CN104573679B - Face identification system based on deep learning under monitoring scene - Google Patents
Face identification system based on deep learning under monitoring scene Download PDFInfo
- Publication number
- CN104573679B CN104573679B CN201510063023.1A CN201510063023A CN104573679B CN 104573679 B CN104573679 B CN 104573679B CN 201510063023 A CN201510063023 A CN 201510063023A CN 104573679 B CN104573679 B CN 104573679B
- Authority
- CN
- China
- Prior art keywords
- face
- module
- layer
- unit
- information
- 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
Abstract
Face identification system based on deep learning under a kind of monitoring scene.The face identification system based on deep learning includes video acquisition unit, Face datection unit, display unit, storage unit under monitoring scene.The image collected information is transferred to Face datection unit and is further processed by video acquisition unit, and simultaneous transmission is stored to storage unit;Face datection unit passes through deep learning, detect the face information in video image, and extract face characteristic information, the face characteristic information extracted carries out characteristic matching with the face in human face data library module, obtain matched face information, it is transferred to display unit to show, simultaneous transmission is stored to storage unit.The device can quickly recognize face and accurately determine corresponding identity information, can realize the early warning of dangerous person, protect the security of the lives and property of people, have important role to the assistance maintenance of China's public security.
Description
Technical field
The invention belongs to field of intelligent video surveillance, are related to the technologies such as pattern-recognition, artificial intelligence, are related specifically to supervise
It controls under scene in terms of Intelligent human-face identification.
Background technology
Face recognition technology is one of biological identification technology of characteristic information progress identification according to face.Face
Identification technology becomes a kind of reliably identification foundation, and with untouchable, concurrency, non-imposed, intuitive etc.
Advantage.Recognition of face at present is mainly used in the fields such as attendance, gate inhibition's authentication, also endless in the application of field of intelligent monitoring
Kind, there are no more complete recognition of face monitoring devices to apply in monitoring scene.
The present invention is directed to apply in field of intelligent monitoring, by monitoring, ratio is identified in the face in target and scene
To matching, the people information searched for.Quickly confirm target identities at a distance, realize the function of intelligent early-warning.
It is gone forward side by side pedestrian by being acquired video information in real time in important protection place such as school, station, supermarket, square, residential quarter etc.
Face identifies, so as to the information of quick lock in discrepancy personnel, determines whether a bad actor, so as to fulfill early warning, protects people's
Personal safety as well as the property safety.With the raising of people's security precautions, it is badly in need of face identification system and is applied to monitoring field, also,
This has important role to strengthening China's security administration level.
Invention content
To solve the above problems, the purpose of the present invention is to provide the recognition of face systems based on deep learning under monitoring scene
System.
This system is able to detect that face and carries out quick analysis matching, identify identity letter early under monitoring scene
Breath, determines whether dangerous person and early early warning.This system have it is friendly, easily acquisition property, that non-imposed, detection is reliable etc. is excellent
Point has higher intelligent level.
The technical solution adopted by the present invention is:The face identification system based on deep learning is adopted including video under monitoring scene
Collect unit, Face datection unit, display unit, storage unit.
Wherein, the video acquisition unit includes embedded high definition camera and video encoder.
Wherein, the Face datection unit includes human face data library module, deep learning module, human face discriminating module, spy
Levy extraction module, face matching module and communication module a.
The human face data library module is to store face to be trained.
The deep learning module establishes process including each layer of neural network.The neural network established, includes input
Layer, output layer, five layers of hidden layer have connection between only adjacent node layer, mutually connectionless between same layer and cross-layer node.
Training process includes preceding to training and backward training.Forward direction training process is unsupervised learning from bottom to top, i.e., is opened from bottom
Begin, past top layer in layer is trained, and in training process, after training study obtains the (n-1)th layer parameter, n-1 layers of output is made
For the input of n-th layer, n-th layer is trained, thus respectively obtains the parameter of each layer;Backward training process is learned for top-down supervision
It practises, i.e. the top-down transmission of training error is finely adjusted parameter.
The human face discriminating module refers to the study by each layer neural network of deep learning, will go out in video image
Existing face information detected.
The characteristic extracting module is according to the face sample learning rule in human face data library module, extracts face and sentences
The characteristic information of face extracted in other module.
The face matching module is the face to be detected and stored face extracted in face discrimination module
Matched process is compared in face in database module.By comparing the people to be detected extracted in human face discriminating module
The characteristic information of face and the face in human face data library module finds out the corresponding identity information of matched face.
The communication module a is responsible for the matched face information that face matching module extracts being transferred to display list
Member is carried out data transmission by pci bus.
Wherein, the display unit includes communication module b, decoder module and display module.
The communication module b is responsible for receiving the matching characteristic information of face of face detection unit transmission and video acquisition list
Member passes through the video information of compressed encoding, is carried out data transmission by pci bus.
The decoder module is responsible for the video information that the video acquisition unit received is transmitted being decoded.
The display module refers to LED display, the identity information for the face that display face matching module obtains and
Real time monitoring video recording.
Wherein, the storage unit is stored for cloud disk, and the collected scene information of video acquisition unit is passed through optical fiber electricity
Cable is transferred to cloud disk and stores, meanwhile, the identity information of the face that face matching module obtains in Face datection unit is carried out
Storage.
Collected video information is passed through network cable transmission to face by the embedded high definition camera of the video acquisition unit
The human face discriminating module of detection unit, at the same through fiber optic cables be transferred to storage unit storage video module carry out data it is standby
Part;Human face data in human face data library module is passed to deep learning module by Face datection unit, carries out neural network oneself
Learning model building, established neural network model parameter are transferred to human face discriminating module;Human face discriminating module utilizes deep learning
The neural network model that module is established carries out human face discriminating to the video information that video acquisition unit transmission comes, and obtains people in video
Face image information;The face information extracted in human face discriminating module is further carried out feature extraction, people by characteristic extracting module
Matching is compared with the face in human face data library module in the face characteristic extracted by face matching module, finds matching degree most
High face information, meanwhile, and the storage matching result module that face matching result is transferred to storage unit stores;Into
And the face characteristic information matched is transferred to display unit by communication module a in a manner of pci bus;Display unit leads to
It crosses communication module b and receives regarding after the matched face information and video acquisition unit compressed encoding that the transmission of face detection unit comes
Frequency information, and using decoder module decode video;The decoded video of decoder module passes to display module;Display module
Show matched face information, while real-time display video information.
The beneficial effects of the invention are as follows:Device the image collected information can quickly recognize face and accurately judge
Go out corresponding identity information, can realize dangerous person's by non-contact, completion information identity inspection confirmation with open arms
Corresponding measure is taken in early warning ahead of time, protects the security of the lives and property of people, maintains have important work to the assistance of China's public security
With.
Description of the drawings
Attached drawing is used to provide further understanding of the present invention, and a part for constitution instruction, the reality with the present invention
Example for explaining the present invention, is not construed as limiting the invention together.In the accompanying drawings:
Fig. 1 is present system figure.
Fig. 2 is process chart of the present invention.
Specific embodiment
Illustrate each detailed problem involved by technical solution of the present invention with reference to specific example.It should be noted that
It is that described example is intended merely to facilitate the understanding of the present invention, is not intended to limit the scope of the present invention.
As shown in Figure 1, the technical solution adopted by the present invention is:Face identification system based on deep learning under monitoring scene
Including video acquisition unit 1, Face datection unit 2, display unit 3, storage unit 4.
Wherein, the video acquisition unit 1 includes embedded high definition camera 11 and video encoder 12.
Wherein, the Face datection unit 2 includes human face data library module 21, deep learning module 22, human face discriminating mould
Block 23, characteristic extracting module 24, face matching module 25 and communication module a 26.
The human face data library module 21 is to store face to be trained.
The deep learning module 22 establishes process including each layer of neural network.The neural network established, comprising defeated
Enter layer, output layer, five layers of hidden layer, have connection between only adjacent node layer, mutually without even between same layer and cross-layer node
It connects.Training process includes preceding to training and backward training.Forward direction training process is unsupervised learning from bottom to top, i.e., from bottom
Start, past top layer training in layer, in training process, after training study obtains the (n-1)th layer parameter, by n-1 layers of output
As the input of n-th layer, n-th layer is trained, thus respectively obtains the parameter of each layer;Backward training process is top-down supervision
Study, the i.e. top-down transmission of training error are finely adjusted parameter.
The human face discriminating module 23 refers to the study by each layer neural network of deep learning, will be in video image
The face information of appearance detected.
The characteristic extracting module 24 is according to the face sample learning rule in human face data library module 21, extracts people
The characteristic information of the face extracted in face discrimination module 23.
The face matching module 25 be in face discrimination module 23 face to be detected that extracts with it is stored
Matched process is compared in face in human face data library module 21.It is obtained by comparing extraction in human face discriminating module 23
The characteristic information of face to be detected and the face in human face data library module 21 finds out the corresponding identity information of matched face.
The communication module a 26 is responsible for the matched face information that the extraction of face matching module 25 obtains being transferred to aobvious
Show unit 3, carried out data transmission by pci bus.
Wherein, the display unit 3 includes communication module b 31, decoder module 32 and display module 33.
The communication module b 31 is responsible for receiving the characteristic information of the matching face of 2 transmission of face detection unit and video is adopted
Collect video information of the unit 1 by compressed encoding, carried out data transmission by pci bus.
The decoder module 32 is responsible for the video information that the video acquisition unit 1 received is transmitted being decoded.
The display module 33 refers to LED display, the identity information of face that display face matching module 25 obtains
And real time monitoring video recording.
Wherein, the storage unit 4 is stored for cloud disk, and 1 collected scene information of video acquisition unit is passed through optical fiber
Cable transmission is to cloud disk and stores, meanwhile, by the identity information of the face that face matching module 25 obtains in Face datection unit 2
It is stored.
Collected video information is passed through network cable transmission to people by the embedded high definition camera 11 of the video acquisition unit 1
The human face discriminating module 23 of face detection unit 2, while the storage video module 41 for being transferred to through fiber optic cables storage unit 4 carries out
Data backup;Human face data in human face data library module 21 is passed to deep learning module 22 by Face datection unit 2, carries out god
Self study modeling through network, established neural network model parameter are transferred to human face discriminating module 23;Human face discriminating module
23 using deep learning module 22 establish neural network model to video acquisition unit 1 transmission come video information progress face
Differentiate, obtain human face image information in video;The face information that characteristic extracting module 24 will be extracted in human face discriminating module 23
Further carry out feature extraction, face matching module 25 is by the face in the face characteristic extracted and human face data library module 21
Matching is compared, finds the highest face information of matching degree, meanwhile, and face matching result is transferred to storage unit 4
Storage matching result module 41 is stored;And then the face characteristic information matched is total with PCI by communication module a 26
Line mode is transferred to display unit 3;Display unit 3 by communication module b 31 receive face detection unit 2 transmit come matching
1 compressed encoding of face information and video acquisition unit after video information, and using decoder module 32 carry out to video decode;
32 decoded video of decoder module passes to display module 3;Display module 3 shows matched face information, while real-time display
Video information.
As shown in Fig. 2, the deep learning module 22 that the present invention uses for 5 layers of neural net layer, in input layer 221 and exports
Be of five storeys hidden layer between layer 227, respectively wave filter group layer 222, correcting layer 223, local contrast normalization layer 224, average pond
With sub-sampling layer 225, maximum pond and sub-sampling layer 226.
Wherein, wave filter group layer 222 described in deep learning module 22 includes convolution filter C, activation primitive, can train increasing
Beneficial G.Wherein, activation primitive we use non-linear transform function sigmoid.It is y=G* that then this layer, which corresponds to input and output,
sigmoid(C).Wherein, in convolution filter group, the present invention carries out convolutional filtering using 96 groups of kernel functions.
Correcting layer 223 described in deep learning module 22 is that wave filter group layer output result is simply corrected, use
It is the operation that takes absolute value, is the meaningless negative value for avoiding occurring in image procossing.
Local contrast normalization layer 224 described in deep learning module 22 refers to taking mean value and variance to upper strata output result
Normalization, i.e. characteristics of image normalize.
Average pondization and sub-sampling layer 225 described in deep learning module 22 are so that the feature of extraction has miniature deformation
Robustness is averaged using sampling window all values, and obtained value is transferred to next sample level.
Maximum pondization and sub-sampling layer 226 described in deep learning module 22 are to realize the feature extracted to translation
Invariance takes average maximum value using sampling window all values, and obtained value is transferred to next sample level.
Any person skilled in the art is changed or is modified as possibly also with the technology contents of the disclosure above
With the equivalent embodiment of variation.But it is every without departing from technical solution of the present invention content, technical spirit according to the present invention to
Any simple modification, equivalent variations and the remodeling that upper embodiment is made still fall within the protection domain of technical solution of the present invention.
Claims (1)
1. the face identification system based on deep learning under monitoring scene, which is characterized in that examined including video acquisition unit, face
Survey unit, the matching display unit of display real time monitoring video recording, storage unit;Video acquisition unit respectively with Face datection list
Member, storage unit connection;Face datection unit connects respectively with display unit, storage unit;The video acquisition unit includes
Embedded high definition camera and video encoder;The Face datection unit includes human face data library module, deep learning module, people
Face discrimination module, characteristic extracting module, face matching module;The human face data library module is to store face to be trained;Institute
Storage unit is stated as cloud disk;The embedded high definition camera of the video acquisition unit passes collected video information by cable
The defeated human face discriminating module to Face datection unit, at the same by fiber optic cables be transferred to the storage video module of storage unit into
Row data backup;Human face data in human face data library module is passed to deep learning module by Face datection list, carries out nerve net
The self study modeling of network, established neural network model parameter are transferred to human face discriminating module;Human face discriminating module is using deeply
It spends the neural network model that study module is established and human face discriminating is carried out to the video information that video acquisition unit transmission comes, depending on
Human face image information in frequency;And then characteristic extracting module extraction face characteristic information, the face that face matching module will extract
Matching is compared with the face in human face data library module in characteristic information, finds the highest face information of matching degree, by logical
Letter module pci bus mode is transferred to matching display unit, the face identity information shown by matching display unit, and passes
Matching result storage is carried out to storage unit;
The deep learning module is 5 layers of neural net layer, and be of five storeys hidden layer between input layer and output layer, respectively filters
Device group layer, correcting layer, local contrast normalization layer, average pond and sub-sampling layer, maximum pond and sub-sampling layer;
The wave filter group layer includes convolution filter C, activation primitive, can train gain G, and activation primitive uses nonlinear transformation
Function sigmoid, then it is y=G*sigmoid (C) that this layer, which corresponds to input and output, wherein, in convolution filter group, using 96 groups of cores
Function carries out convolutional filtering;
The correcting layer is that wave filter group layer output result is corrected, using the operation that takes absolute value, to avoid image
The meaningless negative value occurred in processing;
The local contrast normalization layer is to take mean value and normalized square mean to upper strata output result, i.e. characteristics of image normalizes;
The average pondization and sub-sampling layer are so that the feature of extraction has robustness to miniature deformation, using sample window
Mouth all values are averaged, and obtained value is transferred to next sample level;
The maximum pondization and sub-sampling layer are invariance of the feature extracted of realization to translation, using sample window
Mouth all values take average maximum value, and obtained value is transferred to next sample level.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510063023.1A CN104573679B (en) | 2015-02-08 | 2015-02-08 | Face identification system based on deep learning under monitoring scene |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510063023.1A CN104573679B (en) | 2015-02-08 | 2015-02-08 | Face identification system based on deep learning under monitoring scene |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104573679A CN104573679A (en) | 2015-04-29 |
CN104573679B true CN104573679B (en) | 2018-06-22 |
Family
ID=53089701
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510063023.1A Active CN104573679B (en) | 2015-02-08 | 2015-02-08 | Face identification system based on deep learning under monitoring scene |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104573679B (en) |
Families Citing this family (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105100735A (en) * | 2015-08-31 | 2015-11-25 | 张慧 | People intelligent monitoring management system and management method |
CN105139503A (en) * | 2015-10-12 | 2015-12-09 | 北京航空航天大学 | Lip moving mouth shape recognition access control system and recognition method |
CN106713255B (en) * | 2015-11-17 | 2020-07-24 | 腾讯科技(深圳)有限公司 | User information management method and system |
US10049307B2 (en) | 2016-04-04 | 2018-08-14 | International Business Machines Corporation | Visual object recognition |
CN106204779B (en) * | 2016-06-30 | 2018-08-31 | 陕西师范大学 | Check class attendance method based on plurality of human faces data collection strategy and deep learning |
CN106407369A (en) * | 2016-09-09 | 2017-02-15 | 华南理工大学 | Photo management method and system based on deep learning face recognition |
CN108206929A (en) * | 2016-12-16 | 2018-06-26 | 北京华泰科捷信息技术股份有限公司 | A kind of contactless personnel information acquisition device and its acquisition method |
CN106682650A (en) * | 2017-01-26 | 2017-05-17 | 北京中科神探科技有限公司 | Mobile terminal face recognition method and system based on technology of embedded deep learning |
CN108230491A (en) * | 2017-07-20 | 2018-06-29 | 深圳市商汤科技有限公司 | Access control method and device, system, electronic equipment, program and medium |
CN107403165B (en) * | 2017-08-01 | 2021-12-24 | 杰创智能科技股份有限公司 | Data management architecture of intelligent face recognition system and use method |
CN107392182B (en) * | 2017-08-17 | 2020-12-04 | 宁波甬慧智能科技有限公司 | Face acquisition and recognition method and device based on deep learning |
CN110119656A (en) * | 2018-02-07 | 2019-08-13 | 中国石油化工股份有限公司 | Intelligent monitor system and the scene monitoring method violating the regulations of operation field personnel violating the regulations |
CN108387581A (en) * | 2018-02-24 | 2018-08-10 | 温州宝德电气有限公司 | A kind of general-purpose machines visual identity detection device based on deep learning |
CN108388877A (en) * | 2018-03-14 | 2018-08-10 | 广州影子控股股份有限公司 | The recognition methods of one boar face |
CN108597566A (en) * | 2018-04-17 | 2018-09-28 | 广东南海鹰视通达科技有限公司 | Mobile electron medical records system based on recognition of face and implementation method |
CN108830980A (en) * | 2018-05-22 | 2018-11-16 | 重庆大学 | Security protection integral intelligent robot is received in Study of Intelligent Robot Control method, apparatus and attendance |
CN108960205A (en) * | 2018-08-06 | 2018-12-07 | 广州开瑞信息科技有限公司 | A kind of Intelligent human-face recognition methods and system |
CN109657617A (en) * | 2018-12-20 | 2019-04-19 | 江苏图越电子科技有限公司 | A kind of face recognition device and its identifying system |
CN112539772B (en) * | 2020-11-02 | 2023-04-07 | 上海大学 | Positioning method of Sagnac distributed optical fiber sensing system based on convolutional neural network integrated learning |
CN113192240A (en) * | 2021-03-12 | 2021-07-30 | 广州朗国电子科技有限公司 | Deep learning-based identification module identification method, equipment and medium |
CN114038077A (en) * | 2021-10-29 | 2022-02-11 | 金陵科技学院 | Temperature measurement system is signed in with special use of epidemic situation prevention and control in classroom |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103824054A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascaded depth neural network-based face attribute recognition method |
CN104268521A (en) * | 2014-09-23 | 2015-01-07 | 朱毅 | Image recognition method based on convolutional neural network in non-finite category |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1905824A4 (en) * | 2005-05-31 | 2009-02-25 | Olympus Corp | A transgenic cell and a method for cell analysis |
CN101739712B (en) * | 2010-01-25 | 2012-01-18 | 四川大学 | Video-based 3D human face expression cartoon driving method |
CN102622585A (en) * | 2012-03-06 | 2012-08-01 | 同济大学 | Back propagation (BP) neural network face recognition method based on local feature Gabor wavelets |
CN102900305A (en) * | 2012-07-13 | 2013-01-30 | 太仓博天网络科技有限公司 | Keyless safe box system based on human face recognition |
CN102915434B (en) * | 2012-09-26 | 2016-05-25 | 上海交通大学 | A kind of face identification system based on low-power-consumption embedded platform |
-
2015
- 2015-02-08 CN CN201510063023.1A patent/CN104573679B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103824054A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascaded depth neural network-based face attribute recognition method |
CN104268521A (en) * | 2014-09-23 | 2015-01-07 | 朱毅 | Image recognition method based on convolutional neural network in non-finite category |
Also Published As
Publication number | Publication date |
---|---|
CN104573679A (en) | 2015-04-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104573679B (en) | Face identification system based on deep learning under monitoring scene | |
CN105868689B (en) | A kind of face occlusion detection method based on concatenated convolutional neural network | |
CN103984915B (en) | Pedestrian's recognition methods again in a kind of monitor video | |
CN109819208A (en) | A kind of dense population security monitoring management method based on artificial intelligence dynamic monitoring | |
CN112149638B (en) | Personnel identity recognition system construction and use method based on multi-modal biological characteristics | |
CN106919921B (en) | Gait recognition method and system combining subspace learning and tensor neural network | |
CN106327738B (en) | A kind of intelligent hierarchical monitoring system | |
CN106778556A (en) | Intelligent identity identification system based on dynamic human face identification and Gait Recognition fusion | |
CN105427421A (en) | Entrance guard control method based on face recognition | |
CN108009482A (en) | One kind improves recognition of face efficiency method | |
CN102629320A (en) | Ordinal measurement statistical description face recognition method based on feature level | |
CN206224519U (en) | Intelligent identity identification system based on dynamic human face identification and Gait Recognition fusion | |
CN202771441U (en) | Face recognition monitoring system | |
CN103593648B (en) | Face recognition method for open environment | |
CN104751143A (en) | Person and credential verification system and method based on deep learning | |
CN116563797B (en) | Monitoring management system for intelligent campus | |
CN111667599A (en) | Face recognition card punching system and method | |
CN110070055A (en) | A kind of capital construction scene safety detecting system and method based on deep learning | |
CN107480625A (en) | A kind of authentication identifying method based on recognition of face | |
CN112435414A (en) | Security monitoring system based on face recognition and monitoring method thereof | |
CN204926116U (en) | Picture system is declared in safety inspection containing video analysis | |
CN106341658A (en) | Intelligent city security state monitoring system | |
CN103294981A (en) | Intelligent face recognition system | |
CN113591619A (en) | Face recognition verification device based on video and verification method thereof | |
CN110443134B (en) | Face recognition tracking system based on video stream and working method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information |
Address after: 300457 10 North Zone, TEDA 57, Second Avenue, Binhai New Area Development Zone, Tianjin, Applicant after: Tianjin Aisikeer Technology Co., Ltd. Address before: 300457 building 20, No. sixth, 1 Avenue, Tianjin Binhai New Area Development Zone Applicant before: Tianjin Aisikeer Technology Co., Ltd. |
|
CB02 | Change of applicant information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |