CN103325122A - Pedestrian retrieval method based on bidirectional sequencing - Google Patents

Pedestrian retrieval method based on bidirectional sequencing Download PDF

Info

Publication number
CN103325122A
CN103325122A CN2013102773599A CN201310277359A CN103325122A CN 103325122 A CN103325122 A CN 103325122A CN 2013102773599 A CN2013102773599 A CN 2013102773599A CN 201310277359 A CN201310277359 A CN 201310277359A CN 103325122 A CN103325122 A CN 103325122A
Authority
CN
China
Prior art keywords
pedestrian
measured
similarity
checked
neighbour
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
Application number
CN2013102773599A
Other languages
Chinese (zh)
Other versions
CN103325122B (en
Inventor
胡瑞敏
冷清明
梁超
王亦民
陈军
张华�
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN201310277359.9A priority Critical patent/CN103325122B/en
Publication of CN103325122A publication Critical patent/CN103325122A/en
Application granted granted Critical
Publication of CN103325122B publication Critical patent/CN103325122B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a pedestrian retrieval method based on bidirectional sequencing, belonging to the technical field of monitoring video retrieval. According to the method, firstly, the initial sequencing of a people set to be tested is carried out through feature extraction and scale learning of a pedestrian object, then each row of people in the people set to be tested is reversely inquired, the bidirectional content similarity and neighbor similarity of pedestrian to be inquired and pedestrian to be tested are calculated, and finally each pedestrian to be tested in the people set is rearranged according to the bidirectional content similarity and the neighbor similarity. According to the method, a two-way matching idea is introduced, the pedestrian is rearranged through the content and neighbor similarities, a more accurate pedestrian retrieval result can be obtained, and the pedestrian appearance change caused by the environmental change is more robust.

Description

Pedestrian's search method based on Bidirectional sort
Technical field
The invention belongs to monitor video retrieval technique field, relate in particular to a kind of pedestrian's search method based on Bidirectional sort.
Background technology
Monitor video pedestrian retrieval is the technology of the specific pedestrian's object of coupling under the non-overlapping multi-cam of irradiation area.In the actual video investigation, the investigator is main to come quick lock in, investigation and tracking suspicion target according to moving frame and track with a group traveling together's object.The video investigative mode that Traditional Man is browsed need to expend a large amount of manpowers and time, easily affects adversely the opportunity of solving a case.The heavy recognition technology of pedestrian is convenient to the video investigator and is found quickly and accurately suspicion goal activities picture and track, and public security department is improved case-solving rate, safeguards that life property safety of people is significant.
Existing pedestrian's retrieval (claiming that again the pedestrian heavily identifies) method can be divided into two classes: first kind method is mainly constructed the visual signature of robust, and then the distance function of Application standard (such as Euclidean distance etc.) carries out similarity measurement.For example document 1(is referring to M.Farenzena, L.Bazzani, A.Perina, V.Murino, and M.Cristani, " Person re-identification by symmetry-driven accumulation of local features ", IEEE Conf.on Computer Vision and Pattern Recognition (CVPR), PP.2360 – 2367,2010.) proposition is based on the heavily recognition methods of pedestrian of many local features couplings of symmetry division.At first, utilizing the color characteristic clue that health is carried out horizontal and vertical cuts apart; Secondly, extract multiple color feature and the textural characteristics of each cut zone, and the color characteristic and the textural characteristics that extract based on horizontal axis weighting obtain the comprehensive visual feature; At last, utilize above-mentioned comprehensive visual feature to carry out expression and the coupling of object.
The Equations of The Second Kind method without strict demand, is mainly carried out more accurately distance metric by learning a suitable yardstick for latent structure.For example document 2(is referring to M.Kostinger, M.Hirzer, P.Wohlhart, P.M.Roth, and H.Bischof, " Large scale metric learning from equivalence constraints ", in Computer Vision and Pattern Recognition (CVPR), PP.2288-2295,2012.) difference vector of similar sample and the difference vector of different samples are expressed as different Gaussian distribution; Then, the ratio of employing probability is measured the distance between the sample; Finally, the ratio of Gaussian distribution is converted to the form of mahalanobis distance, thereby learn a suitable mahalanobis distance function.
It all is according to the distance of inquiring about pedestrian's object and all pedestrian's object macroscopic featuress to be measured collection to be measured to be sorted that above line people retrieves method.Yet under the actual video monitoring environment, the factors such as the visual angle of different cameras, illumination, aberration are different, cause often having significant difference with the macroscopic features of a group traveling together under multi-cam, thereby so that result for retrieval are inaccurate.
Summary of the invention
Deficiency for prior art exists the invention provides a kind of pedestrian's search method based on Bidirectional sort, and the method can promote the accuracy of mating with a group traveling together under the multi-cam.
For solving the problems of the technologies described above, the present invention adopts following technical scheme:
A kind of pedestrian's search method based on Bidirectional sort comprises step:
Step 1 is measured pedestrian p to be checked and pedestrian to be measured and is collected G={g i| i=1 ..., each pedestrian g to be measured among the n} iSimilarity, obtain pedestrian to be measured and collect each pedestrian g to be measured among the G iThe forward ranking results, obtain pedestrian p to be checked and pedestrian g to be measured according to the forward ranking results iForward content similarity
Figure BDA00003457178500021
N is that pedestrian to be measured collects pedestrian's quantity to be measured among the G;
Step 2 makes up pedestrian g to be measured iPedestrian to be measured collection
Figure BDA00003457178500022
Measure pedestrian g to be measured iWith pedestrian's collection to be measured
Figure BDA00003457178500023
In each pedestrian's similarity, obtain pedestrian's collection to be measured In each pedestrian's sorting by reversals result, and obtain pedestrian p to be checked and pedestrian g to be measured according to the sorting by reversals result iThe reverse content similarity
Figure BDA00003457178500025
Step 3, the neighbour's collection and the neighbour far away that obtain pedestrian p to be checked according to the forward ranking results collect, and obtain pedestrian g to be measured according to the sorting by reversals result iNeighbour collection and adjacent collection far away, based on pedestrian p to be checked and pedestrian g to be measured iNeighbour collection and adjacent collection far away obtain pedestrian p to be checked and pedestrian g to be measured iNeighbour's similarity, described neighbour's similarity comprises neighbour's similarity S Near(p, g i) and adjacent similarity S far away Far(p, g i);
Step 4 considers pedestrian p to be checked and pedestrian g to be measured iForward content similarity The reverse content similarity
Figure BDA00003457178500027
With neighbour's similarity, pedestrian to be measured is collected each pedestrian g to be measured among the G iRearrangement.
All measure similarity based on macroscopic features in above-mentioned steps one and the step 2.
Above-mentioned pedestrian p to be checked and pedestrian g to be measured iForward content similarity
Figure BDA00003457178500028
According to pedestrian g to be measured iSequence number in the forward ranking results obtains.
Above-mentioned pedestrian p to be checked and pedestrian g to be measured iThe reverse content similarity
Figure BDA00003457178500031
Obtain according to the sequence number of pedestrian p to be checked in the sorting by reversals result.
The neighbour of above-mentioned pedestrian p to be checked collection is: in the forward ranking results with k pedestrian's to be measured of pedestrian p similarity minimum to be checked set n k(p), wherein, k is value rule of thumb.
The far away adjacent collection of above-mentioned pedestrian p to be checked is: in the forward ranking results with k' pedestrian's to be measured of pedestrian p similarity maximum to be checked set n K'(p), wherein, k' is value rule of thumb.
Above-mentioned pedestrian g to be measured iNeighbour collection be: among the sorting by reversals result with pedestrian g to be measured iThe k of similarity minimum pedestrian's to be measured set n k(g i), wherein, k is value rule of thumb.
Above-mentioned pedestrian g to be measured iFar away adjacent collection be: among the sorting by reversals result with pedestrian g to be measured iThe k' of similarity maximum pedestrian's to be measured set n K'(g i), wherein, k' is value rule of thumb.
Above-mentioned pedestrian p to be checked and pedestrian g to be measured iNeighbour's similarity S Near(p, g i) collect n according to the neighbour of pedestrian p to be checked k(p) and pedestrian g to be measured iThe neighbour collect n k(g i) obtain, be specially n k(p) and n k(g i) common factor in the number of element.
Above-mentioned pedestrian p to be checked and pedestrian g to be measured iFar away adjacent similarity S Far(p, g i) collect n according to the neighbour far away of pedestrian p to be checked K'(p) and pedestrian g to be measured iFar away adjacent collection n K'(g i) obtain, be specially n K'(p) and n K'(g i) common factor in the number of element.
A kind of embodiment of above-mentioned steps four is:
According to pedestrian p to be checked and pedestrian g to be measured iForward content similarity
Figure BDA00003457178500032
The reverse content similarity
Figure BDA00003457178500033
Obtain comprehensive similarity S with neighbour's similarity *(g i), based on comprehensive similarity S *(g i) pedestrian to be measured is collected G={g i| i=1 ..., each pedestrian g to be measured among the n} iResequence described comprehensive similarity S * ( g i ) = S p g i × S g i p α + S near ( p , g i ) + S p g i × S g i p α + S far ( p , g i ) , Wherein, S Near(p, g i) and S Far(p, g i) be respectively pedestrian p to be checked and pedestrian g to be measured iNeighbour's similarity and adjacent similarity far away,
Figure BDA00003457178500035
With
Figure BDA00003457178500036
Be respectively pedestrian p to be checked and pedestrian g to be measured iForward content similarity and reverse content similarity, α is one and prevents that denominator from being 0 constant, 0≤α≤1.
Compared with prior art, the present invention has the following advantages and beneficial effect:
1) the present invention introduces the thought of bi-directional matching, resets pedestrian to be measured by content and neighbour's similarity, can obtain more accurately pedestrian's result for retrieval;
2) the present invention has not only considered the similarity in pedestrian's characteristics of objects space, has also considered the similarity in neighbour space, and the pedestrian's appearance that causes for environmental change changes more robust.
Description of drawings
Fig. 1 is the inventive method process flow diagram.
Embodiment
The present invention is based on pedestrian's search method of Bidirectional sort, at first, the feature extraction by pedestrian's object and scale learning initially sort to pedestrian's collection to be measured; Then, Query pedestrian to be measured concentrates each pedestrian, and calculates pedestrian to be checked and pedestrian's to be measured two-way content similarity and neighbour's similarity; At last, reset pedestrian to be measured according to two-way content similarity and neighbour's similarity and concentrate each pedestrian to be measured.
Pedestrian p to be checked is pedestrian's object that camera A takes among the present invention, and pedestrian to be measured is pedestrian's object that camera B takes, and pedestrian's object that camera B takes consists of pedestrian's collection to be measured.The present invention is exactly pedestrian's object that to find with pedestrian p to be checked from pedestrian's object that camera B takes be same people.
Fig. 1 is the inventive method process flow diagram, below in conjunction with Fig. 1 the implementation of the inventive method is specified.
Following implementation adopts MATLAB7 as Simulation Experimental Platform, tests at pedestrian's retrieve data collection VIPeR commonly used.The VIPeR data set has 632 the pedestrian's images pair under two cameras, has the differences such as obvious visual angle, illumination between two cameras.Describe this implementation in detail below in conjunction with each step.
1, concentrates each pedestrian to be measured initially to sort to pedestrian to be measured, and obtain pedestrian to be checked and concentrated each pedestrian's to be measured of pedestrian to be measured forward content similarity.
1. feature extraction
Pedestrian to be checked is expressed as p, and pedestrian to be measured collects G={g i| i=1 ..., n}, n are the sizes of pedestrian's collection to be measured, namely pedestrian to be measured concentrates element number.In this implementation, select at random among pedestrian's retrieve data collection VIPeR a half data to do sample training, second half does test, so, and n=316.
Pedestrian's image carries out piecemeal by the window of 8 * 16 pixels with 8 * 8 step-length, extracts color characteristic and textural characteristics on each piece.Color characteristic uses RGB color characteristic (wherein, R represents red, and G represents green, and B represents indigo plant) and HSV feature (wherein, H represents colourity, and S represents saturation degree, and V represents brightness), and every kind of mark sheet is shown as 24 dimensions.Textural characteristics uses the LBP(local binary patterns), be expressed as 59 dimensions.The feature of all pieces adds up and the PCA(principal component analysis (PCA)) to the macroscopic features of 400 Wesys in expression pedestrian object.
2. scale learning
Utilize step 1. in 1 feature extracting method represent training sample, then, by scale learning method study mahalanobis distance function.The scale learning method is generally carried out on the basis of mahalanobis distance, and the distance matrix metric that obtains according to sample training replaces the covariance matrix M in the mahalanobis distance.
Given two pedestrian's object images O aAnd O b, both distance D (O a, O b) be defined as:
D(O a,O b)＝(O a-O b) TM(O a-O b) （1）
Wherein, M is a semi-definite matrix, (O a-O b) T(O a-O b) transposition.At minimum above-mentioned distance D (O a, O b) when finding the solution M, use random Gradient Descent to carry out Algorithm Learning.
3. initial forward ordering
Utilize the 2. distance metric function D (O of middle distance learning method acquisition of step a, O b) tolerance pedestrian p to be checked and pedestrian g to be measured iSimilarity, calculate respectively pedestrian p to be checked and each pedestrian g to be measured iApart from d (p, g i).According to d (p, g i) size pedestrian to be measured is collected each pedestrian g to be measured among the G iSort, obtain initial forward ranking results tabulation
Figure BDA00003457178500051
Figure BDA00003457178500052
I pedestrian to be measured in the tabulation of forward ranking results, wherein, d ( p , g 1 0 ) ≤ . . . ≤ d ( p , g i 0 ) ≤ . . . ≤ d ( p , g n 0 ) .
Adopt score value S (p, g 1) expression pedestrian p to be checked and pedestrian g to be measured iBetween forward content similarity, its value equals pedestrian g to be measured iSequence number in the tabulation of forward ranking results.Score value S (p, g 1) write a Chinese character in simplified form into
2, obtain the reverse content similarity that pedestrian to be checked and pedestrian to be measured concentrate each pedestrian to be measured.
Pedestrian to be measured is collected pedestrian g to be measured among the G iPedestrian's collection to be measured that structure is corresponding
Figure BDA00003457178500055
Pedestrian's collection to be measured
Figure BDA00003457178500056
Comprise that pedestrian p to be checked and pedestrian to be measured collect among the G except g iOuter pedestrian's object.
Utilize the 2. distance metric function D (O of middle distance learning method acquisition of step 1 substep a, O b) tolerance pedestrian g to be measured iWith pedestrian's collection to be measured
Figure BDA00003457178500061
In the similarity of each pedestrian's object, according to similarity from small to large, obtain pedestrian's collection to be measured
Figure BDA00003457178500062
In sorting by reversals the results list of each pedestrian's object, and the sequence number of pedestrian p to be checked on sorting by reversals the results list represented pedestrian p to be checked and pedestrian g to be measured iThe reverse content similarity, be designated as S (g 1, p), be abbreviated as
Figure BDA00003457178500063
Obtain pedestrian p to be checked and pedestrian g to be measured in conjunction with forward content similarity and reverse content similarity iContent similarity S Cn(g i):
S cn ( g i ) = S P g i × S g i p - - - ( 2 )
3, obtain neighbour's similarity that pedestrian to be checked and pedestrian to be measured concentrate each pedestrian to be measured.
According to the forward ranking results tabulation that step 1 generates, get before the tabulation of forward ranking results the neighbour that k pedestrian to be measured consist of pedestrian p to be checked and collect n k(p), get the far away adjacent collection n that rear k' the pedestrian to be measured of forward ranking results tabulation consists of pedestrian p to be checked K '(p).
According to sorting by reversals the results list that step 2 generates, negate consists of pedestrian g to be measured to front k the pedestrian's object of ranking results tabulation iThe neighbour collect n k(g i), negate consists of pedestrian g to be measured to rear k' the pedestrian's object of ranking results tabulation iFar away adjacent collection n K'(g i).
K and k' be value rule of thumb, and in this implementation, the k value is preferably 20～40, k' span and is preferably 80～120.Make k=30, k'=100.Calculate pedestrian p to be checked and pedestrian g to be measured iK neighbour space and the similarity between k' adjacent air space far away.Pedestrian p to be checked and pedestrian g to be measured iNeighbour's similarity S Near(p, g i) be n k(p) and n k(g i) common factor in the number of element, as follows:
S near(p,g i)＝|n k(p)∩n k(g i)| （3）
Pedestrian p to be checked and pedestrian g to be measured iFar away adjacent similarity S Far(p, g i) be n K'(p) and n K'(g i) common factor in the number of element, as follows:
S far(p,g i)＝|n k'(p)∩n k'(g i)| （4）
4, comprehensive pedestrian p to be checked and pedestrian g to be measured iContent similarity and neighbour's similarity, pedestrian to be measured is collected each pedestrian g to be measured among the G iRearrangement.
Comprehensive pedestrian p to be checked and pedestrian g to be measured iContent similarity and neighbour's similarity obtain pedestrian p to be checked and pedestrian g to be measured iFinal similarity S *(g i):
S * ( g i ) = S cn ( g i ) S cx ( g i ) = S p g i × S g i p α + | n k ( p ) ∩ n k ( g i ) | + S p g i × S g i p α + | n k ′ ( p ) ∩ n k ′ ( g i ) | - - - ( 5 )
Wherein:
S Cn(g i) be used for describing pedestrian p to be checked and pedestrian g to be measured iThe content similarity;
S Cx(g i) be used for describing pedestrian p to be checked and pedestrian g to be measured iNeighbour's similarity;
α is one and prevents that denominator from being 0 constant, and 0≤α≤1 makes α=0.01 in this implementation.
According to final similarity S *(g i) size resets pedestrian to be measured and collect each pedestrian g to be measured among the G i, final similarity S *(g i) less, the pedestrian g to be measured of its correspondence iSort more forward.
Adopt CMC pedestrian to retrieve evaluation index and estimate the inventive method.The CMC value refers to return the probability that correct pedestrian's object is arranged among the front r result in N the inquiry.When returning front r as a result the time, the CMC value is higher, and expression pedestrian retrieval performance is better.Repeat the test process 10 times of this implementation, calculate the average CMC value that repeats for its 10 times, contrast mahalanobis distance scale learning and the average CMC value that reorders see Table 1.Can find from table 1, the pedestrian of the present invention heavily retrieval performance of recognition methods obviously is better than contrasting algorithm.
1,10,25,50 average CMC value as a result time the before table 1 returns respectively on VIPeR

Claims (7)

1. the pedestrian's search method based on Bidirectional sort is characterized in that, comprises step:
Step 1 is measured pedestrian p to be checked and pedestrian to be measured and is collected G={g i| i=1 ..., each pedestrian g to be measured among the n} iSimilarity, obtain pedestrian to be measured and collect each pedestrian g to be measured among the G iThe forward ranking results, obtain pedestrian p to be checked and pedestrian g to be measured according to the forward ranking results iForward content similarity
Figure FDA00003457178400011
N is that pedestrian to be measured collects pedestrian's quantity to be measured among the G;
Step 2 makes up pedestrian g to be measured iPedestrian to be measured collection Measure pedestrian g to be measured iWith pedestrian's collection to be measured
Figure FDA00003457178400013
In each pedestrian's similarity, obtain pedestrian's collection to be measured
Figure FDA00003457178400014
In each pedestrian's sorting by reversals result, and obtain pedestrian p to be checked and pedestrian g to be measured according to the sorting by reversals result iThe reverse content similarity
Figure FDA00003457178400015
Step 3, the neighbour's collection and the neighbour far away that obtain pedestrian p to be checked according to the forward ranking results collect, and obtain pedestrian g to be measured according to the sorting by reversals result iNeighbour collection and adjacent collection far away, based on pedestrian p to be checked and pedestrian g to be measured iNeighbour collection and adjacent collection far away obtain pedestrian p to be checked and pedestrian g to be measured iNeighbour's similarity, described neighbour's similarity comprises neighbour's similarity S Near(p, g i) and adjacent similarity S far away Far(p, g i);
Step 4 considers pedestrian p to be checked and pedestrian g to be measured iForward content similarity The reverse content similarity With neighbour's similarity, pedestrian to be measured is collected each pedestrian g to be measured among the G iRearrangement.
2. the pedestrian's search method based on Bidirectional sort as claimed in claim 1 is characterized in that:
All measure similarity based on macroscopic features in step 1 and the step 2.
3. the pedestrian's search method based on Bidirectional sort as claimed in claim 1 is characterized in that:
Described pedestrian p to be checked and pedestrian g to be measured iForward content similarity
Figure FDA00003457178400018
According to pedestrian g to be measured iSequence number in the forward ranking results obtains.
4. the pedestrian's search method based on Bidirectional sort as claimed in claim 1 is characterized in that:
Described pedestrian p to be checked and pedestrian g to be measured iThe reverse content similarity
Figure FDA00003457178400019
Obtain according to the sequence number of pedestrian p to be checked in the sorting by reversals result.
5. the pedestrian's search method based on Bidirectional sort as claimed in claim 1 is characterized in that:
The neighbour of described pedestrian p to be checked collection is: in the forward ranking results with k pedestrian's to be measured of pedestrian p similarity minimum to be checked set n k(p);
The far away adjacent collection of described pedestrian p to be checked is: in the forward ranking results with k' pedestrian's to be measured of pedestrian p similarity maximum to be checked set;
Described pedestrian g to be measured iNeighbour collection be: among the sorting by reversals result with pedestrian g to be measured iThe k of similarity minimum pedestrian's to be measured set n k(g i);
Described pedestrian g to be measured iFar away adjacent collection be: among the sorting by reversals result with pedestrian g to be measured iThe k' of similarity maximum pedestrian's to be measured set n K'(g i);
Above-mentioned k and k' are empirical value.
6. the pedestrian's search method based on Bidirectional sort as claimed in claim 1 is characterized in that:
Described pedestrian p to be checked and pedestrian g to be measured iNeighbour's similarity S Near(p, g i) collect n according to the neighbour of pedestrian p to be checked k(p) and pedestrian g to be measured iThe neighbour collect n k(g i) obtain, be specially n k(p) and n k(g i) common factor in the number of element;
Described pedestrian p to be checked and pedestrian g to be measured iFar away adjacent similarity S Far(p, g i) collect n according to the neighbour far away of pedestrian p to be checked K'(p) and pedestrian g to be measured iFar away adjacent collection n K'(g i) obtain, be specially n K'(p) and n K'(g i) common factor in the number of element.
7. the pedestrian's search method based on Bidirectional sort as claimed in claim 1 is characterized in that:
Step 4 is specially:
According to pedestrian p to be checked and pedestrian g to be measured iForward content similarity
Figure FDA00003457178400021
The reverse content similarity
Figure FDA00003457178400022
Obtain comprehensive similarity S with neighbour's similarity *(g i), based on comprehensive similarity S *(g i) pedestrian to be measured is collected G={g i| i=1 ..., each pedestrian g to be measured among the n} iResequence described comprehensive similarity S * ( g i ) = S p g i × S g i p α + S near ( p , g i ) + S p g i × S g i p α + S far ( p , g i ) , Wherein, S Near(p, g i) and S Far(p, g i) be respectively pedestrian p to be checked and pedestrian g to be measured iNeighbour's similarity and adjacent similarity far away,
Figure FDA00003457178400024
With
Figure FDA00003457178400025
Be respectively pedestrian p to be checked and pedestrian g to be measured iForward content similarity and reverse content similarity, α is one and prevents that denominator from being 0 constant, 0≤α≤1.
CN201310277359.9A 2013-07-03 2013-07-03 Based on the pedestrian retrieval method of Bidirectional sort Active CN103325122B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310277359.9A CN103325122B (en) 2013-07-03 2013-07-03 Based on the pedestrian retrieval method of Bidirectional sort

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310277359.9A CN103325122B (en) 2013-07-03 2013-07-03 Based on the pedestrian retrieval method of Bidirectional sort

Publications (2)

Publication Number Publication Date
CN103325122A true CN103325122A (en) 2013-09-25
CN103325122B CN103325122B (en) 2016-01-20

Family

ID=49193845

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310277359.9A Active CN103325122B (en) 2013-07-03 2013-07-03 Based on the pedestrian retrieval method of Bidirectional sort

Country Status (1)

Country Link
CN (1) CN103325122B (en)

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103793721A (en) * 2014-03-04 2014-05-14 武汉大学 Pedestrian repeat recognition method and system based on area related feedback
CN103984915A (en) * 2014-02-28 2014-08-13 中国计量学院 Pedestrian re-recognition method in monitoring video
CN104200206A (en) * 2014-09-09 2014-12-10 武汉大学 Double-angle sequencing optimization based pedestrian re-identification method
CN104462550A (en) * 2014-12-25 2015-03-25 武汉大学 Pedestrian re-recognition method based on similarity and dissimilarity fusion ranking optimization
CN103793702B (en) * 2014-02-28 2017-02-01 武汉大学 Pedestrian re-identifying method based on coordination scale learning
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
CN107833239A (en) * 2017-10-26 2018-03-23 辽宁工程技术大学 A kind of searching of optimal matching method for tracking target based on weighted model constraint
CN108921064A (en) * 2018-06-21 2018-11-30 西安理工大学 Pedestrian based on multi-feature fusion recognition methods again
CN109299664A (en) * 2018-08-27 2019-02-01 华中科技大学 A kind of method for reordering that pedestrian identifies again
CN109684493A (en) * 2019-01-04 2019-04-26 重庆大学 A kind of image rearrangement sequence method based on k neighborhood distribution score
CN109858308A (en) * 2017-11-30 2019-06-07 株式会社日立制作所 Video frequency searching device, video retrieval method and storage medium
CN110197154A (en) * 2019-05-30 2019-09-03 汇纳科技股份有限公司 Pedestrian recognition methods, system, medium and the terminal again of fusion site texture three-dimensional mapping
CN110377774A (en) * 2019-07-15 2019-10-25 腾讯科技（深圳）有限公司 Carry out method, apparatus, server and the storage medium of personage's cluster
CN112419166A (en) * 2020-09-24 2021-02-26 南京晓庄学院 Image defogging method based on combination of local region segmentation and SCN

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102521340A (en) * 2011-12-08 2012-06-27 中国科学院自动化研究所 Method for analyzing TV video based on role
US20120173549A1 (en) * 2006-11-22 2012-07-05 Jean-Yves Bouguet Methods and apparatus for retrieving images from a large collection of images

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120173549A1 (en) * 2006-11-22 2012-07-05 Jean-Yves Bouguet Methods and apparatus for retrieving images from a large collection of images
CN102521340A (en) * 2011-12-08 2012-06-27 中国科学院自动化研究所 Method for analyzing TV video based on role

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
DANIEL CARLOS GUIMARÃES PEDRONETTE ET AL: "Exploiting Contextual Spaces for Image Re-Ranking and Rank Aggregation", 《THE 1ST ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA RETRIEVAL》 *
XIAOHUI SHEN ET AL: "Object retrieval and localization with spatially-constrained similarity measure and k-NN re-ranking", 《IEEE COMPUTER VISION AND PATTERN RECOGNITION》 *
ZHONG WU ET AL: "Scalable Face Image Retrieval with Identity-Based Quantization and Multireference Reranking", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 *

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103984915A (en) * 2014-02-28 2014-08-13 中国计量学院 Pedestrian re-recognition method in monitoring video
CN103793702B (en) * 2014-02-28 2017-02-01 武汉大学 Pedestrian re-identifying method based on coordination scale learning
CN103793721B (en) * 2014-03-04 2017-05-10 武汉大学 Pedestrian repeat recognition method and system based on area related feedback
CN103793721A (en) * 2014-03-04 2014-05-14 武汉大学 Pedestrian repeat recognition method and system based on area related feedback
CN104200206A (en) * 2014-09-09 2014-12-10 武汉大学 Double-angle sequencing optimization based pedestrian re-identification method
CN104462550B (en) * 2014-12-25 2017-07-11 武汉大学 Pedestrian's recognition methods again of sorting consistence is merged based on similitude and dissimilarity
CN104462550A (en) * 2014-12-25 2015-03-25 武汉大学 Pedestrian re-recognition method based on similarity and dissimilarity fusion ranking optimization
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
CN107833239B (en) * 2017-10-26 2021-05-18 辽宁工程技术大学 Optimization matching target tracking method based on weighting model constraint
CN107833239A (en) * 2017-10-26 2018-03-23 辽宁工程技术大学 A kind of searching of optimal matching method for tracking target based on weighted model constraint
CN109858308A (en) * 2017-11-30 2019-06-07 株式会社日立制作所 Video frequency searching device, video retrieval method and storage medium
CN108921064A (en) * 2018-06-21 2018-11-30 西安理工大学 Pedestrian based on multi-feature fusion recognition methods again
CN108921064B (en) * 2018-06-21 2021-09-10 西安理工大学 Pedestrian re-identification method based on multi-feature fusion
CN109299664A (en) * 2018-08-27 2019-02-01 华中科技大学 A kind of method for reordering that pedestrian identifies again
CN109684493A (en) * 2019-01-04 2019-04-26 重庆大学 A kind of image rearrangement sequence method based on k neighborhood distribution score
CN110197154A (en) * 2019-05-30 2019-09-03 汇纳科技股份有限公司 Pedestrian recognition methods, system, medium and the terminal again of fusion site texture three-dimensional mapping
CN110377774A (en) * 2019-07-15 2019-10-25 腾讯科技（深圳）有限公司 Carry out method, apparatus, server and the storage medium of personage's cluster
CN110377774B (en) * 2019-07-15 2023-08-01 腾讯科技（深圳）有限公司 Method, device, server and storage medium for person clustering
CN112419166A (en) * 2020-09-24 2021-02-26 南京晓庄学院 Image defogging method based on combination of local region segmentation and SCN
CN112419166B (en) * 2020-09-24 2024-01-05 南京晓庄学院 Image defogging method based on combination of local region segmentation and SCN

Also Published As

Publication number Publication date
CN103325122B (en) 2016-01-20

Similar Documents

Publication Publication Date Title
CN103325122B (en) Based on the pedestrian retrieval method of Bidirectional sort
CN106250870B (en) A kind of pedestrian's recognition methods again of joint part and global similarity measurement study
Zhao et al. Learning mid-level filters for person re-identification
CN107506703A (en) A kind of pedestrian's recognition methods again for learning and reordering based on unsupervised Local Metric
CN105808732B (en) A kind of integrated Target attribute recognition and precise search method based on depth measure study
CN108520226B (en) Pedestrian re-identification method based on body decomposition and significance detection
CN109948425A (en) A kind of perception of structure is from paying attention to and online example polymerize matched pedestrian's searching method and device
CN103413145B (en) Intra-articular irrigation method based on depth image
CN103793702A (en) Pedestrian re-identifying method based on coordination scale learning
CN109064502A (en) The multi-source image method for registering combined based on deep learning and artificial design features
CN101551809A (en) Search method of SAR images classified based on Gauss hybrid model
CN104298992B (en) A kind of adaptive scale pedestrian recognition methods again based on data-driven
CN103186775A (en) Human body motion recognition method based on mixed descriptor
CN108764269A (en) A kind of cross datasets pedestrian recognition methods again based on space-time restriction incremental learning
CN101551851A (en) Infrared image target identification method
CN104268514A (en) Gesture detection method based on multi-feature fusion
CN108960142B (en) Pedestrian re-identification method based on global feature loss function
CN104966075B (en) A kind of face identification method and system differentiating feature based on two dimension
CN102495998A (en) Static object detection method based on visual selective attention computation module
CN108875448B (en) Pedestrian re-identification method and device
CN104484679B (en) Non- standard rifle shooting warhead mark image automatic identifying method
CN109766748A (en) A kind of pedestrian based on projective transformation and dictionary learning knows method for distinguishing again
CN108830179A (en) Merge the pedestrian detection algorithm of Color Image Edge and depth direction histogram
CN102930291A (en) Automatic K adjacent local search heredity clustering method for graphic image
Kanji Unsupervised part-based scene modeling for visual robot localization

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant