CN106875417A - A kind of multi-object tracking method associated across time domain based on high-order figure - Google Patents
A kind of multi-object tracking method associated across time domain based on high-order figure Download PDFInfo
- Publication number
- CN106875417A CN106875417A CN201710015550.4A CN201710015550A CN106875417A CN 106875417 A CN106875417 A CN 106875417A CN 201710015550 A CN201710015550 A CN 201710015550A CN 106875417 A CN106875417 A CN 106875417A
- Authority
- CN
- China
- Prior art keywords
- order
- subgraph
- time domain
- summit
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
Abstract
The invention discloses a kind of multi-object tracking method associated across time domain based on high-order figure, the testing result of each frame in video is obtained according to multi-target detection method first；Then by these detection responses and the restricted function F (v on structure high-order sidei,vj) build a common high-order figure across time domain；Afterwards for the local path section set under each time domain for being included in the common high-order figure of rapid extraction, common high-order figure is first changed into random uniformity high-order figure using the optimization method of RANSAC style, further change into common second order figure, subgraph search finally is carried out to ordinary second order figure, multiple orbit segments in each subgraph are coupled together according to the sequencing of time domain again, target track long is formed, so that the multiple target tracking in complex scene has good robustness.The present invention makes full use of the movable information of multiple target in complex scene and the presentation information associated across time domain, solve adjacent objects it is apparent similar when there is the tracking failure problem that identity is exchanged or local association mistake is caused.
Description
Technical field
The invention belongs to field of intelligent video surveillance, more particularly to a kind of multiple target associated across time domain based on high-order figure with
Track method.
Background technology
In recent years, multiple target tracking algorithm is increasingly paid attention to by scholar in computer vision field, multiple target tracking
Algorithm main purpose is while position and mark the positions of multiple mobile targets, to be linked in sequence each marker bit according to time domain, is entered
And obtain the motion track long of multiple targets.In other words, target following technology is actually to be obtained using the video sequence of input
Detection information is obtained, and certain association process is carried out to it, obtain the pursuit path of target.The multiple target video of current main flow with
Track technology may be summarized to be two parts:Target detection and target following.Wherein, the main task of target detection part is from regarding
Target interested is detected in frequency image.Visual monitor and monitoring system generally use Still camera, from so-called static
Background in isolate moving object, realize the detection of target.This system is generally used and builds background model mode, using threshold
Value is calculated, and detects foreground target.Target following part main task is will to detect that the target for obtaining is associated treatment.It is based on
The multiple target tracking algorithm of Track association is the focus algorithm of domestic and foreign scholars research.The cardinal principle of Track association algorithm be by
The multiple short and small tracking fragment obtained after target detection carries out the association of many levels, final to obtain the continuous, smooth of target
Pursuit path.
The content of the invention
It is an object of the invention to propose a kind of multi-object tracking method associated across time domain based on high-order figure, first basis
The multi-target detection method of current main-stream obtains the testing result of each frame in video, then by these detection responses and structure high-order
Restricted function F (the v on sidei,vj) build a common high-order figure across time domain；Afterwards in order in the common high-order figure of rapid extraction
Comprising high-order subgraph, common high-order figure is first changed into random uniformity high-order figure using the optimization method of RANSAC-style,
Common second order figure is further changed into, subgraph search, then each height that will be finally given finally are carried out to ordinary second order figure
Multiple orbit segments in figure are coupled together according to the sequencing of time domain, form target track long, and the method makes in miscellaneous scene
Multiple target tracking there is good robustness, efficiently solved adjacent objects it is apparent similar when occur identity exchange or office
The problems such as tracking that portion's associated errors are caused fails.
To achieve these goals, the technical scheme is that used it is a kind of based on high-order figure across time domain associate it is many
Method for tracking target, comprises the following steps：
Step A, video sequence long is divided into N equal portions, according to the offline inspection of multiple target in the time segment such as each
Result carries out the structure of common high-order figure H respectively；
Step B, samples are connected by progressively sampling and set up multiple candidates complete random uniformity high-order figure to commonly
High-order figure H's is approximate；
Step C, for improve high-order subgraph search element efficiency carry out random uniformity high-order figure to ordinary second order figure H turn
Change；
Step D, suitable weights measure function is selected to scan for the subgraph in ordinary second order figure；
Step E, the situation that physical limit is not met according to the subgraph for more than greedy algorithm solution searching, further according to part
The temporal order that orbit segment occurs is attached, so as to obtain target track long.
Further, it is specifically as follows comprising step in the step A：
Because serially the place of carrying out comprehends and takes very much to a video sequence long, the requirement of real-time is unsatisfactory for, therefore use
The method of parallel processing by different level, will this video sequence long be divided into N equal portions, Ln frames are included per equal portions, if last group of time
Section is then incorporated in previous group less than Ln frames.Then restricted function F (v to each group of time period respectively according to high-order sidei,vj)∈
{ 0,1 } builds across the when domain high order figure H=(V, E, α) of oneself, and wherein V represents the summit (pursuit path section) in high-order figure, E generations
High-order side in table high-order figure, α represents the probability for belonging to summit same high-order side.
Further, it is specifically as follows comprising step in the step B：
Step B-1, in order to all high-order subgraphs comprising target trajectory section in rapid extraction high-order figure below are, it is necessary to elder generation
Common high-order figure is carried out into approximate processing, commonly random uniformity high-order figure will be converted into by high-order figure.First by progressively adopting
The method of sample obtains multiple candidate's connection samples to set up random uniformity high-order figure RCH={ S1,...,Si... }, wherein Si
It is i-th connection sample, each connection sample includes L summit, and this L summit is from comprising ViSame high-order side in
Randomly select, and meet restricted function F (vi,vj)=1, j=1 ..., L-1.
Step B-2, due to be in B-1 randomly select meet the summit of restricted function, therefore there is insecure connection sample
This.In order to exclude above-mentioned bad connection sample, can be by two kinds of confidence function Ca(s) and CmS () is taken into consideration to obtain
Reliable connection sample, and then set up RCH, wherein C comprising all important high-order subgraphs in former high-order figure HaS () is used for measuring
The presentation similitude of target, CmS () is used for the kinematic similarity of metric objective.
Further, it is specifically as follows comprising step in the step C：
Completed to ordinary second order figure H'=(V', E', W') using the reliable random uniformity high-order figure obtained in step B-2
Conversion, wherein W' is the probability that two summits belong to a line in second order figure.In order to determine in second order figure two summits it
Between probability W', it is necessary to set up point set neighbour figure Ω=(ν, ε), in point set neighbour's figure, and if only if restricted function F
(vi,vjDuring)=1, there is a line between both sides.Then RCH to ordinary second order figure H' is completed using Clique graph search algorithms
Conversion.
Further, it is specifically as follows comprising step in the step D：
Each vertex v in step C is obtained ordinary second order figure H' firstpAs starting point, it is necessary to search forIndividual top
Point, and according to the weights measure function Γ (v of definitionpUN(vp)) so that the point set of these point compositions is obtained in that the power of maximum
Value measure function value, the subgraph for so being searched hasIndividual summit.In order to avoid degenerating occurs in vertex number in subgraph
Problem is, it is necessary to the subgraph to searching for sets a smallest dimensionAndIn addition, making yi=1 is
The indicator variable of subgraph vertex set U, works as yiWhen=1, vertex v is representediBelong to the subgraph of this search；Work as yi=0 is not belonging to
The subgraph.
Further, it is specifically as follows comprising step in the step E：
It is due to there may be the situation for not meeting physical limit between the different subgraphs of search in step D, such as same
Summit (pursuit path section) can not possibly simultaneously belong to two or more different targets, true this violates physics.In order to eliminate
This conflict situations, can be processed using greedy algorithm, and the subgraph that will first obtain is ranked up according to fiducial probability size, obtains
Subgraphs sequence T after to sequence.Make againIt is the subgraph set of post processing, according to greedy algorithm the vertex v for conflictingiPlus
In entering to have the subgraph of lap, i.e. viUT'→T'.Finally post processing subgraph is connected each summit according to temporal order
Get up to obtain target track long.
In general, by the contemplated above technical scheme of the present invention compared with prior art, can obtain following has
Beneficial effect：
1st, present invention employs step A and B, by a video signal process that time-consuming and capacity is big into N equal portions, with simultaneously
Capable mode replaces serially being processed, and greatly increases the processing speed of video, has reached the requirement of real-time；To it
In each time slice each build a high-order figure, because the side of high-order figure is made up of the summit more than 2, so
Multiple orbit segments in time slice can be carried out across time domain association, make full use of the higher order relationship between orbit segment as association
Condition, can so greatly reduce the probability that mistake occurs in local association, have good robustness to neighbouring apparent similar target；
2nd, present invention employs step C, step D and step E, in order to the random uniformity that rapid extraction step B is obtained is high
The relevant high-order subgraph of institute included in rank figure, is further converted to ordinary second order figure, and search using Clique figures by high-order figure
Rope algorithm and weights measure function are processed, and can so make subgraph search improved efficiency more than hundred times, finally by greedy algorithm
The high-order subgraph set that treatment is obtained can effectively remove the situation for not meeting physical limit, it is ensured that the uniqueness of track and continuity
Property.
Brief description of the drawings
Fig. 1 is the multi-object tracking method flow chart associated across time domain based on high-order figure of the invention.
Specific embodiment
In order that the object, technical solutions and advantages of the present invention are clearer, below in conjunction with drawings and Examples to this
Invention is described in further detail.
As shown in figure 1, following is a specific embodiment, its specific steps is followed successively by implementation schematic diagram of the invention：
Step A, video sequence long is divided into N equal portions, according to the offline inspection of multiple target in the time segment such as each
Result carries out the structure of common high-order figure H respectively；
Because serially the place of carrying out comprehends and takes very much to a video sequence long, the requirement of real-time is unsatisfactory for, therefore use
The method of parallel processing by different level, will this video sequence long be divided into N equal portions, Ln frames are included per equal portions, if last group of time
Section is then incorporated in previous group less than Ln frames.Then restricted function F (v to each group of time period respectively according to high-order sidei,vj)∈
{ 0,1 } builds across the when domain high order figure H=(V, E, α) of oneself, and wherein V represents the summit in high-order figure, and E is represented in high-order figure
High-order side, α represents the probability for belonging to summit same high-order side.
Step B, samples are connected by progressively sampling and set up multiple candidates arrive completing random uniformity high-order figure (RCH)
Common high-order figure H's is approximate；
In order to all high-order subgraphs comprising target trajectory section in rapid extraction high-order figure below will be, it is necessary to first will be common high
Rank figure carries out approximate processing, commonly will be converted into random uniformity high-order figure by high-order figure.Method first by progressively sampling
Multiple candidate's connection samples are obtained to set up random uniformity high-order figure RCH={ S1,...,Si... }, wherein SiIt is i-th company
Sample is connect, each connection sample includes L summit, and this L summit is from comprising ViSame high-order side in randomly select
, and meet restricted function F (vi,vj)=1, j=1 ..., L-1.
Due to it is above-mentioned be randomly select meet the summit of restricted function, therefore there is insecure connection sample.In order to arrange
Except above-mentioned bad connection sample, can be by two kinds of confidence functions：A kind of is the presentation similarity measurement of target：
Wherein χa(vi,vi+1) represent color histogram similarity, χw(vi,vi+1) represent texture similarity, χs(vi,vi+1)
Represent space similarity；Another kind is the kinematic similarity measurement of target：
Wherein dfp(vi,vi+1) and dbp(vi,vi+1) forward prediction and back forecast are represented respectively.By two kinds of confidence function knots
Close and consider to obtain reliable connection sample, so set up comprising in former high-order figure H all important high-order subgraphs it is random consistent
Property high-order figure RCH.
Step C, for improve high-order subgraph search element efficiency carry out random uniformity high-order figure to ordinary second order figure H turn
Change；
Completed to ordinary second order figure H'=(V', E', W') using the reliable random uniformity high-order figure obtained in step B
Conversion, wherein W' is the probability that two summits belong to a line in second order figure.In order to determine in second order figure between two summits
Probability W', set up point set neighbour figure Ω=(ν, ε), wherein ν be be connected sample S between confidence level more than threshold value λ
Vertex set, ε is the restricted function F (v that whether high-order side is met between two summitsi,vj)=1, i.e., in point set neighbour's figure,
And if only if restricted function F (vi,vjDuring)=1, there is a line between both sides.Then completed using Clique graph search algorithms
Conversions of the RCH to ordinary second order figure H'.
Step D, suitable weights measure function is selected to scan for the subgraph in ordinary second order figure；
Each vertex v in step C is obtained ordinary second order figure H' firstpAs starting point, it is necessary to search forIndividual top
Point, and according to the weights measure function Γ (v of definitionpUN(vp)) so that the point set of these point compositions is obtained in that the power of maximum
Value measure function value, the subgraph for so being searched hasIndividual summit, especially, in order to be able to make vertex number relatively fewer
Subgraph can also smoothly search, can using set average side right weight values be used as weights measure function：
Wherein vertex set U=vpUN(vp).In order to avoid degenerate problem occurs in vertex number in subgraph, to the son searched for
Figure sets a smallest dimensionAndIn addition, making y={ y1,...,yn}∈RnIt is subgraph summit
The indicator variable of set U, works as yiWhen=1, vertex v is representediBelong to the subgraph of this search；Work as yi=0 is not belonging to the subgraph.
Step E, the situation that physical limit is not met according to the subgraph for more than greedy algorithm solution searching, further according to part
The temporal order that orbit segment occurs is attached, so as to obtain target track long.
It is due to there may be the situation for not meeting physical limit between the different subgraphs of search in step D, such as same
Summit (pursuit path section) can not possibly simultaneously belong to two or more different targets, true this violates physics.In order to eliminate
This conflict situations, can be processed using greedy algorithm, and the subgraph that will first obtain is ranked up according to fiducial probability size, obtains
Subgraphs sequence T={ T after to sequence1,...,Tn}.Make againIt is the subgraph set of post processing, is searched for i-th
SubgraphSo directly ΤiIt is put into set Τ ', otherwise according to greedy algorithm ΤiAddition has
In the subgraph of lap, i.e. TiUT'→Tj' in.Finally the post processing subgraph for obtaining is connected each summit according to temporal order
Pick up to obtain target track long.
The above is not intended to limit the invention, all any modifications within the spirit and principles in the present invention, made,
Equivalent, improvement etc., should be included within the scope of the present invention.
Claims (6)
1. a kind of multi-object tracking method associated across time domain based on high-order figure, it is characterised in that as follows including step：
Step A, N equal portions are divided into by video sequence long, according to the offline inspection result of multiple target in the time segment such as each
The structure of common high-order figure H is carried out respectively；
Step B, random uniformity high-order figure to common high-order is completed by progressively sampling and setting up multiple candidates' connection samples
Figure H's is approximate；
Step C, for the efficiency that lifting high-order subgraph searches element carries out random uniformity high-order figure to the conversion of ordinary second order figure H；
Step D, selects suitable weights measure function to scan for the subgraph in ordinary second order figure；
Step E, the situation of physical limit is not met according to the subgraph for more than greedy algorithm solution searching, further according to local path
The temporal order that section occurs is attached, so as to obtain target track long.
2. a kind of multi-object tracking method associated across time domain based on high-order figure according to claim 1, it is characterised in that
Comprise the following steps in the step A：
Because serially the place of carrying out comprehends and takes very much to a video sequence long, the requirement of real-time is unsatisfactory for, therefore use layering
The method of secondary parallel processing, will this video sequence long be divided into N equal portions, per equal portions include Ln frames, if last group of time period is not
Sufficient Ln frames, then be incorporated in previous group, then to each group of time period respectively according to the restricted function F (v for constituting high-order figure sidei,vj)
∈ { 0,1 } builds across when domain high order figure H=(V, E, α), and wherein V represents the summit in high-order figure, and E represents the high-order in high-order figure
Side, α represents the probability for belonging to summit same high-order side.
3. a kind of multi-object tracking method associated across time domain based on high-order figure according to claim 1, it is characterised in that
Comprise the following steps in the step B：
Step B-1, in order to all high-order subgraphs comprising target trajectory section in step C rapid extraction high-order figures are, it is necessary to elder generation
Common high-order figure is carried out into approximate processing, commonly random uniformity high-order figure will be converted into by high-order figure, first by progressively adopting
The method of sample obtains multiple candidate's connection samples to set up random uniformity high-order figure RCH={ S1,...,Si... }, wherein Si
It is i-th connection sample, each connection sample includes L summit, and this L summit is from comprising ViSame high-order figure side
In randomly select, and meet restricted function F (vi,vj)=1, j=1 ..., L-1；
Step B-2, due to be in B-1 randomly select meet the summit of restricted function, therefore there is insecure connection sample, be
The above-mentioned bad connection sample of exclusion, can by two kinds of confidence function Ca(s) and CmS () is taken into consideration to obtain reliability
Connection sample, and then set up RCH, wherein C comprising all important high-order subgraphs in former high-order figure HaS () is used for metric objective
Presentation similitude, CmS () is used for the kinematic similarity of metric objective.
4. a kind of multi-object tracking method associated across time domain based on high-order figure according to claim 1, it is characterised in that
Comprise the following steps in the step C：
Completed to ordinary second order figure H'=(V', E', W') using the reliable random uniformity high-order figure RCH obtained in step B-2
Conversion, wherein W' is the probability that two summits belong to a line in second order figure, in order to determine in second order figure two summits it
Between probability W', it is necessary to set up point set neighbour figure Ω=(ν, ε), in point set neighbour's figure, and if only if restricted function F
(vi,vjDuring)=1, there is a line between both sides, then complete RCH to ordinary second order figure H' using Clique graph search algorithms
Conversion.
5. a kind of multi-object tracking method associated across time domain based on high-order figure according to claim 1, it is characterised in that
Comprise the following steps in the step D：
Each vertex v in step C is obtained ordinary second order figure H' firstpAs starting point, it is necessary to search forIndividual summit, together
Mono- weights measure function Γ (v of Shi DingyipUN(vp)) so that the point set of these point compositions is obtained in that the weights of maximum are estimated
Functional value, the subgraph for so being searched hasIndividual summit, it is right in order to avoid degenerate problem occurs in vertex number in subgraph
The subgraph of search sets a smallest dimensionAndIn addition, making y={ y1,...,yn}∈RnFor
The indicator variable of subgraph vertex set U, works as yiWhen=1, vertex v is representediBelong to the subgraph of this search；Work as yi=0 is not belonging to
The subgraph.
6. a kind of multi-object tracking method associated across time domain based on high-order figure according to claim 1, it is characterised in that
E comprises the following steps in the step：
Due to there may be the situation for not meeting physical limit between the different subgraphs of search in step D, such as same summit
Two or more different targets can not possibly simultaneously be belonged to, it is true this violates physics, in order to eliminate this conflict situations, can adopt
Processed with greedy algorithm, the subgraph that will first obtain is ranked up according to fiducial probability size, the subgraph sequence after being sorted
Row T, then makeIt is the subgraph set of post processing, according to greedy algorithm the vertex v for conflictingiAddition has the son of lap
In figure, i.e. vi∪ T' → T', then each summit are coupled together post processing subgraph according to temporal order and obtain target rail long
Mark.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710015550.4A CN106875417B (en) | 2017-01-10 | 2017-01-10 | One kind is based on high-order figure across the associated multi-object tracking method of time domain |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710015550.4A CN106875417B (en) | 2017-01-10 | 2017-01-10 | One kind is based on high-order figure across the associated multi-object tracking method of time domain |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106875417A true CN106875417A (en) | 2017-06-20 |
| CN106875417B CN106875417B (en) | 2019-10-08 |
Family
ID=59165544
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710015550.4A Active CN106875417B (en) | 2017-01-10 | 2017-01-10 | One kind is based on high-order figure across the associated multi-object tracking method of time domain |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106875417B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111739053A (en) * | 2019-03-21 | 2020-10-02 | 四川大学 | Online multi-pedestrian detection tracking method under complex scene |
| CN111932038A (en) * | 2020-09-24 | 2020-11-13 | 浙江口碑网络技术有限公司 | Trajectory generation method and apparatus, computer device and computer-readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530604A (en) * | 2013-09-27 | 2014-01-22 | 中国人民解放军空军工程大学 | Robustness visual tracking method based on transductive effect |
| CN104008392A (en) * | 2014-05-09 | 2014-08-27 | 南京邮电大学 | Multi-objective tracking method based on continuous minimum-energy appearance model |
| CN105261040A (en) * | 2015-10-19 | 2016-01-20 | 北京邮电大学 | Multi-target tracking method and apparatus |
| CN105260741A (en) * | 2015-09-29 | 2016-01-20 | 刘伟锋 | Digital image marking method based on higher-order graph structure p-Laplacian sparse codes |
| CN105699964A (en) * | 2016-02-29 | 2016-06-22 | 无锡南理工科技发展有限公司 | Road multi-target tracking method based on automobile anti-collision radar |
-
2017
- 2017-01-10 CN CN201710015550.4A patent/CN106875417B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103530604A (en) * | 2013-09-27 | 2014-01-22 | 中国人民解放军空军工程大学 | Robustness visual tracking method based on transductive effect |
| CN104008392A (en) * | 2014-05-09 | 2014-08-27 | 南京邮电大学 | Multi-objective tracking method based on continuous minimum-energy appearance model |
| CN105260741A (en) * | 2015-09-29 | 2016-01-20 | 刘伟锋 | Digital image marking method based on higher-order graph structure p-Laplacian sparse codes |
| CN105261040A (en) * | 2015-10-19 | 2016-01-20 | 北京邮电大学 | Multi-target tracking method and apparatus |
| CN105699964A (en) * | 2016-02-29 | 2016-06-22 | 无锡南理工科技发展有限公司 | Road multi-target tracking method based on automobile anti-collision radar |
Non-Patent Citations (3)
| Title |
|---|
| GUANG HAN ET AL: "Robust Object Tracking Based on Local Region Sparse Appearance Model", 《NEUROCOMPUTING》 * |
| HAIRONG LIU ET AL: "Efficient structure detection via random consensus graph", 《2012 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| XINCHU SHI ET AL: "Multi-target Tracking by Rank-1 Tensor Approximation", 《2013 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111739053A (en) * | 2019-03-21 | 2020-10-02 | 四川大学 | Online multi-pedestrian detection tracking method under complex scene |
| CN111739053B (en) * | 2019-03-21 | 2022-10-21 | 四川大学 | Online multi-pedestrian detection tracking method under complex scene |
| CN111932038A (en) * | 2020-09-24 | 2020-11-13 | 浙江口碑网络技术有限公司 | Trajectory generation method and apparatus, computer device and computer-readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106875417B (en) | 2019-10-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110598554B (en) | Multi-person posture estimation method based on counterstudy | |
| Wu et al. | Exploit the unknown gradually: One-shot video-based person re-identification by stepwise learning | |
| CN105894542B (en) | A kind of online method for tracking target and device | |
| CN111161311A (en) | Visual multi-target tracking method and device based on deep learning | |
| CN113516012B (en) | Pedestrian re-identification method and system based on multi-level feature fusion | |
| Yijing et al. | Fig fruit recognition method based on YOLO v4 deep learning | |
| CN110210379A (en) | A kind of lens boundary detection method of combination critical movements feature and color characteristic | |
| CN109544592A (en) | For the mobile moving object detection algorithm of camera | |
| CN110674886B (en) | Video target detection method fusing multi-level features | |
| CN111882586A (en) | Multi-actor target tracking method oriented to theater environment | |
| CN106875417B (en) | One kind is based on high-order figure across the associated multi-object tracking method of time domain | |
| Wang et al. | Boundary sensitive and category sensitive network for temporal action proposal generation | |
| Bashar et al. | Multiple object tracking in recent times: A literature review | |
| CN113011399B (en) | Video abnormal event detection method and system based on generation cooperative discrimination network | |
| Yan et al. | Unloc: A unified framework for video localization tasks | |
| Mu et al. | Salient object detection in low contrast images via global convolution and boundary refinement | |
| Pillai et al. | Fine-Tuned EfficientNetB4 Transfer Learning Model for Weather Classification | |
| Zhang et al. | Improved YOLOv8 Insulator Fault Detection Algorithm Based on BiFormer | |
| Bozorgtabar et al. | MSMCT: Multi-state multi-camera tracker | |
| CN113538351A (en) | External insulation equipment defect degree evaluation method fusing multi-parameter electric signals | |
| Zhang et al. | What makes for good multiple object trackers? | |
| Yuan et al. | Salient Object Detection Based on Improved PoolNet. | |
| Mori et al. | Behavior prediction from trajectories in a house by estimating transition model using stay points | |
| Pal et al. | A video-based human activity and motion direction classification framework for agricultural fields | |
| Zhang et al. | Multivideo object cosegmentation for irrelevant frames involved videos |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20170620 Assignee: Nanjing Nanyou Information Industry Technology Research Institute Co. Ltd. Assignor: Nanjing Post & Telecommunication Univ. Contract record no.: X2019980001257 Denomination of invention: Multi-target tracking method based on high-order graph cross-time domain association Granted publication date: 20191008 License type: Common License Record date: 20191224 |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210511 Address after: Room 507, 6-3 Xingzhi Road, Nanjing Economic and Technological Development Zone, Jiangsu Province, 210000 Patentee after: NANJING NANYOU INSTITUTE OF INFORMATION TEACHNOVATION Co.,Ltd. Address before: 210003, 66 new model street, Gulou District, Jiangsu, Nanjing Patentee before: NANJING University OF POSTS AND TELECOMMUNICATIONS |
|
| EC01 | Cancellation of recordation of patent licensing contract | ||
| EC01 | Cancellation of recordation of patent licensing contract |
Assignee: NANJING NANYOU INSTITUTE OF INFORMATION TECHNOVATION Co.,Ltd. Assignor: NANJING University OF POSTS AND TELECOMMUNICATIONS Contract record no.: X2019980001257 Date of cancellation: 20220304 |