CN109344842A - A kind of pedestrian's recognition methods again based on semantic region expression - Google Patents
A kind of pedestrian's recognition methods again based on semantic region expression Download PDFInfo
- Publication number
- CN109344842A CN109344842A CN201810930107.4A CN201810930107A CN109344842A CN 109344842 A CN109344842 A CN 109344842A CN 201810930107 A CN201810930107 A CN 201810930107A CN 109344842 A CN109344842 A CN 109344842A
- Authority
- CN
- China
- Prior art keywords
- feature
- pedestrian
- image
- descriptor
- component
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- 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
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 kind of pedestrian's recognition methods again based on semantic region expression, it include: given piece image, the position detection of pedestrian's difference component in image is come out, obtain the set comprising different component areas, the high-layer semantic information being had according to pedestrian, use Scale invariant local mode descriptor and color histogram descriptor, it extracts the feature in each component area and is cascaded, it reuses principal component analysis and dimensionality reduction, the set of obtaining widget provincial characteristics is carried out to primitive character;The feature descriptor more complete to pedestrian is obtained according to the set of component area feature, the global characteristics in conjunction with image and local feature;Training metric function is lost using triple, the metric matrix that training obtains can improve the separability between sample by original maps feature vectors into new feature space;The similitude between pedestrian image feature is calculated according to the metric matrix that study obtains, to realize the identification again to pedestrian.Invention not only avoids the interference of image background, the comparison of similitude between corresponding region is also achieved, effectively improves the robustness and reliability of feature.
Description
Technical field
The present invention relates to image procossing, technical field of computer vision more particularly to it is a kind of based on semantic region expression
Pedestrian's recognition methods again.
Background technique
Growing with smart city and public safety demand, intelligent video monitoring system has obtained extensive concern
And research, and the numerous areas such as security protection, industrial production are had application to, play important role in daily life.In intelligence
The object being primarily upon in field of video monitoring is exactly pedestrian, and most basic task is exactly in large-scale video surveillance network
Long-term and stable tracking is carried out to target.But the considerations of for privacy and maintenance cost etc., camera head monitor net
Network is difficult to cover whole region, to blind area of monitoring occur, when target passes through blind area of monitoring, can not just carry out to target continuous
Tracking, therefore just will appear tracking target disappear in the monitoring visual field of a camera, later in other cameras go out
Existing situation.And it is exactly the process that lost target is given for change again in other monitoring visual fields in a monitoring camera visual field
Pedestrian's weight identification technology.The technology passes through the target handoff for solving the problems, such as the non-overlap camera ken, thus by the different kens
Track connect, realize long-term to the pedestrian target and stable tracking under entirely monitoring network.In addition, pedestrian identifies again
The search function that pedestrian may be implemented quickly searches target pedestrian from history video recording, efficiently fast instead of the mode that manpower is checked
Speed filters out the most possible video image comprising target suspect.Therefore the technology has broad application prospects and studies
Value.
High-precision biological recognition system is had been achieved with compared to other, such as recognition of face, what pedestrian identified again
Research faces illumination variation, visual angle change, background interference, blocks, the interference for the factors such as picture quality is low.For example, 1) supervise
The picture for controlling video is more fuzzy, in addition also it is difficult to ensure that face is the shooting of face video camera, so can not be in monitor video
It is middle to use face recognition technology, it can only be identified using the appearance information of human body;2) due to the shadow of the factors such as visual angle, illumination
It rings, appearance of the same pedestrian under different scenes can have biggish difference, and the appearance without same pedestrian is then possible to compare
It is similar;3) it blocks and will lead to a part of invisible of intended body with factors such as background interferences, be only difficult to by visible part to mesh
Mark distinguishes.These factors all bring very big challenge to pedestrian's weight identification technology, therefore how to extract reliable spy
Sign indicates that so that it is had preferable robustness to these factors is to solve the problems, such as this committed step.In conjunction with having in image
Semantic region expression characteristic, the invention proposes a kind of effective feature extraction modes, identify calculation again for improving pedestrian
The performance of method.
Liao et al. combines color histogram feature, size constancy textural characteristics, for improving the robust of feature
Property.Su et al. does not describe pedestrian with the characteristic of visual angle change using high-rise attribute information, to improve the accuracy of algorithm.
Zhao et al. introduces conspicuousness information during object matching, and Cheng et al. proposes the convolutional Neural net an of multithread
Network model, while the global and local information of learning objective.Wu et al. by the full articulamentum feature of convolutional neural networks and tradition
Color characteristic combine, to improve the discrimination power of full articulamentum feature.
In the implementation of the present invention, the discovery prior art has the following disadvantages and deficiency by inventor:
The method of the prior art often carries out the comparison of target similitude according to the global characteristics of image, but does not consider the overall situation
Many disturbing factors such as background information can be included in feature;Existing method usually calculates the similar of same position between image pair
Property, then the adduction of similitude is carried out, do not account for the problem of pedestrian is misaligned in two images.
Summary of the invention
The present invention provides a kind of pedestrian's recognition methods again based on semantic region expression, and the present invention is by deeply excavating figure
With the region of semantic information as in, a kind of pedestrian's weight recognizer based on semantic region expression of robust is devised, is used for
The precision of algorithm retrieval is improved, described below:
A kind of pedestrian's recognition methods again based on semantic region expression, the described method comprises the following steps:
Given piece image, the position detection of pedestrian's difference component in image is come out, and is obtained comprising different component areas
Set:
The high-layer semantic information being had according to pedestrian uses Scale invariant local mode descriptor and color histogram
Descriptor extracts the feature in each component area and is cascaded, reuse principal component analysis to primitive character into
Row dimensionality reduction, the set of obtaining widget provincial characteristics;
It is obtained according to the set of component area feature, the global characteristics in conjunction with image and local feature more complete to pedestrian
Feature descriptor;
Training metric function is lost using triple, the metric matrix that training obtains can be by original maps feature vectors
Into new feature space, the separability between sample is improved;
The similitude between pedestrian image feature is calculated according to the metric matrix that study obtains, to realize the weight to pedestrian
Identification.
Further, the set of the component area feature specifically:
FP={ fP, 1..., fP, i..., fP, M}
Wherein, FPThat indicate is the set of component area feature, fP, iWhat is indicated is the feature of i-th of segmentation subregion.Into
One step, the feature descriptor specifically:
F={ FP, FG, FL}
Wherein, FPFor component area feature;FGFor global characteristics;FLFor local features.
The similarity specifically:
Wherein, Fa, FbThe characteristics of image descriptor being illustrated respectively under the different monitoring visual field,<,>expression is inner product
Operation,What is indicated is the mahalanobis distance between characteristics of image,Table
That show is the bilinearity distance between characteristics of image, WMAnd WBIt is metric matrix.
The beneficial effect of the technical scheme provided by the present invention is that:
1, the semantic information that the present invention has according to pedestrian body extracts the spy in display foreground in different components respectively
Sign, not only avoids the interference of image background, while also achieving the comparison of similitude between corresponding region, to effectively improve
The robustness and reliability of feature.
2, the present invention is in addition to using the component area information for being more concerned about local detail, also using can be in large scale
The global and local provincial characteristics of image is described.Three kinds of features describe picture from different scale layers time are upper, be incorporated into so as to
To obtain to the more comprehensive feature descriptor of iamge description.
3, the present invention using triple loss progress metric learning, improve sample characteristics in new feature space can
Indexing is conducive to the raising of pedestrian's weight recognizer performance.
Detailed description of the invention
Fig. 1 is a kind of flow chart of pedestrian's recognition methods again based on semantic region expression;
Fig. 2 is the comparing result schematic diagram of the present invention with other algorithms.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, embodiment of the present invention is made below further
Ground detailed description.
Embodiment 1
The embodiment of the present invention proposes a kind of pedestrian's recognition methods again based on semantic region expression, referring to Fig. 1, this method
The following steps are included:
101: the high-layer semantic information being had according to pedestrian extracts the feature of different body parts in display foreground, and
The global characteristics and local feature for combining image obtain the feature description more complete to pedestrian;
102: losing training metric function using triple, keep the feature of identical pedestrian distance in new mapping space most
Measure small, distance is big as far as possible between the feature of different new persons.
This method specifically includes that semantic region is extracted, the image characteristics extraction in semantic region, image it is global and local
Region Feature Extraction, the calculating of similarity between image pattern utilize ternary loss to carry out five parts of training of metric learning.
In conclusion the embodiment of the present invention deeply excavates the region in image with semantic information through the above steps, mention
The high precision of algorithm retrieval.
Embodiment 2
The scheme in embodiment 1 is further introduced below with reference to specific calculation formula, example, it is as detailed below
Description:
One, the image characteristics extraction in semantic region
Pedestrian's weight recognizer comes out the pedestrian detection in monitor video first with pedestrian detection algorithm, further according to figure
Similitude as between finds correct matched target.In the rectangle frame obtained due to pedestrian detection algorithm other than pedestrian, also
Comprising certain background information, if the global characteristics using image are matched, this feature can include many background informations
Interference, to be brought greater impact to matched precision.
Additionally due to the presence of the factors such as visual angle and the transformation of pedestrian's posture, the position of pedestrian is misaligned between image,
It is possible that a certain region for piece image occur belongs to the upper part of the body of pedestrian, and belong in the same area of another piece image another
The lower part of the body of a pedestrian, to the comparison of mistake occur.
Therefore, the embodiment of the present invention is extracted in image in pedestrian's difference body part region according to the semantic information of pedestrian
Feature, in avoiding image background information interfere while, also achieve the comparison of corresponding region similitude between image, thus
Effectively improve the robustness of feature.
Given piece image, carries out component segmentation to wherein pedestrian first, and the position of pedestrian's difference component in image is examined
It measures and, so as to obtain the set comprising different component areas:
RP={ rP, 1..., rP, i..., rP, M}
Wherein, RPThat indicate is the set of component area, rP, iWhat is indicated is any component area, each type
Component area be expressed as a r, what subscript P was indicated is component area, what M was indicated be share M component area, such as: work as M
When equal to 4, the set of component area then includes: head, trunk, left leg, right leg region.
In each subregion, feature, sliding window stepping in the horizontal direction are extracted using the sliding window of 16 × 8 pixel sizes
Length is 8 pixels, and stepping length is 4 pixels in the vertical direction.In each sliding window, respectively extract color histogram feature and
Textural characteristics.For color histogram feature, from HSV (form and aspect, saturation degree, brightness) color space and LAB, (color-is right respectively
It is vertical) two kinds of color histogram is extracted in color space, one is three channels are cascaded, extract 48-bins (group away from)
Histogram, another histogram for individually extracting 8-bins (group away from) to each Color Channel respectively.
For Texture similarity, SILTP (Scale invariant local mode) descriptor is used.It finally will be in each region
Feature cascade, and dimensionality reduction is carried out to primitive character using PCA (principal component analysis) algorithm.It is available for M component area
The feature of corresponding M component area indicates are as follows:
FP={ fP, 1..., fP, i..., fP, M}
Wherein, FPThat indicate is the set of component area feature, fP, iWhat is indicated is the feature of i-th of component area.
Two, the global and local Region Feature Extraction of image
In order to obtain more stable, comprehensive feature description, other than the feature in semantic region, the embodiment of the present invention
Also introduce the feature in global area and regional area.What wherein global area indicated is entire image, with symbol RGCarry out table
Show.Regional area then indicates the subregion by image according to divided in horizontal direction at several horizontal bands, with symbol RL=
{rL, 1..., rL, i..., rL, CIndicate.Wherein, RLIndicate the set of regional area, rL, iIndicate i-th of local subregion,
C indicates the number of local subregion.
For every class region, each subregion is traversed using the method for sliding window, it is straight that color is equally extracted in each sliding window
Square figure feature and SILTP textural characteristics, and the feature that sliding windows all in each subregion are extracted cascades, and carries out PCA spy
Dimensionality reduction is levied, the feature in the subregion is obtained.
Use FGIt indicates global characteristics, uses FL={ fL, 1..., fL, i..., fL, CIndicate local features, wherein
FLIt is the set of local features, fL, iWhat is indicated is the feature of i-th of local subregion.
Last three kinds of different types of regions can indicate with set R, wherein R={ RP, RG, RL}.Wrapped in set
It includes: component area, global area and regional area.
Pass through the component area feature F using pedestrianPAnd global characteristics FGWith local features FL, can be from difference
Picture is described on scale level, therefore new feature can be made to describe more comprehensively, reliably, and with F={ FP, FG, FLAs expression
The feature descriptor of the image.
Three, between image pattern similarity calculating
In order to meet the distance between identical sample of identity target sample the distance between different less than identity, this hair
Bright embodiment is lost using triple, obtains new metric matrix by training and has original maps feature vectors to one
In the feature space of discrimination power.
In new feature space, the distance between identical sample can be reduced, without between same target feature away from
From that will be zoomed out, so that being more easily discriminated between sample, to improve the accuracy rate that pedestrian identifies again.In new mapping space
Interior, the similitude between sample indicates are as follows:
Wherein, Fa, FbThe characteristics of image descriptor being illustrated respectively under the different monitoring visual field,<,>expression is inner product
Operation.What is indicated is the mahalanobis distance of two features,It indicates
Be two features bilinearity distance.WMAnd WBIt is the metric matrix to be obtained by study.
Four, the training for carrying out metric learning is lost using ternary.
In order to obtain more satisfactory metric matrix by training, triple loss is expressed as follows:
g(Xi, Yi)≥α+g(Xi, Yj),
Wherein, X, Y respectively indicate the characteristics of image under the different monitoring visual field, and i and j respectively indicate the identity mark of different pedestrians
Label.g(Xi, Yi) that indicate is superficial similarities of the same target under the different monitoring visual field, g (Xi, Yj) what is indicated is different
Superficial similarities of the target under the different monitoring visual field.α indicates interval threshold, that is, requires the phase between same target sample
It is α bigger than the superficial similarities score between not same target like property score.
In order to utilize the characteristic information training metric matrix of image in entire data set, triple loss building are as follows:
Wherein, []+It is indicator function, when the numerical value in function is less than or equal to 0, output 0, when the numerical value in function is big
When 0, the numerical value is exported.The formula requires the similarity score between any one identical sample than all dissimilar samples
Between similarity score the big α of average value.
Due to requiring when similitude is bigger between sample, score is higher, therefore WMBe negative positive semidefinite matrix, additionally utilizes
l2,1Norm does the regularization term of objective function, last objective function are as follows:
Wherein, R (W)=| | WM||2,1+||WB||2,1For regularization term, S-The positive semidefinite matrix set being negative, λ are weight
?.In order to keep objective function minimum, optimal solution W is sought using alternating direction multipliers method.Therefore objective function is equivalent to:
Wherein, W1, W2, W3It is the variable of three different subproblems, works as W3Be negative positive semidefinite matrix when, S (W3) it is zero, it is on the contrary
It is then infinite.Each variable be by fixing its dependent variable, what re-optimization solved.Wherein W1According to the gradient of dual form
It is solved, W2By fixing its dependent variable, solved using neighbour's operator.W3It is to be obtained by European mapping solution.Through
After crossing multiple iteration optimization, an available optimal metric matrix.
In test phase, it is special to extract image respectively for a given image and all images to be retrieved for needing to inquire
Sign, and the metric matrix obtained using study, calculate the similitude between image pair.According to the similarity score between image pair
It is ranked up, and then finds most like image, to realize that pedestrian identifies again.
In conclusion the embodiment of the present invention is in addition to also using energy using the component area information for being more concerned about local detail
Enough global and local provincial characteristics that image is described in large scale.Three kinds of features describe picture from different scale layers time are upper,
It is incorporated into available to the more comprehensive feature descriptor of iamge description.
Embodiment 3
Feasibility verifying is carried out to the scheme in Examples 1 and 2 below with reference to Fig. 1, described below:
Fig. 2 gives this method and identifies the test result on data set VIPeR again in general pedestrian, and ordinate is CMC tired
Matching score is counted, abscissa is ranking.What wherein Rank1 was indicated is the probability that correct matched target ranking is in first,
What Rank10 was indicated is the probability that correct matched target is in front of ranking ten.From figure 2 it can be seen that by utilizing pedestrian's
Semantic component information, this method achieve best result in different evaluation indexes.
Results of property best at present under the data set is achieved on Rank1, and is achieved on Rank10 and be more than
90% precision shows that most correct matched sample standard deviations appear in the top ten list of search result, therefore can be
Largely reduce manpower and material resources, and rapidly and efficiently find correct matched target.
It will be appreciated by those skilled in the art that attached drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (4)
1. a kind of pedestrian's recognition methods again based on semantic region expression, which is characterized in that the described method comprises the following steps:
Given piece image, the position detection of pedestrian's difference component in image is come out, the collection comprising different component areas is obtained
It closes:
The high-layer semantic information being had according to pedestrian is described using Scale invariant local mode descriptor and color histogram
Symbol, extracts the feature in each component area and is cascaded, reuse principal component analysis and drop to primitive character
Dimension, the set of obtaining widget provincial characteristics;
The feature more complete to pedestrian is obtained according to the set of component area feature, the global characteristics in conjunction with image and local feature
Descriptor;
Training metric function is lost using triple, the metric matrix that training obtains can be by original maps feature vectors to newly
Feature space in, improve sample between separability;
The similitude between pedestrian image feature is calculated according to the metric matrix that study obtains, to realize the knowledge again to pedestrian
Not.
2. a kind of pedestrian's recognition methods again based on semantic region expression according to claim 1, which is characterized in that described
The set of component area feature specifically:
FP={ fP, 1..., fP, i..., fP, M}
Wherein, FPThat indicate is the set of component area feature, fP, iWhat is indicated is the feature of i-th of segmentation subregion.
3. a kind of pedestrian's recognition methods again based on semantic region expression according to claim 1, which is characterized in that described
Feature descriptor specifically:
F={ FP, FG, FL}
Wherein, FPFor component area feature;FGFor global characteristics;FLFor local features.
4. a kind of pedestrian's recognition methods again based on semantic region expression according to claim 1, which is characterized in that described
Similarity specifically:
Wherein, Fa, FbThe characteristics of image descriptor being illustrated respectively under the different monitoring visual field,<,>expression is inner product fortune
It calculates,What is indicated is the mahalanobis distance between characteristics of image,It indicates
It is the bilinearity distance between characteristics of image, WMAnd WBIt is metric matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810930107.4A CN109344842A (en) | 2018-08-15 | 2018-08-15 | A kind of pedestrian's recognition methods again based on semantic region expression |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810930107.4A CN109344842A (en) | 2018-08-15 | 2018-08-15 | A kind of pedestrian's recognition methods again based on semantic region expression |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109344842A true CN109344842A (en) | 2019-02-15 |
Family
ID=65291563
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810930107.4A Pending CN109344842A (en) | 2018-08-15 | 2018-08-15 | A kind of pedestrian's recognition methods again based on semantic region expression |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109344842A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109977798A (en) * | 2019-03-06 | 2019-07-05 | 中山大学 | The exposure mask pond model training identified again for pedestrian and pedestrian's recognition methods again |
CN110427814A (en) * | 2019-06-24 | 2019-11-08 | 深圳云天励飞技术有限公司 | A kind of bicyclist recognition methods, device and equipment again |
CN110490253A (en) * | 2019-08-19 | 2019-11-22 | 浙江工业大学 | A kind of image similarity calculation method based on spatial color EMD |
CN110516623A (en) * | 2019-08-29 | 2019-11-29 | 中新智擎科技有限公司 | A kind of face identification method, device and electronic equipment |
CN111046732A (en) * | 2019-11-11 | 2020-04-21 | 华中师范大学 | Pedestrian re-identification method based on multi-granularity semantic analysis and storage medium |
CN111191602A (en) * | 2019-12-31 | 2020-05-22 | 深圳云天励飞技术有限公司 | Pedestrian similarity obtaining method and device, terminal equipment and readable storage medium |
CN111914844A (en) * | 2019-05-07 | 2020-11-10 | 杭州海康威视数字技术股份有限公司 | Image identification method and device, electronic equipment and storage medium |
CN112200009A (en) * | 2020-09-15 | 2021-01-08 | 青岛邃智信息科技有限公司 | Pedestrian re-identification method based on key point feature alignment in community monitoring scene |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106250870A (en) * | 2016-08-16 | 2016-12-21 | 电子科技大学 | A kind of pedestrian's recognition methods again combining local and overall situation similarity measurement study |
CN106485253A (en) * | 2016-09-14 | 2017-03-08 | 同济大学 | A kind of pedestrian of maximum particle size structured descriptor discrimination method again |
US20170091439A1 (en) * | 2015-09-30 | 2017-03-30 | Apple Inc. | System and Method for Person Reidentification |
CN106874884A (en) * | 2017-03-03 | 2017-06-20 | 中国民航大学 | Human body recognition methods again based on position segmentation |
CN107506703A (en) * | 2017-08-09 | 2017-12-22 | 中国科学院大学 | A kind of pedestrian's recognition methods again for learning and reordering based on unsupervised Local Metric |
CN107506700A (en) * | 2017-08-07 | 2017-12-22 | 苏州经贸职业技术学院 | Pedestrian's recognition methods again based on the study of broad sense similarity measurement |
CN107563344A (en) * | 2017-09-18 | 2018-01-09 | 天津师范大学 | A kind of pedestrian's recognition methods again that study is estimated based on semantic region |
CN107944340A (en) * | 2017-10-25 | 2018-04-20 | 宁波大学 | A kind of combination is directly measured and the pedestrian of indirect measurement recognition methods again |
CN108108754A (en) * | 2017-12-15 | 2018-06-01 | 北京迈格威科技有限公司 | The training of identification network, again recognition methods, device and system again |
CN108345900A (en) * | 2018-01-12 | 2018-07-31 | 浙江大学 | Pedestrian based on color and vein distribution characteristics recognition methods and its system again |
-
2018
- 2018-08-15 CN CN201810930107.4A patent/CN109344842A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170091439A1 (en) * | 2015-09-30 | 2017-03-30 | Apple Inc. | System and Method for Person Reidentification |
CN106250870A (en) * | 2016-08-16 | 2016-12-21 | 电子科技大学 | A kind of pedestrian's recognition methods again combining local and overall situation similarity measurement study |
CN106485253A (en) * | 2016-09-14 | 2017-03-08 | 同济大学 | A kind of pedestrian of maximum particle size structured descriptor discrimination method again |
CN106874884A (en) * | 2017-03-03 | 2017-06-20 | 中国民航大学 | Human body recognition methods again based on position segmentation |
CN107506700A (en) * | 2017-08-07 | 2017-12-22 | 苏州经贸职业技术学院 | Pedestrian's recognition methods again based on the study of broad sense similarity measurement |
CN107506703A (en) * | 2017-08-09 | 2017-12-22 | 中国科学院大学 | A kind of pedestrian's recognition methods again for learning and reordering based on unsupervised Local Metric |
CN107563344A (en) * | 2017-09-18 | 2018-01-09 | 天津师范大学 | A kind of pedestrian's recognition methods again that study is estimated based on semantic region |
CN107944340A (en) * | 2017-10-25 | 2018-04-20 | 宁波大学 | A kind of combination is directly measured and the pedestrian of indirect measurement recognition methods again |
CN108108754A (en) * | 2017-12-15 | 2018-06-01 | 北京迈格威科技有限公司 | The training of identification network, again recognition methods, device and system again |
CN108345900A (en) * | 2018-01-12 | 2018-07-31 | 浙江大学 | Pedestrian based on color and vein distribution characteristics recognition methods and its system again |
Non-Patent Citations (2)
Title |
---|
DAPENG CHEN ET AL.: "Similarity Learning on an Explicit Polynomial Kernel Feature Map for Person Re-Identification", 《2015CVPR》 * |
李鹏: "视频浓缩系统中的行人目标再辨识技术", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109977798A (en) * | 2019-03-06 | 2019-07-05 | 中山大学 | The exposure mask pond model training identified again for pedestrian and pedestrian's recognition methods again |
CN109977798B (en) * | 2019-03-06 | 2021-06-04 | 中山大学 | Mask pooling model training and pedestrian re-identification method for pedestrian re-identification |
CN111914844A (en) * | 2019-05-07 | 2020-11-10 | 杭州海康威视数字技术股份有限公司 | Image identification method and device, electronic equipment and storage medium |
CN110427814A (en) * | 2019-06-24 | 2019-11-08 | 深圳云天励飞技术有限公司 | A kind of bicyclist recognition methods, device and equipment again |
CN110490253A (en) * | 2019-08-19 | 2019-11-22 | 浙江工业大学 | A kind of image similarity calculation method based on spatial color EMD |
CN110516623A (en) * | 2019-08-29 | 2019-11-29 | 中新智擎科技有限公司 | A kind of face identification method, device and electronic equipment |
CN110516623B (en) * | 2019-08-29 | 2022-03-22 | 中新智擎科技有限公司 | Face recognition method and device and electronic equipment |
CN111046732A (en) * | 2019-11-11 | 2020-04-21 | 华中师范大学 | Pedestrian re-identification method based on multi-granularity semantic analysis and storage medium |
CN111046732B (en) * | 2019-11-11 | 2023-11-28 | 华中师范大学 | Pedestrian re-recognition method based on multi-granularity semantic analysis and storage medium |
CN111191602A (en) * | 2019-12-31 | 2020-05-22 | 深圳云天励飞技术有限公司 | Pedestrian similarity obtaining method and device, terminal equipment and readable storage medium |
CN111191602B (en) * | 2019-12-31 | 2023-06-13 | 深圳云天励飞技术有限公司 | Pedestrian similarity acquisition method and device, terminal equipment and readable storage medium |
CN112200009A (en) * | 2020-09-15 | 2021-01-08 | 青岛邃智信息科技有限公司 | Pedestrian re-identification method based on key point feature alignment in community monitoring scene |
CN112200009B (en) * | 2020-09-15 | 2023-10-17 | 青岛邃智信息科技有限公司 | Pedestrian re-identification method based on key point feature alignment in community monitoring scene |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109344842A (en) | A kind of pedestrian's recognition methods again based on semantic region expression | |
CN110414368B (en) | Unsupervised pedestrian re-identification method based on knowledge distillation | |
CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
CN107423737B (en) | The video quality diagnosing method that foreign matter blocks | |
CN104268583B (en) | Pedestrian re-recognition method and system based on color area features | |
CN105701467B (en) | A kind of more people's abnormal behaviour recognition methods based on human figure feature | |
CN103996046B (en) | The personal identification method merged based on many visual signatures | |
US20210034840A1 (en) | Method for Recognzing Face from Monitoring Video Data | |
CN104978567B (en) | Vehicle checking method based on scene classification | |
CN105528794A (en) | Moving object detection method based on Gaussian mixture model and superpixel segmentation | |
CN106991370B (en) | Pedestrian retrieval method based on color and depth | |
CN111242027B (en) | Unsupervised learning scene feature rapid extraction method fusing semantic information | |
CN104376334B (en) | A kind of pedestrian comparison method of multi-scale feature fusion | |
CN109086672A (en) | A kind of recognition methods again of the pedestrian based on reinforcement learning adaptive piecemeal | |
CN109800624A (en) | A kind of multi-object tracking method identified again based on pedestrian | |
CN106846378B (en) | A kind of across the video camera object matching and tracking of the estimation of combination topology of spacetime | |
CN104751136A (en) | Face recognition based multi-camera video event retrospective trace method | |
CN104881671B (en) | A kind of high score remote sensing image Local Feature Extraction based on 2D Gabor | |
CN104077605A (en) | Pedestrian search and recognition method based on color topological structure | |
CN102521565A (en) | Garment identification method and system for low-resolution video | |
WO2021082168A1 (en) | Method for matching specific target object in scene image | |
CN109271932A (en) | Pedestrian based on color-match recognition methods again | |
Park et al. | Detection of construction workers in video frames for automatic initialization of vision trackers | |
CN110008828A (en) | Pairs of constraint ingredient assay measures optimization method based on difference regularization | |
CN109214430B (en) | Pedestrian re-identification method based on feature space topological distribution |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190215 |