CN108681704A - A kind of face identification system based on deep learning - Google Patents
A kind of face identification system based on deep learning Download PDFInfo
- Publication number
- CN108681704A CN108681704A CN201810460351.9A CN201810460351A CN108681704A CN 108681704 A CN108681704 A CN 108681704A CN 201810460351 A CN201810460351 A CN 201810460351A CN 108681704 A CN108681704 A CN 108681704A
- Authority
- CN
- China
- Prior art keywords
- face
- module
- network
- recognition
- deep learning
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A kind of face identification system based on deep learning, belongs to deep learning and technical field of image processing.Including the ends PC network training module, system on chip and periphery peripheral module：The ends PC network training module includes data set processing, CNN training, initialization, gradient passback, backpropagation and propagated forward unit；Network structure is built in the network module of the ends PC first, while acquiring enough face training sets, trains an available deep learning recognition of face network；System on chip includes image capture module, fifo control modules, deep learning face recognition module, personnel information storage module and image display.400FPS can be reached in recognition speed under 50MHz clocks；Accuracy of identification is up to 99.25% than quasi- eye recognition precision；Training set covers various illumination conditions, can complete identification function under most illumination condition, possess good robustness.
Description
Technical field
The present invention relates to a kind of face identification systems based on deep learning, belong to digital image processing techniques field.
Background technology
Face has the characteristics that height is nonrigid, and there is a large amount of details for embodying individual difference.Recognition of face is to pass through
The facial image detected from still image or dynamic video is compared with the facial image in database, finds therewith
The process of matched face is commonly used in the purpose of identification and identification, is the project for belonging to living things feature recognition field.
There are many traditional face identification methods, can be broadly divided into two classes, and one kind is the method based on part, is such as utilized
Local description Gabor, local binary patterns etc. are identified；Another kind of is based on global method, includes the face of classics
The streams such as recognizer, such as eigenface method, Fisher face sub-space learning algorithm and locality preserving projections algorithm
Row learning algorithm.
Principal Component Analysis is mainly still concentrated in the research of field of face identification now both at home and abroad, LBP algorithms, and
On deep learning algorithm, these three methods can be better than other algorithms in the success rate and realization difficulty of recognition of face.
Deep learning possesses very strong robustness and high discrimination in field of face identification, and recognition effect is far super
Other algorithms.But recognition of face is realized in FPGA platform at this stage mainly or uses Principal Component Analysis and LBP algorithms.
Although both algorithms compared with deep learning algorithm, are more easy to realize in FPGA platform, due to algorithm
The limitation (poor robustness, discrimination are low, and pretreatment consumes when increasing) of itself, in the speed and precision of recognition of face, this two
Kind algorithm cannot reach the technology requirement for identifying and adapting in real time frame by frame various illumination conditions.Document is shown, at present in FPGA
The scientific achievement detection speed of realization recognition of face is mostly within 10FPS on platform, and most fast achievement can only also reach at this stage
To 25FPS.And deep learning algorithm is mostly to utilize CPU, GPU, TPU are realized that the research in FPGA platform is also relatively
It is few, it is designed for the even more opposite blank of the deep learning network of recognition of face using FPGA, also there is some scientific research personnel
It makes some attempts in this respect, but never ripe scientific achievement.
The present invention realizes a kind of face identification system of deep learning in FPGA platform, the system 50MHz when
Clock operation at frequencies can reach 400FPS in recognition speed, some remote super existing achievements.Accuracy of identification is than anthropomorphic outlook
Other precision, up to 99.25%.Training set covers various illumination conditions, and identification work(can be completed under most illumination condition
Can, possess good robustness.
Invention content
It is an object of the invention to be directed in FPGA platform face recognition technology have that real-time is poor, accuracy of detection is low and
A kind of defect of poor robustness, it is proposed that face identification system based on deep learning.
It is a kind of based on the face identification system of deep learning by the ends PC network training module, system on chip and periphery peripheral hardware mould
Block forms：
Wherein, the ends PC network training module includes：Data set processing unit, CNN training units, initialization unit, gradient
Back propagation unit, backpropagation unit and propagated forward unit；According to actual needs, net is built in the network module of the ends PC first
Network structure, while enough face training sets are acquired, an available deep learning recognition of face network is trained, wherein；People
Face identifies that the activation primitive that network uses is sigmoid functions；
System on chip includes that image capture module, fifo control modules, deep learning face recognition module, personal information are deposited
Store up module and image display；
Periphery peripheral module includes camera submodule, VGA D/A converter modules and local display；
Wherein, include mainly multi-path camera and camera connecting line in camera submodule；
The connection relation of each module is as follows in a kind of face identification system based on deep learning：
Multichannel data channel is connected to image by camera connecting line and adopted by camera submodule multi-path camera in the block
Collecting module, image capture module is connected to fifo control modules, and fifo control modules are connected to deep learning face recognition module,
It is connected to image display simultaneously, deep learning face module is connected to personnel information storage module, and personal information stores mould
Block is integrated in fifo control modules, is controlled by fifo control modules, and result is output to image display；Image is aobvious
Show that module again by digital data transmission to VGA D/A converter modules, is output to local display after digital-to-analogue conversion；
Modules function is as follows：
Multi-path camera completes the acquisition function of image, and is transferred to image capture module；
The function of image capture module is to receive the input data of video flowing, and be transferred to fifo control modules；
Fifo control modules cache the vision signal transmitted, to ensure the integrality of output image data, and it is right
The picture progress transmitted is down-sampled in real time, and the pixel size for being downsampled to suitable Network Recognition is stored in fifo；
The function of deep learning face recognition module is to realize the acceleration processing of face identification functions, and input is by drop
The facial image of sampling after module completes recognition of face, exports face result；
The information of personnel is deposited in personnel information storage module, it receives the knot from deep learning face recognition module
Fruit, and personnel's specifying information is output to image display, which is integrated in fifo control modules；It is controlled by fifo
Module controls, and is output to image display；
The function of image display be receive by caching video stream data, generate specification video flowing sequential and
With corresponding pixel data, the VGA D/A converter modules being output to outside piece, in addition the module is also with data automatic aligning work(
Can, after local sequential can be overcome to generate problem of misalignment is shown caused by input video flow data lag；
The modules function of periphery peripheral module is as follows：
The function of camera is acquisition outer scene, generates stable digital video frequency flow；
VGA D/A converter module functions are the digital signal and row field sync signal of reception image display outflow, and
Digital signal is subjected to digital-to-analogue conversion, is transferred to local display；The function of local display is acquisition image and identification knot
Fruit real-time display；
A kind of course of work of the face identification system based on deep learning, including build recognition of face network, training of human
Face identification network builds four parts of FPGA platform and network parameter update, is as follows：
Step 1 builds the convolutional neural networks for recognition of face in the ends PC network training module；
Step 2 is trained recognition of face network with the face database being collected in advance, obtains trained network ginseng
Number；
Step 3 builds recognition of face network based on FPGA platform；
In step 4, the recognition of face network for putting up the network parameter steps for importing 3 of step 2 training output.
Advantageous effect
A kind of face identification system based on deep learning has the advantages that compared with prior art：
The design that 1. deep learning is not combined with recognition of face in the prior art and is realized in FPGA platform,
And the present invention is successfully realized in FPGA platform based on deep learning recognition of face network；
2. compared with existing face recognition technology, recognition speed is improved：The place of existing correlation face identification system
Reason speed is generally below 25FPS, the present invention by using the CNN models in depth learning technology field to the image collected into
Row training, substantially increases image processing speed；
3. compared with existing recognition of face, identification accuracy is improved, and is embodied as：Based on FPGA platform reality
The method of existing recognition of face is mostly Principal Component Analysis and LBP algorithms, and the precision of identification can only achieve 80% or so；And this
The deep learning algorithm discrimination that invention uses can reach 99% or more, than the other precision of personification outlook.
Description of the drawings
Fig. 1 is a kind of structure diagram of the face identification system based on deep learning of the present invention；
Fig. 2 is the schematic network structure in a kind of face identification system embodiment based on deep learning of the present invention.
Specific implementation mode
The present invention will be further described with reference to the accompanying drawings and examples and detailed description.
Embodiment 1
The present embodiment describes the structure of a kind of face identification system based on deep learning of the present invention and specific
Implement.
As shown in Figure 1, being a kind of structure diagram of the face identification system based on deep learning.According to the structure diagram,
The present embodiment is made.
The present embodiment is realized that whole system processing image size is using company of purple light Tontru PGT180H devices
540X480 pixels.
First, one 6 layers of CNN convolutional neural networks of matlab project trainings are used in the network training module of the ends PC,
Here we use the CMU_PIE face databases of Carnegie Mellon University and the face database of part oneself acquisition.By test
The test of collection, 6 layer network can reach 99.25% to 20 people's discriminations in face database.
6 layers of convolutional neural networks are by one layer of input layer, two layers of convolutional layer, two layers of pond layer and one layer of full articulamentum group
At.Input layer is that the image of 32x32 pixel sizes is connected to the full articulamentum of last layer, and export knot after 4 hidden layers
Fruit.Network uses activation primitive for sigmoid functions；
Wherein, 4 hidden layers include first layer convolutional layer, first layer pond layer, second layer convolutional layer and second layer pond layer；
6 layer network structures are put up in FPGA platform, and network architecture parameters are trained as shown in Fig. 2, importing.
For embodiment system camera using MT9V034 gray level image sensors, this is a kind of global exposure CMOS biographies
Sensor, when both full-pixel (752Hx480V) is exported, speed is 60FPS, and SoC systems on chip output it in initialization and are configured to
Continuous data (including field/row synchronize) then device automatic collection and is input to the Image Acquisition mould of system by 540x480 sizes
Block；
The example has built one 6 layers of convolutional neural networks in FPGA platform.The network structure and the training of the ends PC
Network structure is identical, by one layer of input layer, two layers of convolutional layer, two layers of pond layer and one layer of full articulamentum composition, wherein
Sigmoid activation primitives are realized using 6 sections of piecewise fittings.Under the clock frequency of 50MHz, the system is complete in 2.4ms
At the identification of face, the number of identification is 20 people.After completing algorithm process, the result of identification is transferred to fifo control modules,
The output of personal information is carried out by fifo control module controllers information storage modules；
The real-time cache image that camera acquires and personal information are transferred to image display by fifo control modules,
Image display generates corresponding row field sync signal, and peripheral hardware VGA D/A converter modules are given in output.VGA D/A converter modules
Conversion of the digital signal to analog signal is completed, and connects VGA display screens, completes real-time acquisition display and the recognition result of image
Display.
The operating process of the present invention：System electrification loads bit files, completes the configuration of system.Identification target enters image
Pickup area, target identification result carry out real-time display in local display.
System described in this example builds the recognition of face net of 20 people in the case where Image Acquisition speed is sufficiently fast
Network can complete recognition of face using the clock frequency of 50MHz as system clock in 2.4ms, recognition speed can reach
400FPS realizes the real-time face identification of 60FPS since the frame number of MT9V034 gray level image sensors limits in this example,
Non-delay, compared to the processing speed of existing design (being less than 25FPS), real-time is improved.And it is high on recognition accuracy
Up to 99.25%, eye recognition accuracy rate of comparing, remote super some current designs based on principal component analysis and LBP algorithms.Not
This example is tested under same light environment, recognition of face can be accurately completed, to the strong robustness of illumination.
The above is presently preferred embodiments of the present invention, and the present invention should not be limited to disclosed in the embodiment and attached drawing
Content.It is every not depart from the lower equivalent or modification completed of spirit disclosed in this invention, both fall within the scope of protection of the invention.
Claims (1)
1. a kind of face identification system based on deep learning, it is characterised in that：Including the ends PC network training module, system on chip
With periphery peripheral module；
Wherein, the ends PC network training module includes：Data set processing unit, CNN training units, initialization unit, gradient passback
Unit, backpropagation unit and propagated forward unit；According to actual needs, network knot is built first in the network module of the ends PC
Structure, while enough face training sets are acquired, an available deep learning recognition of face network is trained, wherein；Face is known
The activation primitive that other network uses is sigmoid functions；
System on chip includes image capture module, fifo control modules, deep learning face recognition module, personal information storage mould
Block and image display；
Periphery peripheral module includes camera submodule, VGA D/A converter modules and local display；
Wherein, include mainly multi-path camera and camera connecting line in camera submodule；
The connection relation of each module is as follows in a kind of face identification system based on deep learning：
Multichannel data channel is connected to Image Acquisition mould by camera submodule multi-path camera in the block by camera connecting line
Block, image capture module are connected to fifo control modules, and fifo control modules are connected to deep learning face recognition module, simultaneously
It is connected to image display, deep learning face module is connected to personnel information storage module, personnel information storage module collection
It in fifo control modules, is controlled by fifo control modules, result is output to image display；Image shows mould
Block is output to local display again by digital data transmission to VGA D/A converter modules after digital-to-analogue conversion；
Modules function is as follows：
Multi-path camera completes the acquisition function of image, and is transferred to image capture module；
The function of image capture module is to receive the input data of video flowing, and be transferred to fifo control modules；
Fifo control modules cache the vision signal transmitted, to ensure the integrality of output image data, and to transmitting
Picture carry out in real time it is down-sampled, the pixel size for being downsampled to suitable Network Recognition is stored in fifo；
The function of deep learning face recognition module is to realize the acceleration processing of face identification functions, and input is by down-sampled
Facial image, module complete recognition of face after, export face result；
The information of personnel is deposited in personnel information storage module, it receive from deep learning face recognition module as a result, simultaneously
Personnel's specifying information is output to image display, which is integrated in fifo control modules；By fifo control modules
Control, is output to image display；
The function of image display is the video stream data received by caching, generates the video flowing sequential of specification and matches phase
The pixel data answered, the VGA D/A converter modules being output to outside piece, in addition the module also have the function of data automatic aligning, can
To overcome the caused display problem of misalignment of input video flow data lag after local sequential generation；
The modules function of periphery peripheral module is as follows：
The function of camera is acquisition outer scene, generates stable digital video frequency flow；
VGA D/A converter module functions are to receive the digital signal and row field sync signal of image display outflow, and will count
Word signal carries out digital-to-analogue conversion, is transferred to local display；The function of local display is that acquisition image and recognition result are real
When show；
A kind of course of work of the face identification system based on deep learning, including build recognition of face network, training face knowledge
Other network builds four parts of FPGA platform and network parameter update, is as follows：
Step 1 builds the convolutional neural networks for recognition of face in the ends PC network training module；
Step 2 is trained recognition of face network with the face database being collected in advance, obtains trained network parameter；
Step 3 builds recognition of face network based on FPGA platform；
In step 4, the recognition of face network for putting up the network parameter steps for importing 3 of step 2 training output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810460351.9A CN108681704A (en) | 2018-05-15 | 2018-05-15 | A kind of face identification system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810460351.9A CN108681704A (en) | 2018-05-15 | 2018-05-15 | A kind of face identification system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108681704A true CN108681704A (en) | 2018-10-19 |
Family
ID=63806132
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810460351.9A Pending CN108681704A (en) | 2018-05-15 | 2018-05-15 | A kind of face identification system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108681704A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109274941A (en) * | 2018-10-23 | 2019-01-25 | 合肥博焱智能科技有限公司 | Multichannel video decoding Face datection recognition methods based on FPGA |
| CN109446996A (en) * | 2018-10-31 | 2019-03-08 | 北京智慧眼科技股份有限公司 | Facial recognition data processing unit and processing method based on FPGA |
| CN109815929A (en) * | 2019-01-31 | 2019-05-28 | 青岛科技大学 | Face identification method based on convolutional neural networks |
| CN111079717A (en) * | 2020-01-09 | 2020-04-28 | 西安理工大学 | Face recognition method based on reinforcement learning |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093175A (en) * | 2011-10-27 | 2013-05-08 | 上海德致伦电子科技有限公司 | Face identification system on chip based on field programmable gate array (FPGA) |
| CN105589938A (en) * | 2015-12-13 | 2016-05-18 | 公安部第三研究所 | Image retrieval system and retrieval method based on FPGA |
| CN105956660A (en) * | 2016-05-16 | 2016-09-21 | 浪潮集团有限公司 | Neural network chip realization method used for real-time image identification |
| CN106845380A (en) * | 2017-01-13 | 2017-06-13 | 江苏信息职业技术学院 | Face identification system based on mechanical arm |
| CN107423680A (en) * | 2017-06-06 | 2017-12-01 | 广西师范大学 | Face identification device based on FPGA |
-
2018
- 2018-05-15 CN CN201810460351.9A patent/CN108681704A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093175A (en) * | 2011-10-27 | 2013-05-08 | 上海德致伦电子科技有限公司 | Face identification system on chip based on field programmable gate array (FPGA) |
| CN105589938A (en) * | 2015-12-13 | 2016-05-18 | 公安部第三研究所 | Image retrieval system and retrieval method based on FPGA |
| CN105956660A (en) * | 2016-05-16 | 2016-09-21 | 浪潮集团有限公司 | Neural network chip realization method used for real-time image identification |
| CN106845380A (en) * | 2017-01-13 | 2017-06-13 | 江苏信息职业技术学院 | Face identification system based on mechanical arm |
| CN107423680A (en) * | 2017-06-06 | 2017-12-01 | 广西师范大学 | Face identification device based on FPGA |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109274941A (en) * | 2018-10-23 | 2019-01-25 | 合肥博焱智能科技有限公司 | Multichannel video decoding Face datection recognition methods based on FPGA |
| CN109274941B (en) * | 2018-10-23 | 2020-12-18 | 合肥博焱智能科技有限公司 | Multi-channel video decoding face detection and identification method based on FPGA |
| CN109446996A (en) * | 2018-10-31 | 2019-03-08 | 北京智慧眼科技股份有限公司 | Facial recognition data processing unit and processing method based on FPGA |
| CN109446996B (en) * | 2018-10-31 | 2021-01-22 | 智慧眼科技股份有限公司 | Face recognition data processing device and method based on FPGA |
| CN109815929A (en) * | 2019-01-31 | 2019-05-28 | 青岛科技大学 | Face identification method based on convolutional neural networks |
| CN111079717A (en) * | 2020-01-09 | 2020-04-28 | 西安理工大学 | Face recognition method based on reinforcement learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108681704A (en) | A kind of face identification system based on deep learning | |
| He | Research of a sign language translation system based on deep learning | |
| Deep et al. | Hybrid model featuring CNN and LSTM architecture for human activity recognition on smartphone sensor data | |
| CN107480178B (en) | Pedestrian re-identification method based on cross-modal comparison of image and video | |
| CN107092894A (en) | A kind of motor behavior recognition methods based on LSTM models | |
| CN109886225A (en) | A kind of image gesture motion on-line checking and recognition methods based on deep learning | |
| US11386598B2 (en) | Method and system for sensor data recognition using data enrichment for the learning process | |
| US11526704B2 (en) | Method and system of neural network object recognition for image processing | |
| CN110458046B (en) | Human motion trajectory analysis method based on joint point extraction | |
| CN107609525A (en) | Remote Sensing Target detection method based on Pruning strategy structure convolutional neural networks | |
| CN104851099A (en) | Method for image fusion based on representation learning | |
| CN112001347B (en) | Action recognition method based on human skeleton morphology and detection target | |
| CN113158727A (en) | Bimodal fusion emotion recognition method based on video and voice information | |
| CN109409219A (en) | Indoor occupant locating and tracking algorithm based on depth convolutional network | |
| WO2022156654A1 (en) | Text data processing method and apparatus | |
| CN108171734A (en) | A kind of method and device of ORB feature extracting and matchings | |
| CN111680660B (en) | Human behavior detection method based on multi-source heterogeneous data stream | |
| CN108921037A (en) | A kind of Emotion identification method based on BN-inception binary-flow network | |
| Chalasani et al. | Egocentric gesture recognition for head-mounted ar devices | |
| Makarov et al. | Russian sign language dactyl recognition | |
| Kasukurthi et al. | American sign language alphabet recognition using deep learning | |
| CN109583334A (en) | A kind of action identification method and its system based on space time correlation neural network | |
| CN116895098A (en) | Video human body action recognition system and method based on deep learning and privacy protection | |
| CN105608662A (en) | FPGA-based dynamic target identification system and identification method thereof | |
| Özbay et al. | 3D Human Activity Classification with 3D Zernike Moment Based Convolutional, LSTM-Deep Neural Networks. |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181019 |
|
| WD01 | Invention patent application deemed withdrawn after publication |