CN104463117B - A kind of recognition of face sample collection method and system based on video mode - Google Patents
A kind of recognition of face sample collection method and system based on video mode Download PDFInfo
- Publication number
- CN104463117B CN104463117B CN201410720464.XA CN201410720464A CN104463117B CN 104463117 B CN104463117 B CN 104463117B CN 201410720464 A CN201410720464 A CN 201410720464A CN 104463117 B CN104463117 B CN 104463117B
- Authority
- CN
- China
- Prior art keywords
- face
- characteristic
- pixel region
- recognition
- face characteristic
- 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
- G06V40/161—Detection; Localisation; Normalisation
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of recognition of face sample collection method and system based on video mode, obtaining first has the image-region of movable information in the picture frame of video, then described image region is detected to obtain face pixel region, then to face pixel region in the picture frame containing face into line trace, obtain the history image frame of same face, the history image frame of the same face high to quality carries out face characteristic extraction again, the face characteristic needed for recognition of face is obtained, finally stores the face characteristic.The process of entire recognition of face sample collection does not need to sample collection worker manual operation, and the speed of acquisition is fast, efficient.
Description
Technical field
The present invention relates to technical field of face recognition, more particularly to a kind of recognition of face sample collection based on video mode
Method and system.
Background technology
Recognition of face is not only the research hotspot of artificial intelligence field, and in public safety field, research also has important
Realistic meaning.Face recognition technology, which includes facial image sample collection, sample image pretreatment, classifier training and sample, to be known
These other sport technique segments, and sample collection work is the foundation stone of recognition of face, meaning is self-evident.
Recognition of face sample collection work is all based on the mode of image in the past.For example, first allow volunteer orderly successively
In some shooting point；Then, sample collection worker repeatedly claps volunteer according to differences such as posture, illumination, expressions
It takes the photograph；Finally, sample collection worker carries out Screening Treatment according to the image that shooting obtains, and establishes recognition of face sample database.It is this
Although the mode based on image is intuitive, but still there are 3 points of deficiencies：First, make a large amount of volunteer orderly in clapping successively
It takes the photograph and is repeatedly taken pictures, this quite time-consuming effort of way；Second, recognition of face has face sample strictly in resolution ratio
Requirement, therefore, can the image of shooting reach criterion of identification, and it is still necessary to sample collection worker is wanted to check one by one；Third, sample
Arrangement is a long process, and sample collection worker inevitably generates work fatigue when arranging, therefore, the face information of storage
There is also the risks of mistake.
It is in the prior art based on video mode there are also recognition of face sample collection work.For example, it is taken the photograph with monitoring
For camera in a certain fixed point shooting video, sample collection worker intercepts the picture frame containing face in video on backstage, then right
Picture frame containing face is identified according to posture, illumination, expression etc., and the picture frame met is put into recognition of face sample
Library.This sample collection technology based on video mode, does not need to that volunteer is orderly successively repeatedly to be clapped in shooting point
It takes the photograph, but sample collection worker is still needed to choose the picture frame containing face manually, and whether manually check the picture frame
Reach criterion of identification, then carry out sample arrangement.Whole process is time-consuming and laborious, and efficiency is low.
Invention content
For this purpose, the technical problems to be solved by the invention are the recognition of face sample collection based on video in the prior art
Method speed is slow, time-consuming and efficiency is low, so as to propose a kind of recognition of face sample collection method based on video mode and be
System.
In order to solve the above technical problems, the present invention provides following technical solution：
A kind of recognition of face sample collection method based on video mode, includes the following steps：
Obtaining has the image-region of movable information in the picture frame of video；
Described image region is detected to obtain face pixel region；
To face pixel region in the picture frame containing face into line trace, the history image frame of same face is obtained；
The history image frame of the same face high to quality carries out face characteristic extraction, obtains the face needed for recognition of face
Feature；
Store the face characteristic.
The above-mentioned recognition of face sample collection method based on video mode, it is described obtain video picture frame in have movement
The image-region step of information includes：
Background model is established using the initial image frame in the video；
Foreground model is established to current image frame；
Calculate the difference of the background model and the foreground model；
The pixel region that the difference between foreground model and background model is more than predetermined threshold value is obtained, as with movement
The image-region of information.
The above-mentioned recognition of face sample collection method based on video mode, it is described to detect described image region to obtain face
Pixel region step includes：
Image pyramid is established to current image frame；
It is loaded into each posture faceform of off-line training；
Calculate the feature on each tomographic image pyramid；
Calculate response of the feature on each posture faceform of off-line training；
Obtain maximum response and reach the picture frame of Face datection threshold value, extract in current image frame face pixel region and
The attitude information of face.
The above-mentioned recognition of face sample collection method based on video mode, it is described to people in the picture frame containing face
Face pixel region carries out tracking step and includes：
Establish face tracking model and face Track Initiation node；
Calculate the sound of the textural characteristics and edge feature of face pixel region in current image frame on face tracking model
It should；
Face pixel region coordinate corresponding to peak response in current image frame is added in the tail portion node of face track；
If the face pixel region of current image frame reaches the boundary of picture frame, terminate tracking, obtain from face track
Start node is to the history image frame of the same face of face track tail portion node.
The above-mentioned recognition of face sample collection method based on video mode, the history figure of the same face high to quality
Include as frame carries out face characteristic extraction step：
The picture frame high to quality carries out illumination pretreatment；
Calculate the face characteristic of face pixel region in the picture frame after illumination pretreatment；
Principle components analysis and independent component analysis are carried out to the face characteristic；
Obtain the face characteristic needed for recognition of face.
The above-mentioned recognition of face sample collection method based on video mode, the storage face characteristic step include：
The face characteristic is compared with each face characteristic in Face Sample Storehouse；
If the face characteristic is close with some face characteristic in Face Sample Storehouse, by the face characteristic and correspondence
The storage of face pixel region in the similar face feature samples message sample library；
If the face characteristic differs greatly with face characteristic each in Face Sample Storehouse, a face characteristic letter is created
Cease face characteristic and corresponding face pixel region described in sample library storage.
A kind of recognition of face sample acquisition system based on video mode, including：
Image-region acquisition module, for obtaining the image-region in the picture frame of video with movable information；
Face detection module, for detecting described image region to obtain face pixel region；
Face tracking module, for, into line trace, obtaining same people to face pixel region in the picture frame containing face
The history image frame of face；
Face characteristic acquisition module carries out face characteristic extraction for the history image frame of the same face high to quality,
Obtain the face characteristic needed for recognition of face；
Face characteristic memory module, for storing the face characteristic.
The above-mentioned recognition of face sample acquisition system based on video mode, described image region acquisition module include：
Background Modeling submodule, for establishing background model using the initial image frame in the video；
Foreground model setting up submodule, for establishing foreground model to current image frame；
Difference computational submodule, for calculating the difference of the background model and the foreground model；
Image-region acquisition submodule, the difference for obtaining between foreground model and background model are more than predetermined threshold value
Pixel region, as the image-region with movable information.
The above-mentioned recognition of face sample acquisition system based on video mode, in the face detection module：
Image pyramid setting up submodule, for establishing image pyramid to current image frame；
Off-line model is loaded into submodule, for being loaded into each posture faceform of off-line training；
Feature calculation submodule, for calculating the feature on each tomographic image pyramid；
Response computational submodule, for calculating response of the feature on each posture faceform of off-line training
Value；
Face pixel region acquisition submodule, for obtaining the picture frame that maximum response reaches Face datection threshold value, carries
Take the attitude information of face pixel region and face in current image frame.
The above-mentioned recognition of face sample acquisition system based on video mode, the face tracking module include：
Face tracking model foundation submodule, for establishing face tracking model and face Track Initiation node；
Characteristic response computational submodule, it is special for calculating the textural characteristics of face pixel region and edge in current image frame
Levy the response on face tracking model；
Peak response handles submodule, for by the face pixel region coordinate corresponding to peak response in current image frame
Add in the tail portion node of face track；
The history image frame acquisition submodule of same face, for reaching image in the face pixel region of current image frame
During the boundary of frame, terminate tracking, obtain from face Track Initiation node to the history of the same face of face track tail portion node
Picture frame.
The above-mentioned recognition of face sample acquisition system based on video mode, the face characteristic acquisition module include：
Submodule is pre-processed, for carrying out illumination pretreatment to the high picture frame of quality；
Face characteristic computational submodule, for calculating the face of face pixel region in the picture frame after illumination pretreatment
Feature；
Human face analysis submodule, for carrying out Principle components analysis and independent component analysis to the face characteristic；
Face characteristic acquisition submodule, for obtaining the face characteristic needed for recognition of face.
The above-mentioned recognition of face sample acquisition system based on video mode, the face characteristic memory module include：
Face characteristic compares submodule, for will be in face characteristic described in face characteristic acquisition module and Face Sample Storehouse
Each face characteristic be compared；
Face characteristic sub-module stored, if for the face characteristic and some face characteristic phase in Face Sample Storehouse
Closely, then the face characteristic and corresponding face pixel region are stored to the similar face feature samples message sample library
In；If the face characteristic differs greatly with face characteristic each in Face Sample Storehouse, a face characteristic information sample is created
Face characteristic described in this library storage and corresponding face pixel region.
The above technical solution of the present invention has the following advantages over the prior art：
(1) a kind of recognition of face sample collection method and system based on video mode of the present invention, obtain first
In the picture frame of video there is the image-region of movable information, then detect described image region to obtain face pixel region,
Then the history image frame of same face is obtained, then confront into line trace to face pixel region in the picture frame containing face
The history image frame for measuring high same face carries out face characteristic extraction, obtains the face characteristic needed for recognition of face, finally deposits
Store up the face characteristic.The process of entire recognition of face sample collection does not need to sample collection worker manual operation, acquisition
Speed is fast, efficient.
Description of the drawings
In order to make the content of the present invention more clearly understood, it below according to specific embodiments of the present invention and combines
Attached drawing, the present invention is described in further detail, wherein
Fig. 1 is a kind of recognition of face sample collection method flow diagram based on video mode of one embodiment of the invention；
Fig. 2 is the method flow diagram that a kind of acquisition of one embodiment of the invention has the image-region of movable information；
Fig. 3 is a kind of method flow diagram of Face datection of one embodiment of the invention；
Fig. 4 is a kind of method flow diagram of face tracking of one embodiment of the invention；
Fig. 5 is the method flow diagram that a kind of face characteristic of one embodiment of the invention obtains；
Fig. 6 is a kind of recognition of face sample collection information network topology diagram of one embodiment of the present of invention；
Fig. 7 is a kind of recognition of face sample collection method flow diagram of one embodiment of the invention；
Fig. 8 is a kind of recognition of face sample acquisition system block diagram based on video mode of one embodiment of the invention.
Reference numeral is expressed as in figure：1- commonly acquires front end, 2- intelligent electronic devices, 3- front-end processor, 4- first
Server, 5- second servers, the first clients of 6-, the second clients of 7-.
Specific embodiment
Embodiment 1
The present embodiment provides a kind of recognition of face sample collection method based on video mode, as shown in Figure 1, including as follows
Step：
S1：Obtaining has the image-region of movable information in the picture frame of video, as shown in Fig. 2, including：
Each pixel or the background model of pixel region are established first with the initial image frame of video, and initial image frame can be with
Choose the 1st frame or preceding 3 frame.
Foreground model is established to pixel each in current image frame or pixel region.
Calculate the difference between each pixel or the background model and foreground model of pixel region.
A predetermined threshold value 20*20 pixel is determined, if the connected region for generating the pixel region of the difference is less than institute
Predetermined threshold value is stated, illustrates that current frame image does not generate movable information, it is impossible to which there are faces, delete the figure that can not possibly generate face
As frame, background model is updated, if the connected region for generating the pixel region of the difference is more than or equal to the predetermined threshold value, is said
Bright current frame image generates movable information, it is understood that there may be face, retaining present frame and obtaining the pixel region and be used as has fortune
The image-region of dynamic information.
S2：Described image region is detected to obtain face pixel region, as shown in figure 3, including：
Image pyramid is established to current frame image, the image-region with movable information obtained in step S1 is carried out
It scales and generates image pyramid.
Each posture faceform of off-line training is loaded into, the face picture of Face datection is used in the multi-pose Face model
Plain region is 20*20 pixels.Each posture faceform of the off-line training is that advance off-line training is good, in the method only
Being loaded into can use.The off-line training process of each posture faceform is：First by the figure containing real human face of acquisition
As storing to face positive sample library；The image for not containing face of acquisition is stored to face negative example base；Pass through sample collection work
Author manually demarcates, and determines accurate coordinates and posture direction of the face in image is acquired in positive sample library；It is sat according to face
Information less face pixel region of extraction background interference from acquisition image is marked, and calculates the feature of the pixel region, institute
It states and is characterized as Haar features；The feature is integrated, obtains each posture faceform of off-line training.
The feature on each tomographic image pyramid is calculated, it is described to be characterized as Haar features.
Response of the feature on each posture faceform of off-line training is calculated, by by each tomographic image pyramid
On feature filtering operation is carried out on each posture faceform of off-line training so as to the value that meets with a response.
Response described in comprehensive analysis obtains maximum response, if maximum response is less than Face datection threshold value, exports
Non-face mark deletes current image frame；If maximum response is more than or equal to Face datection threshold value, face mark is exported, is carried
Take the attitude information of face pixel region and face in current image frame.
S3：To face pixel region in the picture frame containing face into line trace, the history image frame of same face is obtained,
As shown in figure 4, including：
Face tracking model and face Track Initiation node are established, is first judged in the current image frame obtained in step s 2
Whether face pixel region has had face trace model：If it is not, the texture and edge feature of calculating human face region are simultaneously
Face tracking model is formed, and face pixel region coordinate is added in into face Track Initiation node, the face tracking model can
To be made of family's Haar wave filters；If fruit has, continue next step.
Calculate the sound of the textural characteristics and edge feature of face pixel region in current image frame on face tracking model
It should；
Face pixel region coordinate corresponding to peak response in current image frame is added in the tail portion node of face track；
If the face pixel region of current image frame has reached the boundary of picture frame, terminate tracking, obtain from face track
Start node is to the history image frame of the same face of face track tail portion node；Otherwise, face tracking is established back to described
Model and face Track Initiation node step.
S4：The history image frame of the same face high to quality carries out face characteristic extraction, obtains needed for recognition of face
Face characteristic, as shown in figure 5, including：
The picture frame high to quality carries out illumination pretreatment.The acquisition methods of the high picture frame of the quality are to step S3
The history image frame of the same face of middle acquisition carries out face evaluation processing：According to the resolution ratio of face, human face posture, clarity
Overall merit is carried out to pledge to the face pixel region of each picture frame in the history image frame of same face with symmetry
Amount scoring, each picture frame in the history image frame of the same face of the quality score are ranked up, and obtain the matter
Amount scoring reaches the picture frame picture frame high as quality of given threshold.The picture frame high to quality carry out illumination pretreatment it
Face pixel region is zoomed to the face resolution ratio of limitation afterwards, it is general to choose 96*96 pixels.
Calculate the face characteristic of face pixel region in the picture frame after illumination pretreatment.Set 5 scales, 8 directions
40 Gabor cores, convolution is done by face pixel region in picture frame of the Gabor verifications after photo-irradiation treatment, then into
The processing of row piecemeal obtains face characteristic.
Principle components analysis and independent component analysis are carried out to the face characteristic.Described in being reduced by Principle components analysis
Linear Redundancy degree between face characteristic promotes the discrimination between the face characteristic, realization pair by independent component analysis
The dimensionality reduction of the face characteristic.
Obtain the face characteristic needed for recognition of face.
S5：The face characteristic is stored, including：
The face characteristic obtained in step S4 is compared with each face characteristic in Face Sample Storehouse, than
Pair when Euclidean distance may be used be compared.
If the face characteristic is close with some face characteristic in Face Sample Storehouse, by the face characteristic and correspondence
The storage of face pixel region in the similar face feature samples message sample library；
If the face characteristic differs greatly with face characteristic each in Face Sample Storehouse, a face characteristic letter is created
Cease face characteristic and corresponding face pixel region described in sample library storage.
Before step S1, step is further included：
S0：Obtain the video of shooting.According to the required minimum face imaging of the inside and outside parameter of video camera and recognition of face
Resolution ratio determines that camera pedestal sets up an office and shooting operating point, then obtains the video of video camera shooting.Minimum face imaging resolution
Two eye pupil of face of recognition of face can be limited away from the face pixel region not less than 50*50 pixels or recognition of face not
Less than 96*96 pixels, face pixel region is chosen in the present embodiment and is not less than 96*96 pixels.Generally, video camera set up from
Ground level is 2.5-4 meters, and monitoring width is 3 meters or so, and camera pitch angle is within 15 degree, and the focal length of camera lens is 25-50mm.
A kind of recognition of face sample collection method based on video mode described in the present embodiment obtains the figure of video first
As having the image-region of movable information in frame, described image region is then detected to obtain face pixel region, then to containing
There is in the picture frame of face face pixel region into line trace, obtain the history image frame of same face, then to high same of quality
The history image frame of one face carries out face characteristic extraction, obtains the face characteristic needed for recognition of face, finally stores the people
Face feature.The process of entire recognition of face sample collection does not need to sample collection worker manual operation, and the speed of acquisition is fast, effect
Rate is high.
Embodiment 2
The present embodiment provides a kind of recognition of face sample collection methods based on video mode.Recognition of face sample collection is believed
Breath network topological diagram is as shown in fig. 6, mainly have acquisition front end, back-end server and three, customer account management end part according to function.
There are common acquisition 2 two kinds of forms in front end 1 and intelligent electronic device in acquisition front end, and front-end processor 3 exports head end video file, general
Video image after logical acquisition 1 exports coding of front end, the output of intelligent electronic device 2 include the groups of people that intelligent front end is extracted
Face information refers specifically to the information of the forms such as video frame, face snap area image or face characteristic data by concentration.Rear end
Server includes first server 4 and second server 5, and the video of front-end collection is carried out face characteristic and carried by first server 4
It takes, aspect ratio pair is then carried out with the face in face sample database, second server 5 is used for according to comparison result to sample
Database carries out dilatation operation.Customer account management end includes the first client 6 and the second client 7 etc., is mainly used for manually to people
Face identification sample database such as is increased, is deleted, being changed, being looked at the operations.
Recognition of face has strict requirements to the facial image of acquisition in resolution ratio, it is therefore desirable to be set according to recognition of face
Meter person first determines the adaptable minimum facial image resolution ratio of algorithm institute, and then sample worker estimates according to camera interior and exterior parameter
Face imaging resolution is calculated, suitable camera frame is finally selected to set up an office and shooting operating point.Due to Generic face recognizer
It is required that the imaging resolution of human face region is 96*96, in the present embodiment, using million grades of high-definition cameras.Camera is from monitored space
The distance U (rice) in domain center can be according to the focal length f (millimeter) of video camera, monitoring width W (rice) and Sensor target surface sizes a (millis
Rice) it estimates：Empirical equation is approximately equal to f*W/a for U.Generally, it is 2.5-4 meters that video camera, which sets up height from the ground, prison
It is 3 meters or so to control width, and camera pitch angle is within 15 degree, and the focal length of camera lens is 25-50mm.During if there is backlight, polarisation,
Camera is needed to support the functions such as low-light (level) and wide dynamic.If for night-time scene, need to support auto iris.
Recognition of face sample collection method flow diagram is as shown in Figure 7：
First, the minimum face resolution ratio limitation needed for recognition of face is determined, for example requirement is for the face two of recognition of face
Eye pupil not less than 50*50 pixels or requires the face of recognition of face to be not less than 96*96 pixels in pixel away from distance.
Then, the multi-pose Face detection model of off-line training, such as 20*20 are loaded into face sample automated collection systems
Pixel；And the face characteristic extraction model in some fixed resolution, such as 96*96 pixels.
Then, face sample automated collection systems are opened, and each frame in video is analyzed and processed.Processing procedure
Include the following steps：
T1：Video concentration establishes each pixel or the background model of pixel region, just first with the initial frame of video
Beginning frame can choose the 1st frame or preceding 3 frame.After Background Modeling, each pixel or the foreground model of pixel region are calculated,
If it find that differing greatly between each pixel or the prospect and background model of pixel region, then assigned for the pixel or pixel region
Prospect is given to identify.If pixel or pixel region containing prospect mark have certain scale, such as the company containing prospect mark
Logical region is more than 20*20 pixels, then exports movable information mark, otherwise, then exports without motion message identification, and without motion is believed
It ceases the pixel of mark or the background model of pixel region is updated.
T2：Face datection processing, the first picture frame to video concentration module output or the administrative division map containing movable information
As frame zooms in and out and generates image pyramid；Then, by each layer pyramid diagram picture and the multi-pose Face model of off-line training
On do filtering operation so as to the image that meets with a response, for example, the multi-pose Face model of off-line training is family's Haar wave filters, then
Each layer pyramid is allowed to pass through this family Haar wave filters；Finally, these filter responses of comprehensive analysis, determine the motion identification
Region whether be face and face specific posture.
T3：Face tracking processing, face tracking are mainly used for generating face movement locus.It is defeated in face detection module first
Face tracking model and face Track Initiation node are established on the face pixel region gone out, face tracking model is filtered by family Haar
Then the compositions such as wave device, when the image analysis period arrives, generate topography around previous frame face pixel region, allow
This topography, then, will be corresponding to peak response according to response of the topography on model by face tracking model
Coordinate be added to the node tail of face track.
T4：Face evaluation is handled, when the movement locus of certain face reaches image boundary, using being buffered in video buffer pond
Image information and face historical track information, obtain the history image frame of the face, and investigate the resolution ratio of history face, appearance
State, clarity and symmetry provide the quality score of history facial image frame.The quality score is ranked up, and exports that
A little scorings reach the history facial image frame corresponding to recognition of face requirement.
T5：Face characteristic extraction process first carries out history facial image frame illumination pretreatment and by facial image frame
Zoom to the face resolution ratio of face characteristic model limitation, such as 96*96 pixels；Then, 40 Gabor cores, such as 5 are set
Convolution is done, then piecemeal obtains the feature of face in scale, 8 directions with these Gabor collecting images；Then pass through main component
Analysis reduces the Linear Redundancy degree between feature；Pass through the discrimination between independent component analysis lifting feature again；Finally, it obtains
Face characteristic needed for recognition of face.
T6：Face alignment is with entering library module, by each individual in the face characteristic of history facial image frame and Face Sample Storehouse
Face feature is compared, such as Euclidean distance, if the feature of some sample and the face characteristic " distance " of acquisition are relatively near, adopts
The face and its feature of collection are appended to the information area that the face is stored in sample database, otherwise, then to adopt in Face Sample Storehouse
The face and feature of collection create an information storage area.
A kind of recognition of face sample collection method based on video mode provided in this embodiment completes recognition of face sample
During this automatic collection, volunteer only need to by monitoring point position, and sample collection worker need to only ensure network it is normal and
The normal power supply of shooting point position, it is both time saving or laborsaving in this way for sample collection worker；Since camera fixed scene is set up
It is to be designed according to face recognition algorithms, and in gatherer process, camera interior and exterior parameter is constant, so the face of shooting exists
Meet recognition of face requirement in resolution ratio；Face can be oriented from video automatically, can also select suitable recognition of face
Face, additionally it is possible to automatic dilatation is carried out to original recognition of face sample database according to the result of recognition of face, so, to greatest extent
The data input for reducing Face Sample Storehouse administrative staff.
Embodiment 3
The present embodiment provides a kind of recognition of face sample acquisition system based on video mode, as shown in figure 8, including：
Image-region acquisition module, for obtaining the image-region in the picture frame of video with movable information；
Face detection module, for detecting described image region to obtain face pixel region；
Face tracking module, for, into line trace, obtaining same people to face pixel region in the picture frame containing face
The history image frame of face；
Face characteristic acquisition module carries out face characteristic extraction for the history image frame of the same face high to quality,
Obtain the face characteristic needed for recognition of face；
Face characteristic memory module, for storing the face characteristic.
Described image region acquisition module includes：
Background Modeling submodule, for establishing background model using the initial image frame in the video；
Foreground model setting up submodule, for establishing foreground model to current image frame；
Difference computational submodule, for calculating the difference of the background model and the foreground model；
Image-region acquisition submodule, the difference for obtaining between foreground model and background model are more than predetermined threshold value
Pixel region, as the image-region with movable information.
In the face detection module：
Image pyramid setting up submodule, for establishing image pyramid to current image frame；
Off-line model is loaded into submodule, for being loaded into each posture faceform of off-line training；
Feature calculation submodule, for calculating the feature on each tomographic image pyramid；
Response computational submodule, for calculating response of the feature on each posture faceform of off-line training
Value；
Face pixel region acquisition submodule, for obtaining the picture frame that maximum response reaches Face datection threshold value, carries
Take the attitude information of face pixel region and face in current image frame.
The face tracking module includes：
Face tracking model foundation submodule, for establishing face tracking model and face Track Initiation node；
Characteristic response computational submodule, it is special for calculating the textural characteristics of face pixel region and edge in current image frame
Levy the response on face tracking model；
Peak response handles submodule, for by the face pixel region coordinate corresponding to peak response in current image frame
Add in the tail portion node of face track；
The history image frame acquisition submodule of same face, for reaching image in the face pixel region of current image frame
During the boundary of frame, terminate tracking, obtain from face Track Initiation node to the history of the same face of face track tail portion node
Picture frame.
The face characteristic acquisition module includes：
Submodule is pre-processed, for carrying out illumination pretreatment to the high picture frame of quality；
Face characteristic computational submodule, for calculating the face of face pixel region in the picture frame after illumination pretreatment
Feature；
Human face analysis submodule, for carrying out Principle components analysis and independent component analysis to the face characteristic；
Face characteristic acquisition submodule, for obtaining the face characteristic needed for recognition of face.
The face characteristic memory module includes：
Face characteristic compares submodule, for will be in face characteristic described in face characteristic acquisition module and Face Sample Storehouse
Each face characteristic be compared；
Face characteristic sub-module stored, if for the face characteristic and some face characteristic phase in Face Sample Storehouse
Closely, then the face characteristic and corresponding face pixel region are stored to the similar face feature samples message sample library
In；If the face characteristic differs greatly with face characteristic each in Face Sample Storehouse, a face characteristic information sample is created
Face characteristic described in this library storage and corresponding face pixel region.
A kind of recognition of face sample acquisition system based on video mode described in the present embodiment obtains the figure of video first
As having the image-region of movable information in frame, described image region is then detected to obtain face pixel region, then to containing
There is in the picture frame of face face pixel region into line trace, obtain the history image frame of same face, then to high same of quality
The history image frame of one face carries out face characteristic extraction, obtains the face characteristic needed for recognition of face, finally stores the people
Face feature.The process of entire recognition of face sample collection does not need to sample collection worker manual operation, and the speed of acquisition is fast, effect
Rate is high.
It should be understood by those skilled in the art that, the embodiment of the present invention can be provided as method, system or computer program
Product.Therefore, the reality in terms of complete hardware embodiment, complete software embodiment or combination software and hardware can be used in the present invention
Apply the form of example.Moreover, the computer for wherein including computer usable program code in one or more can be used in the present invention
The computer program production that usable storage medium is implemented on (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.)
The form of product.
The present invention be with reference to according to the method for the embodiment of the present invention, the flow of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that it can be realized by computer program instructions every first-class in flowchart and/or the block diagram
The combination of flow and/or box in journey and/or box and flowchart and/or the block diagram.These computer programs can be provided
The processor of all-purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices is instructed to produce
A raw machine so that the instruction generation that the processor set by computer or the processing of other programmable datas performs is used to implement
The device of function specified in one flow of flow chart or multiple flows and/or one box of block diagram or multiple boxes.
These computer program instructions, which may also be stored in, can guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works so that the instruction generation being stored in the computer-readable memory includes referring to
Enable the manufacture of device, the command device realize in one flow of flow chart or multiple flows and/or one box of block diagram or
The function of being specified in multiple boxes.
These computer program instructions can be also loaded into computer or other programmable data processing devices so that counted
Series of operation steps are performed on calculation machine or other programmable devices to generate computer implemented processing, so as in computer or
The instruction offer performed on other programmable devices is used to implement in one flow of flow chart or multiple flows and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Although preferred embodiments of the present invention have been described, but those skilled in the art once know basic creation
Property concept, then can make these embodiments other change and modification.So appended claims be intended to be construed to include it is excellent
It selects embodiment and falls into all change and modification of the scope of the invention.
Claims (10)
- A kind of 1. recognition of face sample collection method based on video mode, which is characterized in that include the following steps：Obtaining has the image-region of movable information in the picture frame of video；Described image region is detected to obtain face pixel region；To face pixel region in the picture frame containing face into line trace, the history image frame of same face is obtained；The history image frame of the same face high to quality carries out face characteristic extraction, and the face obtained needed for recognition of face is special Sign；Store the face characteristic；The storage face characteristic step includes：The face characteristic is compared with each face characteristic in Face Sample Storehouse；If the face characteristic is close with some face characteristic in Face Sample Storehouse, by the face characteristic and corresponding people In the storage to the similar face feature samples message sample library of face pixel region；If the face characteristic differs greatly with face characteristic each in Face Sample Storehouse, a face characteristic information sample is created Face characteristic described in this library storage and corresponding face pixel region.
- 2. the recognition of face sample collection method according to claim 1 based on video mode, which is characterized in that described to obtain The image-region step with movable information in the picture frame of video is taken to include：Background model is established using the initial image frame in the video；Foreground model is established to current image frame；Calculate the difference of the background model and the foreground model；The pixel region that the difference between foreground model and background model is more than predetermined threshold value is obtained, as with movable information Image-region.
- 3. the recognition of face sample collection method according to claim 1 or 2 based on video mode, which is characterized in that institute Detection described image region is stated to obtain face pixel region step to include：Image pyramid is established to current image frame；It is loaded into each posture faceform of off-line training；Calculate the feature on each tomographic image pyramid；Calculate response of the feature on each posture faceform of off-line training；The picture frame that maximum response reaches Face datection threshold value is obtained, extracts face pixel region and face in current image frame Attitude information.
- 4. according to any recognition of face sample collection methods based on video mode of claim 1-2, which is characterized in that Face pixel region progress tracking step includes in the described pair of picture frame containing face：Establish face tracking model and face Track Initiation node；Calculate the response of the textural characteristics and edge feature of face pixel region in current image frame on face tracking model；Face pixel region coordinate corresponding to peak response in current image frame is added in the tail portion node of face track；If the face pixel region of current image frame reaches the boundary of picture frame, terminate tracking, obtain from face Track Initiation Node is to the history image frame of the same face of face track tail portion node.
- 5. according to any recognition of face sample collection methods based on video mode of claim 1-2, which is characterized in that The history image frame of the same face high to quality carries out face characteristic extraction step and includes：The picture frame high to quality carries out illumination pretreatment；Calculate the face characteristic of face pixel region in the picture frame after illumination pretreatment；Principle components analysis and independent component analysis are carried out to the face characteristic；Obtain the face characteristic needed for recognition of face.
- 6. a kind of recognition of face sample acquisition system based on video mode, which is characterized in that including：Image-region acquisition module, for obtaining the image-region in the picture frame of video with movable information；Face detection module, for detecting described image region to obtain face pixel region；Face tracking module, for, into line trace, obtaining same face to face pixel region in the picture frame containing face History image frame；Face characteristic acquisition module carries out face characteristic extraction for the history image frame of the same face high to quality, obtains Face characteristic needed for recognition of face；Face characteristic memory module, for storing the face characteristic；The face characteristic memory module includes：Face characteristic compare submodule, for by face characteristic described in face characteristic acquisition module with it is each in Face Sample Storehouse A face characteristic is compared；Face characteristic sub-module stored, if close with some face characteristic in Face Sample Storehouse for the face characteristic, It will be in the face characteristic and the storage to the similar face feature samples message sample library of corresponding face pixel region；If institute It states face characteristic to differ greatly with face characteristic each in Face Sample Storehouse, then creates a face characteristic information sample library storage The face characteristic and corresponding face pixel region.
- 7. the recognition of face sample acquisition system according to claim 6 based on video mode, which is characterized in that the figure As region acquisition module includes：Background Modeling submodule, for establishing background model using the initial image frame in the video；Foreground model setting up submodule, for establishing foreground model to current image frame；Difference computational submodule, for calculating the difference of the background model and the foreground model；Image-region acquisition submodule, the difference for obtaining between foreground model and background model are more than the picture of predetermined threshold value Plain region, as the image-region with movable information.
- 8. the recognition of face sample acquisition system based on video mode described according to claim 6 or 7, which is characterized in that institute It states in face detection module：Image pyramid setting up submodule, for establishing image pyramid to current image frame；Off-line model is loaded into submodule, for being loaded into each posture faceform of off-line training；Feature calculation submodule, for calculating the feature on each tomographic image pyramid；Response computational submodule, for calculating response of the feature on each posture faceform of off-line training；Face pixel region acquisition submodule, for obtaining the picture frame that maximum response reaches Face datection threshold value, extraction is worked as The attitude information of face pixel region and face in preceding picture frame.
- 9. according to any recognition of face sample acquisition systems based on video mode of claim 6-7, which is characterized in that The face tracking module includes：Face tracking model foundation submodule, for establishing face tracking model and face Track Initiation node；Characteristic response computational submodule exists for calculating the textural characteristics of face pixel region and edge feature in current image frame Response on face tracking model；Peak response handles submodule, for the face pixel region coordinate corresponding to peak response in current image frame to be added in The tail portion node of face track；The history image frame acquisition submodule of same face, for reaching picture frame in the face pixel region of current image frame During boundary, terminate tracking, obtain from face Track Initiation node to the history image of the same face of face track tail portion node Frame.
- 10. according to any recognition of face sample acquisition systems based on video mode of claim 6-7, feature exists In the face characteristic acquisition module includes：Submodule is pre-processed, for carrying out illumination pretreatment to the high picture frame of quality；Face characteristic computational submodule, it is special for calculating the face of face pixel region in the picture frame after illumination pretreatment Sign；Human face analysis submodule, for carrying out Principle components analysis and independent component analysis to the face characteristic；Face characteristic acquisition submodule, for obtaining the face characteristic needed for recognition of face.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410720464.XA CN104463117B (en) | 2014-12-02 | 2014-12-02 | A kind of recognition of face sample collection method and system based on video mode |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410720464.XA CN104463117B (en) | 2014-12-02 | 2014-12-02 | A kind of recognition of face sample collection method and system based on video mode |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104463117A CN104463117A (en) | 2015-03-25 |
| CN104463117B true CN104463117B (en) | 2018-07-03 |
Family
ID=52909133
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410720464.XA Active CN104463117B (en) | 2014-12-02 | 2014-12-02 | A kind of recognition of face sample collection method and system based on video mode |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104463117B (en) |
Families Citing this family (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447459B (en) * | 2015-11-18 | 2019-03-22 | 上海海事大学 | A kind of unmanned plane detects target and tracking automatically |
| CN105426730A (en) * | 2015-12-28 | 2016-03-23 | 小米科技有限责任公司 | Login authentication processing method and device as well as terminal equipment |
| CN105654055A (en) * | 2015-12-29 | 2016-06-08 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Method for performing face recognition training by using video data |
| CN105678289A (en) * | 2016-03-07 | 2016-06-15 | 谭圆圆 | Control method and device of unmanned aerial vehicle |
| CN105930822A (en) * | 2016-05-11 | 2016-09-07 | 北京格灵深瞳信息技术有限公司 | Human face snapshot method and system |
| CN107346417B (en) * | 2017-06-13 | 2020-09-01 | 浪潮金融信息技术有限公司 | Face detection method and device |
| CN107633209B (en) * | 2017-08-17 | 2018-12-18 | 平安科技（深圳）有限公司 | Electronic device, the method for dynamic video recognition of face and storage medium |
| CN107644219B (en) * | 2017-10-10 | 2021-04-02 | Oppo广东移动通信有限公司 | Face registration method and related product |
| CN109859234B (en) * | 2017-11-29 | 2023-03-24 | 深圳Tcl新技术有限公司 | Video human body trajectory tracking method and device and storage medium |
| CN108269333A (en) * | 2018-01-08 | 2018-07-10 | 平安科技（深圳）有限公司 | Face identification method, application server and computer readable storage medium |
| CN108256479B (en) * | 2018-01-17 | 2023-08-01 | 百度在线网络技术（北京）有限公司 | Face tracking method and device |
| CN108171207A (en) * | 2018-01-17 | 2018-06-15 | 百度在线网络技术（北京）有限公司 | Face identification method and device based on video sequence |
| CN110390229B (en) * | 2018-04-20 | 2022-03-04 | 杭州海康威视数字技术股份有限公司 | Face picture screening method and device, electronic equipment and storage medium |
| CN109614841B (en) * | 2018-04-26 | 2023-04-18 | 杭州智诺科技股份有限公司 | Rapid face detection method in embedded system |
| CN108491832A (en) * | 2018-05-21 | 2018-09-04 | 广西师范大学 | A kind of embedded human face identification follow-up mechanism and method |
| CN109034247B (en) * | 2018-07-27 | 2021-04-23 | 北京以萨技术股份有限公司 | Tracking algorithm-based higher-purity face recognition sample extraction method |
| CN109145771B (en) * | 2018-08-01 | 2020-11-20 | 武汉普利商用机器有限公司 | Face snapshot method and device |
| CN111241868B (en) * | 2018-11-28 | 2024-03-08 | 杭州海康威视数字技术股份有限公司 | Face recognition system, method and device |
| CN110210290A (en) * | 2019-04-22 | 2019-09-06 | 平安科技（深圳）有限公司 | Face picture acquisition method, device and computer equipment |
| CN110263704B (en) * | 2019-06-18 | 2021-07-20 | 浙江宇视科技有限公司 | Face data acquisition method, device, server, video acquisition device and medium |
| CN110458026A (en) * | 2019-07-11 | 2019-11-15 | 成都市喜爱科技有限公司 | A kind of face acquisition method, device, system, equipment and medium |
| CN110910549A (en) * | 2019-11-15 | 2020-03-24 | 江苏高泰软件技术有限公司 | Campus personnel safety management system based on deep learning and face recognition features |
| CN111046806A (en) * | 2019-12-12 | 2020-04-21 | 天地伟业技术有限公司 | Heterogeneous image face recognition target library generation method |
| CN111209818A (en) * | 2019-12-30 | 2020-05-29 | 新大陆数字技术股份有限公司 | Video individual identification method, system, equipment and readable storage medium |
| CN111652070B (en) * | 2020-05-07 | 2023-07-28 | 南京航空航天大学 | Face sequence collaborative recognition method based on monitoring video |
| CN112164097B (en) * | 2020-10-20 | 2024-03-29 | 南京莱斯网信技术研究院有限公司 | Ship video detection sample collection method |
| CN112487222B (en) * | 2020-11-30 | 2021-11-30 | 江苏正赫通信息科技有限公司 | Method for quickly searching and effectively storing similar human faces |
| CN113705510A (en) * | 2021-09-02 | 2021-11-26 | 广州市奥威亚电子科技有限公司 | Target identification tracking method, device, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101419670A (en) * | 2008-11-21 | 2009-04-29 | 复旦大学 | Video monitoring method and system based on advanced audio/video encoding standard |
| CN102542492A (en) * | 2012-01-10 | 2012-07-04 | 清华大学 | System and method for evaluating effect of visual advertisement |
| US8542941B2 (en) * | 2007-07-24 | 2013-09-24 | Nikon Corporation | Imaging device, image detecting method and focus adjusting method |
| CN103605969A (en) * | 2013-11-28 | 2014-02-26 | Tcl集团股份有限公司 | Method and device for face inputting |
| CN103942577A (en) * | 2014-04-29 | 2014-07-23 | 上海复控华龙微系统技术有限公司 | Identity identification method based on self-established sample library and composite characters in video monitoring |
-
2014
- 2014-12-02 CN CN201410720464.XA patent/CN104463117B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8542941B2 (en) * | 2007-07-24 | 2013-09-24 | Nikon Corporation | Imaging device, image detecting method and focus adjusting method |
| CN101419670A (en) * | 2008-11-21 | 2009-04-29 | 复旦大学 | Video monitoring method and system based on advanced audio/video encoding standard |
| CN102542492A (en) * | 2012-01-10 | 2012-07-04 | 清华大学 | System and method for evaluating effect of visual advertisement |
| CN103605969A (en) * | 2013-11-28 | 2014-02-26 | Tcl集团股份有限公司 | Method and device for face inputting |
| CN103942577A (en) * | 2014-04-29 | 2014-07-23 | 上海复控华龙微系统技术有限公司 | Identity identification method based on self-established sample library and composite characters in video monitoring |
Non-Patent Citations (1)
| Title |
|---|
| "视频监控下的人脸跟踪与识别系统";桑海峰等，;《计算机工程与应用》;20130116(第12期);第175-179页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104463117A (en) | 2015-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104463117B (en) | A kind of recognition of face sample collection method and system based on video mode | |
| CN108830252B (en) | Convolutional neural network human body action recognition method fusing global space-time characteristics | |
| CN110532970B (en) | Age and gender attribute analysis method, system, equipment and medium for 2D images of human faces | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| CN109522853B (en) | Face datection and searching method towards monitor video | |
| CN104091171B (en) | Vehicle-mounted far infrared pedestrian detecting system and method based on local feature | |
| CN107145845A (en) | The pedestrian detection method merged based on deep learning and multi-characteristic points | |
| CN110210276A (en) | A kind of motion track acquisition methods and its equipment, storage medium, terminal | |
| CN109447169A (en) | The training method of image processing method and its model, device and electronic system | |
| CN107844785B (en) | A kind of method for detecting human face based on size estimation | |
| CN102932605B (en) | Method for selecting camera combination in visual perception network | |
| CN103324677B (en) | Hierarchical fast image global positioning system (GPS) position estimation method | |
| Ren et al. | A novel squeeze YOLO-based real-time people counting approach | |
| CN107909081A (en) | The quick obtaining and quick calibrating method of image data set in a kind of deep learning | |
| CN107798313A (en) | A kind of human posture recognition method, device, terminal and storage medium | |
| CN105138954A (en) | Image automatic screening, query and identification system | |
| CN109508671A (en) | A kind of video accident detection system and method based on Weakly supervised study | |
| CN107315795B (en) | The instance of video search method and system of joint particular persons and scene | |
| CN105096300B (en) | Method for checking object and equipment | |
| CN104182765A (en) | Internet image driven automatic selection method of optimal view of three-dimensional model | |
| CN104504395A (en) | Method and system for achieving classification of pedestrians and vehicles based on neural network | |
| CN109871821A (en) | The pedestrian of adaptive network recognition methods, device, equipment and storage medium again | |
| CN109766873A (en) | pedestrian re-identification method based on hybrid deformable convolution | |
| CN110110755A (en) | Based on the pedestrian of PTGAN Regional disparity and multiple branches weight recognition detection algorithm and device | |
| CN110176024A (en) | Method, apparatus, equipment and the storage medium that target is detected in video |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |