CN105678241B - A kind of cascade two dimensional image face pose estimation - Google Patents
A kind of cascade two dimensional image face pose estimation Download PDFInfo
- Publication number
- CN105678241B CN105678241B CN201511024409.8A CN201511024409A CN105678241B CN 105678241 B CN105678241 B CN 105678241B CN 201511024409 A CN201511024409 A CN 201511024409A CN 105678241 B CN105678241 B CN 105678241B
- Authority
- CN
- China
- Prior art keywords
- human face
- characteristic point
- image
- angle
- posture
- 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.)
- Active
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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The invention discloses a kind of cascade two dimensional image face pose estimations, it include: Step 1: detecting and splitting the human face region on the two-dimension human face image of input, predetermined size is normalized to, its textural characteristics is extracted, which is indicated with feature vector；Step 2: the feature vector of the two-dimension human face image according to obtained in step 1 calculates it the different attitude angles a possibility that；Step 3: selecting the attitude angle with maximum likelihood as the posture initial estimation result of the two-dimension human face image of the input, the facial modeling model that corresponding posture is selected according to the attitude angle with maximum likelihood, detects the characteristic point position on the facial image；Step 4: using the method based on characteristic point, further accurately estimation inputs the human face posture angle in facial image according to the characteristic point detected in step 3.The present invention overcomes the disadvantage of Attitude estimation method respectively based on appearance features and based on characteristic point, estimates attitude angle from thick to thin by cascade mode, improves the precision of Attitude estimation.
Description
Technical field
The present invention relates to Computer Applied Technology and computer vision field, in particular to a kind of cascade two dimensional image people
Face Attitude estimation method.
Background technique
The posture of face refers to rotation angle of the number of people in three dimensions relative to the inherent reference axis of itself, including around
The pitch angle (α) of X-axis rotation, the deflection angle (β) rotated around Y-axis and tilt angle (γ) about the z axis.Human face posture information
Very important effect is played in the applications such as human-computer interaction, recognition of face, blinkpunkt detection.
It is based on currently, estimating that wherein the method for the 3 d pose angle of face specifically includes that from single two-dimensional facial image
The method of characteristic point and method based on appearance features.
Method based on characteristic point is first from detecting key feature points (in such as pupil of both eyes on the facial image of input
The heart, nose and the left and right corners of the mouth) position, then according to the relative positional relationship of these characteristic points estimate face attitude angle.
This process is often realized by means of a three-dimensional face averaging model.The major defect of method based on characteristic point is spy
The precision of sign point detection can decline with the increase of human face posture angle change range, and then influence the essence of human face modeling
Degree.
Method based on appearance features directly extracts feature from facial image, then in the feature and face appearance extracted
Classification or homing method are established between state angle, to realize the estimation to input facial image posture.Based on appearance features
Method be not necessarily to locating human face's characteristic point, be easy to implement, may to the discrimination of attitude angle but influenced by training data
Lower (such as with 5 degree or 10 degree for interval).
Generally speaking, there are respective advantage and disadvantage for existing face pose estimation, such as when attitude angle becomes larger
Evaluated error also becomes larger or the separating capacity of Attitude estimation is limited larger by training data, estimates to affect human face posture
The precision of meter.
Summary of the invention
It is an object of the invention to overcome the above-mentioned deficiency in the presence of the prior art, a kind of raising human face posture is provided and is estimated
Count the cascade two dimensional image face pose estimation of precision.
In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is that:
A kind of cascade two dimensional image face pose estimation, the present invention in, according to specified angle interval, prestore each
The facial modeling model of attitude angle；Further include following steps:
Step 1: detecting and splitting the human face region on the two-dimension human face image of input, pre- scale is normalized to
It is very little, its textural characteristics is extracted, with feature vector F ∈ RNIndicate the textural characteristics；
Step 2: the feature vector of the two-dimension human face image according to obtained in step 1 calculate its in different attitude angles can
It can property；
Step 3: the attitude angle with maximum likelihood is selected initially to estimate as the posture of the two-dimension human face image of the input
Meter is as a result, select the facial modeling model of corresponding posture according to the attitude angle with maximum likelihood, detection
Characteristic point position on the two-dimension human face image；
Step 4: further accurately being estimated according to the characteristic point detected in step 3 using the method based on characteristic point
Human face posture angle in meter input facial image.
The step 2 specifically:
Calculate the posture regression coefficient under each characteristic dimension according to the following formula first,
Wherein, φX(α)、φY(β) and φZ(γ) respectively indicates the p rank about pitch angle, about the q rank of deflection angle and pass
R rank multinomial in inclination angle,For the posture regression parameter of the n-th dimensional feature；
Then a possibility that input facial image belongs to different attitude angles is calculated according to the posture regression coefficient, calculated public
Formula is as follows:
Parameter in step 2 learns to obtain using gradient descent algorithm according to training data.
The characteristic point position on facial image is detected in the step 3 specifically:
Take the moving shape model (ASM) obtained by the data training of corresponding attitude angle or based on cascade recurrence
Facial feature points detection device is detected.
The step 4 specifically:
Using the method for the characteristic point on fitting one average three-dimensional face model to input two dimensional image, by adjusting appearance
State angle come after optimizing on the projecting characteristic points to two dimensional image on three-dimensional face model with feature on input two-dimension human face image
Registration between point.
Compared with prior art, beneficial effects of the present invention:
The present invention combines the Attitude estimation method based on appearance features and based on characteristic point, can overcome these two kinds of methods
Respective disadvantage estimates attitude angle by cascade mode from thick to thin, effectively increases the precision of Attitude estimation.
Detailed description of the invention:
Fig. 1 is the method flow diagram in the embodiment of the present invention.
Specific embodiment
The present invention is described in further detail With reference to embodiment.But this should not be interpreted as to the present invention
The range of above-mentioned theme is only limitted to embodiment below, all that model of the invention is belonged to based on the technology that the content of present invention is realized
It encloses.
The present invention proposes a kind of cascade two-dimension human face image Attitude estimation method.This method is combined based on apparent sum
Method based on characteristic point, to effectively overcome simple using the angular range for being limited to training data based on apparent method
The shortcomings that, and simple using the method based on characteristic point, when angle change range is larger, Attitude estimation accuracy decline is lacked
Point improves the precision of Attitude estimation.
Fig. 1 is the cascade two dimensional image face pose estimation flow chart shown in the embodiment of the present invention, is specifically included
Following steps:
In order to solve in the method based on characteristic point, when using single three-dimensional face averaging model, existing characteristic point
The precision of detection can decline with the increase of human face posture angle change range, and then the precision for influencing human face modeling is asked
Topic, presets the extract facial feature model of each attitude angle；And then human face posture estimation is carried out using following steps；First
According to specified angle interval (5 degree, in practical application are divided into the present embodiment between specified angle, can be as needed at 1 degree to 15 degree
Between select the specified angle interval, actually 2 degree, 3 degree, 4 degree, 6 degree, 7 degree, 10 degree can realize the object of the invention well),
Prestore the facial modeling model of each attitude angle；And then include the following steps
Step 1: detecting and splitting the human face region on the two-dimension human face image of input, pre- scale is normalized to
It is very little, its textural characteristics is extracted, with feature vector F ∈ RNIndicate the textural characteristics；
Step 2: the feature vector of the two-dimension human face image according to obtained in step 1 calculate its in different attitude angles can
It can property；
Step 3: the attitude angle with maximum likelihood is selected initially to estimate as the posture of the two-dimension human face image of the input
Count result, it is assumed that the attitude angle is the maximum possible attitude angle of the two-dimension human face image, so that further selection is corresponding
The facial modeling model of attitude angle, in conjunction on selected facial modeling model inspection two-dimension human face image
Characteristic point position；
Step 4: further accurately being estimated according to the characteristic point detected in step 3 using the method based on characteristic point
Human face posture angle in meter input facial image.
Specifically, the step 2 specifically:
Calculate the posture regression coefficient under each characteristic dimension according to the following formula first,
Wherein, φX(α)、φY(β) and φZ(γ) respectively indicates the p rank about pitch angle, about the q rank of deflection angle and pass
R rank multinomial in inclination angle,For the posture regression parameter of the n-th dimensional feature；
Then a possibility that input facial image belongs to different attitude angles is calculated according to the posture regression coefficient, calculated public
Formula is as follows:
Parameter in step 2 learns to obtain using gradient descent algorithm according to training data.
The characteristic point position on facial image is detected in the step 3 specifically:
Take the moving shape model (ASM) obtained by the data training of corresponding attitude angle or based on cascade recurrence
Facial feature points detection device is detected.
The step 4 specifically: using the characteristic point on fitting one average three-dimensional face model to input two dimensional image
Method, by adjusting attitude angle come after optimizing on the projecting characteristic points to two dimensional image on three-dimensional face model with input two
Tie up the registration on facial image between characteristic point.
Illustrate the present invention below with reference to a specific example.
The method of the present invention includes a training process, according to one group of training data study human face posture angle and its texture
Regression function between feature.Here we are indicated using following regression function between attitude angle and the n-th dimensional feature
Regression relation:
φX(α)、φY(β) and φZ(γ) respectively indicates the p rank about pitch angle, the q rank about deflection angle and wherein,
About the r rank multinomial at inclination angle,For the posture regression parameter of the n-th dimensional feature.It needs through training data
The parameter of habit includes the coefficient in posture regression parameter and three multinomials.These parameters can be obtained by gradient descent method
It arrives.
Test phase, first with the attitude angle of the secondary new input facial image of above-mentioned regression model estimation one.Specifically,
The human face region inputted on facial image is detected and split first, specified size is normalized to, then extracts its line
Feature is managed, note this feature is F ∈ RN, Fn indicates its n-th dimensional feature component.Later, using trained regression model, root is calculated
A possibility that belonging to different attitude angles according to every one-dimensional characteristic component input facial image.Calculation formula is as follows:
Finally, selecting posture initial estimation result of the attitude angle with maximum possible as the input facial image.
Then the feature point detector of corresponding posture is selected according to obtained attitude angle, detection inputs on facial image
Characteristic point (such as pupil center, nose and the corners of the mouth) position.When it is implemented, the data training by corresponding attitude angle can be taken
Obtained moving shape model (ASM) or the facial feature points detection device returned based on cascade.
The characteristic point that last basis obtains further accurately estimates the people in input picture using the method based on characteristic point
Face attitude angle.When it is implemented, can be using the characteristic point on fitting one average three-dimensional face model to input X-Y scheme
The method of picture.Specifically, being thrown by adjusting the attitude angle of three-dimensional face model to optimize the characteristic point on three-dimensional face model
Registration after on shadow to two dimensional image and on input two-dimension human face image between characteristic point, which, which can be used, is based on
Particle group optimizing or the method declined based on gradient.When optimization starts, three-dimensional face model is standard front face posture.Optimize
Cheng Shi, three-dimensional face model attitude angle adjusted are to input the attitude angle of facial image.
Advantages of the present invention are as follows: combine the Attitude estimation method based on appearance features and based on characteristic point.Based on apparent
The Attitude estimation mode of feature is suitble to carry out rough sort to posture, although and the method based on characteristic point finer can be estimated
Attitude angle, but need to detect human face characteristic point first.However, accurately detecting any attitude with a unified model
Under human face characteristic point, it is difficult to realize at present.Therefore the present invention, make full use of based on the attitude prediction of appearance features as a result,
Selection is directed to the facial modeling model of particular pose angular range, is then estimated more with the method based on characteristic point again
Accurate attitude angle.In this way, estimating attitude angle from thick to thin by cascade mode, any attitude feature has both been overcome
Influence of the point location difficulty to Attitude estimation, and effectively increase the precision of Attitude estimation.
A specific embodiment of the invention is described in detail above in conjunction with attached drawing, but the present invention is not restricted to
Embodiment is stated, in the spirit and scope for not departing from claims hereof, those skilled in the art can make
Various modifications or remodeling out.
Claims (4)
1. a kind of cascade two dimensional image face pose estimation characterized by comprising
Step 1: detecting and splitting the human face region on the two-dimension human face image of input, predetermined size is normalized to, is mentioned
Its textural characteristics is taken, with feature vector F ∈ RNIndicate the textural characteristics；
Step 2: the feature vector of the two-dimension human face image according to obtained in step 1 calculates it in the possibility of different attitude angles
Property；
Step 3: selecting the attitude angle with maximum likelihood as the posture initial estimation knot of the two-dimension human face image of the input
Fruit selects the features of human face images location model of corresponding posture according to the attitude angle with maximum likelihood, detection
Characteristic point position on the two-dimension human face image；
Step 4: further accurately being estimated using the method based on characteristic point defeated according to the characteristic point detected in step 3
Enter the human face posture angle in facial image；
The step 2 specifically:
Calculate the posture regression coefficient under each characteristic dimension according to the following formula first,
Wherein,WithRespectively indicate the p rank about pitch angle, the q rank about deflection angle and about inclining
The r rank multinomial at oblique angle,For the posture regression parameter of the n-th dimensional feature；
Then a possibility that input facial image belongs to different attitude angles is calculated according to the posture regression coefficient, calculation formula is such as
Under:
Wherein,
FnIndicate the n-th dimensional feature component of texture feature vector F.
2. cascade two dimensional image face pose estimation according to claim 1, which is characterized in that in step 2
Parameter learns to obtain using gradient descent algorithm according to training data.
3. cascade two dimensional image face pose estimation according to claim 1, which is characterized in that the step 3
Characteristic point position on middle detection facial image specifically:
Take the moving shape model obtained by the data training of corresponding attitude angle or the face characteristic returned based on cascade
Spot detector is detected.
4. cascade two dimensional image face pose estimation according to claim 1, which is characterized in that the step 4
Specifically:
Using the method for the characteristic point on fitting one average three-dimensional face model to input two dimensional image, by adjusting attitude angle
Degree come after optimizing on the projecting characteristic points to two dimensional image on three-dimensional face model with characteristic point on input two-dimension human face image it
Between registration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201511024409.8A CN105678241B (en) | 2015-12-30 | 2015-12-30 | A kind of cascade two dimensional image face pose estimation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201511024409.8A CN105678241B (en) | 2015-12-30 | 2015-12-30 | A kind of cascade two dimensional image face pose estimation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105678241A CN105678241A (en) | 2016-06-15 |
| CN105678241B true CN105678241B (en) | 2019-02-26 |
Family
ID=56189872
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201511024409.8A Active CN105678241B (en) | 2015-12-30 | 2015-12-30 | A kind of cascade two dimensional image face pose estimation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105678241B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106355147A (en) * | 2016-08-26 | 2017-01-25 | 张艳 | Acquiring method and detecting method of live face head pose detection regression apparatus |
| CN106682598B (en) * | 2016-12-14 | 2021-02-19 | 华南理工大学 | Multi-pose face feature point detection method based on cascade regression |
| CN108470328A (en) * | 2018-03-28 | 2018-08-31 | 百度在线网络技术（北京）有限公司 | Method and apparatus for handling image |
| CN108460383B (en) * | 2018-04-11 | 2021-10-01 | 四川大学 | Image significance refinement method based on neural network and image segmentation |
| CN109636804A (en) * | 2018-10-10 | 2019-04-16 | 浙江大学 | One kind being based on the cascade human body image dividing method of more granularities |
| CN109671108B (en) * | 2018-12-18 | 2020-07-28 | 重庆理工大学 | Single multi-view face image attitude estimation method capable of rotating randomly in plane |
| CN111222469B (en) * | 2020-01-09 | 2022-02-15 | 浙江工业大学 | Coarse-to-fine human face posture quantitative estimation method |
| CN114155565A (en) * | 2020-08-17 | 2022-03-08 | 顺丰科技有限公司 | Face feature point coordinate acquisition method and device, computer equipment and storage medium |
| CN113869186B (en) * | 2021-09-24 | 2022-12-16 | 合肥的卢深视科技有限公司 | Model training method and device, electronic equipment and computer readable storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1831846A (en) * | 2006-04-20 | 2006-09-13 | 上海交通大学 | Face posture identification method based on statistical model |
| CN102043966A (en) * | 2010-12-07 | 2011-05-04 | 浙江大学 | Face recognition method based on combination of partial principal component analysis (PCA) and attitude estimation |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4093273B2 (en) * | 2006-03-13 | 2008-06-04 | オムロン株式会社 | Feature point detection apparatus, feature point detection method, and feature point detection program |
-
2015
- 2015-12-30 CN CN201511024409.8A patent/CN105678241B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1831846A (en) * | 2006-04-20 | 2006-09-13 | 上海交通大学 | Face posture identification method based on statistical model |
| CN102043966A (en) * | 2010-12-07 | 2011-05-04 | 浙江大学 | Face recognition method based on combination of partial principal component analysis (PCA) and attitude estimation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105678241A (en) | 2016-06-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105678241B (en) | A kind of cascade two dimensional image face pose estimation | |
| CN105930767B (en) | A kind of action identification method based on human skeleton | |
| US10055851B2 (en) | Determining dimension of target object in an image using reference object | |
| CN105701820B (en) | A kind of point cloud registration method based on matching area | |
| TWI662433B (en) | Living body detection method and device, and identity authentication method and device | |
| CN106055091B (en) | A kind of hand gestures estimation method based on depth information and correcting mode | |
| CN105389539B (en) | A kind of three-dimension gesture Attitude estimation method and system based on depth data | |
| CN104573614B (en) | Apparatus and method for tracking human face | |
| CN108717531B (en) | Human body posture estimation method based on Faster R-CNN | |
| JP5722502B2 (en) | Planar mapping and tracking for mobile devices | |
| JP4950787B2 (en) | Image processing apparatus and method | |
| CN109359514B (en) | DeskVR-oriented gesture tracking and recognition combined strategy method | |
| CN105759967B (en) | A kind of hand overall situation attitude detecting method based on depth data | |
| CN108230383A (en) | Hand three-dimensional data determines method, apparatus and electronic equipment | |
| CN108182397B (en) | Multi-pose multi-scale human face verification method | |
| CN102834845A (en) | Method and arrangement for multi-camera calibration | |
| JP2014127208A (en) | Method and apparatus for detecting object | |
| CN104487915A (en) | Maintaining continuity of augmentations | |
| CN109035330A (en) | Cabinet approximating method, equipment and computer readable storage medium | |
| JP6487642B2 (en) | A method of detecting a finger shape, a program thereof, a storage medium of the program, and a system for detecting a shape of a finger. | |
| CN107564059A (en) | Object positioning method, device and NI Vision Builder for Automated Inspection based on RGB D information | |
| CN107949851A (en) | The quick and robust control policy of the endpoint of object in scene | |
| JP2008146329A (en) | Face feature point detection device and method | |
| JP7230345B2 (en) | Information processing device and information processing program | |
| CN105741326B (en) | A kind of method for tracking target of the video sequence based on Cluster-Fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |