CN108960099A - Face tilts angle estimating method, system, equipment and storage medium - Google Patents

Face tilts angle estimating method, system, equipment and storage medium Download PDF

Info

Publication number
CN108960099A
CN108960099A CN201810653661.2A CN201810653661A CN108960099A CN 108960099 A CN108960099 A CN 108960099A CN 201810653661 A CN201810653661 A CN 201810653661A CN 108960099 A CN108960099 A CN 108960099A
Authority
CN
China
Prior art keywords
image
mentioned
relative difference
angle
face
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
Application number
CN201810653661.2A
Other languages
Chinese (zh)
Other versions
CN108960099B (en
Inventor
徐勇
刘宏
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.)
Peking University Shenzhen Graduate School
Shenzhen Graduate School Harbin Institute of Technology
Original Assignee
Peking University Shenzhen Graduate School
Shenzhen Graduate School Harbin Institute of 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 Peking University Shenzhen Graduate School, Shenzhen Graduate School Harbin Institute of Technology filed Critical Peking University Shenzhen Graduate School
Priority to CN201810653661.2A priority Critical patent/CN108960099B/en
Publication of CN108960099A publication Critical patent/CN108960099A/en
Application granted granted Critical
Publication of CN108960099B publication Critical patent/CN108960099B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06V40/161—Detection; Localisation; Normalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

It tilts estimation method, system, equipment and the storage medium of angle the invention discloses a kind of face, comprising the following steps: facial image is divided into the first image and the second image with specific mode；Calculate the relative difference of the pixel value between the first image and the second image；Corresponding face is calculated according to the relative difference to tilt angle.What face of the invention tilted the estimation method of angle, system, equipment and storage medium has the beneficial effect that calculating corresponding face by the relative difference according to pixel value tilts angle, the step of simplifying the estimation of face tilt angle, improves face and tilts and angle calculation speed and efficiently avoid the influence of illumination variation.

Description

Face tilts angle estimating method, system, equipment and storage medium
Technical field
It tilts angle the present invention relates to pattern-recognition and technical field of computer vision more particularly to a kind of face Estimation method, system, equipment and storage medium.
Background technique
Face tilt angle estimation have in fields such as recognition of face, video tracking, fatigue detecting, human-computer interactions it is important Application value.Some recognitions of face are to require to obtain positive facial image, and camera is then needed according to face inclination angle at this time Degree is adjusted self-view to obtain positive facial image.If video tracking can obtain face tilt angle parameter, just The posture that video camera can be dynamically adjusted is allowed to always in the optimal observation position of monitored object.In addition, utilizing face Tilt angle parameter can also make multiple surveillance cameras coordinate over time and space, realize the company to monitored object Continuous tracking.A major issue in intelligent human-machine interaction research is the coke for requiring accurately to judge people's at a time attention Point, to make computer more fully understand the behavior of people and make corresponding reaction.Driver is likely to occur during driving Fatigue may cause thing when situation is serious, be estimated by face tilt angle, we it can be found that driver there may be fatigue simultaneously It sounds an alarm in time, so as to avoid the generation of accident.
Currently, most of the technology of estimation face tilt angle is estimated whole angle unified, this kind of skill The operation method of art is complicated, calculates complex steps, causes its estimating speed slow.
Summary of the invention
The main object of the present invention be provide it is a kind of based on tilting angle to the face of image pixel variance analysis Estimation method, system, equipment and storage medium are tilted the estimated efficiency of angle with improving face.
The present invention provides the estimation method that a kind of face tilts angle, comprising the following steps:
Facial image is divided into the first image and the second image with specific mode；
Calculate the relative difference of the pixel value between above-mentioned first image and the second image；
Corresponding face is calculated according to above-mentioned relative difference to tilt angle.
Further, above-mentioned the step of facial image is divided into the first image and the second image with specific mode it Before, further includes:
Judge above-mentioned facial image whether vertical tilt；
If so, being calibrated by affine transformation method to facial image.
Further, above-mentioned to judge whether vertical tilt step includes: above-mentioned facial image
It is obtained in above-mentioned facial image respectively close to the location point at two canthus of nose；
To above-mentioned location point line, and judge line segment whether with horizontal line there are tilt angles.
Further, above-mentioned facial image to be divided into the first image with specific mode and the second image step includes:
The line number m and columns n of facial image matrix are obtained, and judges whether n is even number；
If so, above-mentioned facial image is divided intoAbove-mentioned first image of size and the second image；
If it is not, then giving up first or last column of above-mentioned facial image matrix, and then above-mentioned facial image will be divided equally ForAbove-mentioned first image of size and the second image.
Further, the relative difference step of the pixel value between above-mentioned first image of calculating and the second image includes:
The first image and the second image are divided into several image blocks respectively, wherein the pixel that each image block includes Point quantity is identical；
Each image block in the image array of the image array to above-mentioned first image and the second image is marked respectively Number, and each pixel in above-mentioned image block carries out label；
Calculate the relative difference that the pixel value of each pixel of label is corresponded in above-mentioned first image and the second image；
The phase of the pixel value of the corresponding above-mentioned image block of label is calculated according to the relative difference of the pixel value of each pixel To difference value；
Above-mentioned first image and the second image are calculated according to the above-mentioned relative difference of the pixel value of each above-mentioned image block Pixel value relative difference.
Further, above-mentioned the corresponding face angle step that tilts is calculated according to above-mentioned relative difference to include:
Obtain the above-mentioned relative difference of the pixel value of above-mentioned first image and the second image；
It is left that corresponding face is calculated according to the above-mentioned relative difference of above-mentioned first image and the pixel value of the second image Right deviation rake angle.
Further, above-mentioned the step of facial image is divided into the first image and the second image with specific mode it Before, further includes:
Tilt angle calculation formula is established, step includes:
The relative difference of K history facial image is obtained, and is successively denoted as D1......DK, order matrix
The angle that tilts of K above-mentioned history facial images is obtained, and is successively denoted as α1......αK, order matrix
Equation group Pg=α is established, and calculates the linear relationship between P and α, is solvedFor appointing The unknown facial image of tilt angle of anticipating, calculates its relative difference D firsth, then pass through formulaCalculate it Inclination angle alphah；
Wherein, γ, I are respectively a small positive number and unit matrix, the transposition operation of T representing matrix.
The present invention also proposes that a kind of face tilts the estimating system of angle, comprising:
Facial image divides module, for facial image to be divided into the first image and the second image with specific mode；
First computing module, for calculating the relative difference of the pixel value between above-mentioned first image and the second image；
Second computing module tilts angle for calculating corresponding face according to above-mentioned relative difference.
The present invention also proposes a kind of computer equipment, including memory, processor and storage are on a memory and can be The computer program run on processor, above-mentioned processor are realized as described in any one of embodiment when executing above procedure Method.
The present invention also proposes a kind of computer readable storage medium, is stored thereon with computer program, and the program is processed The method as described in any one of embodiment is realized when device executes.
The present invention has the following beneficial effects: it is left to calculate corresponding face by the relative difference according to pixel value Right deviation rake angle, improves face and tilts angle calculation speed the step of simplifying the estimation of face tilt angle；Work as face The columns of image array gives up first row therein or the last column when being technology, ensure that the first image and the second image Symmetry is tilted the accuracy of angle estimation with improving face；By first being calibrated to the facial image of vertical tilt It is right, facial image deflection existing for vertical direction and left and right directions is removed, is estimated with improving the face angle that tilts The accuracy of meter；In such a way that graded calculates, the relative difference computational accuracy of pixel value is improved, to improve a face left side The accuracy rate of right bank angle estimation；Estimate that the face angle that tilts efficiently avoids according to the relative difference of pixel value The influence of illumination variation.
Detailed description of the invention
Fig. 1 be one embodiment of the invention face tilt angle estimation method flow diagram；
Fig. 2 be one embodiment of the invention face tilt angle estimation method flow diagram；
Fig. 3 be one embodiment of the invention face tilt angle estimation method flow diagram；
Fig. 4 be one embodiment of the invention face tilt angle estimation method flow diagram；
Fig. 5 be one embodiment of the invention face tilt angle estimation method flow diagram；
Fig. 6 be one embodiment of the invention face tilt angle estimation method flow diagram；
Fig. 7 be one embodiment of the invention face tilt angle estimation method flow diagram；
Fig. 8 be one embodiment of the invention face tilt angle estimating system structural schematic diagram；
Fig. 9 is a kind of structural schematic diagram of computer equipment of one embodiment of the invention；
In figure: 1, facial image divides module；2, the first computing module；3, the second computing module；4, computer equipment；5, External equipment；6, processing unit；7, bus；8, network adapter；9, (I/O) interface；10, display；11, system storage； 12, random access memory (RAM)；13, cache memory；14, storage system；15, program/utility；16, program Module.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation description, it is clear that described embodiment is only a part of the embodiments of the present invention, instead of all the embodiments.Base Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its His embodiment, shall fall within the protection scope of the present invention.
Referring to Fig.1, in embodiments of the present invention, it includes following for proposing a kind of the tilt estimation method of angle of face Step:
S1, facial image is divided into the first image and the second image with specific mode；
The relative difference of pixel value between S2, above-mentioned first image of calculating and the second image；
S3, corresponding face is calculated according to above-mentioned relative difference tilt angle.
Such as above-mentioned steps S1, facial image is divided into the first image and the second image with specific mode, wherein it is above-mentioned with Specific mode generally can be for by the central point that first obtains facial image width, then by the central point line of width by face Image is divided into corresponding first image and the second image；It is above-mentioned generally can be for by first wide by facial image with specific mode Degree is split as two sections of equal wide sections by the length of width, is then formed and every section wide section corresponding first on facial image Image and the second image；Above-mentioned generally can be then to press the big of area by the area of acquisition facial image with specific mode It is small to be divided into above-mentioned first image and the second image that the equal and width of height is original facial image half；In this embodiment, Preferably by facial image width is first split as two sections of equal wide sections by the length of width, the then shape on facial image At the first image corresponding with every section wide section and the second image；Specifically, general by being obtained after cutting out or marking line of demarcation The first image and the second image comprising identical pixels point quantity are preferably cut out in this embodiment.
Such as above-mentioned steps S2, the relative difference of the pixel value between above-mentioned first image and the second image is calculated, wherein The relative difference of pixel value between above-mentioned first image and the second image and corresponding face tilt the linear phase of angle It closes；Because facial image is approximately axial symmetry image when face does not tilt in left and right directions, at this point, above-mentioned first image with The relative difference numerical value of pixel value between the second image is smaller, and facial image is non-when left and right directions exists and tilts Axial symmetry image, and the asymmetry of facial image becomes larger with the increase at left and right directions inclination angle, at this point, above-mentioned first figure As the relative difference numerical value with the pixel value between the second image increases with the increase of the angle that tilts of face； Generally, above-mentioned pixel value is the value assigned when original image is digitized by computer, it represents a certain small cube of original copy Average luminance information, or perhaps average reflection (transmission) density information of the small cube, above-mentioned pixel value can be with image The colouring information that middle unit area includes changes and changes.For example, above-mentioned pixel value is 225 when above-mentioned colouring information is white； When above-mentioned colouring information is black, above-mentioned pixel value is 0.
Such as above-mentioned steps S3, corresponding face is calculated according to above-mentioned relative difference and is tilted angle, wherein will be upper It states relative difference and substitutes into corresponding tilt angle calculation formulaCorresponding face tilt angle can be calculated；h For the facial image that the angle that tilts is unknown, α is the angle that tilts, and D is relative difference,For design factor.
Facial image is divided into the first image and the second figure with specific mode above-mentioned in the present embodiment referring to Fig. 2 Before the step of picture, further includes:
S4, judge above-mentioned facial image whether vertical tilt；
S5, if so, being calibrated by affine transformation method to facial image.
Since face inclination includes the inclination of inclination and left and right directions in vertical direction, when above-mentioned facial image is vertical When direction has inclination, if do not calibrated to the inclination of above-mentioned facial image vertical direction, face or so will affect The precision of tilt angle estimation.Therefore, step S4~S5 is needed to be implemented before executing step S1~S3 calibrates above-mentioned face figure The vertical tilt of picture is tilted the precision of angle estimation with improving face；
Such as above-mentioned steps S4, judge above-mentioned facial image whether vertical tilt, wherein generally whether judge human face five-sense-organ Vertical tilt；Such as can by judge pupil and pupil location point line whether with horizontal line there are tilt angle, lead to Cross judge close to nose two canthus location point line whether with horizontal line there are tilt angles, by judging two lips Whether there are tilt angles with horizontal line for the line of the location point at angle, in this embodiment, preferably by judgement close to nose The line of location point at two canthus whether there are tilt angles with horizontal line.
Such as above-mentioned steps S5, if so, being calibrated by affine transformation method to facial image, wherein above by imitative It penetrates converter technique calibration is carried out to facial image and refer to carrying out once linear transformation simultaneously by vector space in facial image Connecting a translation transformation is another vector space in turn in vertical direction, there are the facial images of tilt angle to carry out school It is quasi-；The line segment that is formed by line of location point in above-mentioned facial image after calibration close to two canthus of nose is then One horizontal line, at this point, above-mentioned facial image is calibrated in the inclination of vertical direction.If it is not, not carried out to facial image then Calibration.
In the present embodiment, it is above-mentioned state judge above-mentioned facial image whether vertical tilt the step of before, further includes:
A1, judge whether above-mentioned facial image is color image；
A2, if so, above-mentioned color image is converted to gray level image.
Such as above-mentioned steps A1, judge whether above-mentioned facial image is color image, wherein generally according to facial image packet The colouring information contained judges whether facial image is color image, and above-mentioned color image refers to each pixel value in image All it is divided into tri- primary color components of R, G, B, each primary color component directly determines the intensity of its primary colours, such as picture depth is 24, is used R:G:B=8:8:8 indicates color, then R, G, B respectively occupy 8 to indicate respective primary color component intensity, each primary color component Strength grade be 2^8=256 kind.
As above-mentioned steps A2 is converted to above-mentioned color image if so, above-mentioned color image is converted to gray level image Gray level image is conducive to facilitate the above-mentioned relative difference of the subsequent pixel value for calculating above-mentioned first image and the second image；Its In, above-mentioned gray level image refers to the image of each only one sample color of pixel.Above-mentioned gray level image be typically shown as from Gray scale of the most furvous to most bright white.If it is not, then, continuing to use original facial image.
It in this embodiment, is Y=0.3R+0.59G+0.11B by the formula that above-mentioned color image is converted to gray level image, In, Y indicates gray scale, and R, G, B respectively indicate the color value of red, green, blue.
It is in this embodiment, above-mentioned to judge whether vertical tilt step includes: above-mentioned facial image referring to Fig. 3
S6, it is obtained in above-mentioned facial image respectively close to the location point at two canthus of nose；
S7, to above-mentioned location point line, and judge line segment whether with horizontal line there are tilt angles.
Such as above-mentioned steps S6, obtained in above-mentioned facial image respectively close to the location point at two canthus of nose, wherein one As to obtain two canthus in above-mentioned facial image close to nose respectively by the object detection method based on deep learning Location point, above-mentioned deep learning are generally ResNet100 network, specifically, intercept out a batch from several facial images first Left and right eye image takes out the non-human eye image-region on a large amount of facial images as negative sample as positive sample, utilizes The positive sample of acquisition and negative sample train depth network；When inputting above-mentioned facial image, the depth network after training It can detecte out in above-mentioned facial image close to the location point at two canthus of nose.
Such as above-mentioned steps S7, to above-mentioned location point line, and judge line segment whether with horizontal line there are tilt angle, In, when above-mentioned line segment and horizontal line are there are tilt angle theta, tilt angle theta is referred in above-mentioned facial image close to the two of nose The angle between line segment and horizontal line that the location point line at a canthus is formed, tilt angle theta indicate facial image in Vertical Square To deflection.
In this embodiment, above-mentioned facial image to be divided into the first image with specific mode and the second image step includes:
S8, the line number m and columns n for obtaining facial image matrix, and judge whether n is even number；
S9, if so, above-mentioned facial image is divided intoAbove-mentioned first image of size and the second image；
S10, if it is not, then give up first or last column of above-mentioned facial image matrix, and will be then by above-mentioned facial image It is divided intoAbove-mentioned first image of size and the second image.
Such as above-mentioned steps S8, the line number m and columns n of facial image matrix are obtained, and judges whether n is even number, wherein on State facial image matrix line number m and columns n depend on obtain facial image device model and face and equipment between away from From, execution step S9 or step S10 is corresponded to according to the parity of columns n, above-mentioned facial image is split, above-mentioned face figure The dimensional units of picture are pixel.
Such as above-mentioned steps S9, if so, above-mentioned facial image is divided intoAbove-mentioned first image of size and the second figure Picture, wherein when n is even number, at this point, above-mentioned facial image is divided into a manner of cutting outAbove-mentioned first figure of size Picture and the second image；The line number m and columns n of the image array of the image array and the second image of above-mentioned first image are equal.
Such as above-mentioned steps S10, if it is not, then give up first or last column of above-mentioned facial image matrix, and will then will be upper Facial image is stated to be divided intoAbove-mentioned first image of size and the second image, wherein when n is odd number, at this point, house First or last column of above-mentioned facial image matrix are abandoned, then above-mentioned first image and the are split into above-mentioned facial image Two images, to guarantee the line number m and column of the image array of the image array for dividing obtained above-mentioned first image and the second image Number n is equal.
Wherein, above-mentioned steps S8-S10 can be replaced step b8-b9, wherein above-mentioned steps b8 replaces S8, step b8 Are as follows: above-mentioned facial image matrix is cut out into the facial image submatrix for fixed size；Above-mentioned fixed size is generally columns n Even number and line number m are even number or odd number, and in the present embodiment, above-mentioned fixed size is preferably 64 × 64px；Above-mentioned steps b9 generation For S9 and S10, step b9 are as follows: above-mentioned facial image submatrix is divided into above-mentioned first image and the second image；Wherein, on State respectively mode be by facial image width fixed number equal part or spacing equal part, in the present embodiment, preferably spacing equal part.
Referring to Fig. 4, in the present embodiment, the relative different of the pixel value between above-mentioned first image and the second image is calculated Value step includes:
S11, the first image and the second image be divided into several image blocks respectively, wherein each image block includes Pixel quantity is identical；
S12, respectively each image block in the image array of the image array to above-mentioned first image and the second image into Line label, and each pixel in above-mentioned image block carries out label；
S13, the relative mistake that the pixel value of each pixel of label is corresponded in above-mentioned first image and the second image is calculated Different value；
S14, the pixel value that the corresponding above-mentioned image block of label is calculated according to the relative difference of the pixel value of each pixel Relative difference；
S15, above-mentioned first image and second are calculated according to the above-mentioned relative difference of the pixel value of each above-mentioned image block The relative difference of the pixel value of image.
Such as above-mentioned steps S11, the first image and the second image are divided into several image blocks respectively, wherein respectively will It is above-mentioned that first image and the second picture traverse and height are divided into several image blocks in a manner of fixed number equal part or spacing equal part, In the present embodiment, preferably above-mentioned first image and the second picture traverse are specifically divided into 5 uniformly by fixed number equal part The slice of separation, is highly divided into the slice of 10 even partitions, above-mentioned image block by width slice and height be sliced it is orthogonal and It is formed；The pixel quantity that each image block includes is equal, is the equal sized of above-mentioned each image block.
It is each in the image array of the image array to above-mentioned first image and the second image respectively such as above-mentioned steps S12 A image block carries out label, and each pixel in above-mentioned image block carries out label, wherein and above-mentioned first image is denoted as L, on It states the second image and is denoted as R, each described image block is successively denoted as L in the image array of above-mentioned first image1,...,Lq, above-mentioned Each above-mentioned image block is denoted as R in the image array of two images1,...,Rq；With image block L1And R1For pixel label such as Under, L1In each pixel be denoted as l11,...,l1q, R1In each pixel be denoted as r11,...,r1q。
Such as above-mentioned steps S13, the pixel that each pixel of label is corresponded in above-mentioned first image and the second image is calculated The relative difference of value calculates above-mentioned relative difference using the pixel value of pixel as known parameters, above-mentioned corresponding label it is each A pixel refers to pixel lijCorresponding pixel points rij。
In the present embodiment, the picture of each pixel of label is corresponded in above-mentioned first image of above-mentioned calculating and the second image The calculation formula of the above-mentioned relative difference step of element value are as follows:
dij=(lij-rij)2/(lij+ 0.01),
Wherein, dijFor the above-mentioned relative difference of above-mentioned first image and the pixel value of the corresponding pixel of the second image, lijFor the pixel value of j-th of pixel in i-th of above-mentioned image block of above-mentioned first image, rijFor above-mentioned second image The pixel value of j-th of pixel in i-th of above-mentioned image block, above-mentioned calculation formula is in lijIt is null in special circumstances still It is applicable in.
Such as above-mentioned steps S14, the corresponding above-mentioned image of label is calculated according to the relative difference of the pixel value of each pixel The relative difference of the pixel value of block, wherein to correspond to the relative difference d of the pixel value of each pixel of labelijFor Know that parameter is summed, then averaged using the result of summation as known parameters, final result is the above-mentioned image block of corresponding label Pixel value relative difference di。
Such as above-mentioned steps S15, above-mentioned first is calculated according to the above-mentioned relative difference of the pixel value of each above-mentioned image block The relative difference of the pixel value of image and the second image, wherein with correspond to each image block of label pixel value it is opposite Difference value diIt for known parameters summation, is then averaged using the result of summation as known parameters, final result is above-mentioned first The relative difference Di of the pixel value of image and the second image.
Each picture of label is corresponded in above-mentioned first image of above-mentioned calculating and the second image in this embodiment referring to Fig. 5 Before the step of relative difference of the pixel value of vegetarian refreshments, further includes:
A3, the pixel value that each pixel of label is corresponded in above-mentioned first image and the second image is obtained.
Such as above-mentioned steps A3, the pixel that each pixel of label is corresponded in above-mentioned first image and the second image is obtained Value, wherein above-mentioned pixel value is generally between 0~255.
It is in the present embodiment, above-mentioned that corresponding face lateral tilting oblique angle is calculated according to above-mentioned relative difference referring to Fig. 6 Spending step includes:
The above-mentioned relative difference of the pixel value of S16, above-mentioned first image of acquisition and the second image；
S17, corresponding people is calculated with the above-mentioned relative difference of the pixel value of the second image according to above-mentioned first image Face tilts angle.
Such as above-mentioned steps S16, the above-mentioned relative difference of the pixel value of above-mentioned first image and the second image is obtained, In, above-mentioned relative difference is the calculated result for executing step S13-S15.
Such as above-mentioned steps S17, calculated according to the above-mentioned relative difference of above-mentioned first image and the pixel value of the second image Corresponding face tilts angle out, wherein substitutes into above-mentioned tilt angle as known parameters using above-mentioned relative difference and calculates Formula meter obtains face tilt angle, and specifically, above-mentioned tilt angle calculation formula is
Referring to Fig. 7, further, facial image is divided into the first image and the second image with specific mode above-mentioned Before step, further includes:
S18, tilt angle calculation formula is established, step includes:
S19, the relative difference for obtaining K history facial image, and successively it is denoted as D1......DK, order matrix
S20, the angle that tilts for obtaining K above-mentioned history facial images, and successively it is denoted as α1......αK, order matrix
S21, equation group Pg=α is established, and calculates the linear relationship between P and α, solvedIt is right In the facial image that arbitrary inclination is unknown, its relative difference D is calculated firsth, then pass through formulaIt calculates Its inclination angle alpha outh；
Wherein, γ, I are respectively a small positive number and unit matrix, the transposition operation of T representing matrix.
Such as above-mentioned steps S18, tilt angle calculation formula is established, wherein above-mentioned relative difference and above-mentioned face or so There is linear correlation between tilt angle, therefore, utilizes tilt angle and the corresponding relative different of known facial image Value establishes tilt angle calculation formula.
Such as above-mentioned steps S19, the relative difference of K history facial image is obtained, and is successively denoted as D1......DK, enable MatrixIt wherein, include the relative difference of K history facial image in matrix P.
Such as above-mentioned steps S20, the angle that tilts of K above-mentioned history facial images is obtained, and is successively denoted as α1......αK, order matrixIt wherein, include the angle that tilts of K history facial image in matrix α.
Such as above-mentioned steps S21, equation group Pg=α is established, and calculates the linear relationship between P and α, is solvedThe facial image unknown for arbitrary inclination calculates its relative difference D firsth, so After pass through formulaCalculate its inclination angle alphah, wherein the value of γ has significant impact to the solution of equation group, It is found by largely analyzing with experimental verification, the preferred value range of γ is between 0.01~0.1, and the more preferable value of γ is 0.01,0.02,0.03,0.04,0.05,0.06,0.07,0.08,0.09,0.1,It represents and understands to above-mentioned inclination The fitness of angle calculation formula,Smaller, correspondingly, γ value is more reasonable, and solution is more adapted under current data Above-mentioned tilt angle calculation formula, therefore, most preferably γ be can makeThe numerical value of minimum value is obtained, which protects The solution for having demonstrate,proved tilt angle calculation formula has excellent numerical stability and robustness.
Referring to Fig. 8, the present invention also proposes that a kind of face tilts the estimating system of angle, comprising:
Facial image divides module 1, for facial image to be divided into the first image and the second image with specific mode；
First computing module 2, for calculating the relative difference of the pixel value between above-mentioned first image and the second image；
Second computing module 3 tilts angle for calculating corresponding face according to above-mentioned relative difference.
In facial image segmentation module 1, facial image is divided into first with specific mode after obtaining facial image Image and the second image, wherein it is above-mentioned generally can be for by the central point that first obtains facial image width, so with specific mode Facial image is divided into corresponding first image and the second image by the central point line of width afterwards；It is above-mentioned with specific mode one As can be for by the way that facial image width be first split as two sections of equal wide sections by the length of width, then on facial image Form the first image corresponding with every section wide section and the second image；Above-mentioned generally can be to pass through to obtain face with specific mode Then the area of image is divided into above-mentioned first figure that the equal and width of height is original facial image half by the size of area Picture and the second image；In this embodiment, preferably by first by facial image width by the length of width be split as two sections it is equal Wide section, corresponding with every section wide section the first image and the second image are then formed on facial image；Specifically, general logical It crosses and obtains the first image and the second image comprising identical pixels point quantity after cutting out or marking line of demarcation, it is in this embodiment, excellent It is selected as cutting out.
In the first computing module 2, above-mentioned first is calculated after obtaining the pixel value of above-mentioned first image and the second image The relative difference of pixel value between image and the second image, wherein the pixel between above-mentioned first image and the second image The relative difference of value and corresponding face tilt angle linear correlation；Because when face does not tilt in left and right directions Facial image is approximately axial symmetry image, at this point, the relative different of above-mentioned first image and the pixel value between the second image It is smaller to be worth numerical value, and facial image is non-axis symmetry image, and the asymmetry of facial image when left and right directions exists and tilts Become larger with the increase at left and right directions inclination angle, at this point, the phase of the pixel value between above-mentioned first image and the second image Difference value numerical value is increased with the increase of the angle that tilts of face；Generally, above-mentioned pixel value is original image quilt The value assigned when digitlization by computer, it represents the average luminance information of a certain small cube of original copy, or perhaps the small side Average reflection (transmission) density information of block, above-mentioned pixel value can with the variation of unit area includes in image colouring information and Variation.For example, above-mentioned pixel value is 225 when above-mentioned colouring information is white；When above-mentioned colouring information is black, above-mentioned pixel Value is 0.
In the second computing module 3, root after the relative difference of the pixel value of above-mentioned first image and the second image is obtained It calculates corresponding face according to above-mentioned relative difference to tilt angle, wherein substitute into above-mentioned relative difference corresponding Tilt angle calculation formula can calculate corresponding face tilt angle.
In the present embodiment, further includes: vertical tilt judgment module, vertical tilt calibration module, image color judge mould Block, image color conversion module, location point obtain module, line segment judgment module, columns judgment module, first segmentation submodule, Second segmentation submodule, image block form module, labeling module, the first computational submodule, the second computational submodule, third and calculate Submodule, pixel value obtain module, the first relative difference obtains module, tilt angle computing module, formula establish module, the Two relative differences obtain module, the angle that tilts obtains module, formula setting up submodule.
Above-mentioned vertical tilt judgment module, for judge above-mentioned facial image whether vertical tilt, wherein since face inclines It tiltedly include the inclination of inclination and left and right directions in vertical direction, when above-mentioned facial image is when vertical direction exists and tilts, such as Fruit does not calibrate the inclination of above-mentioned facial image vertical direction, then will affect face and tilt the essence of angle estimation Degree；Generally judge human face five-sense-organ whether vertical tilt；Such as can be by judging the line of location point of pupil and pupil No there are tilt angles with horizontal line, and whether the line of the location point at two canthus by judging close nose is deposited with horizontal line In tilt angle, by judge two labial angles location point line whether with horizontal line there are tilt angles, in this implementation In, preferably by the line judged close to the location point at two canthus of nose, whether there are tilt angles with horizontal line.
Above-mentioned vertical tilt calibration module, for being calibrated by affine transformation method to facial image, wherein above-mentioned logical It crosses affine transformation method calibration is carried out to facial image and refer to carrying out once linear change by vector space in facial image Change and connect a translation transformation be another vector space so as to vertical direction there are the facial image of tilt angle into Row calibration；The line segment that location point in above-mentioned facial image after calibration close to two canthus of nose is formed by line It is then a horizontal line, at this point, above-mentioned facial image is calibrated in the inclination of vertical direction.
Above-mentioned image color judgment module, for judging whether above-mentioned facial image is color image, wherein generally root Judge whether facial image is color image, and above-mentioned color image refers in image according to the colouring information that facial image includes Each pixel value be divided into tri- primary color components of R, G, B, each primary color component directly determines the intensity of its primary colours, such as image Depth is 24, and color is indicated with R:G:B=8:8:8, then the intensity that R, G, B respectively occupy 8 to indicate respective primary color component, often The strength grade of a primary color component is 2^8=256 kind.
Above-mentioned image color conversion module, for above-mentioned color image to be converted to gray level image, by above-mentioned color image Be converted to the above-mentioned relative different that gray level image is conducive to facilitate the subsequent pixel value for calculating above-mentioned first image and the second image Value；Wherein, above-mentioned gray level image refers to the image of each only one sample color of pixel.Above-mentioned gray level image is usually shown For from most furvous to the gray scale of most bright white；It is Y=0.3R+ by the formula that above-mentioned color image is converted to gray level image 0.59G+0.11B, wherein Y indicates gray scale, and R, G, B respectively indicate the color value of red, green, blue.
Above-mentioned location point obtains module, for being obtained in above-mentioned facial image respectively close to the position at two canthus of nose Point, wherein generally obtained in above-mentioned facial image respectively by the object detection method based on deep learning close to nose The location point at two canthus, above-mentioned deep learning are generally ResNet100 network, specifically, first from several facial images A collection of left and right eye image is intercepted out as positive sample, while taking out the non-human eye image-region on a large amount of facial images as negative Sample trains depth network using the positive sample and negative sample of acquisition；When inputting above-mentioned facial image, this is after training Depth network can detecte out in above-mentioned facial image close to the location point at two canthus of nose.
Above-mentioned line segment judgment module, be used for and judge line segment whether with horizontal line there are tilt angles, wherein when above-mentioned line With horizontal line there are tilt angle theta, tilt angle theta refers in above-mentioned facial image close to the position at two canthus of nose section The angle between line segment and horizontal line that point line is formed, tilt angle theta indicate facial image in the deflection of vertical direction.
Above-mentioned columns judgment module for obtaining the line number and columns of facial image matrix, and judges whether n is even number, Wherein, the line number m of above-mentioned facial image matrix and columns n depend on obtaining facial image device model and face and equipment it Between distance, be sent to above-mentioned first segmentation submodule or above-mentioned second segmentation submodule pair according to the parity of columns n is corresponding Above-mentioned facial image is split, and the dimensional units of above-mentioned facial image are pixel.
Above-mentioned first segmentation submodule, for being divided into above-mentioned facial image when n is even numberAbove-mentioned the of size One image and the second image, wherein when n is even number, at this point, above-mentioned facial image is divided into a manner of cutting outGreatly Small above-mentioned first image and the second image；The line number m of the image array of the image array and the second image of above-mentioned first image It is equal with columns n.
Above-mentioned second segmentation submodule, for then giving up the first or last of above-mentioned facial image matrix when n is odd number One column, and will then be divided into above-mentioned facial imageAbove-mentioned first image of size and the second image, wherein when n is When odd number, at this point, giving up first or last column of above-mentioned facial image matrix, then above-mentioned facial image is split into The first image and the second image are stated, to guarantee to divide the image array of obtained above-mentioned first image and the image moment of the second image The line number m and columns n of battle array are equal.
Wherein, above-mentioned columns judgment module, above-mentioned first segmentation submodule and above-mentioned second segmentation submodule can be replaced At cutting out module and equal sub-module；Wherein, above-mentioned columns judgment module, which is substituted for, cuts out module, and above-mentioned module of cutting out is generally used for Above-mentioned facial image matrix is cut out into the facial image submatrix for fixed size；It is even number that above-mentioned fixed size, which is generally columns, And line number is even number or odd number, in the present embodiment, above-mentioned fixed size is preferably 64 × 64px；Above-mentioned first segmentation submodule It is substituted for equal sub-module with above-mentioned second segmentation submodule, above-mentioned equal sub-module is generally used for above-mentioned facial image submatrix is equal It is divided into above-mentioned first image and the second image, wherein above-mentioned mode of dividing equally is by facial image width fixed number equal part or spacing etc. Point, in the present embodiment, preferably spacing equal part.
Above-mentioned image block forms module, for the first image and the second image to be divided into several image blocks respectively, In, first image and the second picture traverse and height are divided into a manner of fixed number equal part or spacing equal part by above-mentioned respectively several A image block, in the present embodiment, preferably fixed number equal part specifically divides above-mentioned first image and the second picture traverse For the slice of 5 even partitions, it is highly divided into the slice of 10 even partitions, above-mentioned image block is cut by width slice and height Piece is orthogonal and is formed；The pixel quantity that each image block includes is equal, is the equal sized of above-mentioned each image block.
Above-mentioned labeling module, in the image array for image array and the second image respectively to above-mentioned first image Each image block carries out label, and each pixel in above-mentioned image block carries out label, wherein and above-mentioned first image is denoted as L, Above-mentioned second image is denoted as R, and each described image block is successively denoted as L in the image array of above-mentioned first image1,...,Lq, above-mentioned Each above-mentioned image block is denoted as R in the image array of second image1,...,Rq；With image block L1And R1For pixel label It is as follows, L1In each pixel be denoted as l11,...,l1q, R1In each pixel be denoted as r11,...,r1q。
Above-mentioned first computational submodule, for calculating each pixel for corresponding to label in above-mentioned first image and the second image The relative difference of the pixel value of point, wherein above-mentioned relative difference is calculated by known parameters of the pixel value of pixel, it is above-mentioned Each pixel of corresponding label refers to pixel lijCorresponding pixel points rij.Above-mentioned first image of above-mentioned calculating and the second figure The calculation formula of the above-mentioned relative difference of the pixel value of each pixel of label is corresponded to as in are as follows: dij=(lij-rij)2/ (lij+0.01), wherein dijFor the above-mentioned relative difference of above-mentioned first image and the pixel value of the corresponding pixel of the second image, lijFor the pixel value of j-th of pixel in i-th of above-mentioned image block of above-mentioned first image, rijFor above-mentioned second image The pixel value of j-th of pixel in i-th of above-mentioned image block, above-mentioned calculation formula is in lijIt is null in special circumstances still It is applicable in.
Above-mentioned second computational submodule, the relative difference for the pixel value according to each pixel calculate corresponding label The relative difference of the pixel value of above-mentioned image block, wherein to correspond to the relative different of the pixel value of each pixel of label Value dijIt for known parameters summation, is then averaged using the result of summation as known parameters, final result is the upper of corresponding label State the relative difference d of the pixel value of image blocki。
Above-mentioned third computational submodule, the above-mentioned relative difference for the pixel value according to each above-mentioned image block calculate The relative difference of the pixel value of above-mentioned first image and the second image, wherein to correspond to the pixel of each image block of label The relative difference d of valueiFor known parameters summation, then averaged using the result of summation as known parameters, final result is The relative difference Di of the pixel value of above-mentioned first image and the second image.
Above-mentioned pixel value obtains module, for obtaining each pixel for corresponding to label in above-mentioned first image and the second image The pixel value of point, wherein above-mentioned pixel value is generally between 0~255.
Above-mentioned first relative difference obtains module, for obtaining above-mentioned first image and the pixel value of the second image State relative difference, wherein above-mentioned relative difference is that above-mentioned first computational submodule, the first computational submodule and third calculate The final calculation result of submodule.
Above-mentioned tilt angle computing module, according to the above-mentioned relative different of above-mentioned first image and the pixel value of the second image Value calculates corresponding face and tilts angle, wherein substitutes into above-mentioned inclination angle by known parameters of above-mentioned relative difference Degree calculation formula meter obtains face tilt angle, and specifically, above-mentioned tilt angle calculation formula is
Above-mentioned formula establishes module, for establishing above-mentioned tilt angle calculation formula, wherein above-mentioned relative difference with it is upper It states face and tilts and there is linear correlation between angle, therefore, utilize tilt angle and the correspondence of known facial image Relative difference establish tilt angle calculation formula.
Above-mentioned second relative difference obtains module, for obtaining the relative difference of K history facial image, and successively It is denoted as D1......DK, order matrixIt wherein, include the relative difference of K history facial image in matrix P.
The above-mentioned angle that tilts obtains module, for obtaining the angle that tilts of K above-mentioned history facial images, and Successively it is denoted as α1......αK, order matrixIt wherein, include tilting for K history facial image in matrix α Angle.
Above-mentioned formula setting up submodule for establishing equation group Pg=α, and calculates the linear relationship between P and α, solvesThe facial image unknown for arbitrary inclination calculates its relative difference D firsth, so After pass through formulaCalculate its inclination angle alphah, wherein the value of γ has significant impact to the solution of equation group, It is found by largely analyzing with experimental verification, the preferred value range of γ is between 0.01~0.1, and the more preferable value of γ is 0.01,0.02,0.03,0.04,0.05,0.06,0.07,0.08,0.09,0.1,It represents and understands to above-mentioned inclination The fitness of angle calculation formula,Smaller, correspondingly, γ value is more reasonable, and solution is more adapted under current data Above-mentioned tilt angle calculation formula, therefore, most preferably γ be can makeThe numerical value of minimum value is obtained, which protects The solution for having demonstrate,proved tilt angle calculation formula has excellent numerical stability and robustness.
Referring to Fig. 9, in embodiments of the present invention, the present invention also provides a kind of computer equipment, above-mentioned computer equipment 4 with The form of universal computing device shows, and the component of computer equipment 4 can include but is not limited to: one or more processor or Person's processing unit 6, system storage 11 connect the bus of different system components (including system storage 11 and processing unit 6) 7。
Bus 7 indicates one of a few class bus structures or a variety of, including memory bus or Memory Controller, outside Enclose bus, graphics acceleration port, processor or the local bus using any bus structures in a variety of bus structures.Citing For, these architectures include but is not limited to industry standard architecture (ISA) bus, and microchannel architecture (MAC) is total Line, enhanced isa bus, Video Electronics Standards Association (VESA) local bus and peripheral component interconnection (PCI) bus.
Computer equipment 4 typically comprises a variety of computer system readable media.These media can be it is any can be by The usable medium that computer equipment 4 accesses, including volatile and non-volatile media, moveable and immovable medium.
System storage 11 may include the computer system readable media of form of volatile memory, such as arbitrary access Memory (RAM) 12 and/or cache memory 13.Computer equipment 4 may further include other movement/it is irremovable , volatile/non-volatile computer decorum storage medium.Only as an example, storage system 15 can be used for reading and writing not removable Dynamic, non-volatile magnetic media (commonly referred to as " hard disk drive ").Although being not shown in Fig. 9, can provide for removable The disc driver of dynamic non-volatile magnetic disk (such as " floppy disk ") read-write, and to removable anonvolatile optical disk (such as CD~ ROM, DVD~ROM or other optical mediums) read-write CD drive.In these cases, each driver can pass through one A or multiple data media interfaces are connected with bus 7.Memory may include at least one program product, program product tool There is one group of (for example, at least one) program module 16, these program modules 16 are configured to perform the function of various embodiments of the present invention Energy.
Program/utility 16 with one group of (at least one) program module 16, can store in memory, for example, Such program module 16 includes --- but being not limited to --- operating system, one or more application program, other program moulds It may include the realization of network environment in block and program data, each of these examples or certain combination.Program module 16 usually execute function and/or method in embodiment described in the invention.
Computer equipment 4 (such as keyboard, sensing equipment, display 10, can also be taken the photograph with one or more external equipments 5 As head etc.) communication, the equipment interacted with the computer equipment 4 can be also enabled a user to one or more to be communicated, and/or with Any equipment (such as network interface card, the modulatedemodulate that the computer equipment 4 is communicated with one or more of the other calculating equipment Adjust device etc.) communication.This communication can be carried out by input/output (I/O) interface 9.Also, computer equipment 4 can be with Pass through network adapter 9 and one or more network (such as local area network (LAN)), wide area network (WAN) and/or public network (such as internet) communication.As shown, network adapter 8 is communicated by bus 7 with other modules of computer equipment 4.It answers When understanding, although being not shown in Fig. 9, other hardware and/or software module can be used in conjunction with computer equipment 4, including not Be limited to: microcode, device driver, redundant processing unit, external disk drive array, RAID system, tape drive and Data backup storage system etc..
Processing unit 6 by the program that is stored in system storage 11 of operation, thereby executing various function application and Data processing, such as realize that face provided by the embodiment of the present invention tilts the estimation method of angle.
That is, above-mentioned processing unit 6 is realized when executing above procedure: by facial image with designated parties after acquisition facial image Formula is divided into the first image and the second image；Calculate the relative different of the pixel value between above-mentioned first image and the second image Value；Corresponding face is calculated according to above-mentioned relative difference to tilt angle.
In embodiments of the present invention, the present invention also proposes a kind of computer readable storage medium, is stored thereon with computer Program realizes that the face provided such as all embodiments of the application tilts the estimation side of angle when the program is executed by processor Method:
That is, realization when being executed by processor to program: dividing equally facial image with specific mode after obtaining facial image For the first image and the second image；Calculate the relative difference of the pixel value between above-mentioned first image and the second image；According to Above-mentioned relative difference calculates corresponding face and tilts angle.It can be using one or more computer-readable media Any combination.Computer-readable medium can be computer gram signal media or computer readable storage medium.It calculates Machine readable storage medium storing program for executing for example may be-but not limited to-the system of electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor, dress It sets or device, or any above combination.The more specific example (non exhaustive list) of computer readable storage medium wraps It includes: there is the electrical connections of one or more conducting wires, portable computer diskette, hard disk, random access memory (RAM) 12, only Read memory (ROM), erasable programmable read-only memory (EPOM or flash memory), optical fiber, portable compact disc read-only memory (CD~ROM), light storage device, magnetic memory device or above-mentioned any appropriate combination.In this document, computer-readable Storage medium can be it is any include or storage program tangible medium, the program can be commanded execution system, device or Device use or in connection.
Computer-readable signal media may include in a base band or as carrier wave a part propagate data-signal, Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including --- but It is not limited to --- electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be Any computer-readable medium other than computer readable storage medium, change computer-readable medium can send, propagate or Transmission is for by the use of instruction execution system, device or device or program in connection.
The computer for executing operation of the present invention can be write with one or more programming languages or combinations thereof Program code, above procedure design language include object oriented program language --- such as Java, Smalltalk, C+ +, further include conventional procedural programming language --- such as " C " language or similar programming language.Program code It can fully execute on the user computer, partly execute, held as an independent software package on the user computer Part executes on the remote computer or holds on a remote computer or server completely on the user computer for row, part Row.In situations involving remote computers, remote computer can pass through the network of any kind --- including local area network (LAN) or wide area network (WAN) --- it is connected to subscriber computer, or, it may be connected to outer computer (such as using because of spy Service provider is netted to connect by internet).
The above description is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all utilizations Equivalent structure or equivalent flow shift made by description of the invention and accompanying drawing content is applied directly or indirectly in other correlations Technical field, be included within the scope of the present invention.

Claims (10)

  1. The angle estimating method 1. a kind of face tilts, which comprises the following steps:
    Facial image is divided into the first image and the second image with specific mode；
    Calculate the relative difference of the pixel value between the first image and the second image；
    Corresponding face is calculated according to the relative difference to tilt angle.
  2. The angle estimating method 2. face according to claim 1 tilts, which is characterized in that described by facial image Before the step of being divided into the first image and the second image with specific mode, further includes:
    Judge the facial image whether vertical tilt；
    If so, being calibrated by affine transformation method to facial image.
  3. The angle estimating method 3. face according to claim 2 tilts, which is characterized in that the judgement face Whether vertical tilt step includes: image
    It is obtained in the facial image respectively close to the location point at two canthus of nose；
    To the location point line, and judge line segment whether with horizontal line there are tilt angles.
  4. The angle estimating method 4. face according to claim 1 tilts, which is characterized in that it is described by facial image with Specific mode is divided into the first image and the second image step includes:
    The line number m and columns n of facial image matrix are obtained, and judges whether n is even number；
    If so, the facial image is divided intoThe first image of size and the second image；
    If it is not, then giving up first or last column of the facial image matrix, and then the facial image will be divided intoThe first image of size and the second image.
  5. The angle estimating method 5. face according to claim 4 tilts, which is characterized in that calculate the first image The relative difference step of pixel value between the second image includes:
    The first image and the second image are divided into several image blocks respectively, wherein the pixel number that each image block includes It measures identical；
    Label is carried out to each image block in the image array of the image array of the first image and the second image respectively, and Each pixel in described image block carries out label；
    Calculate the relative difference that the pixel value of each pixel of label is corresponded in the first image and the second image；
    The relative mistake of the pixel value of corresponding label described image block is calculated according to the relative difference of the pixel value of each pixel Different value；
    The picture of the first image and the second image is calculated according to the relative difference of the pixel value of each described image block The relative difference of element value.
  6. The angle estimating method 6. face according to claim 1 tilts, which is characterized in that described according to described opposite Difference value calculates the corresponding face angle step that tilts
    Obtain the relative difference of the pixel value of the first image and the second image；
    Corresponding face lateral tilting is calculated with the relative difference of the pixel value of the second image according to the first image Rake angle.
  7. The angle estimating method 7. face according to claim 1 tilts, which is characterized in that described by facial image Before the step of being divided into the first image and the second image with specific mode, further includes:
    Tilt angle calculation formula is established, step includes:
    The relative difference of K history facial image is obtained, and is successively denoted as D1.......DK, order matrix
    The angle that tilts of the K history facial images is obtained, and is successively denoted as α1......αK, order matrix
    Equation group Pg=α is established, and calculates the linear relationship between P and α, is solvedFor arbitrarily inclining The unknown facial image of rake angle calculates its relative difference D firsth, then pass through formulaCalculate its inclination Angle [alpha]h；
    Wherein, γ, I are respectively a small positive number and unit matrix, the transposition operation of T representing matrix.
  8. The angle estimation system 8. a kind of face tilts characterized by comprising
    Facial image divides module, for facial image to be divided into the first image and the second image with specific mode；
    First computing module, for calculating the relative difference of the pixel value between the first image and the second image；
    Second computing module tilts angle for calculating corresponding face according to the relative difference.
  9. 9. a kind of computer equipment, can run on a memory and on a processor including memory, processor and storage Computer program, which is characterized in that the processor is realized when executing described program such as any one of claim 1~7 institute The method stated.
  10. 10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor The method as described in any one of claim 1~7 is realized when execution.
CN201810653661.2A 2018-06-22 2018-06-22 Method, system, equipment and storage medium for estimating left and right inclination angles of human face Active CN108960099B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810653661.2A CN108960099B (en) 2018-06-22 2018-06-22 Method, system, equipment and storage medium for estimating left and right inclination angles of human face

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810653661.2A CN108960099B (en) 2018-06-22 2018-06-22 Method, system, equipment and storage medium for estimating left and right inclination angles of human face

Publications (2)

Publication Number Publication Date
CN108960099A true CN108960099A (en) 2018-12-07
CN108960099B CN108960099B (en) 2021-07-06

Family

ID=64486169

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810653661.2A Active CN108960099B (en) 2018-06-22 2018-06-22 Method, system, equipment and storage medium for estimating left and right inclination angles of human face

Country Status (1)

Country Link
CN (1) CN108960099B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070165951A1 (en) * 2006-01-16 2007-07-19 Fujifilm Corporation Face detection method, device and program
CN102999748A (en) * 2012-12-12 2013-03-27 湖北微驾技术有限公司 Refactoring method for optimizing super resolution of facial images
CN104834928A (en) * 2015-05-08 2015-08-12 小米科技有限责任公司 Method for determining identification area in picture and device thereof
CN107358207A (en) * 2017-07-14 2017-11-17 重庆大学 A kind of method for correcting facial image

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070165951A1 (en) * 2006-01-16 2007-07-19 Fujifilm Corporation Face detection method, device and program
CN102999748A (en) * 2012-12-12 2013-03-27 湖北微驾技术有限公司 Refactoring method for optimizing super resolution of facial images
CN104834928A (en) * 2015-05-08 2015-08-12 小米科技有限责任公司 Method for determining identification area in picture and device thereof
CN107358207A (en) * 2017-07-14 2017-11-17 重庆大学 A kind of method for correcting facial image

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
马里千: "基于Ａｄａｂｏｏｓｔ人脸检测融合五官", 《华中科技大学学报（自然科学版）》 *

Also Published As

Publication number Publication date
CN108960099B (en) 2021-07-06

Similar Documents

Publication Publication Date Title
US20210049760A1 (en) Sight line detection method and sight line detection device
CN111524106B (en) Skull fracture detection and model training method, device, equipment and storage medium
CN111563502B (en) Image text recognition method and device, electronic equipment and computer storage medium
EP3576017A1 (en) Method, apparatus, and device for determining pose of object in image, and storage medium
CN108805979B (en) Three-dimensional reconstruction method, device, equipment and storage medium for dynamic model
WO2022156640A1 (en) Gaze correction method and apparatus for image, electronic device, computer-readable storage medium, and computer program product
US11263803B2 (en) Virtual reality scene rendering method, apparatus and device
CN104899563A (en) Two-dimensional face key feature point positioning method and system
CN110490158B (en) Robust face alignment method based on multistage model
CN110197146A (en) Facial image analysis method, electronic device and storage medium based on deep learning
CN107609466A (en) Face cluster method, apparatus, equipment and storage medium
CN111723691A (en) Three-dimensional face recognition method and device, electronic equipment and storage medium
CN111046734A (en) Multi-modal fusion sight line estimation method based on expansion convolution
CN111209811B (en) Method and system for detecting eyeball attention position in real time
CN110555426A (en) Sight line detection method, device, equipment and storage medium
CN110689000A (en) Vehicle license plate identification method based on vehicle license plate sample in complex environment
CN111145215A (en) Target tracking method and device
CN109523558A (en) A kind of portrait dividing method and system
CN113705349A (en) Attention power analysis method and system based on sight estimation neural network
JP2019212148A (en) Information processing device and information processing program
Ling et al. Research on gesture recognition based on YOLOv5
CN116805360B (en) Obvious target detection method based on double-flow gating progressive optimization network
CN114332927A (en) Classroom hand-raising behavior detection method, system, computer equipment and storage medium
CN112017212A (en) Training and tracking method and system of face key point tracking model
CN108960099A (en) Face tilts angle estimating method, system, equipment and storage medium

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