CN110059577A - Pedestrian's attribute information extracting method and device - Google Patents
Pedestrian's attribute information extracting method and device Download PDFInfo
- Publication number
- CN110059577A CN110059577A CN201910232030.8A CN201910232030A CN110059577A CN 110059577 A CN110059577 A CN 110059577A CN 201910232030 A CN201910232030 A CN 201910232030A CN 110059577 A CN110059577 A CN 110059577A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- feature
- attributive character
- preliminary
- sample
- 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
Links
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides pedestrian's attribute information extracting method and devices.This method is executed using convolutional neural networks.Convolutional neural networks include preliminary Feature Selection Model, pedestrian divides Feature Selection Model, pedestrian's attributive character extracts model and full articulamentum.This method comprises: preliminary characteristic extraction step, pedestrian image is input in preliminary Feature Selection Model, the preliminary feature of pedestrian image is obtained；Divide characteristic extraction step, preliminary feature is input to pedestrian and is divided in Feature Selection Model, extracts pedestrian and divide feature；Preliminary feature is input to pedestrian's attributive character and extracted in model, extracts pedestrian's attributive character by attributive character extraction step；Pedestrian is divided feature and merged with pedestrian's attributive character, obtains fusion feature by Fusion Features step；Fusion feature is input in full articulamentum by attribute information prediction steps, obtains pedestrian's attribute information of prediction.The combination for dividing feature and pedestrian's attributive character by pedestrian, improves the accuracy rate of pedestrian's attribute information.
Description
Technical field
This invention relates generally to field of artificial intelligence, relate more specifically to pedestrian's attribute information extracting method and dress
It sets, electronic equipment and computer readable storage medium.
Background technique
In many applications of video structural, the analysis of pedestrian is most important, and the identification especially for people exists
The various fields such as security protection, video frequency searching play central role.
Pedestrian's attribute is important a part in video structural, and the accuracy rate of pedestrian's attribute is right in practical application scene
There is vital effect in promoting security protection working efficiency etc..
Summary of the invention
In order to solve the above-mentioned problems in the prior art, the embodiment of the invention provides pedestrian's attribute information extraction sides
Method and device, electronic equipment and computer readable storage medium.
In a first aspect, the embodiment of the present invention provides a kind of pedestrian's attribute information extracting method, wherein this method utilizes convolution
Neural network executes, wherein convolutional neural networks include preliminary Feature Selection Model, pedestrian segmentation Feature Selection Model, pedestrian
Attributive character extracts model and full articulamentum, this method comprises: preliminary characteristic extraction step, is input to preliminary spy for pedestrian image
Sign is extracted in model, and the preliminary feature of pedestrian image is obtained；Divide characteristic extraction step, preliminary feature is input to pedestrian's segmentation
In Feature Selection Model, extracts pedestrian and divide feature；Preliminary feature is input to pedestrian's attributive character by attributive character extraction step
It extracts in model, extracts pedestrian's attributive character；Pedestrian is divided feature and melted with pedestrian's attributive character by Fusion Features step
It closes, obtains fusion feature；Fusion feature is input in full articulamentum by attribute information prediction steps, and the pedestrian for obtaining prediction belongs to
Property information.
In one example, it includes multilayer convolution and multilayer deconvolution neural network that pedestrian, which divides Feature Selection Model,.
In one example, Fusion Features step include: by pedestrian divide feature and pedestrian's attributive character execute bit-wise addition and/
Or step-by-step multiplication operation, obtain fusion feature.
In one example, Fusion Features step includes: and divides feature according to pedestrian to obtain pedestrian's division position information；Utilize row
Part pedestrian's attributive character in people's division position information mask pedestrian's attributive character, the pedestrian's attribute for obtaining area-of-interest are special
Sign, to obtain fusion feature.
In one example, it includes that pedestrian divides semantic information that pedestrian, which divides feature, and pedestrian's attributive character includes pedestrian's attribute language
Adopted information.Also, Fusion Features step includes: the semanteme divided between semantic information and pedestrian's attribute semantemes information according to pedestrian
Relevance obtains fusion feature.
In one example, convolutional neural networks are through the following steps that being trained and obtaining: preliminary feature extraction training
Sample pedestrian image is input in preliminary Feature Selection Model by step, obtains the preliminary feature of sample of sample pedestrian image；Point
Feature extraction training step is cut, the preliminary feature of sample is input to pedestrian and is divided in Feature Selection Model, sample pedestrian point is extracted
Cut feature；Attributive character extracts training step, and the preliminary feature of sample is input to pedestrian's attributive character and is extracted in model, sample is extracted
The humanized feature of current row；Sample pedestrian is divided feature and merged with sample pedestrian's attributive character by Fusion Features training step,
Obtain samples fusion feature；Attribute information predicts training step, and samples fusion feature is input in full articulamentum, is predicted
Sample pedestrian's attribute information；Loss function calculates step, divides feature, sample pedestrian attributive character and sample according to sample pedestrian
The humanized information of current row calculates the whole loss function of convolutional neural networks；Loss function feedback step, by whole loss function
Feed back convolutional neural networks；Parameter tuning step adjusts the parameter of convolutional neural networks, Zhi Daojuan according to whole loss function
Product neural network convergence.
In one example, it includes: to divide feature calculation segmentation loss function according to sample pedestrian that loss function, which calculates step,；Root
According to sample pedestrian's attributive character computation attribute characteristic loss function；According to sample pedestrian's attribute information computation attribute information loss letter
Number；Summation is weighted to segmentation loss function, attributive character loss function and attribute information loss function, obtains whole loss
Function.
Second aspect, the embodiment of the present invention provide a kind of pedestrian's attribute information extraction element, wherein the device utilizes convolution
Neural fusion, wherein convolutional neural networks include preliminary Feature Selection Model, pedestrian segmentation Feature Selection Model, pedestrian
Attributive character extracts model and full articulamentum, the device include: preliminary characteristic extracting module, is configured to input pedestrian image
Into preliminary Feature Selection Model, the preliminary feature of pedestrian image is obtained；Divide characteristic extracting module, is configured to preliminary spy
Sign is input to pedestrian and divides in Feature Selection Model, extracts pedestrian and divides feature；Attributive character extraction module, being configured to will be first
Step feature is input to pedestrian's attributive character and extracts in model, extracts pedestrian's attributive character；Fusion Features module is configured to go
People's segmentation feature is merged with pedestrian's attributive character, obtains fusion feature；Attribute information prediction module is configured to merge
Feature is input in full articulamentum, obtains pedestrian's attribute information of prediction.
The third aspect, the embodiment of the present invention provide a kind of electronic equipment, and electronic equipment includes: memory, refer to for storing
It enables；And processor, the instruction execution above method for calling memory to store.
Fourth aspect, the embodiment of the present invention provide a kind of computer readable storage medium, and computer readable storage medium is deposited
Computer executable instructions are contained, computer executable instructions when executed by the processor, execute the above method.
It pedestrian's attribute information extracting method provided in an embodiment of the present invention and device, electronic equipment and computer-readable deposits
Storage media greatly improves the accuracy rate of pedestrian's attribute information by the combination of pedestrian's segmentation information and pedestrian's attribute information.
Detailed description of the invention
The following detailed description is read with reference to the accompanying drawings, above-mentioned and other purposes, the feature of embodiment of the present invention
It will become prone to understand with advantage.In the accompanying drawings, several implementations of the invention are shown by way of example rather than limitation
Mode, in which:
Fig. 1 shows the flow chart of pedestrian's attribute information extracting method of embodiment according to the present invention；
Fig. 2 shows the block diagrams of pedestrian's attribute information extraction element of embodiment according to the present invention；
Fig. 3 shows the block diagram of the electronic equipment of embodiment according to the present invention.
In the accompanying drawings, identical or corresponding label indicates identical or corresponding part.
Specific embodiment
The principle and spirit of the invention are described below with reference to several illustrative embodiments.It should be appreciated that providing this
A little embodiments are used for the purpose of making those skilled in the art can better understand that realizing the present invention in turn, and be not with any
Mode limits the scope of the invention.
As shown in Figure 1, an embodiment of the invention provides a kind of pedestrian's attribute information extracting method 100.Method
100 are executed using convolutional neural networks.Convolutional neural networks are current image recognition, common a kind of network in video analysis,
It is made of several convolution units (i.e. convolution kernel), each convolution unit extracts different features.Convolutional neural networks can wrap
Include preliminary Feature Selection Model, pedestrian divides Feature Selection Model, pedestrian's attributive character extracts model and full articulamentum.Method
100 include step S101-S105.
Step S101 is preliminary characteristic extraction step, and pedestrian image is input in preliminary Feature Selection Model, is gone
The preliminary feature of people's image.
It is, for example, possible to use classical network structures as preliminary Feature Selection Model.Only as an example, classical network
Structure such as GoogleNet, VGG, ResNet etc..In some embodiments, it for an image, is first entered into just
It walks in Feature Selection Model, the parameter trained basic model initialization of the preliminary Feature Selection Model.
In some embodiments, preliminary Feature Selection Model may include one or more layers convolutional neural networks.
Step S102 is segmentation characteristic extraction step, and preliminary feature is input to pedestrian and is divided in Feature Selection Model, is mentioned
Pedestrian is taken to divide feature.
In some embodiments, extract pedestrian divide feature may include by the image information of one or more pedestrians from
It is extracted in background image, the image of a part of pedestrian and other pedestrians is split or is partitioned into the partial zones of pedestrian
The image information in domain.The head of the regional area of pedestrian such as pedestrian, the upper part of the body, the lower part of the body etc..
Step S103 is attributive character extraction step, and preliminary feature is input to pedestrian's attributive character and is extracted in model, is mentioned
Take pedestrian's attributive character.
In some embodiments, pedestrian's attributive character may include the attributive character of all pedestrians in image, part row
The attributive character of people or the attributive character of specific one pedestrian.Only as an example, pedestrian's attributive character may include: property
Not, the age, clothing style, clothing length, clothing color, whether be branded as, whether knapsack, the style of packet, hair length, whether
Cycle etc..Above-mentioned pedestrian's attributive character is not exhaustive, can also include other any pedestrian's attributive character.
Although showing step S103 in Fig. 1 to execute after step s 102, it is however noted that, step S102 and
The sequence of S103 is without being limited thereto.Alternatively, step S102 can be executed after step s 103.As another reality
Mode is applied, step S102 and step S103 may be performed simultaneously.The present invention is in this regard with no restrictions.
Step S104 is Fusion Features step, and pedestrian is divided feature and is merged with pedestrian's attributive character, is merged
Feature.
In some embodiments, fusion feature may include pedestrian local attribute feature, and pedestrian local attribute feature can
To be to be also possible to the local features relative to a pedestrian relative to single pedestrian's feature in multiple pedestrians.
Step S105 is attribute information prediction steps, and fusion feature is input in full articulamentum, obtains the pedestrian of prediction
Attribute information.
Pedestrian's attribute information extracting method 100 that embodiments of the present invention provide is by by pedestrian's segmentation information and pedestrian
Attribute information combines, and improves the precision of pedestrian's attribute information using relationship between the two.
As an embodiment of the invention, it may include multilayer convolution and multilayer that pedestrian, which divides Feature Selection Model,
Deconvolution neural network.
As an embodiment of the invention, Fusion Features step S104 may include: that pedestrian is divided feature and row
Humanized feature executes bit-wise addition and/or step-by-step multiplication operation, obtains fusion feature.
As an embodiment of the invention, Fusion Features step S104 may include: to divide feature according to pedestrian to obtain
Obtain pedestrian's division position information；Using part pedestrian's attributive character in pedestrian's division position information mask pedestrian's attributive character,
Pedestrian's attributive character of area-of-interest is obtained, to obtain fusion feature.
For example, if the relevant attribute of analysis cap, so that it may the cap region that pedestrian is divided is analyzed, and incite somebody to action
Other regional coverages such as upper body lower part of the body are lived；If the relevant attribute of analysis bag, so that it may which the packet region that pedestrian is divided carries out
Analysis obtains local message, other regional coverages is lived, the interference in other regions is avoided.
By that in conjunction with pedestrian's attributive character, next processing step can be made to close using pedestrian's segmentation feature as mask
The interested part in pedestrian's attributive character is infused, and other uninterested part overlaids are fallen, avoids the dry of other parts
It disturbs, improves the accuracy of attribute information.
As an embodiment of the invention, it may include that pedestrian divides semantic information that pedestrian, which divides feature, and pedestrian belongs to
Property feature may include pedestrian's attribute semantemes information.Fusion Features step S104 may include: to divide semantic information according to pedestrian
Semantic relevance between pedestrian's attribute semantemes information obtains fusion feature.
Only as an example, pedestrian divides the semanteme that semantic information may include each physical feeling of pedestrian, such as head,
Pedestrian's attribute semantemes information may include the semantic information whether pedestrian is branded as, by the semantic relevance on head and cap,
Segmentation feature can be merged with attributive character.
It is associated by the way that pedestrian is divided feature and pedestrian's attributive character semantically, finds cross-correlation with can be convenient
The pedestrian of connection divides feature and pedestrian's attributive character, to obtain fusion feature using both features.
As an embodiment of the invention, convolutional neural networks can be through the following steps that being trained and obtaining
: sample pedestrian image is input in preliminary Feature Selection Model by preliminary feature extraction training step, obtains sample pedestrian figure
The preliminary feature of the sample of picture；Divide feature extraction training step, the preliminary feature of sample is input to pedestrian and divides feature extraction mould
In type, extracts sample pedestrian and divide feature；Attributive character extracts training step, and it is special that the preliminary feature of sample is input to pedestrian's attribute
Sign is extracted in model, and sample pedestrian attributive character is extracted；Sample pedestrian is divided feature and sample row by Fusion Features training step
Humanized feature is merged, and samples fusion feature is obtained；Attribute information predicts training step, and samples fusion feature is input to
In full articulamentum, sample pedestrian's attribute information of prediction is obtained；Loss function calculate step, according to sample pedestrian divide feature,
Sample pedestrian attributive character and sample pedestrian's attribute information calculate the whole loss function of convolutional neural networks；Loss function feedback
Step, by whole loss function feedback to convolutional neural networks；Parameter tuning step adjusts convolution mind according to whole loss function
Parameter through network, until convolutional neural networks are restrained.
In some embodiments, loss function can be used to estimate the predicted value of model and the inconsistent journey of true value
Degree.It can be a non-negative real-valued function.
For example, the pedestrian that can be obtained according to current convolutional neural networks is divided feature, pedestrian's attributive character and pedestrian and is belonged to
Property information predicted value divide feature with true pedestrian respectively, pedestrian's attributive character and pedestrian's attribute information are compared, obtain
To whole loss function.
In some embodiments, loss function may include hinge loss, mutual entropy loss, Squared Error Loss, figure penalties
Etc..
In some embodiments, it may include: to divide feature calculation point according to sample pedestrian that loss function, which calculates step,
Cut loss function；According to sample pedestrian's attributive character computation attribute characteristic loss function；It is calculated according to sample pedestrian's attribute information
Attribute information loss function；Segmentation loss function, attributive character loss function and attribute information loss function are weighted and are asked
With acquisition whole loss function.
In some embodiments, weighted sum can be to each individual event loss function multiplied by respective weight ratio phase again
Add, obtains population value loss function as a whole.It is alternatively possible to be added after weighting again divided by the loss function for participating in weighting
Item number obtains weighted average loss function as a whole.It is of course also possible to use other any weighting schemes, the present invention exist
In this respect with no restrictions.
In some embodiments, according to whole loss function, the parameter in convolutional neural networks is adjusted, it can be with
So that convolutional neural networks tend to restrain.For example, can determine the convolutional Neural when the value of loss function is lower than a certain threshold value
Network convergence.
In some embodiments, it is described above by whole loss function adjust convolutional neural networks parameter come pair
Network is trained can realize by adjusting the weight of each individual event loss function.
As shown in Fig. 2, an embodiment of the invention provides a kind of pedestrian's attribute information extraction element 200.Device
200 are realized using convolutional neural networks.Wherein, convolutional neural networks include preliminary Feature Selection Model, pedestrian divide feature mention
Modulus type, pedestrian's attributive character extract model and full articulamentum.Device 200 includes module 201-205.
Preliminary characteristic extracting module 201 may be configured to be input to pedestrian image in preliminary Feature Selection Model, obtain
Obtain the preliminary feature of pedestrian image.
Segmentation characteristic extracting module 202 may be configured to for preliminary feature to be input to pedestrian's segmentation Feature Selection Model
In, it extracts pedestrian and divides feature.
Attributive character extraction module 203 may be configured to for preliminary feature to be input to pedestrian's attributive character extraction model
In, extract pedestrian's attributive character.
Fusion Features module 204 may be configured to merge pedestrian's segmentation feature with pedestrian's attributive character, obtain
Fusion feature.
Attribute information prediction module 205 may be configured to be input to fusion feature in full articulamentum, obtain prediction
Pedestrian's attribute information.
As an embodiment of the invention, it may include multilayer convolution and multilayer that pedestrian, which divides Feature Selection Model,
Deconvolution neural network.
As an embodiment of the invention, Fusion Features module 204 can also be configured to: pedestrian is divided feature
Bit-wise addition and/or step-by-step multiplication operation are executed with pedestrian's attributive character, obtains fusion feature.
As an embodiment of the invention, Fusion Features module 204 can also be configured to: be divided according to pedestrian special
Sign obtains pedestrian's division position information；It is special using part pedestrian's attribute in pedestrian's division position information mask pedestrian's attributive character
Sign, obtains pedestrian's attributive character of area-of-interest, to obtain fusion feature.
As an embodiment of the invention, it may include that pedestrian divides semantic information that pedestrian, which divides feature, and pedestrian belongs to
Property feature includes pedestrian's attribute semantemes information.Also, Fusion Features module 204 can also be configured to: divide language according to pedestrian
Semantic relevance between adopted information and pedestrian's attribute semantemes information obtains fusion feature.
As an embodiment of the invention, convolutional neural networks can be to be obtained and being trained with lower module
: preliminary feature extraction training module is configured to for sample pedestrian image being input in preliminary Feature Selection Model, obtains sample
The preliminary feature of the sample of this pedestrian image；Divide feature extraction training module, is configured to the preliminary feature of sample being input to row
People is divided in Feature Selection Model, extracts sample pedestrian and divides feature；Attributive character extracts training module, is configured to sample
Preliminary feature is input to pedestrian's attributive character and extracts in model, extracts sample pedestrian attributive character；Fusion Features training module, matches
It sets and is merged for sample pedestrian to be divided feature with sample pedestrian's attributive character, obtain samples fusion feature；Attribute information
It predicts training module, is configured to for samples fusion feature being input in full articulamentum, obtain sample pedestrian's attribute letter of prediction
Breath；Loss function computing module is configured to divide feature, sample pedestrian attributive character and sample pedestrian category according to sample pedestrian
Property information calculate convolutional neural networks whole loss function；Loss function feedback module is configured to whole loss function
Feed back convolutional neural networks；Parameter adjustment module is configured to adjust the ginseng of convolutional neural networks according to whole loss function
Number, until convolutional neural networks are restrained.
As an embodiment of the invention, loss function computing module can also be configured to: according to sample pedestrian
Divide feature calculation and divides loss function；According to sample pedestrian's attributive character computation attribute characteristic loss function；According to sample row
Humanized information computation attribute information loss function；To segmentation loss function, attributive character loss function and attribute information loss
Function is weighted summation, obtains whole loss function.
The function that modules in device are realized is corresponding with the step in method as described above, specific implementation
The description for method and step above is referred to technical effect, details are not described herein.
As shown in figure 3, an embodiment of the invention provides a kind of electronic equipment 300.Wherein, the electronic equipment
300 include memory 301, processor 302, input/output (Input/Output, I/O) interface 303.Wherein, memory 301,
For storing instruction.Processor 302, pedestrian's attribute of the instruction execution embodiment of the present invention for calling memory 301 to store
Information extracting method.Wherein, processor 302 is connect with memory 301, I/O interface 303 respectively, such as can pass through bus system
And/or bindiny mechanism's (not shown) of other forms is attached.Memory 301 can be used for storing program and data, including this
Pedestrian's attribute information extraction procedure involved in inventive embodiments, processor 302 are stored in the program of memory 301 by operation
Thereby executing the various function application and data processing of electronic equipment 300.
Processor 302 can use digital signal processor (Digital Signal in the embodiment of the present invention
Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), programmable patrol
At least one of volume array (Programmable Logic Array, PLA) example, in hardware realizes, the processor 302
It can be central processing unit (Central Processing Unit, CPU) or there is data-handling capacity and/or instruction
The combination of one or more of the processing unit of other forms of executive capability.
Memory 301 in the embodiment of the present invention may include one or more computer program products, the computer
Program product may include various forms of computer readable storage mediums, such as volatile memory and/or non-volatile deposit
Reservoir.The volatile memory for example may include random access memory (Random Access Memory, RAM) and/
Or cache memory (Cache) etc..The nonvolatile memory for example may include read-only memory (Read-Only
Memory, ROM), flash memory (Flash Memory), hard disk (Hard Disk Drive, HDD) or solid state hard disk
(Solid-State Drive, SSD) etc..
In the embodiment of the present invention, I/O interface 303 can be used for receiving input instruction (such as number or character information, and
Generate key signals input related with the user setting of electronic equipment 300 and function control etc.), it can also be output to the outside various
Information (for example, image or sound etc.).In the embodiment of the present invention I/O interface 303 may include physical keyboard, function button (such as
Volume control button, switch key etc.), mouse, operating stick, trace ball, microphone, one in loudspeaker and touch panel etc.
It is a or multiple.
An embodiment of the invention provides a kind of computer readable storage medium, the computer readable storage medium
Computer executable instructions are stored with, computer executable instructions when executed by the processor, execute any side described above
Method.
Although description operation in a particular order in the accompanying drawings should not be construed as requiring specific shown in
Sequence or serial order operate to execute these operations, or shown in requirement execution whole to obtain desired result.?
In specific environment, multitask and parallel processing be may be advantageous.
Methods and apparatus of the present invention can be completed using standard programming technology, using rule-based logic or its
His logic realizes various method and steps.It should also be noted that herein and the terms used in the claims " device "
" module " is intended to include using the realization of a line or multirow software code and/or hardware realization and/or for receiving input
Equipment.
One or more combined individually or with other equipment can be used in any step, operation or program described herein
A hardware or software module are executed or are realized.In one embodiment, software module use includes comprising computer program
The computer program product of the computer-readable medium of code is realized, can be executed by computer processor any for executing
Or whole described step, operation or programs.
For the purpose of example and description, the preceding description that the present invention is implemented is had been presented for.Preceding description is not poor
Also not the really wanting of act property limits the invention to exact form disclosed, according to the above instruction there is likely to be various modifications and
Modification, or various changes and modifications may be obtained from the practice of the present invention.Select and describe these embodiments and be in order to
Illustrate the principle of the present invention and its practical application, so that those skilled in the art can be to be suitable for the special-purpose conceived
Come in a variety of embodiments with various modifications and utilize the present invention.
Claims (10)
1. a kind of pedestrian's attribute information extracting method, wherein the method is executed using convolutional neural networks, wherein the volume
Product neural network includes preliminary Feature Selection Model, pedestrian divides Feature Selection Model, pedestrian's attributive character extracts model and complete
Articulamentum, which comprises
Pedestrian image is input in the preliminary Feature Selection Model, obtains the pedestrian image by preliminary characteristic extraction step
Preliminary feature；
Divide characteristic extraction step, the preliminary feature is input to the pedestrian and is divided in Feature Selection Model, pedestrian is extracted
Divide feature；
The preliminary feature is input to pedestrian's attributive character and extracted in model, extracts pedestrian by attributive character extraction step
Attributive character；
The pedestrian is divided feature and merged with pedestrian's attributive character, obtains fusion feature by Fusion Features step；
The fusion feature is input in the full articulamentum by attribute information prediction steps, obtains pedestrian's attribute letter of prediction
Breath.
2. according to the method described in claim 1, wherein, it includes multilayer convolution and more that the pedestrian, which divides Feature Selection Model,
Layer deconvolution neural network.
3. according to the method described in claim 1, wherein, the Fusion Features step includes:
The pedestrian is divided into feature and pedestrian's attributive character executes bit-wise addition and/or step-by-step multiplication operation, obtains institute
State fusion feature.
4. according to the method described in claim 1, wherein, the Fusion Features step includes:
Divide feature according to the pedestrian and obtains pedestrian's division position information；
Using part pedestrian's attributive character in pedestrian's attributive character described in pedestrian's division position information mask, obtain feeling emerging
Pedestrian's attributive character in interesting region, to obtain the fusion feature.
5. according to the method described in claim 1, wherein, it includes that pedestrian divides semantic information that the pedestrian, which divides feature, described
Pedestrian's attributive character includes pedestrian's attribute semantemes information, also,
The Fusion Features step includes:
Divide the semantic relevance between semantic information and pedestrian's attribute semantemes information according to the pedestrian, melts described in acquisition
Close feature.
6. according to the method described in claim 1, wherein, the convolutional neural networks are through the following steps that being trained and obtaining
:
Preliminary feature extraction training step, sample pedestrian image is input in the preliminary Feature Selection Model, described in acquisition
The preliminary feature of the sample of sample pedestrian image；
Divide feature extraction training step, the preliminary feature of the sample be input to the pedestrian and is divided in Feature Selection Model,
It extracts sample pedestrian and divides feature；
Attributive character extracts training step, and the preliminary feature of the sample is input to pedestrian's attributive character and is extracted in model,
Extract sample pedestrian attributive character；
The sample pedestrian is divided feature and merged with the sample pedestrian attributive character, obtained by Fusion Features training step
Obtain samples fusion feature；
Attribute information predicts training step, and the samples fusion feature is input in the full articulamentum, the sample of prediction is obtained
The humanized information of current row；
Loss function calculates step, divides feature, the sample pedestrian attributive character and the sample according to the sample pedestrian
Pedestrian's attribute information calculates the whole loss function of the convolutional neural networks；
Loss function feedback step, by whole loss function feedback to the convolutional neural networks；
Parameter tuning step adjusts the parameter of the convolutional neural networks according to the whole loss function, until the convolution
Neural network convergence.
7. according to the method described in claim 6, wherein, the loss function calculates step and includes:
Divide feature calculation according to the sample pedestrian and divides loss function；
According to the sample pedestrian attributive character computation attribute characteristic loss function；
According to the sample pedestrian attribute information computation attribute information loss function；
The segmentation loss function, the attributive character loss function and the attribute information loss function are weighted and are asked
With obtain the whole loss function.
8. a kind of pedestrian's attribute information extraction element, wherein described device is realized using convolutional neural networks, wherein the volume
Product neural network includes preliminary Feature Selection Model, pedestrian divides Feature Selection Model, pedestrian's attributive character extracts model and complete
Articulamentum, described device include:
Preliminary characteristic extracting module is configured to be input to pedestrian image in the preliminary Feature Selection Model, described in acquisition
The preliminary feature of pedestrian image；
Divide characteristic extracting module, be configured to be input to the preliminary feature in pedestrian's segmentation Feature Selection Model,
It extracts pedestrian and divides feature；
Attributive character extraction module is configured to be input to the preliminary feature in pedestrian's attributive character extraction model,
Extract pedestrian's attributive character；
Fusion Features module is configured to merge pedestrian segmentation feature with pedestrian's attributive character, be melted
Close feature；
Attribute information prediction module is configured to for the fusion feature being input in the full articulamentum, obtains the row of prediction
Humanized information.
9. a kind of electronic equipment, the electronic equipment include:
Memory, for storing instruction；And
Processor, for calling the described instruction of the memory storage to execute such as side of any of claims 1-7
Method.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer executable instructions, institute
It states computer executable instructions when executed by the processor, executes such as method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910232030.8A CN110059577B (en) | 2019-03-26 | 2019-03-26 | Pedestrian attribute information extraction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910232030.8A CN110059577B (en) | 2019-03-26 | 2019-03-26 | Pedestrian attribute information extraction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110059577A true CN110059577A (en) | 2019-07-26 |
| CN110059577B CN110059577B (en) | 2022-02-18 |
Family
ID=67316332
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910232030.8A Active CN110059577B (en) | 2019-03-26 | 2019-03-26 | Pedestrian attribute information extraction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110059577B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110969087A (en) * | 2019-10-31 | 2020-04-07 | 浙江省北大信息技术高等研究院 | Gait recognition method and system |
| CN111046950A (en) * | 2019-12-11 | 2020-04-21 | 北京迈格威科技有限公司 | Image processing method and device, storage medium and electronic device |
| CN112232173A (en) * | 2020-10-12 | 2021-01-15 | 济南博观智能科技有限公司 | Pedestrian attribute identification method, deep learning model, equipment and medium |
| WO2021174941A1 (en) * | 2020-03-03 | 2021-09-10 | 平安科技（深圳）有限公司 | Physical attribute recognition method, system, computer device, and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107784282A (en) * | 2017-10-24 | 2018-03-09 | 北京旷视科技有限公司 | The recognition methods of object properties, apparatus and system |
| CN108229267A (en) * | 2016-12-29 | 2018-06-29 | 北京市商汤科技开发有限公司 | Object properties detection, neural metwork training, method for detecting area and device |
| CN108875536A (en) * | 2018-02-06 | 2018-11-23 | 北京迈格威科技有限公司 | Pedestrian's analysis method, device, system and storage medium |
| CN108921054A (en) * | 2018-06-15 | 2018-11-30 | 华中科技大学 | A kind of more attribute recognition approaches of pedestrian based on semantic segmentation |
-
2019
- 2019-03-26 CN CN201910232030.8A patent/CN110059577B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108229267A (en) * | 2016-12-29 | 2018-06-29 | 北京市商汤科技开发有限公司 | Object properties detection, neural metwork training, method for detecting area and device |
| CN107784282A (en) * | 2017-10-24 | 2018-03-09 | 北京旷视科技有限公司 | The recognition methods of object properties, apparatus and system |
| CN108875536A (en) * | 2018-02-06 | 2018-11-23 | 北京迈格威科技有限公司 | Pedestrian's analysis method, device, system and storage medium |
| CN108921054A (en) * | 2018-06-15 | 2018-11-30 | 华中科技大学 | A kind of more attribute recognition approaches of pedestrian based on semantic segmentation |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110969087A (en) * | 2019-10-31 | 2020-04-07 | 浙江省北大信息技术高等研究院 | Gait recognition method and system |
| CN110969087B (en) * | 2019-10-31 | 2023-11-21 | 杭州未名信科科技有限公司 | Gait recognition method and system |
| CN111046950A (en) * | 2019-12-11 | 2020-04-21 | 北京迈格威科技有限公司 | Image processing method and device, storage medium and electronic device |
| CN111046950B (en) * | 2019-12-11 | 2023-09-22 | 北京迈格威科技有限公司 | Image processing method and device, storage medium and electronic device |
| WO2021174941A1 (en) * | 2020-03-03 | 2021-09-10 | 平安科技（深圳）有限公司 | Physical attribute recognition method, system, computer device, and storage medium |
| CN112232173A (en) * | 2020-10-12 | 2021-01-15 | 济南博观智能科技有限公司 | Pedestrian attribute identification method, deep learning model, equipment and medium |
| CN112232173B (en) * | 2020-10-12 | 2023-04-07 | 济南博观智能科技有限公司 | Pedestrian attribute identification method, deep learning model, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110059577B (en) | 2022-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ozcan et al. | Transfer learning-based convolutional neural networks with heuristic optimization for hand gesture recognition | |
| Jung et al. | Discriminative feature learning for unsupervised video summarization | |
| CN110059577A (en) | Pedestrian's attribute information extracting method and device | |
| US10958748B2 (en) | Resource push method and apparatus | |
| Fauvel et al. | XEM: An explainable-by-design ensemble method for multivariate time series classification | |
| CN103530540B (en) | User identity attribute detection method based on man-machine interaction behavior characteristics | |
| US11741711B2 (en) | Video classification method and server | |
| Yan et al. | Hierarchical multi-scale attention networks for action recognition | |
| CN110263733B (en) | Image processing method, nomination evaluation method and related device | |
| Kempe et al. | Detecting tactical patterns in basketball: comparison of merge self-organising maps and dynamic controlled neural networks | |
| CN110599492A (en) | Training method and device for image segmentation model, electronic equipment and storage medium | |
| CN106529511A (en) | Image structuring method and device | |
| Liu et al. | Hybrid design for sports data visualization using AI and big data analytics | |
| JP7292657B2 (en) | DATA PROCESSING METHOD, DATA PROCESSING DEVICE, COMPUTER PROGRAM AND ELECTRONIC DEVICE | |
| Hii et al. | Multigap: Multi-pooled inception network with text augmentation for aesthetic prediction of photographs | |
| CN110147882A (en) | Training method, crowd's method of diffusion, device and the equipment of neural network model | |
| Zhang et al. | Video event classification based on two-stage neural network | |
| CN111400615B (en) | Resource recommendation method, device, equipment and storage medium | |
| Yudistira et al. | Correlation net: Spatiotemporal multimodal deep learning for action recognition | |
| CN113742580B (en) | Recall method and device for target type data, electronic equipment and storage medium | |
| Werling et al. | On-the-job learning with bayesian decision theory | |
| Amirgholipour et al. | PDANet: Pyramid density-aware attention based network for accurate crowd counting | |
| Han et al. | Robust human action recognition using global spatial-temporal attention for human skeleton data | |
| Sun et al. | Triplet attention multiple spacetime-semantic graph convolutional network for skeleton-based action recognition | |
| Peng et al. | Motion boundary emphasised optical flow method for human action recognition |
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 |