
S42, selected area-of-interest equal proportion is zoomed into feature detection window size；
S43, the image for meeting size is asked for into HOG features；
S44, by the HOG sought out with being compared previously by the trained models of SVM, to region to be measured carry out action point
Class；
S45, selected image-region move to right a step-length, until sliding window is moved to the right end of image one's own profession；
S46, selected image-region move down a step-length until sliding window is moved to last column of image；
S47, the size in amplification selected digital image region；
S48, S41 to S47 is repeated, you can the region of traversal all sizes of current image.
7. the tutor auxiliary platform big data intelligent analysis method of feature based identification as described in claim 1, which is characterized in that
S5, the action for choosing maximum probability under the classification of motion, identification maneuver includes the following steps：
S51, traverse sliding window all detection images and establish image pyramid；
S52, current all target windows containing current action classification are collected, and counts and obtain the corresponding window W of highest；
S53, all windows for having obvious overlapping with window W are eliminated；
S54, next window for having highest response certainty factor is moved to, S51-S53 is repeated under current scale；
S55, after the completion of this process, the pyramidal next scale of mobile image, and repeat S51-S55 process.
8. the tutor auxiliary platform big data intelligent analysis method of feature based identification as described in claim 1, which is characterized in that
S6, people's progress face recognition is sent out to action in S5, included the following steps：
S61, the LBP feature extraction modes using equivalent formulations, saltus step of the LBP coding modes " equivalence " between 0 and 1 is asked
Topic；
S62, in the design process of system, transition times in binary coding less than 2 times LBP coding be uniformly classified as one
A equivalent formulations class, improved procedure in this way, binary-coded type can greatly reduce, and will not lose information,
Binary group/cording quantity is also from original 2p+ 2 kinds of p (p-1) is reduced to, wherein, p represents the sampling pixel points in neighborhood
Number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810106613.1A CN108268854B (en) | 2018-02-02 | 2018-02-02 | Teaching assistance big data intelligent analysis method based on feature recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810106613.1A CN108268854B (en) | 2018-02-02 | 2018-02-02 | Teaching assistance big data intelligent analysis method based on feature recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108268854A true CN108268854A (en) | 2018-07-10 |
| CN108268854B CN108268854B (en) | 2022-06-10 |
Family
ID=62773516
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810106613.1A Active CN108268854B (en) | 2018-02-02 | 2018-02-02 | Teaching assistance big data intelligent analysis method based on feature recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108268854B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110031461A (en) * | 2019-02-14 | 2019-07-19 | 江苏恒力化纤股份有限公司 | A kind of polyester filament dye uniformity test method |
| CN112613342A (en) * | 2020-11-27 | 2021-04-06 | 深圳市捷视飞通科技股份有限公司 | Behavior analysis method and apparatus, computer device, and storage medium |
| CN116993643A (en) * | 2023-09-27 | 2023-11-03 | 山东建筑大学 | Unmanned aerial vehicle photogrammetry image correction method based on artificial intelligence |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102243687A (en) * | 2011-04-22 | 2011-11-16 | 安徽寰智信息科技股份有限公司 | Physical education teaching auxiliary system based on motion identification technology and implementation method of physical education teaching auxiliary system |
| US20150363644A1 (en) * | 2014-06-17 | 2015-12-17 | Nantworks, LLC | Activity recognition systems and methods |
| US9230159B1 (en) * | 2013-12-09 | 2016-01-05 | Google Inc. | Action recognition and detection on videos |
| CN105913025A (en) * | 2016-04-12 | 2016-08-31 | 湖北工业大学 | Deep learning face identification method based on multiple-characteristic fusion |
| CN107085721A (en) * | 2017-06-26 | 2017-08-22 | 厦门劢联科技有限公司 | A kind of intelligence based on Identification of Images patrols class management system |
| CN107609517A (en) * | 2017-09-15 | 2018-01-19 | 华中科技大学 | A kind of classroom behavior detecting system based on computer vision |
-
2018
- 2018-02-02 CN CN201810106613.1A patent/CN108268854B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102243687A (en) * | 2011-04-22 | 2011-11-16 | 安徽寰智信息科技股份有限公司 | Physical education teaching auxiliary system based on motion identification technology and implementation method of physical education teaching auxiliary system |
| US9230159B1 (en) * | 2013-12-09 | 2016-01-05 | Google Inc. | Action recognition and detection on videos |
| US20150363644A1 (en) * | 2014-06-17 | 2015-12-17 | Nantworks, LLC | Activity recognition systems and methods |
| CN105913025A (en) * | 2016-04-12 | 2016-08-31 | 湖北工业大学 | Deep learning face identification method based on multiple-characteristic fusion |
| CN107085721A (en) * | 2017-06-26 | 2017-08-22 | 厦门劢联科技有限公司 | A kind of intelligence based on Identification of Images patrols class management system |
| CN107609517A (en) * | 2017-09-15 | 2018-01-19 | 华中科技大学 | A kind of classroom behavior detecting system based on computer vision |
Non-Patent Citations (3)
| Title |
|---|
| VINA AYUMI 等: "Pose-based Human Action Recognition with Extreme Gradient Boosting", 《 2016 IEEE STUDENT CONFERENCE ON RESEARCH AND DEVELOPMENT (SCORED)》 * |
| 文楷 等: "基于视频分析的课堂教学效果自动评价系统", 《技术在线》 * |
| 王克耀: "基于视频的人体运动行为分析", 《万方》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110031461A (en) * | 2019-02-14 | 2019-07-19 | 江苏恒力化纤股份有限公司 | A kind of polyester filament dye uniformity test method |
| CN110031461B (en) * | 2019-02-14 | 2022-03-18 | 江苏恒力化纤股份有限公司 | Polyester filament yarn dyeing uniformity test method |
| CN112613342A (en) * | 2020-11-27 | 2021-04-06 | 深圳市捷视飞通科技股份有限公司 | Behavior analysis method and apparatus, computer device, and storage medium |
| CN116993643A (en) * | 2023-09-27 | 2023-11-03 | 山东建筑大学 | Unmanned aerial vehicle photogrammetry image correction method based on artificial intelligence |
| CN116993643B (en) * | 2023-09-27 | 2023-12-12 | 山东建筑大学 | Unmanned aerial vehicle photogrammetry image correction method based on artificial intelligence |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108268854B (en) | 2022-06-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109147254B (en) | Video field fire smoke real-time detection method based on convolutional neural network | |
| CN110929774B (en) | Classification method, model training method and device for target objects in image | |
| CN112734775B (en) | Image labeling, image semantic segmentation and model training methods and devices | |
| CN102043945B (en) | License plate character recognition method based on real-time vehicle tracking and binary index classification | |
| CN106407903A (en) | Multiple dimensioned convolution neural network-based real time human body abnormal behavior identification method | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN107871101A (en) | A kind of method for detecting human face and device | |
| CN112307919B (en) | Improved YOLOv 3-based digital information area identification method in document image | |
| CN109886269A (en) | A kind of transit advertising board recognition methods based on attention mechanism | |
| CN108268854A (en) | A kind of tutor auxiliary platform big data intelligent analysis method of feature based identification | |
| CN107633226A (en) | A kind of human action Tracking Recognition method and system | |
| CN109002752A (en) | A kind of complicated common scene rapid pedestrian detection method based on deep learning | |
| CN106373146A (en) | Target tracking method based on fuzzy learning | |
| CN108256462A (en) | A kind of demographic method in market monitor video | |
| CN105976397B (en) | A kind of method for tracking target | |
| CN107038416A (en) | A kind of pedestrian detection method based on bianry image modified HOG features | |
| CN107424175B (en) | Target tracking method combined with space-time context information | |
| CN112232371A (en) | American license plate recognition method based on YOLOv3 and text recognition | |
| CN103810500A (en) | Place image recognition method based on supervised learning probability topic model | |
| CN109800756A (en) | A kind of text detection recognition methods for the intensive text of Chinese historical document | |
| CN115936944B (en) | Virtual teaching management method and device based on artificial intelligence | |
| CN110414616A (en) | A kind of remote sensing images dictionary learning classification method using spatial relationship | |
| CN102708384A (en) | Bootstrapping weak learning method based on random fern and classifier thereof | |
| CN113657414B (en) | Object identification method | |
| CN111507353A (en) | Chinese field detection method and system based on character recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |