CN111160226B - Pedestrian gender identification method based on visual angle adaptive feature learning - Google Patents

Pedestrian gender identification method based on visual angle adaptive feature learning Download PDF

Info

Publication number
CN111160226B
CN111160226B CN201911370041.9A CN201911370041A CN111160226B CN 111160226 B CN111160226 B CN 111160226B CN 201911370041 A CN201911370041 A CN 201911370041A CN 111160226 B CN111160226 B CN 111160226B
Authority
CN
China
Prior art keywords
pedestrian
visual angle
adaptive
gender identification
feature
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
Application number
CN201911370041.9A
Other languages
Chinese (zh)
Other versions
CN111160226A (en
Inventor
曾焕强
蔡磊
陈婧
朱建清
曹九稳
王勇涛
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 CN201911370041.9A priority Critical patent/CN111160226B/en
Publication of CN111160226A publication Critical patent/CN111160226A/en
Application granted granted Critical
Publication of CN111160226B publication Critical patent/CN111160226B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a pedestrian gender identification method based on visual angle adaptive feature learning, which comprises the following steps: a visual angle self-adaptive training process and a gender identification process. The invention guides the characteristic learning process of the convolutional neural network by utilizing the visual angle information of the input pedestrian so as to reduce the influence of the visual angle change of the pedestrian on the gender identification of the neural network, and the trained network model has more accurate pedestrian gender identification effect. The invention combines the visual angle information of the pedestrian, solves the defect of the prior convolutional neural network-based pedestrian gender identification problem, and effectively improves the accuracy of the pedestrian gender identification. The invention can be widely applied to intelligent video monitoring scenes, such as superstores, airports, railway stations and the like.

Description

Pedestrian gender identification method based on visual angle adaptive feature learning
Technical Field
The invention relates to computer vision and pattern recognition, in particular to a pedestrian gender recognition method based on visual angle adaptive feature learning.
Background
In recent years, with the active push of "smart cities" and the increasing demand of traffic monitoring, video monitoring will gradually cover various important places such as superstores, airports, train stations, etc. Tens of millions of cameras will provide basic guarantee for urban public safety. In order to meet the requirements of intelligent security, intelligent traffic, intelligent home and the like, a rapid identification technology for people moving in a remote and target non-cooperative state is urgently needed for video monitoring intellectualization, so that the identity of people can be rapidly confirmed under a remote condition, and intelligent management is realized. As an important auxiliary means for rapid identification of pedestrians, the pedestrian gender identification means identifying the gender of passing pedestrians in a monitoring video, and the technology will play an important role in a future intelligent video monitoring system.
The pedestrian gender identification method in the prior art is mainly used for identifying the gender of the pedestrian based on manual features, such as gradient histogram features (HOG) capable of describing the outline and shape of the pedestrian and LBP features capable of describing the texture details of the pedestrian, but the identification accuracy of a single manual feature extraction method is generally not high.
With the rapid development of deep learning, the convolutional neural network is effectively applied to a human-based identification task, and higher identification precision is obtained compared with manual characteristics. However, the method based on the deep convolutional neural network is sensitive to the perspective transformation of the pedestrian, for example, when the perspective of the pedestrian is changed, the convolutional neural network may not be able to correctly identify the gender of the pedestrian at a certain perspective.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a pedestrian gender identification method based on visual angle adaptive feature learning, and effectively improves the accuracy of pedestrian gender identification.
The technical scheme of the invention is as follows:
a pedestrian gender identification method based on visual angle adaptive feature learning comprises a visual angle adaptive training process and a gender identification process;
the visual angle self-adaptive training process comprises the following steps:
1.1 Basic model training step: selecting N training images with gender label attributes, inputting the training images into a convolutional neural network for training until the convolutional neural network is converged to obtain a basic model M;
1.2 View influence score calculation step: dividing the training image into a forward visual angle, a backward visual angle and other visual angles, respectively inputting the training image into a basic model M for extracting characteristics, and calculating the average influence score of the corresponding visual angle according to the extracted characteristics;
1.3 View angle fine-tuning step: adjusting the view angle of the basic model M by using the average influence score of each view angle until the model converges to obtain a feature extraction model P;
the steps of the gender identification process are as follows:
2.1 Inputting the test image and the average influence score obtained from the step 1.2) into a feature extraction model P, and obtaining view angle adaptive features through forward propagation;
2.2 Computing the gender probability of the view angle self-adaptive features by utilizing a Softmax classification function, and outputting a gender identification result.
Preferably, the basic model training steps are as follows:
1.1.1 Randomly selecting N training images with gender label attribute;
1.1.2 Inputting the selected training image into a convolutional neural network for training;
1.1.3 ) repeating the step 1.1.1) and the step 1.1.2) until the convolutional neural network is converged to obtain a basic model M;
the steps of calculating the view influence score are as follows:
1.2.1 Divide the training image into a forward view, a backward view, and other views;
1.2.2 Respectively inputting the training images of three visual angles into the basic model M obtained in the step 1.1.3), and obtaining the pedestrian depth characteristics gamma under the three visual angles through forward propagation frontal 、γ back 、γ other ；
1.2.3 According to the pedestrian depth characteristic gamma at three viewing angles frontal 、γ back 、γ other Calculating corresponding average influence scores
Figure BDA0002339432910000021
The visual angle fine adjustment steps are as follows:
1.3.1 Step 1.2.3) average impact score for each perspective
Figure BDA0002339432910000022
Adjusting the visual angle of the basic model M;
1.3.2 Step 1.3.1) is repeated until the model converges, resulting in the feature extraction model P.
Preferably, the average impact score of the forward viewing angle
Figure BDA0002339432910000031
The calculation process of (2) is as follows:
the average influence score of the jth neuron of the network characteristic output layer is as follows:
Figure BDA0002339432910000032
I j,frontal ＝L(γ frontal\j )-L(γ frontal )；
wherein F represents the depth feature set of the pedestrian with the forward view angle, E (-) is the averaging operation, I j,frontal Denotes the impact score, γ, of the jth neuron frontal D-dimensional feature vector, gamma, representing the M output of the base model frontal\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
Preferably, the average impact score of the back viewing angle
Figure BDA0002339432910000033
The calculation process of (2) is as follows:
the average influence score of the jth neuron of the network characteristic output layer is as follows:
Figure BDA0002339432910000034
I j,back ＝L(γ back\j )-L(γ back )；
wherein F represents the depth feature set of the pedestrian with the back view angle, E (-) is the average operation, I j,back Denotes the impact score, γ, of the jth neuron back D-dimensional feature vector, gamma, representing the M output of the base model back\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
Preferably, the average impact score of other perspectives
Figure BDA0002339432910000035
The calculation process of (2) is as follows:
the average influence score of the jth neuron of the network feature output layer is as follows:
Figure BDA0002339432910000036
I j,othter ＝L(γ othter\j )-L(γ othter )；
wherein F represents the depth feature set of pedestrians at other visual angles, E (-) is the averaging operation, I j,othter Denotes the impact score, γ, of the jth neuron othter D-dimensional feature vector, gamma, representing the M output of the base model othter\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
Preferably, in step 1.3.1), the perspective adjustment of the base model M includes a forward propagation process and a backward propagation process;
in the forward propagation process, performing point multiplication on the characteristic vector and a view angle influence mask, and inhibiting neuron response irrelevant to a forward view angle to obtain a view angle self-adaptive characteristic vector;
in the process of backward propagation, the visual angle self-adaptive characteristic vector is substituted into a Softmax Loss function, and error Loss is calculated; and optimizing network parameters according to the error loss back propagation until the model converges to obtain a feature extraction model P.
Preferably, the viewing angle influencing mask is specifically as follows:
Figure BDA0002339432910000041
wherein m is j,frontal A view angle impact mask representing the jth neuron of the net feature output layer,
Figure BDA0002339432910000042
the mean impact score for the jth neuron is shown.
Preferably, the gender identification process is specifically as follows:
test images and average influence scores obtained in step 1.2.3)
Figure BDA0002339432910000043
Inputting a feature extraction model P, outputting a feature vector through forward propagation, and performing dot product calculation with a view angle influence mask to obtain view angle adaptive features; and calculating the gender probability for the self-adaptive characteristics of the input visual angle by utilizing a Softmax classification function, and outputting a gender identification result.
The invention has the following beneficial effects:
according to the pedestrian gender identification method based on visual angle adaptive feature learning, the characteristic learning process of the convolutional neural network is guided by using the input visual angle information of the pedestrian, so that the influence of the visual angle change of the pedestrian on the gender identification of the neural network is reduced, and the trained network model has a more accurate pedestrian gender identification effect. The invention combines the visual angle information of the pedestrian, solves the defect of the prior convolutional neural network-based pedestrian gender identification problem, and effectively improves the accuracy of the pedestrian gender identification.
The invention can be widely applied to intelligent video monitoring scenes, such as superstores, airports, railway stations and the like.
Drawings
Fig. 1 is a schematic diagram of the principle of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
The invention provides a pedestrian gender identification method based on visual angle adaptive feature learning, which comprises a visual angle adaptive training process and a gender identification process as shown in figure 1.
The visual angle self-adaptive training process comprises the following steps:
1.1 Basic model training step: selecting N training images with gender label attributes, inputting the training images into a convolutional neural network for training until the convolutional neural network is converged to obtain a basic model M;
1.2 View influence score calculation step: dividing the training image into a forward visual angle, a backward visual angle and other visual angles, respectively inputting the training image into a basic model M for extracting characteristics, and calculating the average influence score of the corresponding visual angle according to the extracted characteristics;
1.3 View angle fine-tuning step: and adjusting the view angle of the basic model M by using the average influence score of each view angle until the model converges to obtain a feature extraction model P.
Specifically, in this embodiment, the basic model training steps are specifically as follows:
1.1.1 Randomly selecting N training images with the attribute of the gender label;
1.1.2 Inputting the selected training image into a convolutional neural network for training;
1.1.3 ) repeating the step 1.1.1) and the step 1.1.2) until the convolutional neural network converges to obtain the basic model M.
The steps of calculating the view influence score are as follows:
1.2.1 Divide the training image into a forward view, a backward view, and other views;
1.2.2 Respectively inputting the training images of three visual angles into the basic model M obtained in the step 1.1.3), and obtaining the pedestrian depth characteristics gamma under the three visual angles through forward propagation frontal 、γ back 、γ other ；
1.2.3 According to the pedestrian depth characteristic gamma at three viewing angles frontal 、γ back 、γ other Calculating corresponding average influence scores
Figure BDA0002339432910000051
In this embodiment, the average impact score of the forward viewing angle
Figure BDA0002339432910000052
The calculation process of (2) is as follows:
the average influence score of the jth neuron of the network feature output layer is as follows:
Figure BDA0002339432910000053
I j,frontal ＝L(γ frontal\j )-L(γ frontal )；
wherein F represents the depth feature set of the pedestrian with the forward view angle, E (-) is the averaging operation, I j,frontal Denotes the impact score, γ, of the jth neuron frontal D-dimensional feature vector, gamma, representing the M output of the base model frontal\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
Similarly, average impact score for backward viewing angle
Figure BDA0002339432910000061
The calculation process of (2) is as follows:
the average influence score of the jth neuron of the network characteristic output layer is as follows:
Figure BDA0002339432910000062
I j,back ＝L(γ back\j )-L(γ back )；
wherein F represents the depth feature set of the pedestrian with a back view angle, E (-) is the averaging operation, I j,back Representing the j-th neuronInfluence fraction, gamma back D-dimensional feature vector, gamma, representing the M output of the base model back\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
Similarly, the average impact score for other perspectives
Figure BDA0002339432910000063
The calculation process of (c) is as follows:
the average influence score of the jth neuron of the network characteristic output layer is as follows:
Figure BDA0002339432910000064
I j,othter ＝L(γ othter\j )-L(γ othter )；
wherein F represents the depth feature set of pedestrians at other visual angles, E (-) is the averaging operation, I j,othter Denotes the impact score, γ, of the jth neuron othter D-dimensional feature vector, gamma, representing the M output of the base model othter\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
The visual angle fine adjustment steps are as follows:
1.3.1 Step 1.2.3) average impact score for each perspective
Figure BDA0002339432910000065
And adjusting the visual angle of the basic model M. The visual angle adjustment of the basic model M comprises a forward propagation process and a backward propagation process;
in the forward propagation process, performing point multiplication on the characteristic vector and a view angle influence mask, and inhibiting neuron response irrelevant to a forward view angle to obtain a view angle self-adaptive characteristic vector; the view angle influence mask is specifically as follows:
Figure BDA0002339432910000066
wherein m is j,frontal A view angle impact mask representing the jth neuron of the net feature output layer,
Figure BDA0002339432910000067
the mean impact score for the jth neuron is shown.
In the process of backward propagation, the visual angle self-adaptive characteristic vector is substituted into a Softmax Loss function, and error Loss is calculated; and optimizing network parameters according to the error loss back propagation until the model converges to obtain a feature extraction model P.
1.3.2 Step 1.3.1) is repeated until the model converges, resulting in the feature extraction model P.
The steps of the gender identification process are as follows:
2.1 Inputting the test image and the average influence score obtained from the step 1.2) into a feature extraction model P, and obtaining visual angle adaptive features through forward propagation;
2.2 Computing the gender probability of the view angle self-adaptive features by utilizing a Softmax classification function, and outputting a gender identification result.
In this embodiment, the gender identification process is specifically as follows:
test images and the average influence scores obtained in the step 1.2.3)
Figure BDA0002339432910000071
Inputting a feature extraction model P, outputting a feature vector through forward propagation, and performing dot product calculation with a view angle influence mask to obtain view angle adaptive features; and calculating the gender probability for the self-adaptive characteristics of the input visual angle by utilizing a Softmax classification function, and outputting a gender identification result.
The above examples are provided only for illustrating the present invention and are not intended to limit the present invention. Changes, modifications, etc. to the above-described embodiments are intended to fall within the scope of the claims of the present invention as long as they are in accordance with the technical spirit of the present invention.

Claims (7)

1. A pedestrian gender identification method based on visual angle adaptive feature learning is characterized by comprising a visual angle adaptive training process and a gender identification process;
the visual angle self-adaptive training process comprises the following steps:
1.1 Basic model training step: selecting N training images with gender label attributes, inputting the training images into a convolutional neural network for training until the convolutional neural network is converged to obtain a basic model M;
1.2 View influence score calculation step: dividing the training image into a forward visual angle, a backward visual angle and other visual angles, respectively inputting the training image into a basic model M for extracting characteristics, and calculating the average influence score of the corresponding visual angle according to the extracted characteristics;
1.3 View angle fine-tuning step: adjusting the view angle of the basic model M by using the average influence score of each view angle until the model converges to obtain a feature extraction model P;
the steps of the gender identification process are as follows:
2.1 Inputting the test image and the average influence score obtained from the step 1.2) into a feature extraction model P, and obtaining view angle adaptive features through forward propagation;
2.2 Utilizing a Softmax classification function to calculate the gender probability of the visual angle self-adaptive features, and outputting a gender identification result;
the basic model training steps are as follows:
1.1.1 Randomly selecting N training images with gender label attribute;
1.1.2 Inputting the selected training image into a convolutional neural network for training;
1.1.3 ) repeating the step 1.1.1) and the step 1.1.2) until the convolutional neural network is converged to obtain a basic model M;
the steps of calculating the view influence score are as follows:
1.2.1 Divide the training image into a forward view, a backward view, and other views;
1.2.2 Respectively inputting the training images of three visual angles into the basic model M obtained in the step 1.1.3), and obtaining the pedestrian depth characteristics gamma under the three visual angles through forward propagation frontal 、γ back 、γ other ；
1.2.3 According to the pedestrian depth feature gamma at three viewing angles frontal 、γ back 、γ other Calculating corresponding average influence scores
Figure FDA0003977506690000011
The visual angle fine adjustment steps are as follows:
1.3.1 Step 1.2.3) average impact score for each perspective
Figure FDA0003977506690000021
Adjusting the visual angle of the basic model M;
1.3.2 Step 1.3.1) is repeated until the model converges, resulting in the feature extraction model P.
2. The method of claim 1, wherein the average impact score of forward perspective is based on the perspective adaptive feature learning for pedestrian gender identification
Figure FDA0003977506690000022
The calculation process of (2) is as follows:
the average influence score of the jth neuron of the network feature output layer is as follows:
Figure FDA0003977506690000023
I j,frontal ＝L(γ frontal\j )-L(γ frontal )；
wherein F represents the depth feature set of the pedestrian with the forward view angle, E (-) is the averaging operation, I j,frontal Denotes the impact score, γ, of the jth neuron frontal D-dimensional feature vector, gamma, representing the M output of the base model frontal\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
3. According to claimThe method of pedestrian gender identification based on perspective-adaptive feature learning of claim 1, wherein the average impact score of the back-to-perspective
Figure FDA0003977506690000024
The calculation process of (2) is as follows: />
The average influence score of the jth neuron of the network characteristic output layer is as follows:
Figure FDA0003977506690000025
I j,back ＝L(γ back\j )-L(γ back )；
wherein F represents the depth feature set of the pedestrian with the back view angle, E (-) is the average operation, I j,back Denotes the impact score, γ, of the jth neuron back D-dimensional feature vector, gamma, representing the M output of the base model back\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
4. The pedestrian gender identification method based on perspective adaptive feature learning of claim 1, wherein the average impact score of other perspectives
Figure FDA0003977506690000026
The calculation process of (2) is as follows:
the average influence score of the jth neuron of the network feature output layer is as follows:
Figure FDA0003977506690000031
I j,othter ＝L(γ othter\j )-L(γ othter )；
wherein F represents the depth feature set of the pedestrian at other view angles, E (-) is the averaging operation, I j,othter Means the jth godFractional influence of meridian elements, gamma othter D-dimensional feature vector, gamma, representing the M output of the base model othter\j And L (-) represents a characteristic vector obtained when the jth neuron response of the network characteristic output layer is set to be 0, and represents a Softmax Loss function.
5. The pedestrian gender identification method based on the visual angle adaptive feature learning of claim 1, wherein in the step 1.3.1), the visual angle adjustment of the basic model M comprises a forward propagation process and a backward propagation process;
in the forward propagation process, performing point multiplication on the characteristic vector and a view angle influence mask, and inhibiting neuron response irrelevant to a forward view angle to obtain a view angle self-adaptive characteristic vector;
in the process of backward propagation, the visual angle self-adaptive characteristic vector is substituted into a Softmax Loss function, and error Loss is calculated; and optimizing network parameters according to the error loss back propagation until the model converges to obtain a feature extraction model P.
6. The pedestrian gender identification method based on the perspective adaptive feature learning of claim 5, wherein the perspective influence mask is specifically as follows:
Figure FDA0003977506690000032
wherein m is j,frontal A view angle impact mask representing the jth neuron of the net feature output layer,
Figure FDA0003977506690000033
the mean impact score for the jth neuron is shown.
7. The pedestrian gender identification method based on the visual angle adaptive feature learning of claim 6, wherein the gender identification process is as follows:
test images and average influence scores obtained in step 1.2.3)
Figure FDA0003977506690000034
Inputting a feature extraction model P, outputting a feature vector through forward propagation, and performing dot product calculation with a view angle influence mask to obtain view angle adaptive features; and calculating the gender probability for the self-adaptive characteristics of the input visual angle by utilizing a Softmax classification function, and outputting a gender identification result. />
CN201911370041.9A 2019-12-26 2019-12-26 Pedestrian gender identification method based on visual angle adaptive feature learning Active CN111160226B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911370041.9A CN111160226B (en) 2019-12-26 2019-12-26 Pedestrian gender identification method based on visual angle adaptive feature learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911370041.9A CN111160226B (en) 2019-12-26 2019-12-26 Pedestrian gender identification method based on visual angle adaptive feature learning

Publications (2)

Publication Number Publication Date
CN111160226A CN111160226A (en) 2020-05-15
CN111160226B true CN111160226B (en) 2023-03-31

Family

ID=70556854

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911370041.9A Active CN111160226B (en) 2019-12-26 2019-12-26 Pedestrian gender identification method based on visual angle adaptive feature learning

Country Status (1)

Country Link
CN (1) CN111160226B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108009525A (en) * 2017-12-25 2018-05-08 北京航空航天大学 A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks
WO2018107760A1 (en) * 2016-12-16 2018-06-21 北京大学深圳研究生院 Collaborative deep network model method for pedestrian detection
CN109711281A (en) * 2018-12-10 2019-05-03 复旦大学 A kind of pedestrian based on deep learning identifies again identifies fusion method with feature

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018107760A1 (en) * 2016-12-16 2018-06-21 北京大学深圳研究生院 Collaborative deep network model method for pedestrian detection
CN108009525A (en) * 2017-12-25 2018-05-08 北京航空航天大学 A kind of specific objective recognition methods over the ground of the unmanned plane based on convolutional neural networks
CN109711281A (en) * 2018-12-10 2019-05-03 复旦大学 A kind of pedestrian based on deep learning identifies again identifies fusion method with feature

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
视频监控下的行人性别检测;苏宁等;《现代计算机(专业版)》;20181015(第29期);全文 *

Also Published As

Publication number Publication date
CN111160226A (en) 2020-05-15

Similar Documents

Publication Publication Date Title
CN109145939B (en) Semantic segmentation method for small-target sensitive dual-channel convolutional neural network
CN110163110B (en) Pedestrian re-recognition method based on transfer learning and depth feature fusion
US11195051B2 (en) Method for person re-identification based on deep model with multi-loss fusion training strategy
CN107563372B (en) License plate positioning method based on deep learning SSD frame
CN110781838B (en) Multi-mode track prediction method for pedestrians in complex scene
CN109543695B (en) Population-density population counting method based on multi-scale deep learning
CN110163187B (en) F-RCNN-based remote traffic sign detection and identification method
CN109815826B (en) Method and device for generating face attribute model
CN110276264B (en) Crowd density estimation method based on foreground segmentation graph
CN111191667B (en) Crowd counting method based on multiscale generation countermeasure network
CN110135296A (en) Airfield runway FOD detection method based on convolutional neural networks
CN107145889A (en) Target identification method based on double CNN networks with RoI ponds
CN103605972A (en) Non-restricted environment face verification method based on block depth neural network
CN106096535A (en) A kind of face verification method based on bilinearity associating CNN
CN111178208A (en) Pedestrian detection method, device and medium based on deep learning
CN104504362A (en) Face detection method based on convolutional neural network
CN110390308B (en) Video behavior identification method based on space-time confrontation generation network
CN103971091B (en) Automatic plane number recognition method
CN107657625A (en) Merge the unsupervised methods of video segmentation that space-time multiple features represent
CN109948593A (en) Based on the MCNN people counting method for combining global density feature
CN113963032A (en) Twin network structure target tracking method fusing target re-identification
CN106651915A (en) Target tracking method of multi-scale expression based on convolutional neural network
CN112861970B (en) Fine-grained image classification method based on feature fusion
CN103839033A (en) Face identification method based on fuzzy rule
CN106682681A (en) Recognition algorithm automatic improvement method based on relevance feedback

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
EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20200515

Assignee: Xiamen yunzhixin Intelligent Technology Co.,Ltd.

Assignor: HUAQIAO University

Contract record no.: X2024990000310

Denomination of invention: A Pedestrian Gender Recognition Method Based on Perspective Adaptive Feature Learning

Granted publication date: 20230331

License type: Common License

Record date: 20240627