CN106952293A - A kind of method for tracking target based on nonparametric on-line talking - Google Patents
A kind of method for tracking target based on nonparametric on-line talking Download PDFInfo
- Publication number
- CN106952293A CN106952293A CN201611219002.5A CN201611219002A CN106952293A CN 106952293 A CN106952293 A CN 106952293A CN 201611219002 A CN201611219002 A CN 201611219002A CN 106952293 A CN106952293 A CN 106952293A
- Authority
- CN
- China
- Prior art keywords
- target
- cluster centre
- feature
- template
- nonparametric
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000009826 distribution Methods 0.000 claims abstract description 23
- 238000000895 extractive distillation Methods 0.000 claims abstract 2
- 230000008569 process Effects 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 4
- 238000004821 distillation Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 238000013459 approach Methods 0.000 claims description 2
- 238000009415 formwork Methods 0.000 claims description 2
- 238000004088 simulation Methods 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 239000004744 fabric Substances 0.000 claims 1
- 238000004458 analytical method Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 3
- 230000000903 blocking effect Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
A kind of method for tracking target based on nonparametric on-line talking, its solve technical problem be：A video is given, online target following is carried out to a certain specified interesting target, so that the particular location and boundary rectangle frame of subsequent image frames target are predicted, the small depth network arrived based on big depth e-learning, to interesting target extractive distillation feature.This feature has stronger appearance representation ability, discriminating power and robustness.It is of the invention to be the problem of further solve：History outward appearance to interesting target carries out nonparametric on-line talking, cluster centre based on weighting produces the target appearance template of a reservation historical information, and produce object space distribution with the target appearance template, so as to improve target following accuracy rate, the risk of target drift is reduced, and then reduces error rate.
Description
Technical field
Present invention relates generally to the fields such as Computer Vision, intelligent video monitoring, pattern-recognition, and in particular to be
A kind of method for tracking target based on nonparametric on-line talking.
Background technology
Target following is the most basic and most challenging key problem of field of intelligent video surveillance one.It is used as calculating
The problem of comparing bottom in machine vision algorithm, it is with a wide range of applications.As the motion analysis in field of intelligent video surveillance,
The intellectual analysis application of the high level such as abnormality detection is carried out based on target following.Moving target in scene is carried out
Track is extracted, effective basis can be provided for intellectual analysis higher level in monitoring scene.
Traditional method for tracking target initializes a certain interesting target in the given frame of video first mostly, then trains one
Individual target appearance model, carries out discriminate judgement by display model during tracking or production is predicted, so as to be inferred to
The positional information of next frame interesting target.This kind of method is entered during tracking using the pattern of on-line study to display model
Row is constantly updated, have ignored interesting target easily by blocking, disappearance-reproduction is influenceed, and is easily caused model drifting problem.Together
When, because the learning process of model largely relies on display model, after target is drifted about, it is difficult to object module is reinitialized,
So as to cause tracking to fall flat, poor tracking result is obtained.Traditional fusion Meanshift and Kalman filter tracking
Method, can be modeled to target motion, using the effective forecast function of Kalman filtering, in each frame Meanshift first
Algorithm starts before prediction, and dbjective state is predicted according to conventional movable information, obtains a possible position of target
Point, then carries out the iteration of Meanshift algorithms from the possible position point predicted, and wherein Meanshift algorithms are mainly first
The core color histogram of target is first calculated, the similarity function of a space smoothing is then defined on this basis, and use
The mode of gradient optimizing is quickly positioned to target.Different from mode of the Kalman filtering to target motion modeling, Wo Menti
It is multiple initial to obtain that the method gone out calculates similarity based on the apparent information of target to multiple space lattices of target context
Position candidate, rather than the possible predicted position that Kalman filter is got, so as to add tracker to quick
The robustness and adaptability moved and blocked.Different from Meanshift algorithms expressed using core color histogram target with
Track device, we employ the feature of feature distiller extraction preferably to express the apparent information of target.Different from Meanshift
Algorithm is quickly positioned by the way of gradient optimizing to target, our tracker by domain space learn filtering with
Track device simultaneously realizes the quick detection of target and final positioning using Fast Fourier Transform (FFT).
The content of the invention
The invention provides a kind of method for tracking target, its technical problem solved is：A video is given, to a certain
Interesting target is specified to carry out online target following, so as to predict the particular location and boundary rectangle of subsequent image frames target
Frame.
It is of the invention to be the problem of further solve：The small depth network arrived based on big depth e-learning, to interested
Objective extraction distills feature.This feature has stronger appearance representation ability, discriminating power and robustness.
It is of the invention to be the problem of further solve：History outward appearance to interesting target carries out nonparametric on-line talking,
Cluster centre based on weighting produces the target appearance template of a reservation historical information, and produces mesh with the target appearance template
Spatial distribution is marked, so as to improve target following accuracy rate, the risk of target drift is reduced, and then reduce error rate.
It is of the invention to be the problem of further solve：Reduce hsrdware requirements and algorithm complex, realize quick target with
Track.
To solve the above problems, the invention discloses a kind of method for tracking target based on nonparametric on-line talking, for
Given video initialization target performs this method, obtains target following result.
The step of method for tracking target, comprises the following steps：
Step 1, an input video is given, and the square frame of an interesting target is initialized in the first frame；To the mesh
Mark square frame and extract color histogram feature, and cluster centre is initialized using the method for K averages；
Step 2, the image of spatial distribution is kept to distill characteristic pattern the context area image zooming-out of the square frame, it is assumed that should
There is the priori label distribution of a same scale size in underlying image feature figure；
Step 3, based on image distill characteristic pattern and priori label distribution, by the way of ridge regression domain space without
It is that traditional time domain space learns a filter tracker；
Step 4, for next two field picture, the target's center position predicted based on previous frame target tracker is divided many
Individual grid, color histogram feature is extracted to each grid, by the color histogram feature and the template generated by cluster centre
Similarity is calculated, reference value value spatial probability distribution is obtained.Based on the rough object space probability distribution, similarity is obtained
Maximum preceding k grid.
Step 5, to each in this k grid, one and the first frame context area formed objects image are extracted
Characteristic pattern, by filter tracker and the characteristics of image figure by convolution with FFT, so as to obtain each locus
Confidence level.It is target's center position to take the maximum locus of confidence value.The state of target can also be obtained.Based on the mesh
Mark state, extracts new color histogram feature；
Step 6, new color histogram was compared with the former cluster centre learnt, if similarity is more than necessarily
The color histogram, then be included into the cluster centre most like with it, and update the cluster centre by value；If similarity is less than certain
Value, then newly create a cluster centre.By nonparametric after line incremental clustering obtains cluster centre, by these cluster centres
It is merged into a To Template.Context area characteristic pattern is extracted based on the center, and learns a new tracking filter
Device, the tracking filter of previous frame is updated with the tracking filter；
Step 7, circulation performs step 4-6, carries out online target following.
In addition, this method mainly includes following components, it is exactly specifically：Candidate Submission generation unit, for base
A small amount of relatively reliable candidate samples motion is produced in priori or history space time information；Feature extraction unit, for carrying
Take the feature representation of the corresponding image-region of each candidate samples motion；Observation model judging unit, for motion feature
Classified or returned, judge that the motion belongs to the probability of target or whether is target；Integrated study integrated unit, is used for
Multiple observation model judging units are subjected to integrated study by certain strategy, and multiple tracking results are merged, from
And obtain current image frame target position information and extraneous rectangle frame information；On-time model updating block, for observation model
Judging unit carries out online updating, with the robustness of assurance model.
The motion of the present invention produces process for being possible to model drifting problem occur in object tracking process, produces a small amount of
The higher Candidate Submission of confidence level, the possibility for detecting or correcting again is provided for observation model judging unit below.So drop
It is low block for a long time or disappearance-rendering problems caused by template drift risk.This motion, which produces process, mainly includes target mould
Plate generating process and object space distribution generation process.
To Template generating process mainly use nonparametric on-line talking method and generate a To Template process for：It is right
The new online incremental clustering of color histogram nonparametric, obtains new cluster centre.By these cluster centres according to certain
Weight is merged into a To Template.The weight of each cluster centre is determined by the number of samples in each cluster.In addition, being
More stable To Template is obtained, before the To Template is generated, weight less cluster centre is first excluded.To weight compared with
Big cluster centre is normalized in the way of weight is added, so as to obtain To Template.The wherein calculating side of cluster centre
Formula is as follows：
Here wjIt is the weight of j-th of cluster centre.Specifically, it is exactly to obtain a sample x onlineiAfterwards, the sample is calculated
Sheet and each cluster centre cjSimilarity.Its similarity function s (xi,cj) be：
The computing formula of To Template is as follows：
wiIt is the weight of each cluster centre.
Spatial distribution produces process primarily with respect to next two field picture, the target predicted based on previous frame target tracker
Center, divides multiple grids, and color histogram feature is extracted to each grid, by the color histogram feature with by clustering
The formwork calculation similarity being centrally generated, obtains object space distribution.Spatial distribution based on target obtains similarity maximum
Preceding k grid.
The characteristic extraction procedure of the present invention is that the big deep learning network for learning other by database is small with one
Moldeed depth degree network carries out extraction characteristic simulation, and in the training stage, supervision message is the spy that big every layer of deep learning network is extracted
Levy, simulated as much as possible with small-sized depth network and approach these supervision messages；In test phase, arrived with small-sized depth network extraction
Feature be distillation feature.Feature is distilled using the depth of small network extraction, with stronger feature representation ability and differentiation
Ability.
The observation model training process of the present invention is directed to target corresponding model of on-line study under different video environment, from
And irrational tracking result can be effectively filtered during tracking, it is to avoid tracking result and surrounding in existing method
Caused by environmental information mismatch the problem of tracking result Wrong localization.
The integrated study fusion of the present invention is employed to be clustered to the target appearance information in history, so as to obtain relative
Stable appearance representation cluster centre, enhances the stability of display model expression.
The renewal process of the present invention carries out the training of model just for specific effective sample selection rule during tracking
Practise, it is to avoid all sample rows predicted are calculated, reduce model modification and occur drift or track inaccurate risk.This
The method of invention and corresponding algorithm have higher reasonability and operational efficiency, reduce complexity, can reach in real time
Target following.
The fields such as vision security protection, machine vision tracking, man-machine interface are the method can be widely used in, are that one kind has versatility
Intelligent vision monitoring core methed.
Brief description of the drawings
Figure 1A show the structured flowchart of the Target Tracking System of the present invention；
Figure 1B show Candidate Submission generation unit structured flowchart in Target Tracking System of the present invention；
Fig. 1 C show Target Tracking System feature extraction unit structured flowchart of the present invention；
Fig. 1 D show Target Tracking System observation model judging unit structured flowchart of the present invention；
Fig. 2 show the flow chart of method for tracking target of the present invention.
Embodiment
For the object, technical solutions and advantages of the present invention are more clearly understood, below in conjunction with specific embodiment, and reference
Accompanying drawing, the present invention is described in more detail.
The present invention carries out manually that initialization or auto-initiation be (for example to the first frame general target of video sequence
Detection), then export the target trajectory of subsequent frame, i.e. target position in each frame and profile boundary rectangle frame.
Further, the present invention distills feature based on the small depth network extraction learnt, with stronger expression energy
Power.
Further, the present invention non-ginseng of on-line study from the history space-time target visual feature in video sequence image
Number cluster centre and target cluster template, improve the robustness for the problems such as target following is to blocking, improve accuracy rate, reduce
Error rate.
Further, the present invention differentiates to spatial distribution constraint of the target in current frame image, reduces mistake
Misprediction result, further increases tracking accuracy.
Further, the present invention is low to hardware requirement, and algorithm complex is low.
The structural representation of Target Tracking System of the invention shown in reference picture 1A, the system is obtained comprising video frame images
Take device, target tracker, storage device and display device.
Image acquiring device is used to obtain input picture, and for example it can be a monitoring camera, will obtain one section of image
Sequence and the target frame of the first frame initialization are sent to target tracker.
Target tracker is used to be handled for the image sequence.Tracks of device may be provided at ordinary PC, board,
In graphic process unit or embedded processing box.Target tracker include Candidate Submission generation unit, low-level image feature extraction unit,
Observation model judging unit, integrated study integrated unit and on-time model updating block.
Candidate Submission generation unit receives the target initial block of image sequence and the first frame from image acquiring device, when
When handling the second frame centered on the target initial position of the first frame, a small amount of Candidate Submission is extracted according to certain strategy.
As shown in Figure 1B, Candidate Submission generation unit further comprises：On-line checking filter unit, target distribution probability list
Member and Candidate Submission output unit.
On-line checking filter unit Main Analysis history target tracking data, and obtained by on-line study outside a target
See expression wave filter.Target distribution probability unit mainly calculates target in the range of image overall by on-line checking filter unit
Spatial probability distribution.
Candidate Submission output unit is based primarily upon several positions that destination probability distribution roughly estimates the presence of target most probable
Put, regard these positions as Candidate Submission.
As shown in Figure 1 C, feature extraction unit obtains small character network mould by distilling existing big depth network model
Type, so as to extract the depth characteristic with higher robustness and ability to express more quickly.
As shown in figure iD, observation model judging unit mainly includes model parameter, convolution operation and threshold value arbiter.Model
Parameter mainly learns what is obtained by ridge regression.Main process is the feature by target area and its context area, carries out Fu
In leaf transformation, and based on loop structure (referring specifically to Joao F.Henriques, et.al.High-Speed Tracking
With Kernelized Correlation Filters.PAMI2015.) intensive sampling is carried out, so that rapidly mould is arrived in study
The parameter of type.The feature that quick detection mainly extracts model parameter and region of search carries out Fourier transformation, Ran Houbian
Model parameter and feature after changing carry out convolution operation in domain space, so as to quickly obtain classifying for each position
Divide, i.e. the confidence level figure of target.
Threshold value arbiter, is analyzed confidence level figure, more than certain threshold value and the maximum position of confidence level, by it
It is determined as target；If the maximum of confidence level figure is less than the threshold value, then it is assumed that the target of present frame is blocked, at this moment just not to seeing
Model is surveyed to be updated.
Integrated study integrated unit carries out final decision-making to the multiple candidate target frames finally given and differentiated, filters coincidence
Some higher tracking results of rate, obtain the final position for tracking target and boundary rectangle information (referring specifically to Guibo
Zhu,Jinqiao Wang and Hanqing Lu,“Clustering Ensemble Tracking”,ACCV,2014)。
On-time model updating block is mainly carries out online updating to the pattern function learnt according to certain ratio.With
Threshold value arbiter come judge whether update model parameter, with prevent model occur update storage device, for storing Decision fusion
The final goal tracking result of unit output.Display device, can be a display screen, final in the input picture for being shown in
Target following result is watched for user.
Fig. 2 shows a kind of flow chart of method for tracking target of the invention, and this method comprises the following steps：
Step 200, Target Tracking System is started；
Step 201, start step, while obtaining a series of frame video data, and the target of the first frame is carried out initially
Change square frame；
Step 202, a small amount of candidate target motion is produced；
Step 203, feature is extracted to each candidate target motion；
Step 204, observation model judging unit；
Step 205, judge whether to merge observation model, if fusion link, be then transferred to step 206；Otherwise
It is transferred to step 207；
Step 206, integrated study is carried out to multiple observation models, so as to obtain more preferable result；
Step 207, the classification response figure or confidence level figure of target are calculated using step 205 or 206 result of calculation；
Step 208, judge whether the maximum of confidence level figure is more than a certain threshold value, if it is, into on-time model more
New unit；Otherwise not more new model；
Step 209, when needing on-time model updating block, the target frame output based on step 210 is sentenced to observation model
Disconnected unit carries out online updating；
Step 210, the objective degrees of confidence figure based on step 207 is calculated, so as to export target location and target frame；
To after target frame, step 202 is jumped to, then the target to subsequent frame is tracked prediction；
Step 211, the target frame of history is exported, so as to obtain tracking the track of target；
Step 212, target end tracking system.
The above-described embodiments merely illustrate the principles and effects of the present invention, and the embodiment that part is used, for
For one of ordinary skill in the art, without departing from the concept of the premise of the invention, can also make it is some deformation and
Improve, these belong to protection scope of the present invention.
Claims (6)
1. a kind of method for tracking target based on nonparametric on-line talking, comprises the following steps：
Step 1, an input video is given, and a target frame interested is initialized in the first frame；The target frame is carried
Color histogram feature is taken, and cluster centre is initialized using the method for K averages；
Step 2, the image distillation characteristic pattern of holding spatial distribution is calculated to the context area area image of the target frame, and assumes to be somebody's turn to do
There is the priori label distribution of a same scale size in underlying image feature figure；
Step 3, characteristic pattern and the distribution of priori label are distilled based on image, in domain space rather than biography by the way of ridge regression
The time domain space of system learns a filter tracker；
Step 4, for next two field picture, the target's center position predicted based on previous frame target tracker divides multiple grid
Lattice, color histogram feature is extracted to each grid, by the color histogram feature and the formwork calculation generated by cluster centre
Similarity, obtains reference value value spatial probability distribution, based on the rough object space probability distribution, obtains similarity maximum
Preceding k grid；
Step 5, to each in this k grid, the feature of one and the first frame context area formed objects image are extracted
Figure, by filter tracker and the characteristics of image figure by convolution with FFT, so as to obtain putting for each locus
Reliability.It is target's center position to take the maximum locus of confidence value, and the state of target can also be obtained, based on the target-like
State, extracts new color histogram feature；
Step 6, new color histogram was compared with the former cluster centre learnt, if similarity is more than certain value,
The color histogram is then included into the cluster centre most like with it, and updates the cluster centre；If similarity is less than certain value,
Then newly create a cluster centre.By nonparametric after line incremental clustering obtains cluster centre, these cluster centres are closed
And into a To Template.Context area characteristic pattern, and one new tracking filter of study are extracted based on the center,
The tracking filter of previous frame is updated with the tracking filter；
Step 7, ring performs step 4-6, carries out online target following.
2. according to the method described in claim 1, it is characterised in that step 2 extractive distillation feature is exactly specifically by other
The big deep learning network learnt by database carries out extraction characteristic simulation with a small-sized depth network, in training rank
Section, supervision message is the feature that big every layer of deep learning network is extracted, and is simulated as much as possible with small-sized depth network and approaches this
A little supervision messages；In test phase, with small-sized depth network extraction to feature be distillation feature.
3. according to the method described in claim 1, it is characterised in that obtain the sky of target by the way of grid using step 4
Between a kind of nonparametric on-line talking algorithm for proposing of distribution and step 7, then these cluster centres are closed according to certain weight
And into a To Template.
4. according to the method described in claim 1, it is characterised in that in step 4, obtaining object space distributed process is：Base
The target's center position predicted in previous frame target tracker, divides multiple grids, and color histogram is extracted to each grid
Feature, by the color histogram feature and the To Template calculating similarity generated by cluster centre, obtains object space distribution.
5. according to the method described in claim 1, it is characterised in that in step 7, using nonparametric on-line talking method and life
It is into a To Template process：To the new online incremental clustering of color histogram nonparametric, new cluster centre is obtained, will
These cluster centres are merged into a To Template according to certain weight, and the weight of each cluster centre is by each cluster
Number of samples determine, in addition, in order to obtain more stable To Template, before the To Template is generated, first exclude weight
Less cluster centre, the cluster centre larger to weight is normalized in the way of weight is added, so as to obtain target mould
The calculation of plate, wherein cluster centre is as follows：
Here wjIt is the weight of j-th of cluster centre, is exactly to obtain a sample x online specificallyiAfterwards, calculate the sample with
Each cluster centre cjSimilarity, its similarity function s (xi,cj) be：
The computing formula of To Template is as follows：
wiIt is the weight of each cluster centre.
6. according to the method described in claim 1, it is characterised in that in step 3, based on distillation characteristic pattern and priori label point
Cloth, learns a filter tracker, based on loop structure and Fast Fourier Transform (FFT), so as to fast by the way of ridge regression
Speed study filter tracker is simultaneously used for quickly detecting with it.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611219002.5A CN106952293B (en) | 2016-12-26 | 2016-12-26 | Target tracking method based on nonparametric online clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611219002.5A CN106952293B (en) | 2016-12-26 | 2016-12-26 | Target tracking method based on nonparametric online clustering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106952293A true CN106952293A (en) | 2017-07-14 |
| CN106952293B CN106952293B (en) | 2020-02-28 |
Family
ID=59465781
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611219002.5A Active CN106952293B (en) | 2016-12-26 | 2016-12-26 | Target tracking method based on nonparametric online clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106952293B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108093153A (en) * | 2017-12-15 | 2018-05-29 | 深圳云天励飞技术有限公司 | Method for tracking target, device, electronic equipment and storage medium |
| CN108986146A (en) * | 2017-12-11 | 2018-12-11 | 罗普特（厦门）科技集团有限公司 | A kind of correlation filtering tracking based on background information and adaptive recurrence label |
| CN109215057A (en) * | 2018-07-31 | 2019-01-15 | 中国科学院信息工程研究所 | A kind of high-performance visual tracking method and device |
| CN109785368A (en) * | 2017-11-13 | 2019-05-21 | 腾讯科技（深圳）有限公司 | A kind of method for tracking target and device |
| CN109919921A (en) * | 2019-02-25 | 2019-06-21 | 天津大学 | Based on the influence degree modeling method for generating confrontation network |
| CN110111370A (en) * | 2019-05-15 | 2019-08-09 | 重庆大学 | A kind of vision object tracking methods based on TLD and the multiple dimensioned space-time characteristic of depth |
| CN110633597A (en) * | 2018-06-21 | 2019-12-31 | 北京京东尚科信息技术有限公司 | Driving region detection method and device |
| CN114973167A (en) * | 2022-07-28 | 2022-08-30 | 松立控股集团股份有限公司 | Multi-target tracking method based on off-line clustering and unsupervised contrast learning |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101251893A (en) * | 2008-03-28 | 2008-08-27 | 西安电子科技大学 | Self-adapting multi-dimension veins image segmenting method based on wavelet and average value wander |
| CN101777187A (en) * | 2010-01-15 | 2010-07-14 | 西安电子科技大学 | Video microscopic image cell automatic tracking method based on Meanshift arithmetic |
| US20110113095A1 (en) * | 2009-11-10 | 2011-05-12 | Hamid Hatami-Hanza | System and Method For Value Significance Evaluation of Ontological Subjects of Networks and The Applications Thereof |
| WO2011067790A3 (en) * | 2009-12-02 | 2011-10-06 | Tata Consultancy Services Limited | Cost-effective system and method for detecting, classifying and tracking the pedestrian using near infrared camera |
| CN103761532A (en) * | 2014-01-20 | 2014-04-30 | 清华大学 | Label space dimensionality reducing method and system based on feature-related implicit coding |
| CN104021577A (en) * | 2014-06-19 | 2014-09-03 | 上海交通大学 | Video tracking method based on local background learning |
| US20150074036A1 (en) * | 2013-09-12 | 2015-03-12 | Agt International Gmbh | Knowledge management system |
| WO2016040304A1 (en) * | 2014-09-10 | 2016-03-17 | Bae Systems Information And Electronic Systems Integration Inc. | A method for detection and characterization of technical emergence and associated methods |
| CN105976400A (en) * | 2016-05-10 | 2016-09-28 | 北京旷视科技有限公司 | Object tracking method and device based on neural network model |
| CN106097391A (en) * | 2016-06-13 | 2016-11-09 | 浙江工商大学 | A kind of multi-object tracking method identifying auxiliary based on deep neural network |
-
2016
- 2016-12-26 CN CN201611219002.5A patent/CN106952293B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101251893A (en) * | 2008-03-28 | 2008-08-27 | 西安电子科技大学 | Self-adapting multi-dimension veins image segmenting method based on wavelet and average value wander |
| US20110113095A1 (en) * | 2009-11-10 | 2011-05-12 | Hamid Hatami-Hanza | System and Method For Value Significance Evaluation of Ontological Subjects of Networks and The Applications Thereof |
| WO2011067790A3 (en) * | 2009-12-02 | 2011-10-06 | Tata Consultancy Services Limited | Cost-effective system and method for detecting, classifying and tracking the pedestrian using near infrared camera |
| CN101777187A (en) * | 2010-01-15 | 2010-07-14 | 西安电子科技大学 | Video microscopic image cell automatic tracking method based on Meanshift arithmetic |
| US20150074036A1 (en) * | 2013-09-12 | 2015-03-12 | Agt International Gmbh | Knowledge management system |
| CN103761532A (en) * | 2014-01-20 | 2014-04-30 | 清华大学 | Label space dimensionality reducing method and system based on feature-related implicit coding |
| CN104021577A (en) * | 2014-06-19 | 2014-09-03 | 上海交通大学 | Video tracking method based on local background learning |
| WO2016040304A1 (en) * | 2014-09-10 | 2016-03-17 | Bae Systems Information And Electronic Systems Integration Inc. | A method for detection and characterization of technical emergence and associated methods |
| CN105976400A (en) * | 2016-05-10 | 2016-09-28 | 北京旷视科技有限公司 | Object tracking method and device based on neural network model |
| CN106097391A (en) * | 2016-06-13 | 2016-11-09 | 浙江工商大学 | A kind of multi-object tracking method identifying auxiliary based on deep neural network |
Non-Patent Citations (6)
| Title |
|---|
| CHUNG-CHENG CHIU ET AL.: ""A Robust Object Segmentation System Using a Probability-Based Background Extraction Algorithm"", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
| GEOFFREY HINTON ET AL.: ""Distilling the Knowledge in a Neural Network"", 《ARXIV：1503.02531V1》 * |
| JOÃO F. HENRIQUES ET AL.: ""High-Speed Tracking with Kernelized Correlation Filters"", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 杨恒 等: ""基于随机背景建模的目标检测算法"", 《应用光学》 * |
| 王韦桦: ""智能视觉监控中运动目标检测与行为识别方法"", 《中国博士学位论文全文数据库信息科技辑》 * |
| 葛仕明 等: ""基于相关滤波器组的人眼定位方法"", 《网络新媒体技术》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785368B (en) * | 2017-11-13 | 2022-07-22 | 腾讯科技（深圳）有限公司 | Target tracking method and device |
| CN109785368A (en) * | 2017-11-13 | 2019-05-21 | 腾讯科技（深圳）有限公司 | A kind of method for tracking target and device |
| CN108986146A (en) * | 2017-12-11 | 2018-12-11 | 罗普特（厦门）科技集团有限公司 | A kind of correlation filtering tracking based on background information and adaptive recurrence label |
| CN108093153A (en) * | 2017-12-15 | 2018-05-29 | 深圳云天励飞技术有限公司 | Method for tracking target, device, electronic equipment and storage medium |
| CN110633597A (en) * | 2018-06-21 | 2019-12-31 | 北京京东尚科信息技术有限公司 | Driving region detection method and device |
| CN110633597B (en) * | 2018-06-21 | 2022-09-30 | 北京京东尚科信息技术有限公司 | Drivable region detection method and device |
| CN109215057B (en) * | 2018-07-31 | 2021-08-20 | 中国科学院信息工程研究所 | High-performance visual tracking method and device |
| CN109215057A (en) * | 2018-07-31 | 2019-01-15 | 中国科学院信息工程研究所 | A kind of high-performance visual tracking method and device |
| CN109919921A (en) * | 2019-02-25 | 2019-06-21 | 天津大学 | Based on the influence degree modeling method for generating confrontation network |
| CN109919921B (en) * | 2019-02-25 | 2023-10-20 | 天津大学 | Environmental impact degree modeling method based on generation countermeasure network |
| CN110111370A (en) * | 2019-05-15 | 2019-08-09 | 重庆大学 | A kind of vision object tracking methods based on TLD and the multiple dimensioned space-time characteristic of depth |
| CN110111370B (en) * | 2019-05-15 | 2023-05-30 | 重庆大学 | Visual object tracking method based on TLD and depth multi-scale space-time features |
| CN114973167A (en) * | 2022-07-28 | 2022-08-30 | 松立控股集团股份有限公司 | Multi-target tracking method based on off-line clustering and unsupervised contrast learning |
| CN114973167B (en) * | 2022-07-28 | 2022-11-04 | 松立控股集团股份有限公司 | Multi-target tracking method based on off-line clustering and unsupervised contrast learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106952293B (en) | 2020-02-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106952293A (en) | A kind of method for tracking target based on nonparametric on-line talking | |
| CN109800689B (en) | Target tracking method based on space-time feature fusion learning | |
| CN106897670B (en) | Express violence sorting identification method based on computer vision | |
| Huang et al. | Multiple target tracking by learning-based hierarchical association of detection responses | |
| CN105405154B (en) | Target object tracking based on color-structure feature | |
| Yang et al. | Context-aware visual tracking | |
| CN100573548C (en) | The method and apparatus of tracking bimanual movements | |
| KR102462934B1 (en) | Video analysis system for digital twin technology | |
| CN111932583A (en) | Space-time information integrated intelligent tracking method based on complex background | |
| CN108447080A (en) | Method for tracking target, system and storage medium based on individual-layer data association and convolutional neural networks | |
| CN110569843B (en) | Intelligent detection and identification method for mine target | |
| CN108664844A (en) | The image object semantics of convolution deep neural network identify and tracking | |
| CN106355604A (en) | Target image tracking method and system | |
| CN106570490B (en) | A kind of pedestrian's method for real time tracking based on quick clustering | |
| CN112836640A (en) | Single-camera multi-target pedestrian tracking method | |
| CN108734109B (en) | Visual target tracking method and system for image sequence | |
| Atanasov et al. | Hypothesis testing framework for active object detection | |
| CN111368634B (en) | Human head detection method, system and storage medium based on neural network | |
| CN108320306A (en) | Merge the video target tracking method of TLD and KCF | |
| CN113688797A (en) | Abnormal behavior identification method and system based on skeleton extraction | |
| CN109697727A (en) | Method for tracking target, system and storage medium based on correlation filtering and metric learning | |
| CN106447698A (en) | Multi-pedestrian tracking method and system based on distance sensor | |
| López-Rubio et al. | Anomalous object detection by active search with PTZ cameras | |
| CN106127798B (en) | Dense space-time contextual target tracking based on adaptive model | |
| CN109615641A (en) | Multiple target pedestrian tracking system and tracking based on KCF algorithm |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200927 Address after: Room 108, No. 318, Shuixiu Road, Jinze town (Xichen), Qingpu District, Shanghai 201700 Patentee after: Shanghai Yingpu Technology Co.,Ltd. Address before: 100025, Beijing, Gaobeidian Township North Garden Village North 166 media elite headquarters, 6 floor, room 609, Chaoyang District Patentee before: BEIJING MOVIEBOOK SCIENCE AND TECHNOLOGY Co.,Ltd. |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A Target Tracking Method Based on Non parametric Online Clustering Effective date of registration: 20230425 Granted publication date: 20200228 Pledgee: Bank of Communications Co.,Ltd. Beijing Tongzhou Branch Pledgor: Shanghai Yingpu Technology Co.,Ltd. Registration number: Y2023990000234 |
|
| PP01 | Preservation of patent right | ||
| PP01 | Preservation of patent right |
Effective date of registration: 20231128 Granted publication date: 20200228 |