CN110070062A - A kind of system and method for the recognition of face based on binocular active infrared - Google Patents
A kind of system and method for the recognition of face based on binocular active infrared Download PDFInfo
- Publication number
- CN110070062A CN110070062A CN201910351639.7A CN201910351639A CN110070062A CN 110070062 A CN110070062 A CN 110070062A CN 201910351639 A CN201910351639 A CN 201910351639A CN 110070062 A CN110070062 A CN 110070062A
- Authority
- CN
- China
- Prior art keywords
- face
- image
- recognition
- head portrait
- infrared
- 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.)
- Pending
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
- G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Collating Specific Patterns (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
This application provides a kind of system and method for recognition of face based on binocular active infrared, the system comprises: binocular infrared eye, for taking pictures to target face, to obtain two infrared images；Processing unit includes neural network recognization unit, for obtaining the first image, and it inputs in 2D recognition of face neural network and carries out recognition of face, when target face head portrait and the recognition of face parameter of any face head portrait in registration database in 2D recognition of face neural network are coincide in identifying the first image, the information of the corresponding target person of target face head portrait is obtained；Binocular parallax recognition unit is used to construct 3D rendering according to two infrared images, and 3D rendering is then carried out face characteristic with the first image and is compared；Transmission unit is used for when 3D rendering is identical as the face characteristic of the first image, exports the information of target person.The application carries out recognition of face by binocular active infrared, has not only been able to satisfy low reject rate and misclassification rate, but also can solve illumination and In vivo detection problem.
Description
Technical field
The present invention relates to technical field of computer vision more particularly to a kind of recognitions of face based on binocular active infrared
System and method.
Background technique
Field of face identification is used widely in all trades and professions, but various technologies, cost, the technique of recognition of face
The defects of also emerge one by one in application process.All there is many defects, such as tradition in traditional 2D and 3D recognition of face
For 2D camera when light is weaker, discrimination is not high, is taken a favorable turn using infrared light filling Shi Caihui；In addition, being needed when for In vivo detection
User is wanted to cooperate, and nonetheless, it is also possible to cheat the In vivo detection program of 2D recognition of face by video recording mode；Tradition
3D recognition of face colleague uses structure light, TOF or the binocular camera with characteristic point light filling to be identified, at high cost, yield
It is low.
In field of face identification, while how meeting low reject rate and low misclassification rate, lighting issues and living body inspection are solved
Survey is just to have needed, and on this basis if can reduce cost, reduce process complexity, has an opportunity for face recognition technology to be applied to more
In extensive scene.
Summary of the invention
In order to overcome the above problem, embodiments herein provides a kind of recognition of face based on binocular active infrared
System and method.
In order to achieve the above object, embodiments herein adopts the following technical scheme that
In a first aspect, the application provides a kind of system of recognition of face based on binocular active infrared, comprising: binocular is infrared
Photographic device, for taking pictures to target face, to obtain two infrared images；Processing unit, the processing unit include
Neural network recognization unit, binocular parallax recognition unit and transmission unit, the neural network recognization unit, for obtaining first
Image, and input in 2D recognition of face neural network and carry out recognition of face, the target face head in identifying the first image
When as coincideing with the recognition of face parameter of any face head portrait in registration database in the 2D recognition of face neural network, obtain
The information of the corresponding target person of the target face head portrait；The first image is an image in two infrared images；
The binocular parallax recognition unit, for constructing 3D rendering according to two infrared images, then by the 3D rendering and institute
It states the first image and carries out face characteristic comparison；The face characteristic is for determining that the target is living body；The transmission unit is used
In when the 3D rendering is identical as the face characteristic of the first image, the information of the target person is exported.
In another possible realization, the binocular infrared eye further includes at least one floodlight light compensating lamp, institute
Floodlight light compensating lamp is stated, for carrying out infrared floodlight light filling to the target.
In another possible realization, the compensation wavelength that the floodlight light compensating lamp provides is 850nm or 940nm.
In another possible realization, the neural network recognization unit is specifically used for, and the first image is input to
In the 2D recognition of face neural network, judge whether to identify face head portrait；When the 2D recognition of face neural network recognization
Out in the first image after face head portrait, face head portrait in the first image is obtained, and judge people in the first image
Whether the recognition of face parameter of face head portrait is in the registered face list；When the face of face head portrait in the first image is known
Other parameter obtains the recognition of face parameter of face head portrait in the first image in registered face in the registered face list
The information of corresponding target person in list.
In another possible realization, the binocular parallax recognition unit is specifically used for, and is gone out by neural network recognization
The stereo profile of the 3D rendering, and be the size of the first image by the size adjusting of the 3D rendering, then confirm institute
Whether the profile of the stereo profile and the first image of stating 3D rendering is identical.
Second aspect, the application provide a kind of method of recognition of face based on binocular active infrared, comprising: to target person
Face is taken pictures, to obtain two infrared images；The first image is obtained, and inputs in 2D recognition of face neural network and carries out face
Identification, in identifying the first image in target face head portrait and the 2D recognition of face neural network in registration database
When the recognition of face parameter of any face head portrait is coincide, the information of the corresponding target person of the target face head portrait is obtained；It is described
First image is an image in two infrared images；3D rendering is constructed according to two infrared images, then by institute
3D rendering is stated to compare with the first image progress face characteristic；The face characteristic is for determining that the target is living body；?
When the 3D rendering is identical as the face characteristic of the first image, the information of the target person is exported.
In another possible realization, take pictures described to target, before obtaining two infrared images, packet
It includes: infrared floodlight light filling is carried out to the target.
In another possible realization, the first image of the acquisition, and input in 2D recognition of face neural network and carry out
Recognition of face, target face head portrait and log-on data in the 2D recognition of face neural network in identifying the first image
When the recognition of face parameter of any face head portrait is coincide in library, the information of the corresponding target person of the target face head portrait is obtained,
Specifically include: the first image is input in the 2D recognition of face neural network, judges whether to identify face head portrait；When
The 2D recognition of face neural network recognization goes out in the first image after face head portrait, obtains face head in the first image
Picture, and judge the recognition of face parameter of face head portrait in the first image whether in the registered face list；When described
The recognition of face parameter of face head portrait obtains face head in the first image in the registered face list in first image
The information of recognition of face parameter corresponding target person in registered face list of picture.
It is described that 3D rendering is constructed according to two infrared images in another possible realization, then by the 3D
Image carries out face characteristic with the first image and compares, and specifically includes: going out the vertical of the 3D rendering by neural network recognization
Body profile, and be the size of the first image by the size adjusting of the 3D rendering, then confirm the solid of the 3D rendering
Whether profile and the profile of the first image are identical.
The application carries out recognition of face by binocular active infrared, has not only been able to satisfy low reject rate and misclassification rate, but also can solve
Lighting issues and In vivo detection problem, and the application uses general infrared camera and general floodlight light filling, cost
More have compared with existing structure light, TOF, the binocular camera with characteristic point light filling (also have and be structure light light filling) with technique excellent
Gesture.
Detailed description of the invention
The attached drawing used required in embodiment or description of the prior art is briefly described below.
Fig. 1 is a kind of structural representation of the face identification system based on binocular active infrared provided by the embodiments of the present application
Figure；
Fig. 2 is a kind of framework signal of face identification system based on binocular active infrared provided by the embodiments of the present application
Figure；
Fig. 3 is that 2D recognition of face neural network provided by the embodiments of the present application is trained the schematic diagram with registration process；
Fig. 4 is a kind of flow chart of the method for the recognition of face based on binocular active infrared provided by the embodiments of the present application；
Fig. 5 is a kind of process signal of method of the recognition of face based on binocular active infrared provided by the embodiments of the present application
Figure.
Specific embodiment
Embodiments herein is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar unit or unit with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, and is only used for explaining the application, and should not be understood as the limitation to the application.
Fig. 1 is a kind of structural representation of the face identification system based on binocular active infrared provided by the embodiments of the present application
Figure.As shown in Figure 1, the system includes floodlight compensating lamp, camera 1, camera 2 and processor.
Floodlight light compensating lamp generates infrared light, infrared floodlight light filling is carried out to face, to guarantee camera 1 and camera 2
The infrared image of face can be acquired.Camera 1 and camera 2 obtain the infrared image of two faces by acquisition facial image,
Processor is then sent to be handled.Processor first select wherein an infrared image handled, judge the infrared image
In whether have facial image.If not having facial image in the infrared image, processor if, prompts unidentified target；If this is red
There is facial image in outer image, processor will acquire the face head portrait in the infrared image in facial image, and according to the face
The face characteristic collection of head portrait is compared with the face characteristic collection in database.If the face characteristic collection of the face head portrait does not exist
In database, processor then prompts unidentified target；If a certain face in the face characteristic collection of the face head portrait and database
When feature set is wholly or largely identical, then the information of the corresponding target person of face feature set is obtained.
Then processor constructs 3D rendering according to according to this two infrared images, then above-mentioned infrared image and 3D are schemed
As carrying out face characteristic comparison, to judge whether the target person in the image obtained is living body.The face characteristic wherein compared is
Show that target is the feature of living body.If the face characteristic of comparison is not identical, processor then prompts unidentified target；If
When the face characteristic of comparison is wholly or largely identical, then show the artificial living body of target in the image obtained, then processor is sent out
It send the information of target person to external equipment, shows system identification success.
The application carries out recognition of face by binocular active infrared, has not only been able to satisfy low reject rate and misclassification rate, but also can solve
Lighting issues and In vivo detection problem, and the application uses general infrared camera and general floodlight light filling, cost
More have compared with existing structure light, TOF, the binocular camera with characteristic point light filling (also have and be structure light light filling) with technique excellent
Gesture.
Fig. 2 is a kind of framework signal of face identification system based on binocular active infrared provided by the embodiments of the present application
Figure.As shown in Fig. 2, the system includes binocular infrared eye 1 and processing unit 2.
Binocular infrared eye 1 includes two infrared cameras 11 and 12, for taking pictures to target, obtains two
Infrared image.The wherein binocular infrared camera that infrared camera 11 and infrared camera 12 form, can be obtained by way of calibration
Its internal reference and outer ginseng are taken, to obtain the position orientation relation between them.
There is baseline width between infrared camera 11 and infrared camera 12, for obtaining the red of target from different perspectives
Outer image.Wherein, baseline width can be increased or be reduced as needed, can also adjust the direction of baseline to horizontally or vertically,
Or there is certain angle with level.
Binocular infrared eye 1 further includes floodlight light compensating lamp 13, for carrying out infrared floodlight light filling to target, to guarantee
Infrared camera 11 and infrared camera 12 can collect the infrared image of target.
The light filling wavelength of floodlight light compensating lamp 13 can be 850nm or 940nm, and intensity of illumination can ensure that infrared camera energy
Normal imaging or itself it will ensure that the wavelength of infrared camera normal imaging and the light compensating lamp of intensity of illumination.
Preferably, under certain applications scene, preferred 940nm light compensating lamp, it can increase energy outside camera lens and light compensating lamp
The protective layer for realizing infrared light transmission allows detected personnel to be not felt by the presence of camera lens and light compensating lamp, improves the personnel of being detected
User experience.
Wherein, the quantity of floodlight infrared light compensating lamp 13 can be one or more.
Processing unit 2 includes neural network recognization unit 21, binocular parallax recognition unit 22 and transmission unit 23.
Neural network recognization unit 21 is used to obtain in the infrared image that infrared camera 11 and infrared camera 12 are clapped
One image, and input in 2D recognition of face neural network and carry out recognition of face, when the 2D recognition of face neural network recognization
Out in the image after face head portrait, face head portrait in the image is obtained, and judges the recognition of face ginseng of face head portrait in the image
Whether number is in the registered face list；When the recognition of face parameter of face head portrait in the image is in the registered face list
In, obtain the information of recognition of face parameter corresponding target person in registered face list of face head portrait in the image.
Fig. 3 is that 2D recognition of face neural network provided by the embodiments of the present application is trained the schematic diagram with registration process.
As shown in figure 3,2D recognition of face neural network in the training process, by by existing picture format mainly include band someone
The RGB image of face, the RGB that these RGB files switch to gray level image is turned gray level image, visible light MONO camera acquisition it is visible
The active infrared NIR image with floodlight light filling that light grayscale image MONO image, the active infrared camera with floodlight light filling are shot
These four formats input 2D recognition of face neural network is trained, and the training data that can identify face head portrait is obtained, so that 2D
Recognition of face neural network can identify the face head portrait in image.
Wherein, the training data of acquisition can be the Facial Features such as eyes size, the size of mouth, face contour, acquisition
Facial Features are more, and the result of identification is more accurate.It may include several for each face head portrait or even tens Facial Features
Collection.
It, will while input great amount of images file is trained 2D recognition of face neural network in registration process
Each RGB image file, RGB with face turns gray level image file, visible light grayscale image MONO image file and with floodlight
The log-on data of the active infrared NIR image file of light filling is input in 2D recognition of face neural network, then knows 2D face
The registration information of the image file of the Facial Features collection and input of the face head portrait identified in other neural network is matched one by one, structure
At registered face list.By determining the relationship of the log-on data of face head portrait and the corresponding image file of face head portrait, realize
2D recognition of face neural network can obtain identical registration number according to the different images file of identical face head portrait
According to.
It in one embodiment, include party A-subscriber's face head portrait in the registered face list of 2D recognition of face neural network
Facial Features collection B and image B be named as the registration information of " party A-subscriber ", and Facial Features collection B and registration information " party A-subscriber " are
Pairing relationship.If identified in system input picture C to 2D recognition of face neural network, when 2D recognition of face neural network
It identifies in image C after including face head portrait, obtains Facial Features collection D in image C, and by Facial Features collection D to registrant
It is matched in face list.If Facial Features are completely or exhausted in Facial Features collection B and Facial Features collection D in registered face list
When most of identical, it is registered users that 2D recognition of face neural network recognization, which goes out image C, and obtains the registration letter of " party A-subscriber "
Breath；If when the Facial Features collection not matched with Facial Features collection D in registered face list, prompting unidentified target out
Or provide corresponding status information.
Wherein, the mode that 2D recognition of face neural network obtains registration information can be taken pictures the head of generation using remote terminal
As the head portrait file or existing file progress remote login on file, resident identification card or other certificates.
Binocular parallax recognition unit 22 is used to obtain two infrared images that infrared camera 11 and infrared camera 12 are clapped,
It by neural network or uses conventional methods and is converted into disparity map, and thus export 3D depth map, then to the 3D rendering
Disparity Analysis is carried out, identifies face stereoscopic features, the infrared image then obtained with neural network recognization unit 21 is closed
Key aspect ratio pair.Wherein, the face key feature of comparison come confirm target be living body.
If the key feature for the infrared image that 3D depth map and neural network recognization unit 21 obtain compares not identical,
Binocular parallax recognition unit 22 prompts unidentified target out, or provides corresponding status information；If 3D depth map and nerve net
When the key feature for the infrared image that network recognition unit 21 obtains compares identical, system identification success, binocular parallax identification are indicated
Unit 22 indicates that transmission unit 23 exports the information of the target person obtained in neural network recognization unit 21.
It should be noted that binocular parallax recognition unit 22 obtain in 3D depth map with neural network recognization unit 21 it is red
In the key feature comparison process of outer image, threshold value can be set.When the key feature quantity being compared reaches threshold value, then table
Show and compares successfully；When the key feature quantity being compared is not up to threshold value, then it represents that compare failure.
In one embodiment, the stereo profile of 3D rendering is identified by neural network or image algorithm, and 3D is schemed
The size adjusting of picture is the size of the first image, then confirm stereo profile and the first image of 3D rendering profile whether phase
Together.
Wherein, during identifying face stereoscopic features, to two infrared images progress with floodlight light filling and parallax
Analysis, forms the disparity map of two images, then can form the stereo profile of face by disparity map, pass through the inspection of key position
Survey the profile that confirmation stereo profile is face.
The application constitutes binocular active infrared camera module by two infrared cameras in binocular infrared eye
Recognition of face is carried out, first selects an image to know by neural network recognization unit and carries out recognition of face, determine target face
It whether is registered；Then two images are converted by 3D rendering by binocular parallax recognition unit, by 3D rendering and pass through net
The image of network neural unit carries out key feature comparison, if the key feature compared is identical, system identification success, and Ji Nengman
The low reject rate of foot and misclassification rate, and can solve lighting issues and In vivo detection problem.In addition, use general infrared camera with
And general floodlight light filling, cost and technique and existing structure light, TOF, band characteristic point light filling (being also structure light light filling)
Binocular camera compared to advantageously.
Fig. 4 is a kind of flow chart of the method for the recognition of face based on binocular active infrared provided by the embodiments of the present application.
As shown in figure 4, the application provides a kind of method of recognition of face based on binocular active infrared, this method implements step such as
Under:
Step S401 takes pictures to target face, to obtain two infrared images.
Specifically, the application takes pictures to target by two infrared cameras, obtains two infrared images.Wherein two
The binocular infrared camera of a infrared camera composition, can obtain its internal reference and outer ginseng by way of calibration, with obtain them it
Between position orientation relation.
There is baseline width between two infrared cameras, for obtaining the infrared image of target from different perspectives.Wherein,
Baseline width can be increased or be reduced as needed, can also adjust the direction of baseline to horizontally or vertically, or have with level
Certain angle.
Optionally, before step S401, comprising: carry out infrared floodlight light filling to target.To guarantee two infrared photographies
Head can collect the infrared image of target.
The light filling wavelength for carrying out the floodlight light compensating lamp of infrared floodlight light filling to target can be 850nm or 940nm, illumination
Intensity can ensure that infrared camera energy normal imaging or itself it will ensure that the wavelength of infrared camera normal imaging and illumination are strong
The light compensating lamp of degree.
Preferably, under certain applications scene, preferred 940nm light compensating lamp, it can increase energy outside camera lens and light compensating lamp
The protective layer for realizing infrared light transmission allows detected personnel to be not felt by the presence of camera lens and light compensating lamp, improves the personnel of being detected
User experience.
Wherein, the quantity of floodlight infrared light compensating lamp 13 can be one or more.
Step S403 obtains the first image, and inputs in 2D recognition of face neural network and carry out recognition of face, is identifying
Target face head portrait and the face of any face head portrait in registration database in 2D recognition of face neural network are known in first image
When other parameter is coincide, the information of the corresponding target person of target face head portrait is obtained.
Specifically, 2D recognition of face neural network mainly includes training, registers and identify these three processes.In training process
In, by the way that the RGB image file, the RGB that have face are turned gray level image file, visible light grayscale image MONO image file and band
The active infrared NIR image file of floodlight light filling, which is input in 2D recognition of face neural network, to be trained, and obtains to identify people
The training data of face head portrait enables 2D recognition of face neural network to identify the face head portrait in image.
Wherein, the training data of acquisition can be the Facial Features such as eyes size, the size of mouth, face contour, acquisition
Facial Features are more, and the result of identification is more accurate.It may include several for each face head portrait or even tens Facial Features
Collection.
It, will while input great amount of images file is trained 2D recognition of face neural network in registration process
Each RGB image file, RGB with face turns gray level image file, visible light grayscale image MONO image file and with floodlight
The log-on data of the active infrared NIR image file of light filling is input in 2D recognition of face neural network, then knows 2D face
The registration information of the image file of the Facial Features collection and input of the face head portrait identified in other neural network is matched one by one, structure
At registered face list.By determining the relationship of the log-on data of face head portrait and the corresponding image file of face head portrait, realize
2D recognition of face neural network can obtain identical registration number according to the different images file of identical face head portrait
According to.
It include that A is used in the registered face list of 2D recognition of face neural network in one embodiment in identification process
The Facial Features collection B and image B of family face head portrait are named as the registration information of " party A-subscriber ", and Facial Features collection B and registration information
" party A-subscriber " is pairing relationship.If identified in system input picture C to 2D recognition of face neural network, when 2D face is known
Other neural network recognization goes out in image C include face head portrait after, obtain Facial Features collection D in image C, and by Facial Features collection
D is matched into registered face list.If appearance is special in Facial Features collection B and Facial Features collection D in registered face list
When sign is completely or the overwhelming majority is identical, it is registered users that 2D recognition of face neural network recognization, which goes out image C, and obtains " A use
The registration information at family "；If prompted when the Facial Features collection not matched with Facial Features collection D in registered face list
Unidentified target out provides corresponding status information.
Wherein, the mode that 2D recognition of face neural network obtains registration information can be taken pictures the head of generation using remote terminal
As the head portrait file or existing file progress remote login on file, resident identification card or other certificates.
Step S405 constructs 3D rendering according to two infrared images, and it is special that 3D rendering and the first image are then carried out face
Sign compares.
Step S407 exports the information of target person when 3D rendering is identical as the face characteristic of the first image.
Specifically, in two infrared images for obtaining infrared camera 11 and the bat of infrared camera 12, pass through neural network
Or use conventional methods and be converted into disparity map, and thus export 3D depth map, Disparity Analysis then is carried out to the 3D rendering,
It identifies face stereoscopic features, 3D depth map is subjected to key feature with the first image and is compared.If 3D depth map and the first figure
When the key feature of picture compares not identical, then prompts unidentified target out or provide corresponding status information；If 3D depth map
When comparing identical with the key feature of the first image, system identification success is indicated, then export the face head portrait letter in step S405
Breath.
It should be noted that the key feature ratio of the infrared image obtained in 3D depth map and neural network recognization unit 21
To in the process, threshold value can be set.When the key feature quantity being compared reaches threshold value, then it represents that compare successfully；Work as progress
When the key feature quantity of comparison is not up to threshold value, then it represents that compare failure.
In one embodiment, the stereo profile of 3D rendering is identified by neural network or image algorithm, and 3D is schemed
The size adjusting of picture is the size of the first image, then confirm stereo profile and the first image of 3D rendering profile whether phase
Together.
Wherein, during identifying face stereoscopic features, to two infrared images progress with floodlight light filling and parallax
Analysis, forms the disparity map of two images, then can form the stereo profile of face by disparity map, pass through the inspection of key position
Survey the profile that confirmation stereo profile is face.
Fig. 5 is a kind of process signal of method of the recognition of face based on binocular active infrared provided by the embodiments of the present application
Figure.As shown in figure 5, the process that this method executes is as follows:
Step S501 takes pictures to target face, to obtain two infrared images.
Any one infrared image in step S502A, obtaining step S501 in two infrared images.
Step S503A will carry out face knowledge in the infrared image input 2D recognition of face neural network in step S502A
Not, judge whether there is face head portrait in infrared image；When there is no face head portrait in infrared image, then output system it is unidentified go out mesh
Mark；There is facial image in hot outer head portrait, executes step S504A.
Step S504A, in identification infrared image after target face head portrait, from the acquisition of 2D recognition of face neural network and mesh
Mark the information of people.
Then two infrared images are converted into 3D rendering by two infrared images in step S502B, obtaining step S501.
Step S503B, after confirming the infrared image for having face head portrait from acquisition in step S503A, by the head portrait and 3D
Image carries out face characteristic comparison；When comparison result is not identical, then output system it is unidentified go out target；When comparison result is identical
When, execute step S505.
Step S505, it is defeated when infrared image is identical with the face characteristic comparison result of 3D rendering in verification step S503A
The information of target person in step S504A out.
The application constitutes binocular active infrared camera module by two infrared cameras in binocular infrared eye
Recognition of face is carried out, first selects an image to know by neural network recognization unit and carries out recognition of face, determine target face
It whether is registered；Then two images are converted by 3D rendering by binocular parallax recognition unit, by 3D rendering and pass through net
The image of network neural unit carries out key feature comparison, if the key feature compared is identical, system identification success, and Ji Nengman
The low reject rate of foot and misclassification rate, and can solve lighting issues and In vivo detection problem.In addition, use general infrared camera with
And general floodlight light filling, cost and technique and existing structure light, TOF, band characteristic point light filling (being also structure light light filling)
Binocular camera compared to advantageously.
In the description of this specification, particular features, structures, materials, or characteristics can be real in any one or more
It applies and is combined in a suitable manner in example or example.
Finally, it is stated that: above embodiments are only to illustrate the technical solution of the application, and limit it；Although reference
The application is described in detail in previous embodiment, those skilled in the art should understand that: it still can be right
Technical solution documented by foregoing embodiments is modified or equivalent replacement of some of the technical features；And this
A little modifications or substitutions, the spirit and scope of each embodiment technical solution of the application that it does not separate the essence of the corresponding technical solution.
Claims (9)
1. a kind of system of the recognition of face based on binocular active infrared characterized by comprising
Binocular infrared eye, for taking pictures to target face, to obtain two infrared images；
Processing unit, the processing unit include neural network recognization unit, binocular parallax recognition unit and transmission unit,
The neural network recognization unit for obtaining the first image, and inputs in 2D recognition of face neural network and carries out face
Identification, in identifying the first image in target face head portrait and the 2D recognition of face neural network in registration database
When the recognition of face parameter of any face head portrait is coincide, the information of the corresponding target person of the target face head portrait is obtained；It is described
First image is an image in two infrared images；
The binocular parallax recognition unit, for constructing 3D rendering according to two infrared images, then by the 3D rendering
Face characteristic is carried out with the first image to compare；The face characteristic is for determining that the target is living body；
The transmission unit, for exporting the target when the 3D rendering is identical as the face characteristic of the first image
The information of people.
2. system according to claim 1, which is characterized in that the binocular infrared eye further includes that at least one is general
Light light compensating lamp,
The floodlight light compensating lamp, for carrying out infrared floodlight light filling to the target.
3. system according to claim 2, which is characterized in that the compensation wavelength that the floodlight light compensating lamp provides is 850nm
Or 940nm.
4. system according to claim 1, which is characterized in that the neural network recognization unit is specifically used for,
The first image is input in the 2D recognition of face neural network, judges whether to identify face head portrait；When described
2D recognition of face neural network recognization goes out in the first image after face head portrait, obtains face head portrait in the first image,
And judge the recognition of face parameter of face head portrait in the first image whether in the registered face list；When described first
The recognition of face parameter of face head portrait obtains face head portrait in the first image in the registered face list in image
The information of recognition of face parameter corresponding target person in registered face list.
5. system according to claim 1, which is characterized in that the binocular parallax recognition unit is specifically used for,
Go out the stereo profile of the 3D rendering by neural network recognization, and is described first by the size adjusting of the 3D rendering
Whether the profile of the size of image, the stereo profile and the first image that then confirm the 3D rendering is identical.
6. a kind of method of the recognition of face based on binocular active infrared characterized by comprising
It takes pictures to target face, to obtain two infrared images；
The first image is obtained, and inputs in 2D recognition of face neural network and carries out recognition of face, is identifying the first image
The recognition of face of any face head portrait is joined in registration database in middle target face head portrait and the 2D recognition of face neural network
When number coincide, the information of the corresponding target person of the target face head portrait is obtained；The first image is two infrared figures
An image as in；
3D rendering is constructed according to two infrared images, the 3D rendering and the first image are then subjected to face characteristic
It compares；The face characteristic is for determining that the target is living body；
When the 3D rendering is identical as the face characteristic of the first image, the information of the target person is exported.
7. two infrared to obtain according to the method described in claim 6, it is characterized in that, take pictures described to target
Before image, comprising:
Infrared floodlight light filling is carried out to the target.
8. according to the method described in claim 6, it is characterized in that, the first image of the acquisition, and inputting 2D recognition of face mind
Through carrying out recognition of face in network, target face head portrait and the 2D recognition of face nerve net in identifying the first image
When the recognition of face parameter of any face head portrait is coincide in registration database in network, the corresponding mesh of the target face head portrait is obtained
The information for marking people, specifically includes:
The first image is input in the 2D recognition of face neural network, judges whether to identify face head portrait；
After the 2D recognition of face neural network recognization goes out face head portrait in the first image, obtain in the first image
Face head portrait, and judge the recognition of face parameter of face head portrait in the first image whether in the registered face list；
When the recognition of face parameter of face head portrait in the first image is in the registered face list, first figure is obtained
The information of recognition of face parameter corresponding target person in registered face list of face head portrait as in.
9. according to the method described in claim 6, it is characterized in that, it is described according to two infrared images construct 3D rendering,
Then the 3D rendering is carried out face characteristic with the first image to compare, is specifically included:
Go out the stereo profile of the 3D rendering by neural network recognization, and is described first by the size adjusting of the 3D rendering
Whether the profile of the size of image, the stereo profile and the first image that then confirm the 3D rendering is identical.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910351639.7A CN110070062A (en) | 2019-04-28 | 2019-04-28 | A kind of system and method for the recognition of face based on binocular active infrared |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910351639.7A CN110070062A (en) | 2019-04-28 | 2019-04-28 | A kind of system and method for the recognition of face based on binocular active infrared |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110070062A true CN110070062A (en) | 2019-07-30 |
Family
ID=67369353
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910351639.7A Pending CN110070062A (en) | 2019-04-28 | 2019-04-28 | A kind of system and method for the recognition of face based on binocular active infrared |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110070062A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991266A (en) * | 2019-11-13 | 2020-04-10 | 北京智芯原动科技有限公司 | Binocular face living body detection method and device |
| CN112528949A (en) * | 2020-12-24 | 2021-03-19 | 杭州慧芯达科技有限公司 | Binocular face recognition method and system based on multiband light |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108108704A (en) * | 2017-12-28 | 2018-06-01 | 努比亚技术有限公司 | Face identification method and mobile terminal |
| CN108491805A (en) * | 2018-03-27 | 2018-09-04 | 百度在线网络技术（北京）有限公司 | Identity identifying method and device |
| CN105023005B (en) * | 2015-08-05 | 2018-12-07 | 王丽婷 | Face identification device and its recognition methods |
| CN109086718A (en) * | 2018-08-02 | 2018-12-25 | 深圳市华付信息技术有限公司 | Biopsy method, device, computer equipment and storage medium |
| CN109145750A (en) * | 2018-07-23 | 2019-01-04 | 华迅金安（北京）科技有限公司 | A kind of driver identity rapid authentication method and system |
-
2019
- 2019-04-28 CN CN201910351639.7A patent/CN110070062A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105023005B (en) * | 2015-08-05 | 2018-12-07 | 王丽婷 | Face identification device and its recognition methods |
| CN108108704A (en) * | 2017-12-28 | 2018-06-01 | 努比亚技术有限公司 | Face identification method and mobile terminal |
| CN108491805A (en) * | 2018-03-27 | 2018-09-04 | 百度在线网络技术（北京）有限公司 | Identity identifying method and device |
| CN109145750A (en) * | 2018-07-23 | 2019-01-04 | 华迅金安（北京）科技有限公司 | A kind of driver identity rapid authentication method and system |
| CN109086718A (en) * | 2018-08-02 | 2018-12-25 | 深圳市华付信息技术有限公司 | Biopsy method, device, computer equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 邱晨鹏: "基于双目摄像机的人脸活体检测的研究", 《现代计算机》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991266A (en) * | 2019-11-13 | 2020-04-10 | 北京智芯原动科技有限公司 | Binocular face living body detection method and device |
| CN110991266B (en) * | 2019-11-13 | 2024-02-20 | 北京智芯原动科技有限公司 | Binocular face living body detection method and device |
| CN112528949A (en) * | 2020-12-24 | 2021-03-19 | 杭州慧芯达科技有限公司 | Binocular face recognition method and system based on multiband light |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109558764B (en) | Face recognition method and device and computer equipment | |
| CN103106401B (en) | Mobile terminal iris recognition device with human-computer interaction mechanism | |
| CN103383723B (en) | Method and system for spoof detection for biometric authentication | |
| CN109543640B (en) | Living body detection method based on image conversion | |
| US10521650B2 (en) | 3D imaging recognition by stereo matching of RGB and infrared images | |
| Gupta et al. | Texas 3D face recognition database | |
| JP2020523665A (en) | Biological detection method and device, electronic device, and storage medium | |
| CN107122744A (en) | A kind of In vivo detection system and method based on recognition of face | |
| CN107800965B (en) | Image processing method, device, computer readable storage medium and computer equipment | |
| CN103514440A (en) | Facial recognition | |
| KR101510312B1 (en) | 3D face-modeling device, system and method using Multiple cameras | |
| CN110991266A (en) | Binocular face living body detection method and device | |
| CN109977929A (en) | A kind of face identification system and method based on TOF | |
| CN109255282B (en) | Biological identification method, device and system | |
| CN111597938A (en) | Living body detection and model training method and device | |
| CN106446855A (en) | Real name authentication device | |
| CN109059868A (en) | A kind of binocular distance measuring method based on Adaptive matching window | |
| CN112818722A (en) | Modular dynamically configurable living body face recognition system | |
| US20210256244A1 (en) | Method for authentication or identification of an individual | |
| CN112257641A (en) | Face recognition living body detection method | |
| TWI394085B (en) | Method of identifying the dimension of a shot subject | |
| CN110070062A (en) | A kind of system and method for the recognition of face based on binocular active infrared | |
| CN108550184A (en) | A kind of biological characteristic 3D 4 D datas recognition methods and system based on light-field camera | |
| CN108197549A (en) | Face identification method and terminal based on 3D imagings | |
| JPH0981732A (en) | Area extracting device and direction detecting device using same |
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 | ||
| AD01 | Patent right deemed abandoned | ||
| AD01 | Patent right deemed abandoned |
Effective date of abandoning: 20211029 |