CN105956606A - Method for re-identifying pedestrians on the basis of asymmetric transformation - Google Patents
Method for re-identifying pedestrians on the basis of asymmetric transformation Download PDFInfo
- Publication number
- CN105956606A CN105956606A CN201610255341.2A CN201610255341A CN105956606A CN 105956606 A CN105956606 A CN 105956606A CN 201610255341 A CN201610255341 A CN 201610255341A CN 105956606 A CN105956606 A CN 105956606A
- Authority
- CN
- China
- Prior art keywords
- image
- pedestrian
- test
- asymmetric
- vector
- 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
- 238000000034 method Methods 0.000 title claims abstract description 27
- 230000009466 transformation Effects 0.000 title claims abstract description 14
- 238000012360 testing method Methods 0.000 claims abstract description 42
- 239000013598 vector Substances 0.000 claims abstract description 42
- 239000011159 matrix material Substances 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000006243 chemical reaction Methods 0.000 claims description 28
- 230000008859 change Effects 0.000 claims description 10
- 238000009795 derivation Methods 0.000 claims description 8
- 238000012935 Averaging Methods 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims 1
- 230000006870 function Effects 0.000 description 25
- 230000000694 effects Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for re-identifying pedestrians on the basis of asymmetric transformation. The method comprises steps of: extracting a LOMO characteristic of a pedestrian image set used by training to form a characteristic matrix and constructing an expected sparse reconstruction vector of each training-used test image; constructing an overall objective function required to be optimized; optimizing the objective function by using an alternately optimized framework and obtaining asymmetric transformation matrixes TA and TB after the objective function converges; in a testing stage, reconstructing an image characteristic in a transformed testing set by using the image characteristic in a transformed testing set; and ordering the images in the image set by using a size relation of reconstitution coefficient vectors. The method improves the reliability of sparse reconstruction, enhances the accuracy and the robustness of pedestrian re-identification, solves troubles due to shielding status in the pedestrian re-identification problem, and increases the accuracy of pedestrian re-identification.
Description
Technical field
The present invention relates to a kind of method that pedestrian identifies again, in the case of particularly relating to a kind of multiframe, based on asymmetric change
The pedestrian changed identification method again.
Background technology
It is the image set for known each pictures label that pedestrian identifies problem again, therefrom finds and does not currently mark
Know image the most close picture picture is identified, be broadly divided into two steps: feature extraction and metric learning.Due to light
According to, attitude, block, the change of the factor such as background, cause the existence of class inherited so that pedestrian identify again problem become one non-
Normal challenging problem.
Additionally, during the realization that pedestrian identifies again, general way is to use description or the tool of a robust
Image in image set is ranked up by the metric learning model having identification, but such shortcoming is not account for test
Image and the relativeness of all image similarity in image set, the information about test image is not utilized in a large number, makes
Obtain matching performance and can not reach good effect.
Giuseppe Lisanti proposed to utilize the method for sparse reconstruct to identify problem again to solve pedestrian in 2015, dilute
The method of thin reconstruct is the linear combination that test image approximate is all images in image set, and it is right to utilize reconstructed error
In image set, all images are ranked up.Have huge yet with the sample obtained under photographic head different in pedestrian again mark problem
Big diversity, the sample obtained from a photographic head does not has enough expression energy for the sample of another one photographic head
Power, causes the effect carrying out sparse reconstruct in original space unsatisfactory.
Summary of the invention
For overcoming the deficiency of existing pedestrian identification method again, it is provided that a kind of matching rate is high, the pedestrian of strong robustness marks again
Knowledge method, the present invention proposes a kind of pedestrian based on asymmetric conversion identification method again, the technical scheme is that so
:
A kind of pedestrian based on asymmetric conversion identification method again, including step:
S1: LOMO feature constitutive characteristic matrix G and P of the pedestrian's image set used by extraction training, for each pedestrian
All characteristics of image carry out being averaging operation, obtain representing the feature of this pedestrian；
S2: build the desired sparse reconstruct vector of the test image used by each Zhang Xunlian；
S3: after obtaining the desired sparse reconstruct vector of all test images, builds the overall object function needing to optimize；
S4: using object function described in the framework team of alternative optimization to be optimized, after object function is restrained, it is non-right to obtain
Claim transformation matrix TAAnd TB；
S5: at test phase, the image in image set is used asymmetric transformation matrix TAConvert, by test set
Image use asymmetric transformation matrix TBConvert, and use the characteristics of image in the image set after conversion come to conversion after
Test set in characteristics of image be reconstructed；
S6: use the magnitude relationship of reconstruction coefficient vector that the image in image set is ranked up.
Further, step S1 includes step:
S11: pedestrian's image is processed；
S12: in addition to description of color, also describes texture of described pedestrian's image；
S13: all subwindows that detection is on same level line, choose the value conduct that in rectangular histogram, each position is maximum
This horizontal general characteristic；
S14: feature and texture that color is described son are described sub feature string and be unified into as LOMO feature.
Further, step S11 use Retinex algorithm carry out pedestrian's image procossing.
Further, step S12 use SILTP image texture is described.
Further, the step S2 structure desired sparse reconstruct vector of test image used by each Zhang Xunlian includes step
Rapid:
For the test characteristics of image of training, the image set average characteristics in training finds the spy as its class mark
Levy and record its position, construct a length full 0 vector equal with image set number of features, and this vector will correspond to institute
The position stating the same feature of class mark is set to 1.
Further, in step S3, the structure of object function concretely comprises the following steps: assume in the pedestrian's image set used by training
Test amount of images is n, for test image pkThe structure desired sparse coding vector s of structurek, asymmetric change to be optimized
Change matrix and be respectively TAAnd TB, wherein TAAnd TBBeing initially unit matrix, the object function of structure is
WhereinIt is reconstruction coefficient vector desired for i-th sample, S=[s1,
s2,...sn] it is the matrix of all coefficient vector compositions to be optimized.
Further, step S4 includes step:
S41: initialize TAAnd TBFor unit battle array, object function S is random matrix；
S42: fixing TAAnd TB, S is optimized, due to TAAnd TBBeing fixed, every string of S is the most all mutual
Independent, therefore every string s of SiNow can be updated by solving following formula:
si,j>=0, j=1,2 ..., n；
S43: fixing S and TB, to TAIt is optimized, by allowing object function to TADerivation obtains about TALocal Minimum
Value is to TAIt is updated:
S44: fixing S and TA, to TBIt is optimized, by allowing object function to TBDerivation obtains about TBLocal Minimum
Value is to TBIt is updated:
S45: repeated execution of steps S42 to S44, until object function S restrains, finally needed the asymmetric change of study
Change matrix TAAnd TB。
Further, the characteristics of image in image set after using conversion described in step S5 comes the test set after conversion
In characteristics of image be reconstructed and include step:
S51: characteristic vector image set and test image set being belonged to same person is averaging, and obtains matrix
S52: image will be testedImage is respectively by projection matrix TAProject, will be with image setIn image
By projection matrix TBProject, finally solve following formula and obtain reconstruction coefficient vector si:
si,m>=0, m=1,2 ..., n.
The beneficial effects of the present invention is, first, the present invention is by projecting in transformation space by characteristics of image so that
The reconstructed error belonged between the image of same pedestrian diminishes, and makes to be not belonging between the image of same person pedestrian simultaneously
Reconstructed error becomes big, improves the reliability of sparse reconstruct, improves accuracy and robustness that pedestrian identifies again；Secondly, originally
Invent for the problem likely occurring blocking in pedestrian again mark problem, by all images to everyone before coupling
Be averaged operation, largely solves the puzzlement that in pedestrian's mark problem again, circumstance of occlusion brings；Again, of the present invention
By adding clear and definite constraint during solving reconstruction coefficient vector so that the reconstruction coefficients between two images can have
Matching probability between two images of expression of effect, greatly enhances the accuracy that pedestrian identifies again.
Accompanying drawing explanation
Fig. 1 is present invention pedestrian based on asymmetric conversion identification method flow chart again；
Fig. 2 is objective function optimization flow chart in step S4.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise
Embodiment, broadly falls into the scope of protection of the invention.
Refer to Fig. 1, a kind of pedestrian based on asymmetric conversion identification method again, including step:
S1: LOMO feature constitutive characteristic matrix G and P of the pedestrian's image set used by extraction training, for each pedestrian
All characteristics of image carry out being averaging operation, obtain representing the feature of this pedestrian；
S2: build the desired sparse reconstruct vector of the test image used by each Zhang Xunlian；
S3: after obtaining the desired sparse reconstruct vector of all test images, builds the overall object function needing to optimize；
S4: using object function described in the framework team of alternative optimization to be optimized, after object function is restrained, it is non-right to obtain
Claim transformation matrix TAAnd TB；
S5: at test phase, the image in image set is used asymmetric transformation matrix TAConvert, by test set
Image use asymmetric transformation matrix TBConvert, and use the characteristics of image in the image set after conversion come to conversion after
Test set in characteristics of image be reconstructed；
S6: use the magnitude relationship of reconstruction coefficient vector that the image in image set is ranked up.
Wherein, in step sl, the process of LOMO feature is extracted, including step S11: use Retinex algorithm to pedestrian
Image processes, and this algorithm makes result have a good robustness for illumination variation by processing coloured image:
S12: in addition to description of color, also uses texture descriptor to express the feature of image, used here as
Image texture is described by SILTP；S13: propose the local detail using sliding window to express pedestrian's picture, I
All subwindows that detection is on same level line, choose the value that in rectangular histogram, each position is maximum horizontal always as this
Body characteristics, such operation makes the histogram feature of output obtain, for isogonal effect, can also capturing simultaneously
Local features final sum step S14 of pedestrian: above-mentioned color is described son and texture is described substring and is unified into as LOMO special
Levy.
In step s 2, the step constructing desired sparse coding vector is as follows: for the test characteristics of image of trainingImage set average characteristics in trainingFind the feature as its class markAnd record go to position, construct one long
Spend the full 0 vector equal with image set number of features, and will this vector correspond toPosition be set to 1.
In step s3, build concretely comprising the following steps of object function: assume that having training integrated test amount of images is n, right
In test image pkThe structure desired sparse coding vector s of structurek, asymmetric transformation matrix to be optimized is respectively TAAnd TB。
The object function of structure is
Step S4 includes step:
S41: initialize TAAnd TBFor unit battle array, object function S is random matrix；
S42: fixing TAAnd TB, S is optimized, due to TAAnd TBBeing fixed, every string of S is the most all mutual
Independent, therefore every string s of SiNow can be updated by solving following formula:
si,j>=0, j=1,2 ..., n；
S43: fixing S and TB, to TAIt is optimized, by allowing object function to TADerivation obtains about TALocal Minimum
Value is to TAIt is updated:
S44: fixing S and TA, to TBIt is optimized, by allowing object function to TBDerivation obtains about TBLocal Minimum
Value is to TBIt is updated:
S45: repeated execution of steps S42 to S44, until object function S restrains, finally needed the asymmetric change of study
Change matrix TAAnd TB。
The characteristics of image in image set after using conversion described in step S5 comes the image in the test set after conversion
Feature is reconstructed and includes step:
S51: characteristic vector image set and test image set being belonged to same person is averaging, and obtains matrix
S52: image will be testedImage is respectively by projection matrix TAProject, will be with image setIn image
By projection matrix TBProject, finally solve following formula and obtain reconstruction coefficient vector si:
si,m>=0, m=1,2 ..., n.
The present invention sparse reconstruct pedestrian's identification method again based on asymmetric conversion, one embodiment is as follows:
Step S1: for image set and test image, first carry the feature descriptor LOMO of robust, specifically can join
Examine " S.Liao, Y.Hu, X.Zhu, and S.Z.Li, " Person re-identification by local maximal
occurrence representation and metric learning,”in CVPR,2015.pp.2197-2206”。
Step S2: for all characteristic sets belonging to i-th peopleWithObtain its average characteristicsWithDescription as this people；Survey for training
Sample is originallyFind matched pedestrian's featureGenerate its desired reconstruction coefficient vector, specifically, first generate one
Full 0 vector, the number of features of its a length of image set, then would correspond toThat position be set to 1.
Step S3: after the reconstruction coefficient vector construction complete of each training sample, we construct overall mesh
Scalar functions:
WhereinIt is reconstruction coefficient vector desired for i-th sample, S=[s1,s2,...sn] it is all to be optimized
The matrix of coefficient vector composition.
Step S4: the optimization process of object function is as in figure 2 it is shown, use alternative optimization framework here to object function
It is optimized:
S41: initialize TAAnd TBFor unit battle array, S is random matrix.
S42: fixing TAAnd TB, S is optimized, due to TAAnd TBBeing fixed, every string of S is the most all mutual
Independent, therefore every string s of SiNow can be updated by solving following formula,
si,j>=0, j=1,2 ..., n
This problem can be solved by simple convex optimization tool bag, and here we are entered by CVX tool kit
Row solves, and detailed class sees " M.Grant and S.Boyd.Graph implementations for non-smooth
convex programs.In V.Blondel,S.Boyd,and H.Kimura,editors,Recent Advances in
Learning and Control,Lecture Notes in Control and Information Sciences,pages
95–110,2008.”。
S43: fixing S and TB, to TAIt is optimized, by allowing object function to TADerivation obtains about TALocal Minimum
Value is to TAIt is updated:
S44: fixing S and TA, to TBIt is optimized, by allowing object function to TBDerivation obtains about TBLocal Minimum
Value is to TBIt is updated:
S45: repeated execution of steps S42-S44, it is known that object function convergence, is finally needed the asymmetric conversion of study
Matrix TAAnd TB。
All images are first extracted LOMO feature by step S5: at test phase respectively, for image set and test image
Collection belongs to the characteristic vector of same person and is averaging, and obtains matrix
Then will test image Pi tWith image setIn image respectively by projection matrix TAAnd TBProject, finally
Solve following formula and obtain reconstruction coefficient vector si。
si,m>=0, m=1,2 ..., n.
Step S6: by reconstruction coefficient vector siMagnitude relationship to image setIn image be ranked up.
Pass through above step, it is possible to form pedestrian's identity in the system identified again under photographic head scene.
The above is the preferred embodiment of the present invention, it is noted that for those skilled in the art
For, under the premise without departing from the principles of the invention, it is also possible to make some improvements and modifications, these improvements and modifications are also considered as
Protection scope of the present invention.
Claims (8)
1. pedestrian based on an asymmetric conversion identification method again, it is characterised in that include step:
S1: LOMO feature constitutive characteristic matrix G and P of the pedestrian's image set used by extraction training, for the institute of each pedestrian
There is characteristics of image to carry out being averaging operation, obtain representing the feature of this pedestrian；
S2: build the desired sparse reconstruct vector of the test image used by each Zhang Xunlian；
S3: after obtaining the desired sparse reconstruct vector of all test images, builds the overall object function needing to optimize；
S4: use object function described in the framework team of alternative optimization to be optimized, after object function is restrained, obtain asymmetric change
Change matrix TAAnd TB；
S5: at test phase, the image in image set is used asymmetric transformation matrix TAConvert, by the figure in test set
As using asymmetric transformation matrix TBConvert, and use the characteristics of image in the image set after conversion to come the survey after conversion
The characteristics of image that examination is concentrated is reconstructed；
S6: use the magnitude relationship of reconstruction coefficient vector that the image in image set is ranked up.
2. pedestrian based on asymmetric conversion identification method more as claimed in claim 1, it is characterised in that step S1 includes step
Rapid:
S11: pedestrian's image is processed；
S12: in addition to description of color, also describes texture of described pedestrian's image；
S13: all subwindows of being on same level line of detection, choose value that in rectangular histogram, each position is maximum as this water
The general characteristic of horizontal line；
S14: feature and texture that color is described son are described sub feature string and be unified into as LOMO feature.
3. pedestrian based on asymmetric conversion identification method more as claimed in claim 2, it is characterised in that make in step S11
Pedestrian's image procossing is carried out by Retinex algorithm.
4. pedestrian based on asymmetric conversion identification method more as claimed in claim 2, it is characterised in that make in step S12
With SILTP, image texture is described.
5. pedestrian based on asymmetric conversion identification method more as claimed in claim 1, it is characterised in that step S2 builds every
The desired sparse reconstruct vector of test image used by one training includes step:
For the test characteristics of image of training, the image set average characteristics in training finds the feature as its class mark also
Record its position, construct a length full 0 vector equal with image set number of features, and this vector will correspond to described class
The position of the feature that mark is the same is set to 1.
6. pedestrian based on asymmetric conversion identification method more as claimed in claim 1, it is characterised in that target in step S3
The structure of function concretely comprises the following steps: assume that the pedestrian's image integrated test amount of images used by training is n, for test image pk
The structure desired sparse coding vector s of structurek, asymmetric transformation matrix to be optimized is respectively TAAnd TB, wherein TAAnd TBJust
Beginning to be unit battle array, the object function of structure is
WhereinIt is for the i-th sample phase
The reconstruction coefficient vector hoped, S=[s1,s2,...sn] it is the matrix of all coefficient vector compositions to be optimized.
7. pedestrian based on asymmetric conversion identification method more as claimed in claim 1, it is characterised in that step S4 includes step
Rapid:
S41: initialize TAAnd TBFor unit battle array, object function S is random matrix；
S42: fixing TAAnd TB, S is optimized, due to TAAnd TBBeing fixed, every string of S is the most all separate
, therefore every string s of SiNow can be updated by solving following formula:
si,j>=0, j=1,2 ..., n；
S43: fixing S and TB, to TAIt is optimized, by allowing object function to TADerivation obtains about TALocal minimum to TA
It is updated:
S44: fixing S and TA, to TBIt is optimized, by allowing object function to TBDerivation obtains about TBLocal minimum to TB
It is updated:
S45: repeated execution of steps S42 to S44, until object function S restrains, finally needed the asymmetric conversion square of study
Battle array TAAnd TB。
8. pedestrian based on asymmetric conversion identification method more as claimed in claim 1, it is characterised in that described in step S5
Use the characteristics of image in the image set after conversion that the characteristics of image in the test set after conversion is reconstructed and include step:
S51: characteristic vector image set and test image set being belonged to same person is averaging, and obtains matrix
S52: image will be testedImage is respectively by projection matrix TAProject, will be with image setIn image pass through
Projection matrix TBProject, finally solve following formula and obtain reconstruction coefficient vector si:
si,m>=0, m=1,2 ..., n.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610255341.2A CN105956606B (en) | 2016-04-22 | 2016-04-22 | A kind of pedestrian's identification method again based on asymmetry transformation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610255341.2A CN105956606B (en) | 2016-04-22 | 2016-04-22 | A kind of pedestrian's identification method again based on asymmetry transformation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105956606A true CN105956606A (en) | 2016-09-21 |
| CN105956606B CN105956606B (en) | 2019-09-10 |
Family
ID=56915418
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610255341.2A Active CN105956606B (en) | 2016-04-22 | 2016-04-22 | A kind of pedestrian's identification method again based on asymmetry transformation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105956606B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107330355A (en) * | 2017-05-11 | 2017-11-07 | 中山大学 | A kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again |
| CN107886116A (en) * | 2017-10-30 | 2018-04-06 | 国家新闻出版广电总局广播科学研究院 | The LOMO3D feature extracting methods identified again for video data pedestrian |
| CN108229512A (en) * | 2016-12-22 | 2018-06-29 | 扬州大学 | A kind of two benches recognition methods based on non-negative indication coefficient |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103984922A (en) * | 2014-04-30 | 2014-08-13 | 苏亚 | Face identification method based on sparse representation and shape restriction |
| CN104537356A (en) * | 2015-01-12 | 2015-04-22 | 北京大学 | Pedestrian re-identification method and device for carrying out gait recognition through integral scheduling |
| CN104915643A (en) * | 2015-05-26 | 2015-09-16 | 中山大学 | Deep-learning-based pedestrian re-identification method |
| CN105046195A (en) * | 2015-06-09 | 2015-11-11 | 浙江理工大学 | Human behavior identification method based on asymmetric generalized Gaussian distribution model (AGGD) |
| CN105095475A (en) * | 2015-08-12 | 2015-11-25 | 武汉大学 | Incomplete attribute tagged pedestrian re-identification method and system based on two-level fusion |
-
2016
- 2016-04-22 CN CN201610255341.2A patent/CN105956606B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103984922A (en) * | 2014-04-30 | 2014-08-13 | 苏亚 | Face identification method based on sparse representation and shape restriction |
| CN104537356A (en) * | 2015-01-12 | 2015-04-22 | 北京大学 | Pedestrian re-identification method and device for carrying out gait recognition through integral scheduling |
| CN104915643A (en) * | 2015-05-26 | 2015-09-16 | 中山大学 | Deep-learning-based pedestrian re-identification method |
| CN105046195A (en) * | 2015-06-09 | 2015-11-11 | 浙江理工大学 | Human behavior identification method based on asymmetric generalized Gaussian distribution model (AGGD) |
| CN105095475A (en) * | 2015-08-12 | 2015-11-25 | 武汉大学 | Incomplete attribute tagged pedestrian re-identification method and system based on two-level fusion |
Non-Patent Citations (4)
| Title |
|---|
| JOHN WRIGHT 等: "Robust Face Recognition via Sparse Representation", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| SHENGCAI LIAO 等: "Efficient PSD Constrained Asymmetric Metric Learning for Person Re-Identification", 《2015 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》 * |
| YING-CONG CHEN 等: "An Asymmetric Distance Model for Cross-View Feature Mapping in Person Reidentification", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
| YING-CONG CHEN 等: "Mirror representation for modeling view-specific transform in person re-identification", 《IJCAI15 PROCEEDINGS OF THE 24TH INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108229512A (en) * | 2016-12-22 | 2018-06-29 | 扬州大学 | A kind of two benches recognition methods based on non-negative indication coefficient |
| CN108229512B (en) * | 2016-12-22 | 2021-10-08 | 扬州大学 | Two-stage identification method based on non-negative expression coefficient |
| CN107330355A (en) * | 2017-05-11 | 2017-11-07 | 中山大学 | A kind of depth pedestrian based on positive sample Constraints of Equilibrium identification method again |
| CN107330355B (en) * | 2017-05-11 | 2021-01-26 | 中山大学 | Deep pedestrian re-identification method based on positive sample balance constraint |
| CN107886116A (en) * | 2017-10-30 | 2018-04-06 | 国家新闻出版广电总局广播科学研究院 | The LOMO3D feature extracting methods identified again for video data pedestrian |
| CN107886116B (en) * | 2017-10-30 | 2021-03-26 | 国家新闻出版广电总局广播科学研究院 | LOMO3D feature extraction method for pedestrian re-identification of video data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105956606B (en) | 2019-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | SaliencyGAN: Deep learning semisupervised salient object detection in the fog of IoT | |
| WO2022267641A1 (en) | Image defogging method and system based on cyclic generative adversarial network | |
| Huang et al. | Bidirectional recurrent convolutional networks for multi-frame super-resolution | |
| CN103093444B (en) | Image super-resolution reconstruction method based on self-similarity and structural information constraint | |
| CN112115783A (en) | Human face characteristic point detection method, device and equipment based on deep knowledge migration | |
| CN115240121B (en) | Joint modeling method and device for enhancing local features of pedestrians | |
| WO2023082784A1 (en) | Person re-identification method and apparatus based on local feature attention | |
| CN102509304A (en) | Intelligent optimization-based camera calibration method | |
| CN114973317A (en) | Pedestrian re-identification method based on multi-scale adjacent interaction features | |
| CN113159158B (en) | License plate correction and reconstruction method and system based on generation countermeasure network | |
| CN114677722A (en) | Multi-supervision human face in-vivo detection method integrating multi-scale features | |
| CN111680705A (en) | MB-SSD method and MB-SSD feature extraction network suitable for target detection | |
| CN114140623A (en) | Image feature point extraction method and system | |
| CN105956606A (en) | Method for re-identifying pedestrians on the basis of asymmetric transformation | |
| WO2023206343A1 (en) | Image super-resolution method based on image pre-training strategy | |
| CN107729885B (en) | Face enhancement method based on multiple residual error learning | |
| CN116863223A (en) | Method for classifying remote sensing image scenes by embedding semantic attention features into Swin transform network | |
| CN106845555A (en) | Image matching method and image matching apparatus based on Bayer format | |
| Yuan et al. | Structure flow-guided network for real depth super-resolution | |
| CN111178229B (en) | Deep learning-based vein imaging method and device | |
| CN115984949B (en) | Low-quality face image recognition method and equipment with attention mechanism | |
| CN112330639A (en) | Significance detection method for color-thermal infrared image | |
| CN116957974A (en) | Image rain removing method and device based on hybrid network | |
| CN111126173A (en) | High-precision face detection method | |
| CN116347251A (en) | Image processing super-parameter optimization method, system, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210831 Address after: 510700 room 2001-1, floor 20, building F, No. 98, Xiangxue 8th Road, Huangpu District, Guangzhou, Guangdong Province (office only) Patentee after: Guangzhou ziweiyun Technology Co.,Ltd. Address before: 510275 No. 135 West Xingang Road, Guangdong, Guangzhou Patentee before: SUN YAT-SEN University |