CN108345837A - A kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation - Google Patents
A kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation Download PDFInfo
- Publication number
- CN108345837A CN108345837A CN201810046080.2A CN201810046080A CN108345837A CN 108345837 A CN108345837 A CN 108345837A CN 201810046080 A CN201810046080 A CN 201810046080A CN 108345837 A CN108345837 A CN 108345837A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- image
- feature representation
- feature
- alignmentization
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
- G06F18/2193—Validation; Performance evaluation; Active pattern learning techniques based on specific statistical tests
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Abstract
The invention discloses a kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation.Method is specially：The image data set for training pedestrian to identify again is obtained, and defines algorithm target；In the case where being marked only by pedestrian's ID number, automatic study meets the human body key area of semantic content；The corresponding depth characteristic of depth convolutional neural networks model extraction is used according to these key areas；The feature of respective extracted region is merged as final human region alignmentization feature representation；Similarity between feature representation between calculating pedestrian image, to achieve the purpose that the different pedestrians of identification；It is finally searched in application in actual pedestrian, to extracting feature representation under image line in database and storing, carries out similarity calculation online when pedestrian searches for.The present invention identifies the features such as problem can reach effect robust, accuracy is high, the speed of service is fast again for pedestrian's feature representation study in image or video data, for the pedestrian under monitoring scene.
Description
Technical field
The invention belongs to computer vision fields, particularly a kind of to be learnt based on human region alignmentization feature representation
Pedestrian's recognition methods again.
Background technology
Identification mission is that the pedestrian that different cameras capture is identified and is searched under monitoring scene to pedestrian again.
Pedestrian is not needed for the application of recognition of face, under the scene to cooperate on one's own initiative, also therefore bringing face information can not
With the variation of, human body attitude greatly, camera angle be not fixed etc. it is difficult with challenge.The algorithm that pedestrian identifies again is intended to scheme some pedestrian
As matching with the pedestrian image of other cameras capture, judge whether to be the same pedestrian.Firstly the need of to given two
Pedestrian image carry out feature representation, then carry out characteristic matching again, finally calculate two images similarity come complete identification or
Search.Having many methods at present is learnt to matching, is generally embedded in study matching in the middle of deep neural network and is closed
System.But it is still required in practical application using quickly matching, for example is wished in search process and the institute in database
There is the feature that pedestrian image prestores quickly to be compared, if carry out feature that will be online to each feature in database
Match, needs largely to calculate the time.Therefore the feature representation of study robust is a target that can reach quickly and efficiently,
Can be by the calculating and storage under feature representation into line in practical application, on-line search when, carries out simple matched meter
It calculates, the calculation amount of On-line matching is preferably minimized, the present invention also will be in terms of the study that focus on characteristics of human body's expression.
People generally can ignore background and then compare when whether judge two pedestrian images is the same pedestrian
Whether the different parts of pedestrian are similar, there is the problem of being mutually matched different parts between an image here.At present very
More algorithms are also that image is carried out feature representation, then the feature of two images are compared according to position similar to calculate
Degree.These matched strategies are substantially preset by people, it is assumed that it is suitble to pedestrian to identify problem again.Much it is based on position relationship
Division methods can not cope with well human posture variation etc. challenges.
Invention content
In order to solve the problems in the existing technology, characteristic matching is implicitly embedded into characteristics of human body's expression by the present invention
In study, the human region alignment feature learning method based on deep neural network model is obtained.People are judging two pedestrians
When whether image is the same pedestrian, the attention mechanism of human vision is conformed generally to, it will usually ignore background and then compare
It is whether similar in the different parts of pedestrian, it is asked what different parts were mutually matched between an image specific to there is in algorithm
Topic.
Algorithm proposed by the present invention is to pedestrian image automatic mining and detects human body key area and pattern, to a certain extent
The attention model of human vision is simulated, the pattern match mechanism that view-based access control model inspires is realized, so as to human body attitude
There is certain robustness with camera angle variation.It is desirable that in the learning process of characteristics of human body's expression, according to semantically
Human region position feature is subjected to alignment, can exclude the interference of background in this way can equally solve feature extraction region
The problem of location matches.This method can successfully manage pedestrian's postural change, camera visual angle be different, between image pedestrian position without
The problems such as method is aligned can also reach effect robust, the features such as accuracy is high, the speed of service is fast in practical applications.
To achieve the above object, the technical scheme is that：
Based on pedestrian's recognition methods again of human region alignmentization feature representation study, in given pedestrian image situation
The problem of lower extraction feature carries out pedestrian's identification or identifies again comprising following steps：
The image data set of S1, acquisition for training pedestrian to identify again, every pictures that image data is concentrated include one
Pedestrian, and each pedestrian has unique ID number markup information；
S2, it is based on pedestrian's ID number markup information, several human bodies pass for meeting semantic content is obtained by learning automatically
Key range；
S3, in pedestrian's search process, to the pedestrian image that different cameras capture, closed according to the human body obtained in S2
Key range, the online lower depth characteristic using each human body key area in depth convolutional neural networks every image of model extraction；
The feature of each human body key area is as human region pair final in the image in every image of S4 and then fusion
Neatization feature representation is simultaneously stored in database；
S5, for the human region alignmentization feature representation in query image, line computation its with each pedestrian in database
Similarity between the human region alignmentization feature representation of image, and pedestrian is returned to according to similarity and identifies search result again,
To achieve the purpose that identify pedestrian.
Preferably, the algorithm object definition of pedestrian's recognition methods again is：Give two pedestrian images, judge it includes
Whether pedestrian is the same pedestrian, that is, judges whether the ID number mark of two pedestrian images is identical.
Preferably, in step S1, described is used to train the image data set that pedestrian identifies again, including under monitoring scene
The pedestrian image that different cameras capture, every image include a pedestrian, which has unique corresponding ID number information,
The image that different cameras capture the same pedestrian possesses identical ID number mark, and the image of different pedestrians possesses different
ID number marks；
Preferably, in step S2, it is based on pedestrian's ID number markup information, meets semantic content by learning to obtain automatically
Several human body key areas, specifically include：
S21, image { I are opened to all N in image data set1,I2,…,INBe combined into according to pedestrian's ID number it is different
Triple { (Ia,Ib,Ic), wherein (Ia,Ib) two images have common identity number, (Ia,Ic) two images have different bodies
Part number；
S22, human region detection training is carried out to the image in triple using convolutional neural networks, until detection obtains
K+1 human body key area so that (Ia,Ib) characteristic similarities of two image corresponding regions is higher than (Ia,Ic) two images pair
It includes 1 region M for only including pedestrian's foreground to answer the characteristic similarity in region, K+1 human body key area0With K sub-district
Domain { M1,M2,…,Mk,…,MK}。
Further, in step S3, each human body in depth convolutional neural networks every image of model extraction is used under online
The specific method of the depth characteristic of key area is：
It is crucial according to each human body learnt to every image zooming-out depth characteristic figure T using full convolutional neural networks
Region MiOperation is weighted to depth characteristic and obtains the feature representation F of key areai, i.e.,：
Fi=T ⊙ Mi
Wherein i=0,1 ..., K；⊙ is the multiplying by element.
Further, in step S4, the specific method for merging the feature representation of each human body key area is：
S41, the feature vector f ' in the region is obtained after being converted the feature representation of each human body key areai, i.e.,：
f′i=transform (Fi)
S42, K+1 feature vector is stitched together as final human region alignmentization feature representation vector f,
I.e.：
F=[f '0,f′1,f′2,…,f′k,…,f′K]
Pedestrian image { the I captured from all cameras1,I2,…,IMIn the human region alignmentization mark sheet that extracts
Up to vector { f1,f2,…,fM, it is stored in database.
Further, the concrete methods of realizing of step S5 is as follows：
S51, for query image IqExtract its human region alignmentization feature representation vector fq；
S52, in database { f1,f2,…,fMWhen scanning for, to the feature representation vector f of each storagem, calculate fq
With fmBetween similarity score sqm, as pedestrian image (Iq,Im) pedestrian's recognition result again；
S53, to all similarity score sqmIt is sorted from big to small, returns to the sequence after sequence, correspond to image
{I1,I2,…,IMRanking results return pedestrian identify search result again.
Further, a pair of of image (Iq,Im) between similarity score sqmCalculating using Euclidean distance inverse, meter
Calculating formula is：
Pedestrian's recognition methods again based on the study of human region alignmentization feature representation of the present invention, compared to existing row
People's recognition methods again, has the advantages that：
First, characteristic matching is implicitly embedded into the study of characteristics of human body's expression pedestrian of the invention by recognition methods again,
In practical applications can be by the calculating and storage under feature representation into line, on-line search when, carries out simple matched meter
It calculates, the calculation amount of On-line matching is preferably minimized.
Secondly, the present invention proposes a human region alignment feature learning method based on deep neural network model,
For this method in the case of not by additional mark, our frame can acquire the human body key area for meeting semantic content automatically
Domain.The attention model of human vision is simulated to a certain extent, realizes the pattern match mechanism that view-based access control model inspires, so as to
It is enough that there is certain robustness to human body attitude and camera angle variation.
Finally, recognition methods proposes that deep learning network removes combined optimization feature representation end to end to pedestrian of the invention again
Learning process and human body key area learning process, to obtain more robust as a result, it is possible to successfully manage pedestrian's posture
The problems such as pedestrian position can not be aligned between variation, camera visual angle difference, image.
Pedestrian's recognition methods again based on the study of human region alignmentization feature representation of the present invention, in video monitoring, figure
As in search, the accuracy and efficiency that pedestrian identifies again can be effectively improved, there is good application value.
Description of the drawings
Fig. 1 is the flow diagram of the present invention；
Fig. 2 is the human body key area that the original image in embodiment is arrived with study；
Fig. 3 is that the pedestrian in embodiment identifies search results map again.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
On the contrary, the present invention covers any replacement done in the spirit and scope of the present invention being defined by the claims, repaiies
Change, equivalent method and scheme.Further, in order to make the public have a better understanding the present invention, below to the thin of the present invention
It is detailed to describe some specific detail sections in section description.Part without these details for a person skilled in the art
Description can also understand the present invention completely.
With reference to figure 1, for pedestrian's recognition methods implementing procedure again of the present invention.Lower mask body is described in detail in the preferable reality of the present invention
It applies in example, the specific steps of pedestrian's recognition methods again based on the study of human region alignmentization feature representation：
First, pedestrian's recognizer again is established：
S1：It is used to train the image data set that pedestrian identifies again described in obtaining, image data concentration includes monitoring scene
The pedestrian image that lower difference camera captures, every image include a pedestrian, which has unique corresponding ID number letter
Breath, the image that different cameras capture the same pedestrian possess identical ID number mark, and the image of different pedestrians possesses not
Same ID number mark；
Defining algorithm target is：Give two pedestrian images, judge it includes pedestrian whether be the same pedestrian.
S2：In the case where being marked only by pedestrian's ID number, several human bodies that automatic study meets semantic content close
Key range specifically includes following sub-step：
S21：To all N in image data set image { I1,I2,…,INBe combined into according to pedestrian's ID number it is different
Triple { (Ia,Ib,Ic), wherein in each triple, (Ia,Ib) two images have common identity number, (Ia,Ic) two figures
As having different identity number.Triple is used to be used as model training data, therefore its quantity is as more as possible.
S22：Human region detection training is carried out to the image in triple using convolutional neural networks, until model meets
Detection obtains K+1 human body key area so that (Ia,Ib) characteristic similarities of two image corresponding regions is higher than (Ia,Ic) two
Open the characteristic similarity of image corresponding region.In this K+1 human body key area, including 1 area for only including pedestrian's foreground
Domain M0With K sub-regions { M1,M2,…,Mk,…,MK}。
After obtaining K+1 human body key area, you can be applied to pedestrian's identification process again.At this point,
S3, in pedestrian's search process, pedestrian image { I that all cameras are captured1,I2,…,IM, it is online under by
A pair of every image carries out depth characteristic extraction.In extraction process, using full convolutional neural networks, to every image zooming-out depth
Characteristic pattern T is directed to each human body key area M successively according to the K+1 human body key area obtained in S2i, to depth characteristic
It is weighted operation and obtains the feature representation F of each key areai, i.e.,：
Fi=T ⊙ Mi
Wherein i=0,1 ..., K；⊙ is the multiplying by element.
The feature of each human body key area is as human region pair final in the image in every image of S4 and then fusion
Neatization feature representation is simultaneously stored in database.Its specific method is：
S41, by the feature representation F of each human body key areaiThe feature vector f ' in the region is obtained after being convertedi,
I.e.：
f′i=transform (Fi)
Wherein transform () transformation includes that average pondization operation (Average Pooling) and one layer of linear transformation are complete
Articulamentum (Fully Connected Layer), obtain final vector f 'i.In the present embodiment, dimension may be configured as
S42, by K+1 feature vector f 'iThe human region alignmentization mark sheet being stitched together as 512 final dimensions
Up to vector f, i.e.,：
F=[f '0,f′1,f′2,…,f′k,…,f′K]
To each pedestrian image ImThe calculating for carrying out above-mentioned human region alignmentization feature representation vector, obtains M people
Body region alignmentization feature representation is to fm.Pedestrian image { the I captured from all cameras under online1,I2,…,IMIn extract
Human region alignmentization feature representation vector { f1,f2,…,fMAfter, it can be stored in database, for subsequently in line computation
It calls.
S5, it is searched in application in actual pedestrian, for the human region alignmentization feature representation in query image, online
Its similarity between the human region alignment feature representation of each pedestrian image in database is calculated, and is returned according to similarity
Reversion people identifies search result again, to achieve the purpose that identify pedestrian.Its concrete methods of realizing is as follows：
S51, for query image IqIts human region alignmentization feature representation vector f is extracted according to the method for S4q；
S52, in database { f1,f2,…,fMWhen scanning for, to each pre-stored feature representation vector
fm, calculate fqWith fmBetween similarity score sqm, as pedestrian image (Iq,Im) pedestrian's recognition result again.The present embodiment
In, a pair of of image (Iq,Im) between similarity score sqmCalculating using simple Euclidean distance inverse, calculation formula
For：
S53, to all similarity score sqmIt is sorted from big to small, returns to the sequence after sequence, correspond to image
{I1,I2,…,IMRanking results return pedestrian identify search result again.
The above method is applied in specific embodiment below, so as to those skilled in the art can better understand that this hair
Bright effect.
Embodiment
The implementation method of the present embodiment is as previously mentioned, no longer elaborate specific step, below only for case data
Show its effect.The present invention is implemented on the data set at two with true value mark, respectively：
Market-1501 data sets：The data set includes 32668 images, there is 1501 pedestrians.
CUHK03 data sets：The data set includes 13164 images, there is 1360 pedestrians.
This example carries out automatic study and the hair of experimental verification human body key area first on Market-1501 data sets
Existing, the key area that obtained K is set as 8 is as shown in Figure 2.
Feature representation is obtained according to these key areas, carries out pedestrian image search, the last eight visualization knot of pedestrian's search
Fruit is as shown in Figure 3.
The result of implementation of CUHK03 data sets is essentially identical, repeats no more.Inquiry accuracy rate amount based on two datasets
The results are shown in table below for change：
| Data set | Preceding 1 accuracy rate | Preceding 5 accuracy rate | Preceding 10 accuracy rate | Average Accuracy |
| Market-1501 | 82.3 | 92.4 | 95.6 | 65.7 |
| CUHK03 | 86.5 | 97.8 | 99.7 | 91.5 |
By above technical scheme, the present invention implements to provide based on depth learning technology a kind of based on human region alignment
Change pedestrian's recognition methods again of feature representation study.The present invention can be on various true monitor videos and image data to wherein
Pedestrian carry out feature representation, to realize the purpose for identifying different pedestrians and carrying out pedestrian image search.The present invention is for scheming
Pedestrian's feature representation study in picture or video data, identifies that problem can reach effect Shandong again for the pedestrian under monitoring scene
Stick, the features such as accuracy is high, the speed of service is fast.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
All any modification, equivalent and improvement etc., should all be included in the protection scope of the present invention made by within refreshing and principle.
Claims (8)
1. a kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation, which is characterized in that including following
Step：
The image data set of S1, acquisition for training pedestrian to identify again, every pictures that image data is concentrated include a pedestrian,
And each pedestrian has unique ID number markup information；
S2, it is based on pedestrian's ID number markup information, several human body key areas for meeting semantic content is obtained by learning automatically
Domain；
S3, in pedestrian's search process, to the pedestrian image that different cameras capture, according to the human body key area obtained in S2
Domain, the online lower depth characteristic using each human body key area in depth convolutional neural networks every image of model extraction；
The feature of each human body key area is as human region alignmentization final in the image in every image of S4 and then fusion
Feature representation is simultaneously stored in database；
S5, for the human region alignmentization feature representation in query image, line computation its with each pedestrian image in database
Human region alignmentization feature representation between similarity, and according to similarity return pedestrian identify search result again, to
Achieve the purpose that identify pedestrian.
2. pedestrian's recognition methods again as described in claim 1 based on the study of human region alignmentization feature representation, feature
It is, the algorithm object definition of pedestrian's recognition methods again is：Give two pedestrian images, judge it includes pedestrian whether be same
One pedestrian judges whether the ID number mark of two pedestrian images is identical.
3. pedestrian's recognition methods again as described in claim 1 based on the study of human region alignmentization feature representation, feature
It is, in step S1, described is used to train the image data set that pedestrian identifies again, including different cameras are caught under monitoring scene
The pedestrian image received, every image include a pedestrian, which has unique corresponding ID number information, different cameras to catch
The image for receiving the same pedestrian possesses identical ID number mark, and the image of different pedestrians possesses different ID number marks.
4. pedestrian's recognition methods again as described in claim 1 based on the study of human region alignmentization feature representation, feature
It is, in step S2, is based on pedestrian's ID number markup information, several human bodies for meeting semantic content is obtained by learning automatically
Key area specifically includes：
S21, image { I are opened to all N in image data set1,I2,…,INDifferent ternarys is combined into according to pedestrian's ID number
Group { (Ia,Ib,Ic), wherein (Ia,Ib) two images have common identity number, (Ia,Ic) two images have different identity number；
S22, human region detection training is carried out to the image in triple using convolutional neural networks, until detection obtains K+1
A human body key area so that (Ia,Ib) characteristic similarities of two image corresponding regions is higher than (Ia,Ic) two images correspond to
The characteristic similarity in region, K+1 human body key area include 1 region M for only including pedestrian's foreground0With K sub-regions
{M1,M2,…,Mk,…,MK}。
5. pedestrian's recognition methods again as claimed in claim 4 based on the study of human region alignmentization feature representation, feature
It is, in step S3, the online lower depth using each human body key area in depth convolutional neural networks every image of model extraction
Degree feature specific method be：
Using full convolutional neural networks, to every image zooming-out depth characteristic figure T, according to each human body key area learnt
MiOperation is weighted to depth characteristic and obtains the feature representation F of key areai, i.e.,：
Wherein i=0,1 ..., K；For by the multiplying of element.
6. pedestrian's recognition methods again as claimed in claim 5 based on the study of human region alignmentization feature representation, feature
It is, in step S4, the specific method for merging the feature representation of each human body key area is：
S41, the feature vector f ' in the region is obtained after being converted the feature representation of each human body key areai, i.e.,：
f′i=transform (Fi)
S42, as final human region alignmentization feature representation vector f, i.e., K+1 feature vector is stitched together：
F=[f '0,f′1,f′2,…,f′k,…,f′K]
Pedestrian image { the I captured from all cameras1,I2,…,IMIn the human region alignmentization feature representation that extracts to
Measure { f1,f2,…,fM, it is stored in database.
7. pedestrian's recognition methods again as claimed in claim 6 based on the study of human region alignmentization feature representation, feature
It is, the concrete methods of realizing of step S5 is as follows：
S51, for query image IqExtract its human region alignmentization feature representation vector fq；
S52, in database { f1,f2,…,fMWhen scanning for, to the feature representation vector f of each storagem, calculate fqWith fm
Between similarity score sqm, as pedestrian image (Iq,Im) pedestrian's recognition result again；
S53, to all similarity score sqmIt is sorted from big to small, returns to the sequence after sequence, correspond to image { I1,
I2,…,IMRanking results return pedestrian identify search result again.
8. pedestrian's recognition methods again as claimed in claim 7 based on the study of human region alignmentization feature representation, feature
It is, a pair of of image (Iq,Im) between similarity score sqmCalculating using the inverse of Euclidean distance, calculation formula is：
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810046080.2A CN108345837A (en) | 2018-01-17 | 2018-01-17 | A kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810046080.2A CN108345837A (en) | 2018-01-17 | 2018-01-17 | A kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108345837A true CN108345837A (en) | 2018-07-31 |
Family
ID=62960817
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810046080.2A Pending CN108345837A (en) | 2018-01-17 | 2018-01-17 | A kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108345837A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063649A (en) * | 2018-08-03 | 2018-12-21 | 中国矿业大学 | Pedestrian's recognition methods again of residual error network is aligned based on twin pedestrian |
| CN109325412A (en) * | 2018-08-17 | 2019-02-12 | 平安科技（深圳）有限公司 | Pedestrian recognition method, device, computer equipment and storage medium |
| CN109446953A (en) * | 2018-10-17 | 2019-03-08 | 福州大学 | A kind of recognition methods again of the pedestrian based on lightweight convolutional neural networks |
| CN109583942A (en) * | 2018-11-07 | 2019-04-05 | 浙江工业大学 | A kind of multitask convolutional neural networks customer behavior analysis method based on dense network |
| CN109614894A (en) * | 2018-11-28 | 2019-04-12 | 北京陌上花科技有限公司 | Pedestrian recognition methods and device, storage medium, server again |
| CN109871771A (en) * | 2019-01-21 | 2019-06-11 | 北京轻舟科技有限公司 | A kind of method and system of the automatic detection human body based on single-view videos |
| CN109948587A (en) * | 2019-03-29 | 2019-06-28 | 清华大学 | Vehicle based on high judgement index feature mining recognition methods and device again |
| CN110046632A (en) * | 2018-11-09 | 2019-07-23 | 阿里巴巴集团控股有限公司 | Model training method and device |
| CN110457572A (en) * | 2019-05-23 | 2019-11-15 | 北京邮电大学 | Commodity information recommendation method and electronic equipment based on figure network |
| CN110659589A (en) * | 2019-09-06 | 2020-01-07 | 中国科学院自动化研究所 | Pedestrian re-identification method, system and device based on attitude and attention mechanism |
| CN110796072A (en) * | 2019-10-28 | 2020-02-14 | 桂林电子科技大学 | Target tracking and identity recognition method based on double-task learning |
| CN111428546A (en) * | 2019-04-11 | 2020-07-17 | 杭州海康威视数字技术股份有限公司 | Method and device for marking human body in image, electronic equipment and storage medium |
| CN111680716A (en) * | 2020-05-09 | 2020-09-18 | 浙江大华技术股份有限公司 | Identification comparison method and device, computer equipment and storage medium |
| WO2020233414A1 (en) * | 2019-05-20 | 2020-11-26 | 阿里巴巴集团控股有限公司 | Object recognition method and apparatus, and vehicle |
| CN112990152A (en) * | 2021-05-10 | 2021-06-18 | 中国科学院自动化研究所 | Vehicle weight identification method based on key point detection and local feature alignment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631413A (en) * | 2015-12-23 | 2016-06-01 | 中通服公众信息产业股份有限公司 | Cross-scene pedestrian searching method based on depth learning |
| CN106778527A (en) * | 2016-11-28 | 2017-05-31 | 中通服公众信息产业股份有限公司 | A kind of improved neutral net pedestrian recognition methods again based on triple losses |
-
2018
- 2018-01-17 CN CN201810046080.2A patent/CN108345837A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631413A (en) * | 2015-12-23 | 2016-06-01 | 中通服公众信息产业股份有限公司 | Cross-scene pedestrian searching method based on depth learning |
| CN106778527A (en) * | 2016-11-28 | 2017-05-31 | 中通服公众信息产业股份有限公司 | A kind of improved neutral net pedestrian recognition methods again based on triple losses |
Non-Patent Citations (1)
| Title |
|---|
| LIMING ZHAO ET AL: "Deeply-Learned Part-Aligned Representations for Person Re-Identification", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063649B (en) * | 2018-08-03 | 2021-05-14 | 中国矿业大学 | Pedestrian re-identification method based on twin pedestrian alignment residual error network |
| CN109063649A (en) * | 2018-08-03 | 2018-12-21 | 中国矿业大学 | Pedestrian's recognition methods again of residual error network is aligned based on twin pedestrian |
| CN109325412A (en) * | 2018-08-17 | 2019-02-12 | 平安科技（深圳）有限公司 | Pedestrian recognition method, device, computer equipment and storage medium |
| CN109325412B (en) * | 2018-08-17 | 2023-11-24 | 平安科技（深圳）有限公司 | Pedestrian recognition method, device, computer equipment and storage medium |
| CN109446953A (en) * | 2018-10-17 | 2019-03-08 | 福州大学 | A kind of recognition methods again of the pedestrian based on lightweight convolutional neural networks |
| CN109583942A (en) * | 2018-11-07 | 2019-04-05 | 浙江工业大学 | A kind of multitask convolutional neural networks customer behavior analysis method based on dense network |
| CN110046632B (en) * | 2018-11-09 | 2023-06-02 | 创新先进技术有限公司 | Model training method and device |
| CN110046632A (en) * | 2018-11-09 | 2019-07-23 | 阿里巴巴集团控股有限公司 | Model training method and device |
| CN109614894A (en) * | 2018-11-28 | 2019-04-12 | 北京陌上花科技有限公司 | Pedestrian recognition methods and device, storage medium, server again |
| CN109871771A (en) * | 2019-01-21 | 2019-06-11 | 北京轻舟科技有限公司 | A kind of method and system of the automatic detection human body based on single-view videos |
| CN109948587A (en) * | 2019-03-29 | 2019-06-28 | 清华大学 | Vehicle based on high judgement index feature mining recognition methods and device again |
| CN111428546A (en) * | 2019-04-11 | 2020-07-17 | 杭州海康威视数字技术股份有限公司 | Method and device for marking human body in image, electronic equipment and storage medium |
| CN111428546B (en) * | 2019-04-11 | 2023-10-13 | 杭州海康威视数字技术股份有限公司 | Method and device for marking human body in image, electronic equipment and storage medium |
| WO2020233414A1 (en) * | 2019-05-20 | 2020-11-26 | 阿里巴巴集团控股有限公司 | Object recognition method and apparatus, and vehicle |
| CN110457572B (en) * | 2019-05-23 | 2022-05-24 | 北京邮电大学 | Commodity information recommendation method based on graph network and electronic equipment |
| CN110457572A (en) * | 2019-05-23 | 2019-11-15 | 北京邮电大学 | Commodity information recommendation method and electronic equipment based on figure network |
| CN110659589B (en) * | 2019-09-06 | 2022-02-08 | 中国科学院自动化研究所 | Pedestrian re-identification method, system and device based on attitude and attention mechanism |
| CN110659589A (en) * | 2019-09-06 | 2020-01-07 | 中国科学院自动化研究所 | Pedestrian re-identification method, system and device based on attitude and attention mechanism |
| CN110796072B (en) * | 2019-10-28 | 2023-04-07 | 桂林电子科技大学 | Target tracking and identity recognition method based on double-task learning |
| CN110796072A (en) * | 2019-10-28 | 2020-02-14 | 桂林电子科技大学 | Target tracking and identity recognition method based on double-task learning |
| CN111680716A (en) * | 2020-05-09 | 2020-09-18 | 浙江大华技术股份有限公司 | Identification comparison method and device, computer equipment and storage medium |
| CN111680716B (en) * | 2020-05-09 | 2023-05-12 | 浙江大华技术股份有限公司 | Identification comparison method, device, computer equipment and storage medium |
| CN112990152A (en) * | 2021-05-10 | 2021-06-18 | 中国科学院自动化研究所 | Vehicle weight identification method based on key point detection and local feature alignment |
| CN112990152B (en) * | 2021-05-10 | 2021-07-30 | 中国科学院自动化研究所 | Vehicle weight identification method based on key point detection and local feature alignment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108345837A (en) | A kind of pedestrian's recognition methods again based on the study of human region alignmentization feature representation | |
| KR102596897B1 (en) | Method of motion vector and feature vector based fake face detection and apparatus for the same | |
| Chen et al. | An end-to-end system for unconstrained face verification with deep convolutional neural networks | |
| CN111126360A (en) | Cross-domain pedestrian re-identification method based on unsupervised combined multi-loss model | |
| CN108805077A (en) | A kind of face identification system of the deep learning network based on triple loss function | |
| CN104751108A (en) | Face image recognition device and face image recognition method | |
| Barnich et al. | Frontal-view gait recognition by intra-and inter-frame rectangle size distribution | |
| CN105718873A (en) | People stream analysis method based on binocular vision | |
| CN111241932A (en) | Automobile exhibition room passenger flow detection and analysis system, method and storage medium | |
| CN111460884A (en) | Multi-face recognition method based on human body tracking | |
| CN113705290A (en) | Image processing method, image processing device, computer equipment and storage medium | |
| CN108875505A (en) | Pedestrian neural network based recognition methods and device again | |
| Devi et al. | Automatic face recognition system using pattern recognition techniques: a survey | |
| CN116386118B (en) | Drama matching cosmetic system and method based on human image recognition | |
| CN113706550A (en) | Image scene recognition and model training method and device and computer equipment | |
| CN115830637B (en) | Method for re-identifying blocked pedestrians based on attitude estimation and background suppression | |
| Manjula et al. | Face detection identification and tracking by PRDIT algorithm using image database for crime investigation | |
| CN109308584A (en) | A kind of noninductive attendance system and method | |
| Ma et al. | A gait recognition method based on the combination of human body posture and human body contour | |
| CN112418078B (en) | Score modulation method, face recognition device and medium | |
| Cai et al. | Person-specific face spoofing detection for replay attack based on gaze estimation | |
| Kumar et al. | Muzzle point pattern recognition system using image pre-processing techniques | |
| Li et al. | Intelligent terminal face spoofing detection algorithm based on deep belief network | |
| Shylaja et al. | Illumination Invariant Novel Approaches for Face Recognition | |
| Cheng et al. | Automatic Data Cleaning System for Large-Scale Location Image Databases Using a Multilevel Extractor and Multiresolution Dissimilarity Calculation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180731 |