CN109063535A - It is a kind of based on combined depth study pedestrian recognize again and pedestrian's gender classification method - Google Patents

It is a kind of based on combined depth study pedestrian recognize again and pedestrian's gender classification method Download PDF

Info

Publication number
CN109063535A
CN109063535A CN201810541294.7A CN201810541294A CN109063535A CN 109063535 A CN109063535 A CN 109063535A CN 201810541294 A CN201810541294 A CN 201810541294A CN 109063535 A CN109063535 A CN 109063535A
Authority
CN
China
Prior art keywords
pedestrian
parameter
gender classification
structures
gender
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.)
Granted
Application number
CN201810541294.7A
Other languages
Chinese (zh)
Other versions
CN109063535B (en
Inventor
朱建清
曾焕强
陈婧
蔡灿辉
杜永兆
傅玉青
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huaqiao University
Original Assignee
Huaqiao University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huaqiao University filed Critical Huaqiao University
Priority to CN201810541294.7A priority Critical patent/CN109063535B/en
Publication of CN109063535A publication Critical patent/CN109063535A/en
Application granted granted Critical
Publication of CN109063535B publication Critical patent/CN109063535B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The present invention relates to it is a kind of based on combined depth study pedestrian recognize again with pedestrian's gender classification method, can predict pedestrian's identity and pedestrian's gender simultaneously.Firstly, the identical depth network of two structures of building, is respectively used to pedestrian and recognizes again and pedestrian's Gender Classification；Secondly, utilize the parameter set of each layer in the identical depth network of parameter correlation canonical two structures of item constraint, so that be unlikely to that excessive deviation occurs in the optimization process of the two parameter, to avoid over-fitting, so that promote pedestrian simultaneously recognizes accuracy rate with pedestrian's Gender Classification again.

Description

It is a kind of based on combined depth study pedestrian recognize again and pedestrian's gender classification method
Technical field
The present invention relates to intelligent video monitoring, machine vision and machine learning, in particular to a kind of to be based on combined depth The pedestrian of habit recognizes and pedestrian's gender classification method again.
Background technique
The pedestrian largely based on depth network is emerged in recent years to recognize again or pedestrian's Gender Classification algorithm.But these Algorithm often recognizes pedestrian again or pedestrian's Gender Classification is as two mutually independent tasks, and there is no combine the two Study, natural recognize again to pedestrian leave leeway with the promotion of the accuracy rate of pedestrian's Gender Classification.
Some depth networks for being successfully applied to recognition of face, such as ResNet, GoogLeNet, DenseNet may not Pedestrian can be directly applied to recognize again or pedestrian's Gender Classification task, because the database size in terms of pedestrian is well below people Database size in terms of face.Depth network is easy to appear over-fitting on small-scale pedestrian's database, to limit Pedestrian recognizes again or the accuracy rate of pedestrian's Gender Classification algorithm.
Summary of the invention
The purpose of the present invention is to provide a kind of pedestrians based on combined depth study to recognize again and pedestrian Gender Classification side Method can be used for predicting pedestrian's identity and pedestrian's gender simultaneously, and can avoid the over-fitting of depth network, and then be conducive to mention simultaneously It rises pedestrian and recognizes the accuracy rate with pedestrian's Gender Classification again.
To achieve the above object, the technical solution adopted by the present invention is that:
It is a kind of based on combined depth study pedestrian recognize again with pedestrian's gender classification method, specifically includes the following steps:
Step 1, the identical depth network of two structures of building, are respectively used to pedestrian and recognize again and pedestrian's Gender Classification；
Step 2 utilizes the parameter set of each layer in the identical depth network of the described two structures of parameter correlation canonical item constraint.
Further, it is recognized again in the step 1 for pedestrian identical with described two structures of pedestrian's Gender Classification The corresponding parameter set of depth network is respectively H1={ H11,H12,…,H1nAnd H2={ H21,H22,…,H2n, wherein n is depth net Convolutional layer number in network, then parameter correlation regular terms R (H both1,H2) it is defined as follows:
Wherein,Indicate point multiplication operation, H1iExpression parameter collection H1In i-th of convolutional layer parameter, H2iExpression parameter collection H2 In i-th of convolutional layer parameter,The Frobenius norm (not this black norm of Luo Beini) of representing matrix；
Based on gradient descent algorithm to H1And H2It optimizes, i.e., the update rule of parameter is as follows in back-propagating:
Wherein, L1And L2Respectively indicate the Softmax loss function recognized again for pedestrian with pedestrian's Gender Classification；λ >=0, For control parameter correlation regular terms R (H1,H2) weight；T+1 and t indicates iteration twice adjacent in gradient descent algorithm； Derivative operation is sought in expression.
It can be respectively used to go, it is evident that after parameter correlation regular terms defined in introducing formula (1) from formula (2) The parameter set that people recognizes depth network identical with described two structures of pedestrian's Gender Classification again is interaction, it is intended to be increased Add the correlation between the corresponding parameter set of the identical depth network of described two structures (Deep Network), to limit correspondence Departure degree of the parameter set in optimization process, to avoid the over-fitting of depth network.
In view of gender information having the same inevitable between the pedestrian with common identity information, and there are different sexes It can not identity information having the same between the pedestrian of information.This explanation, pedestrian recognizes again to be appointed with pedestrian's Gender Classification two There is correlation between business.After adopting the above scheme, the present invention learns the identical depth network difference of two structures simultaneously It is recognized again for pedestrian and pedestrian's Gender Classification.During e-learning, parameter correlation canonical two depths of item constraint are utilized The parameter set in network is spent, and then avoids the over-fitting of depth network, is recognized again and pedestrian's Gender Classification to improve pedestrian Accuracy rate.
Detailed description of the invention
Fig. 1 is that the pedestrian based on combined depth study in the present invention recognizes and pedestrian's gender classification method schematic diagram again；
Fig. 2 is the structural schematic diagram of CBLR unit in the present invention；
Fig. 3 is the structural schematic diagram of depth network of the invention.
Specific embodiment
Present invention discloses it is a kind of based on combined depth study pedestrian recognize again with pedestrian's gender classification method, pass through structure Build the identical depth network of two structures be respectively used to pedestrian recognize again with pedestrian's Gender Classification, during e-learning, Using the parameter set of each layer in the identical depth network of parameter correlation canonical two structures of item constraint, and then avoid depth network Over-fitting, so that improve pedestrian recognizes accuracy rate with pedestrian's Gender Classification again.
As a preferred embodiment, as shown in Figure 1, a kind of pedestrian based on combined depth study recognizes again and pedestrian Gender classification method, the specific steps are as follows:
Step 1, the identical depth network of two structures of building, are respectively used to pedestrian and recognize again and pedestrian's Gender Classification；
As shown in Fig. 2, for ease of description, the present invention by convolutional layer (Conv), batch normalization layer (BatchNorm) and Leaky ReLU activation primitive is integrated into CBLR unit, is used uniformly the filter of 3 × 3 sizes in the present embodiment in all CBLR units Wave device, and the negative semiaxis slope of Leaky ReLU activation primitive is set as 0.15, and is operated using 1 pixel zero padding.Based in Fig. 2 CBLR unit, the present embodiment using VGGNet structure building depth network (i.e. depth network in Fig. 1).As shown in figure 3, institute The depth network of building includes four maximum pond layers (MP1, MP2, MP3 and MP4), and is pacified respectively before MP1, MP2, MP3 and MP4 4 CBLR units (CBLR1-4), 3 CBLR units (CBLR5-7), 3 CBLR units (CBLR8-10) and 3 CBLR are put Unit (CBLR11-13).
Step 2 utilizes the parameter set of each layer in the identical depth network of parameter correlation canonical two structures of item constraint；
Assuming that recognizing depth network identical with described two structures of pedestrian's Gender Classification again for pedestrian in step 1 (Deep Network) corresponding parameter set is respectively H1={ H11,H12,…,H1nAnd H2={ H21,H22,…,H2n, wherein n is Convolutional layer number in depth network, then the parameter correlation regular terms of the two is defined as follows:
Wherein,Indicate point multiplication operation, H1iExpression parameter collection H1In i-th of convolutional layer parameter, H2iExpression parameter collection H2 In i-th of convolutional layer parameter,The Frobenius norm (not this black norm of Luo Beini) of representing matrix；Currently, in depth During network training, gradient descent algorithm optimization is commonly used, based on gradient descent algorithm to H1And H2It optimizes, i.e., backward The update rule of parameter in propagation is as follows:
Wherein, L1And L2Respectively indicate the Softmax loss function recognized again for pedestrian with pedestrian's Gender Classification；λ >=0, For control parameter correlation regular terms R (H1,H2) weight；T+1 and t indicates iteration twice adjacent in gradient descent method；Table Show and seeks derivative operation；It can be used respectively, it is evident that after parameter correlation regular terms defined in introducing formula (1) from formula (2) It is to interact in the parameter set that pedestrian recognizes depth network identical with described two structures of pedestrian's Gender Classification again, purport Increasing the correlation between the corresponding parameter set of the identical depth network (Deep Network) of described two structures, with limitation Departure degree of the corresponding parameter set in optimization process, to avoid the over-fitting of depth network.
In view of gender information having the same inevitable between the pedestrian with common identity information, and there are different sexes It can not identity information having the same between the pedestrian of information.This explanation, pedestrian recognizes again to be appointed with pedestrian's Gender Classification two There is correlation between business.After adopting the above scheme, the present invention learns the identical depth network difference of two structures simultaneously It is recognized again for pedestrian and pedestrian's Gender Classification.During e-learning, parameter correlation canonical two depths of item constraint are utilized The parameter set in network is spent, and then avoids the over-fitting of depth network, is recognized again and pedestrian's gender to improve pedestrian simultaneously The accuracy rate of classification.
The above is only the embodiment of the present invention, is not intended to limit the scope of the present invention, therefore all Any subtle modifications, equivalent variations and modifications to the above embodiments according to the technical essence of the invention still fall within this In the range of inventive technique scheme.

Claims (2)

1. a kind of pedestrian based on combined depth study recognizes and pedestrian's gender classification method again, which is characterized in that specifically include Following steps:
Step 1, the identical depth network of two structures of building, are respectively used to pedestrian and recognize again and pedestrian's Gender Classification；
Step 2 utilizes the parameter set of each layer in the identical depth network of the described two structures of parameter correlation canonical item constraint.
2. it is according to claim 1 it is a kind of based on combined depth study pedestrian recognize again with pedestrian's gender classification method, It is characterized by: recognizing depth net identical with described two structures of pedestrian's Gender Classification again for pedestrian in the step 1 The corresponding parameter set of network is respectively H1={ H11,H12,…,H1nAnd H2={ H21,H22,…,H2n, wherein n is to roll up in depth network Lamination number, then parameter correlation regular terms R (H both1,H2) it is defined as follows:
Wherein,Indicate point multiplication operation, H1iExpression parameter collection H1In i-th of convolutional layer parameter, H2iExpression parameter collection H2In i-th The parameter of a convolutional layer,The Frobenius norm of representing matrix；
Based on gradient descent algorithm to H1And H2It optimizes, i.e., the update rule of parameter is as follows in back-propagating:
Wherein, L1And L2Respectively indicate the Softmax loss function recognized again for pedestrian with pedestrian's Gender Classification；λ >=0 is used for Control parameter correlation regular terms R (H1,H2) weight；T+1 and t indicates iteration twice adjacent in gradient descent algorithm；It indicates Seek derivative operation.
CN201810541294.7A 2018-05-30 2018-05-30 Pedestrian re-identification and pedestrian gender classification method based on joint deep learning Active CN109063535B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810541294.7A CN109063535B (en) 2018-05-30 2018-05-30 Pedestrian re-identification and pedestrian gender classification method based on joint deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810541294.7A CN109063535B (en) 2018-05-30 2018-05-30 Pedestrian re-identification and pedestrian gender classification method based on joint deep learning

Publications (2)

Publication Number Publication Date
CN109063535A true CN109063535A (en) 2018-12-21
CN109063535B CN109063535B (en) 2021-09-28

Family

ID=64819809

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810541294.7A Active CN109063535B (en) 2018-05-30 2018-05-30 Pedestrian re-identification and pedestrian gender classification method based on joint deep learning

Country Status (1)

Country Link
CN (1) CN109063535B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110516569A (en) * 2019-08-15 2019-11-29 华侨大学 A kind of pedestrian's attribute recognition approach of identity-based and non-identity attribute interactive learning
CN111435452A (en) * 2019-01-11 2020-07-21 百度在线网络技术（北京）有限公司 Model training method, device, equipment and medium
CN111695415A (en) * 2020-04-28 2020-09-22 平安科技（深圳）有限公司 Construction method and identification method of image identification model and related equipment

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101290658A (en) * 2007-04-18 2008-10-22 中国科学院自动化研究所 Gender recognition method based on gait
KR20160017152A (en) * 2014-07-31 2016-02-16 경성대학교 산학협력단 Gender Classification Based on Binary Haar Cascade
US20160171546A1 (en) * 2013-12-14 2016-06-16 Fly Fleet LLC Situation Aware Travel Information
CN105718873A (en) * 2016-01-18 2016-06-29 北京联合大学 People stream analysis method based on binocular vision
CN106960181A (en) * 2017-02-28 2017-07-18 中科唯实科技（北京）有限公司 A kind of pedestrian's attribute recognition approach based on RGBD data
CN106960176A (en) * 2017-02-22 2017-07-18 华侨大学 A kind of pedestrian's gender identification method based on transfinite learning machine and color characteristic fusion
CN107480631A (en) * 2017-08-11 2017-12-15 中山大学 A kind of pedestrian's recognition methods again based on visual angle association with depth e-learning

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101290658A (en) * 2007-04-18 2008-10-22 中国科学院自动化研究所 Gender recognition method based on gait
US20160171546A1 (en) * 2013-12-14 2016-06-16 Fly Fleet LLC Situation Aware Travel Information
KR20160017152A (en) * 2014-07-31 2016-02-16 경성대학교 산학협력단 Gender Classification Based on Binary Haar Cascade
CN105718873A (en) * 2016-01-18 2016-06-29 北京联合大学 People stream analysis method based on binocular vision
CN106960176A (en) * 2017-02-22 2017-07-18 华侨大学 A kind of pedestrian's gender identification method based on transfinite learning machine and color characteristic fusion
CN106960181A (en) * 2017-02-28 2017-07-18 中科唯实科技（北京）有限公司 A kind of pedestrian's attribute recognition approach based on RGBD data
CN107480631A (en) * 2017-08-11 2017-12-15 中山大学 A kind of pedestrian's recognition methods again based on visual angle association with depth e-learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
LEI CAI: "《Multi-view joint learning network for pedestrian gender classification》", 《2017 INTERNATIONAL SYMPOSIUM ON INTELLIGENT SIGNAL PROCESSING AND COMMUNICATION SYSTEM》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111435452A (en) * 2019-01-11 2020-07-21 百度在线网络技术（北京）有限公司 Model training method, device, equipment and medium
CN111435452B (en) * 2019-01-11 2023-11-03 百度在线网络技术（北京）有限公司 Model training method, device, equipment and medium
CN110516569A (en) * 2019-08-15 2019-11-29 华侨大学 A kind of pedestrian's attribute recognition approach of identity-based and non-identity attribute interactive learning
CN111695415A (en) * 2020-04-28 2020-09-22 平安科技（深圳）有限公司 Construction method and identification method of image identification model and related equipment
CN111695415B (en) * 2020-04-28 2024-04-12 平安科技（深圳）有限公司 Image recognition method and related equipment

Also Published As

Publication number Publication date
CN109063535B (en) 2021-09-28

Similar Documents

Publication Publication Date Title
CN109376913A (en) The prediction technique and device of precipitation
CN107992982B (en) A kind of Default Probability prediction technique of the unstructured data based on deep learning
US10832136B2 (en) Passive pruning of filters in a convolutional neural network
CN108010514B (en) Voice classification method based on deep neural network
CN106982359B (en) Binocular video monitoring method and system and computer readable storage medium
CN106471526B (en) For handling the method and system of image
CN110175628A (en) A kind of compression algorithm based on automatic search with the neural networks pruning of knowledge distillation
CN103345656B (en) A kind of data identification method based on multitask deep neural network and device
CN109214441A (en) A kind of fine granularity model recognition system and method
CN104700828A (en) Deep long-term and short-term memory recurrent neural network acoustic model establishing method based on selective attention principles
CN108133188A (en) A kind of Activity recognition method based on motion history image and convolutional neural networks
CN110580496A (en) Deep migration learning system and method based on entropy minimization
CN107679580A (en) A kind of isomery shift image feeling polarities analysis method based on the potential association of multi-modal depth
CN106845499A (en) A kind of image object detection method semantic based on natural language
CN109063535A (en) It is a kind of based on combined depth study pedestrian recognize again and pedestrian's gender classification method
CN107330355B (en) Deep pedestrian re-identification method based on positive sample balance constraint
CN106203363A (en) Human skeleton motion sequence Activity recognition method
CN105512680A (en) Multi-view SAR image target recognition method based on depth neural network
CN104217214A (en) Configurable convolutional neural network based red green blue-distance (RGB-D) figure behavior identification method
CN104123007A (en) Multidimensional weighted 3D recognition method for dynamic gestures
CN108596327A (en) A kind of seismic velocity spectrum artificial intelligence pick-up method based on deep learning
CN110300095A (en) A kind of deep learning network inbreak detection method based on improvement learning rate
CN103065158A (en) Action identification method of independent subspace analysis (ISA) model based on relative gradient
CN104537684A (en) Real-time moving object extraction method in static scene
CN114037945A (en) Cross-modal retrieval method based on multi-granularity feature interaction

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
GR01 Patent grant
GR01 Patent grant