CN106898015A - A kind of multi thread visual tracking method based on the screening of self adaptation sub-block - Google Patents
A kind of multi thread visual tracking method based on the screening of self adaptation sub-block Download PDFInfo
- Publication number
- CN106898015A CN106898015A CN201710035427.9A CN201710035427A CN106898015A CN 106898015 A CN106898015 A CN 106898015A CN 201710035427 A CN201710035427 A CN 201710035427A CN 106898015 A CN106898015 A CN 106898015A
- Authority
- CN
- China
- Prior art keywords
- sub
- block
- target
- candidate subchunk
- carried out
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
Abstract
The invention belongs to vision tracking field, and in particular to a kind of multi thread visual tracking method based on the screening of self adaptation sub-block, comprise the following steps：(1) target area is carried out into conspicuousness detection, candidate subchunk set is obtained with reference to uniform piecemeal；(2) multi-scale sampling is carried out to candidate subchunk, determines the larger sub-block of frequency domain response and corresponding yardstick, update candidate subchunk set；(3) estimation is carried out to the sub-block in candidate subchunk set, is merged by the multi thread of sub-block, it is determined that the center of tracking target；(4) by the current location of target, the corresponding Gaussian kernel of each sub-block locations is updated, the sub-block that will be unsatisfactory for requiring is reinitialized.The method of the present invention can remove the interference of background, make full use of the vision of middle level features restricted and upper language prior-constrained, so that the positioning of target is more accurate, also have the advantages that step is simple, amount of calculation is small, be adapted to carry out visual target tracking in the case of with blocking.
Description
Technical field
The invention belongs to vision tracking field, regarded more particularly, to a kind of multi thread based on the screening of self adaptation sub-block
Feel tracking.
Background technology
In practical application scene, tracking object often causes apparent part change, the table of target occur because the reason such as blocking
Seeing model needs to embody the apparent change of target.Therefore, in the apparent modeling of target generally using local feature (such as sub-block).
In Chinese invention patent《A kind of new video tracing method based on self adaptation piecemeal》In disclose it is a kind of from
Method of partition is adapted to, the adjust automatically piecemeal using video and the content information of image, the feature according to the target area for tracking
Strategy, improves video tracking effect.In Chinese invention patent《A kind of new video tracing method based on partition strategy》In
Disclose the partition strategy of a seed block, the color histogram according to fritter with its around fritter color histogram between phase
Judge whether current block and surrounding fritter block like degree, be detected the fritter for blocking in particle filter below
The weight shared by it is reduced during relationship change coefficient between calculating block in algorithm, so that it is accurate to tracking to reduce eclipse phenomena
The influence of true property.In Chinese invention patent《A kind of real-time video tracing method》In, by tracking Target Segmentation into the side of sub-block
Formula compresses characteristics of image, and by building multiple dimensioned candidate region with the dimensional variation for adapting to track target and quick shifting
It is dynamic, finally calculate the correlation between characteristic vector to reach the mesh of video tracking using KCF (coring correlation filter) algorithm
's.
In these methods, use local feature to apparent modeling the situation such as to adapt to block, by uniform piecemeal
Target is divided into sub-block or fritter by method, and the quantity of sub-block and the locus of sub-block are usually fixed.It is evenly dividing
The sub-block quantity for arriving is more, in follow-up trace flow, either using particle filter tracking framework, or KCF tracking box
Frame, is required for processing each sub-block, and the amount of calculation of algorithm is directly proportional to the quantity of sub-block.And in these sub-blocks, deposit
In substantial amounts of information redundancy, and the background information that interference is produced to tracking result can be introduced, the drift during vision can be caused to track
Shifting problem, influences the accuracy of tracking result.
Due to there is drawbacks described above and deficiency, this area is needed badly to make and is further improved, and designs a kind of vision
Method for tracking target, can overcome the defect for being fixed mode method of partition to tracking target area in the prior art,
The need for meet the target following under complicated tracking scene.
The content of the invention
For the disadvantages described above or Improvement requirement of prior art, the invention provides a kind of based on the screening of self adaptation sub-block
Multi thread visual tracking method, the method is a kind of for the mesh for the complicated tracking scene such as blocking and improving Vision Tracking performance
Mark apparent expression and object localization method, sub-block divided and screened according to priori, with more vision significance and
Frequency domain response sub-block high represents target, merge it is apparent, spatially and temporally motion of the information to target estimate, so as to improve
The accuracy and the adaptability to different scenes of vision tracking.The method can be used in the system such as the magnitude of traffic flow or society's monitoring
Target person or vehicle are monitored automatically.
To achieve the above object, according to one aspect of the present invention, there is provided a kind of based on many of self adaptation sub-block screening
Clue visual tracking method, it is characterised in that specifically include following steps：
S1. image information is gathered, image information divide to obtain target area, target area is uniformly divided
Block, and the picture of target area is carried out into conspicuousness detection, candidate subchunk set is obtained according to visual signature significance degree；
S2. carry out multi-scale sampling to the candidate subchunk in the candidate subchunk set that is obtained in step S1, and to sample into
Row core correlation filtering, retains the larger sub-block of candidate subchunk set intermediate frequency domain response, determines the yardstick of these sub-blocks, updates candidate
Sub-block set；
S3. using the sub-block in candidate subchunk set as tracking target, estimation is carried out to sub-block, calculates each sub-block
Apparent clue aci, spatial distribution clue sciWith movement locus clue mci, thus it is calculated the current location of tracking target；
S4. the current location of the tracking target by obtaining in step S3, to each sub-block in current goal region
Put corresponding Gaussian kernel to be updated, the sub-block that will be unsatisfactory for space cohesion or Movement consistency is reinitialized.
It is further preferred that in step sl, carry out dividing the process for obtaining candidate subchunk set specifically including to sub-block
Following steps：
S1.1 gathers a two field picture, obtains target area, and carry out uniform piecemeal to target area；
S1.2 is input into target area picture as conspicuousness detection algorithm, obtains the Saliency maps of target area；
S1.3 carries out Gaussian smoothing removal noise spot to the Saliency maps obtained in step S1.2；
S1.4 takes the maximum of points position (x, y) of Saliency maps, the partition determining maximum point according to step S1.1
Sub-block where position (x, y), is added into candidate subchunk set；
Be updated for Saliency maps by S1.5, the sub-block in removal step 1.4 where maximum of points position, repeat step
S1.4, obtains new sub-block, and registration calculating is carried out with the sub-block in candidate subchunk set, if respectively less than threshold tau, will new son
Block adds candidate subchunk set；
Preferably, in step S1.2, the conspicuousness detection algorithm is FT algorithms.
Preferably, when yardstick screening is carried out to the candidate subchunk after sampling, following steps are specifically included：
S2.1 determines the region of search scope and region of search template size of the sub-block in candidate subchunk set；
S2.2 is according to region of search scope and region of search template size initialization Gaussian kernel；
S2.3 is zoomed in and out the domain samples of each sub-block by region of search template size, and extracts HOG features, is carried out
FFT obtains the multiple dimensioned sample of sub-block；
The multiple dimensioned sample of sub-block is carried out correlation filtering by S2.4 with corresponding Gaussian kernel, obtains frequency domain response, removal
The relatively low sub-block of frequency domain response, updates candidate subchunk set；
S2.5 calculates the PSR values of each sample according to equation below, and the PSR values to all sub-blocks under each yardstick are asked
With, wherein：
Wherein gmaxIt is the peak value of response, s is the secondary lobe region of response, μsAnd σsThe respectively average and standard in secondary lobe region
Difference；
S2.6 analyzes the summing value of PSR values, and the yardstick of the maximum sub-block of PSR values summing value is defined as into best scale.
Preferably, in step S3, comprising the following steps that for estimation is carried out to the sub-block in candidate subchunk set：
S3.1 judges the motion of sub-block, rejected in candidate subchunk set displacement in adjacent two field picture it is excessive or
Direction of motion relative mean displacement direction in adjacent image frame offsets excessive sub-block；
S3.2 calculates the apparent clue ac of each sub-blocki, spatial distribution clue sciWith movement locus clue mci, formula is such as
Under：
aci=PSRi
Wherein PSRiIt is i-th piece after normalization of PSR values, M represents the number by remaining sub-block after step S3.1,
(xi,yi) represent i-th piece of coordinate, (xs,ys) represent all pieces of center, σsFor criterion distance is poor, (Δ xi,Δyi) table
Show i-th piece of motion vector, (Δ xm,Δym) represent all pieces of average motion vector, σmFor motion criteria is poor；
S3.3 calculates wi=rciwr+sciws+mciwm, wherein (wr,ws,wm) represent three kinds of weights of clue；
S3.4 makesObtain tracking the current location of target.
Specifically, the flow of visual target tracking is generally divided into apparent modeling, target estimation and positioning, model modification
Etc. several stages.It is as follows that the present invention tracks the technical scheme that each stage used in vision：
The target apparent modelling phase.To enable the middle level features (i.e. sub-block) being indicated apparent to target to have more expression
Power, conspicuousness detection is carried out in the frame of video the 1st first to the target area for obtaining by hand, and the Saliency maps for obtaining are drawn as sub-block
The priori divided.The conspicuousness of image reflects the significance level of image different parts visual signature, often visually more prominent
The part for going out is easier to be captured, and is more beneficial for the positioning of target.Conspicuousness detection algorithm can use existing algorithm, the present invention
In illustrated by taking FT conspicuousness detection algorithms as an example.The Saliency maps obtained by FT algorithms there may be exceptional value, be removal
Saliency maps are carried out Gaussian smoothing by noise spot.Quickly uniform piecemeal, significance value after Gaussian smoothing are carried out to target area
Sub-block corresponding to the best part elects the 1st candidate subchunk as.To the Saliency maps weight behind the 1st candidate subchunk region of removal
Multiple aforesaid operations, obtain the 2nd sub-block.To ensure the diversity of sub-block, it is to avoid sub-block concentrates on certain local cell of target
Domain, calculates the 2nd sub-block and the 1st registration of candidate subchunk, if registration is higher than certain threshold value, the 2nd sub-block can quilt
Elect candidate subchunk as.Repeat the above steps, obtain candidate subchunk collection.
For adapt to that Moving Objects in practical application often occur draws near or dimensional variation from the close-by examples to those far off, to each
Candidate subchunk carries out multi-scale sampling using image pyramid.Followed by the son that frequency domain response determines to be best suitable in present frame
Block yardstick.To the multiple dimensioned sample of sub-block, the HOG features of image are extracted, carry out FFT, obtain the sample of frequency domain.Frequency domain
Image pattern carries out correlation filtering with the Gaussian kernel of correspondence position, asks for response and further sub-block is screened, and calculate
The PSR values of block.PSR values summation to all pieces under current scale, the maximum yardstick of value is the yardstick of current goal.
The sub-block obtained by the screening in this stage, can protrude the vision significance of target, and it is current to adapt to target
Dimensional variation.The quantity of sub-block has compared to uniform piecemeal and significantly reduces.
Target estimation and positioning stage.Different types of clue contains the different attribute of object, based on single line
The vision tracking of rope is often limited to tracking the adaptability of scene, and the vision tracking based on middle level features sub-block is even more such as
This.In order to the screening technique of the apparent modelling phase sub-block of target matches, apparent clue, the sky of sub-block of present invention synthesis sub-block
Between be distributed the movement locus clue of clue and sub-block and realize the estimation of target.Wherein, the apparent clue of sub-block shows son
The vision reliability of block, the spatial distribution of sub-block reflects cohesion degree of the sub-block to target's center position, the movement locus of sub-block
Clue constrains the Movement consistency of each sub-block.In three kinds of clues based on vision reliability, with spatial distribution and kinematic constraint
Supplemented by, the final positioning of target is carried out by way of multi thread is weighted.
The model modification stage.Model modification includes two aspects in the present invention.One is to obtain target by estimation
Behind current location, the corresponding Gaussian kernel of each sub-block locations of current goal can be updated.Two is to being unsatisfactory in space
The sub-block of poly- property or Movement consistency re-starts initialization.
In general, by the contemplated above technical scheme of the present invention compared with prior art, with advantages below and
Beneficial effect：
(1) partition and screening technique that the present invention is provided, can find the middle level features that can more preferably represent target, go
Except the interference of background；The target estimation framework of multi thread, can make full use of the restricted and high-rise language of the vision of middle level features
That says is prior-constrained so that the positioning of target is more accurate.
(2) method of the present invention, by after multiple dimensioned screening sample, eliminating a large amount of invalid sub-blocks, reduces information superfluous
It is remaining, the amount of calculation of algorithm can be reduced, improve the accuracy and speed of result of calculation.
(3) method of the present invention step is simple and amount of calculation is small, tracking result is accurate, is adapted in the case of with blocking
Carry out visual target tracking.
Brief description of the drawings
Fig. 1 is partition schematic diagram of the invention；
Fig. 2 be along partition described in Fig. 1 after, carry out sub-block and yardstick using image pyramid and KCF tracking and screen
Schematic diagram；
Fig. 3 is multi thread target estimation and positioning schematic diagram；
The step of Fig. 4 is multi thread visual tracking method of the invention flow chart.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as additionally, technical characteristic involved in invention described below each implementation method
Not constituting conflict each other can just be mutually combined.
The step of multi thread visual tracking method as shown in Figure 4 flow chart, wherein, step 1 and step 2 are in Target Modeling
Stage implements, and step 3 is implemented in target estimation and positioning stage.
Step 1 partition, as shown in Figure 1.
Step 1.1 gathers the 1st two field picture, and target area is obtained by hand.Uniform piecemeal is carried out to target area.
Step 1.2 is input into target area picture as conspicuousness detection algorithm (such as FT algorithms), obtains Target Photo
Saliency maps.
Step 1.3 carries out Gaussian smoothing removal noise spot to Saliency maps.
Step 1.4 takes Saliency maps maximum of points position (x, y), it is determined that block where (x, y), is added into candidate subchunk collection
Close.Saliency maps update.
Step 1.5 repeat step 1.4, obtains new sub-block.Registration calculating is carried out with the sub-block in candidate subchunk set, such as
Fruit is respectively less than threshold tau, then new sub-block is added into candidate subchunk set.
Step 2 sub-block scale is screened, as shown in Figure 2.
Step 2.1 determines the region of search scope and region of search template size of sub-block.
Step 2.2 is according to region of search scope and region of search template size initialization Gaussian kernel.
Step 2.3 is zoomed in and out the domain samples of each sub-block by region of search template size, and extracts HOG features,
Carry out the multiple dimensioned sample that FFT obtains sub-block.
The multiple dimensioned sample of sub-block is carried out correlation filtering by step 2.4 with corresponding Gaussian kernel, obtains frequency domain response, is protected
The sub-block that frequency domain response is larger is stayed, candidate subchunk set is updated.
Step 2.5 calculates the PSR (peak-secondary lobe ratios, Peak-to-Sidelobe of each sample according to equation below
Ratio) value, the PSR values to all sub-blocks under each yardstick are sued for peace.Wherein gmaxIt is the peak value of response,
S is the secondary lobe region of response, μsAnd σsThe respectively average and standard deviation in secondary lobe region.
Step 2.6 determines best scale, that is, cause the maximum yardstick of PSR values summing value.
Step 3 multi thread estimation is positioned with target, as shown in Figure 3.
Step 3.1 judges the motion of sub-block.Rejected in candidate subchunk set displacement in consecutive frame it is excessive or
Direction of motion relative mean displacement direction in consecutive frame offsets excessive sub-block.
Step 3.2 calculates the apparent clue ac of each sub-blocki, spatial distribution clue sciWith movement locus clue mci, formula
It is as follows：
aci=PSRi
Wherein PSRiIt is i-th piece after normalization of PSR values, M represents the number by remaining sub-block after step 3.1, (xi,
yi) represent i-th piece of coordinate, (xs,ys) represent all pieces of center, σsFor criterion distance is poor, (Δ xi,Δyi) represent the
The motion vector of i blocks, (Δ xm,Δym) represent all pieces of average motion vector, σmFor motion criteria is poor.
Step 3.3 calculates wi=rciwr+sciws+mciwm, wherein (wr,ws,wm) represent three kinds of weights of clue, this example
In take (0.8,01,0.1).
Step 3.4 makesObtain target's center position.
Step 4 model modification
Gaussian kernel of the step 4.1 according to each correspondence sub-block locations of present frame target update that step 3.4 is obtained.
Step 4.2 is unsatisfactory for space cohesion or the sub-block of Movement consistency is reinitialized.
As it will be easily appreciated by one skilled in the art that the foregoing is only presently preferred embodiments of the present invention, it is not used to
The limitation present invention, all any modification, equivalent and improvement made within the spirit and principles in the present invention etc., all should include
Within protection scope of the present invention.
Claims (5)
1. it is a kind of based on self adaptation sub-block screening multi thread visual tracking method, it is characterised in that specifically include following steps：
S1. image information is gathered, image information divide to obtain target area, uniform piecemeal is carried out to target area, and
The picture of target area is carried out into conspicuousness detection, candidate subchunk set is obtained according to visual signature significance degree；
S2. multi-scale sampling is carried out to the candidate subchunk in the candidate subchunk set that is obtained in step S1, and core is carried out to sampling
Correlation filtering, retains the larger sub-block of candidate subchunk set intermediate frequency domain response, determines the yardstick of these sub-blocks, updates candidate subchunk
Set；
S3. estimation is carried out to the sub-block in candidate subchunk set, calculates the apparent clue ac of each sub-blocki, spatial distribution line
Rope sciWith movement locus clue mci, thus it is calculated the current location of tracking target；
S4. the current location of the tracking target by obtaining in step S3, to each sub-block locations pair in current goal region
The Gaussian kernel answered is updated, and the sub-block that will be unsatisfactory for space cohesion or Movement consistency is reinitialized.
2. the method for claim 1, it is characterised in that in step sl, to sub-block divide obtaining candidate subchunk
The process of set specifically includes following steps：
S1.1 gathers a two field picture, obtains target area, and carry out uniform piecemeal to target area；
S1.2 is input into target area picture as conspicuousness detection algorithm, obtains the Saliency maps of target area；
S1.3 carries out Gaussian smoothing removal noise spot to the Saliency maps obtained in step S1.2；
S1.4 takes the maximum of points position (x, y) of Saliency maps, the partition determining maximum point position according to step S1.1
Sub-block where (x, y), candidate subchunk set is added by the sub-block；
Be updated for Saliency maps by S1.5, the sub-block in removal step 1.4 where maximum of points position, repeat step S1.4,
New sub-block is obtained, registration calculating is carried out with the sub-block in candidate subchunk set, if respectively less than threshold tau, new sub-block is added
Enter candidate subchunk set.
3. method as claimed in claim 1 or 2, it is characterised in that in step S1.2, the conspicuousness detection algorithm is FT
Algorithm.
4. method as claimed in claim 3, it is characterised in that in step 2, yardstick sieve is carried out to the candidate subchunk after sampling
When selecting, following steps are specifically included：
S2.1 determines the region of search scope and region of search template size of the sub-block in candidate subchunk set；
S2.2 is according to region of search scope and region of search template size initialization Gaussian kernel；
S2.3 is zoomed in and out the domain samples of each sub-block by region of search template size, and extracts HOG features, carries out FFT
Conversion obtains the multiple dimensioned sample of sub-block；
The multiple dimensioned sample of sub-block is carried out correlation filtering by S2.4 with corresponding Gaussian kernel, obtains frequency domain response, removes frequency domain
The relatively low sub-block of response, updates candidate subchunk set；
S2.5 calculates the PSR values of each sample according to equation below, and the PSR values to all sub-blocks under each yardstick are sued for peace, its
In：
Wherein gmaxIt is the peak value of response, s is the secondary lobe region of response, μsAnd σsThe respectively average and standard deviation in secondary lobe region；
S2.6 analyzes the summing value of PSR values, and the yardstick of the maximum sub-block of PSR values summing value is defined as into best scale.
5. method as claimed in claim 4, it is characterised in that in step S3, transported to the sub-block in candidate subchunk set
Move comprising the following steps that for estimation：
S3.1 judges the motion of sub-block, displacement in adjacent two field picture is rejected in candidate subchunk set excessive or adjacent
Direction of motion relative mean displacement direction in picture frame offsets excessive sub-block；
S3.2 calculates the apparent clue ac of each sub-blocki, spatial distribution clue sciWith movement locus clue mci, formula is as follows：
aci=PSRi
Wherein PSRiIt is i-th piece after normalization of PSR values, M represents the number by remaining sub-block after step S3.1, (xi,yi)
Represent i-th piece of coordinate, (xs,ys) represent all pieces of center, σsFor criterion distance is poor, (Δ xi,Δyi) represent i-th piece
Motion vector, (Δ xm,Δym) represent all pieces of average motion vector, σmFor motion criteria is poor；
S3.3 calculates wi=rciwr+sciws+mciwm, wherein (wr,ws,wm) represent three kinds of weights of clue；
S3.4 makesObtain tracking the current location of target.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710035427.9A CN106898015B (en) | 2017-01-17 | 2017-01-17 | A kind of multi thread visual tracking method based on the screening of adaptive sub-block |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710035427.9A CN106898015B (en) | 2017-01-17 | 2017-01-17 | A kind of multi thread visual tracking method based on the screening of adaptive sub-block |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106898015A true CN106898015A (en) | 2017-06-27 |
| CN106898015B CN106898015B (en) | 2019-09-24 |
Family
ID=59197898
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710035427.9A Expired - Fee Related CN106898015B (en) | 2017-01-17 | 2017-01-17 | A kind of multi thread visual tracking method based on the screening of adaptive sub-block |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106898015B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107767401A (en) * | 2017-10-16 | 2018-03-06 | 武汉沃德自动化技术有限公司 | Infrared target method for real time tracking and device based on core correlation filtering |
| CN107833240A (en) * | 2017-11-09 | 2018-03-23 | 华南农业大学 | The target trajectory extraction of multi-track clue guiding and analysis method |
| CN108053425A (en) * | 2017-12-25 | 2018-05-18 | 北京航空航天大学 | A kind of high speed correlation filtering method for tracking target based on multi-channel feature |
| CN108198209A (en) * | 2017-12-22 | 2018-06-22 | 天津理工大学 | It is blocking and dimensional variation pedestrian tracking algorithm |
| CN108961226A (en) * | 2018-06-21 | 2018-12-07 | 安徽工业大学 | A kind of method of insulator target following in transmission line-oriented inspection video |
| CN108985153A (en) * | 2018-06-05 | 2018-12-11 | 成都通甲优博科技有限责任公司 | A kind of face recognition method and device |
| CN109146918A (en) * | 2018-06-11 | 2019-01-04 | 西安电子科技大学 | A kind of adaptive related objective localization method based on piecemeal |
| CN109685831A (en) * | 2018-12-20 | 2019-04-26 | 山东大学 | Method for tracking target and system based on residual error layering attention and correlation filter |
| CN109711431A (en) * | 2018-11-27 | 2019-05-03 | 哈尔滨工业大学（深圳） | The method for tracking target of local block convolution, system and storage medium at one |
| CN109864806A (en) * | 2018-12-19 | 2019-06-11 | 江苏集萃智能制造技术研究所有限公司 | The Needle-driven Robot navigation system of dynamic compensation function based on binocular vision |
| CN110246155A (en) * | 2019-05-17 | 2019-09-17 | 华中科技大学 | One kind being based on the alternate anti-shelter target tracking of model and system |
| CN110322473A (en) * | 2019-07-09 | 2019-10-11 | 四川大学 | Target based on significant position is anti-to block tracking |
| CN111860189A (en) * | 2020-06-24 | 2020-10-30 | 北京环境特性研究所 | Target tracking method and device |
| CN112614154A (en) * | 2020-12-08 | 2021-04-06 | 深圳市优必选科技股份有限公司 | Target tracking track obtaining method and device and computer equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105118071A (en) * | 2015-08-04 | 2015-12-02 | 山东大学 | Novel video tracking method based on self-adaptive partitioning |
| CN105654139A (en) * | 2015-12-31 | 2016-06-08 | 北京理工大学 | Real-time online multi-target tracking method adopting temporal dynamic appearance model |
-
2017
- 2017-01-17 CN CN201710035427.9A patent/CN106898015B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105118071A (en) * | 2015-08-04 | 2015-12-02 | 山东大学 | Novel video tracking method based on self-adaptive partitioning |
| CN105654139A (en) * | 2015-12-31 | 2016-06-08 | 北京理工大学 | Real-time online multi-target tracking method adopting temporal dynamic appearance model |
Non-Patent Citations (2)
| Title |
|---|
| ZHENG ZHU等: "STD: A Stereo Tracking Dataset for Evaluating Binocular Tracking Algorithms", 《PROCEEDINGS OF THE 2016 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND BIOMIMETICS》 * |
| 潘振福等: "多尺度估计的核相关滤波器目标跟踪方法", 《激光与光电子学进展》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107767401B (en) * | 2017-10-16 | 2020-01-14 | 武汉沃德自动化技术有限公司 | Infrared target real-time tracking method and device based on nuclear correlation filtering |
| CN107767401A (en) * | 2017-10-16 | 2018-03-06 | 武汉沃德自动化技术有限公司 | Infrared target method for real time tracking and device based on core correlation filtering |
| CN107833240A (en) * | 2017-11-09 | 2018-03-23 | 华南农业大学 | The target trajectory extraction of multi-track clue guiding and analysis method |
| CN107833240B (en) * | 2017-11-09 | 2020-04-17 | 华南农业大学 | Target motion trajectory extraction and analysis method guided by multiple tracking clues |
| CN108198209A (en) * | 2017-12-22 | 2018-06-22 | 天津理工大学 | It is blocking and dimensional variation pedestrian tracking algorithm |
| CN108198209B (en) * | 2017-12-22 | 2020-05-01 | 天津理工大学 | People tracking method under the condition of shielding and scale change |
| CN108053425A (en) * | 2017-12-25 | 2018-05-18 | 北京航空航天大学 | A kind of high speed correlation filtering method for tracking target based on multi-channel feature |
| CN108985153A (en) * | 2018-06-05 | 2018-12-11 | 成都通甲优博科技有限责任公司 | A kind of face recognition method and device |
| CN109146918A (en) * | 2018-06-11 | 2019-01-04 | 西安电子科技大学 | A kind of adaptive related objective localization method based on piecemeal |
| CN108961226A (en) * | 2018-06-21 | 2018-12-07 | 安徽工业大学 | A kind of method of insulator target following in transmission line-oriented inspection video |
| CN109711431A (en) * | 2018-11-27 | 2019-05-03 | 哈尔滨工业大学（深圳） | The method for tracking target of local block convolution, system and storage medium at one |
| CN109864806A (en) * | 2018-12-19 | 2019-06-11 | 江苏集萃智能制造技术研究所有限公司 | The Needle-driven Robot navigation system of dynamic compensation function based on binocular vision |
| CN109685831B (en) * | 2018-12-20 | 2020-08-25 | 山东大学 | Target tracking method and system based on residual layered attention and correlation filter |
| CN109685831A (en) * | 2018-12-20 | 2019-04-26 | 山东大学 | Method for tracking target and system based on residual error layering attention and correlation filter |
| CN110246155A (en) * | 2019-05-17 | 2019-09-17 | 华中科技大学 | One kind being based on the alternate anti-shelter target tracking of model and system |
| CN110246155B (en) * | 2019-05-17 | 2021-05-18 | 华中科技大学 | Anti-occlusion target tracking method and system based on model alternation |
| CN110322473A (en) * | 2019-07-09 | 2019-10-11 | 四川大学 | Target based on significant position is anti-to block tracking |
| CN111860189A (en) * | 2020-06-24 | 2020-10-30 | 北京环境特性研究所 | Target tracking method and device |
| CN111860189B (en) * | 2020-06-24 | 2024-01-19 | 北京环境特性研究所 | Target tracking method and device |
| CN112614154A (en) * | 2020-12-08 | 2021-04-06 | 深圳市优必选科技股份有限公司 | Target tracking track obtaining method and device and computer equipment |
| CN112614154B (en) * | 2020-12-08 | 2024-01-19 | 深圳市优必选科技股份有限公司 | Target tracking track acquisition method and device and computer equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106898015B (en) | 2019-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106898015A (en) | A kind of multi thread visual tracking method based on the screening of self adaptation sub-block | |
| KR100224752B1 (en) | Target tracking method and apparatus | |
| JP2870415B2 (en) | Area division method and apparatus | |
| EP1859411B1 (en) | Tracking objects in a video sequence | |
| CN100530238C (en) | Segmentation unit for and method of determining a second segment and image processing apparatus | |
| CN110517288A (en) | Real-time target detecting and tracking method based on panorama multichannel 4k video image | |
| CN108876820B (en) | Moving target tracking method under shielding condition based on mean shift | |
| CN108446694B (en) | Target detection method and device | |
| US6704433B2 (en) | Human tracking device, human tracking method and recording medium recording program thereof | |
| CN109086724B (en) | Accelerated human face detection method and storage medium | |
| EP2128818A1 (en) | Method of moving target tracking and number accounting | |
| JPH09218950A (en) | System for detecting object from multi-eye image | |
| CN107182036A (en) | The adaptive location fingerprint positioning method merged based on multidimensional characteristic | |
| CN106204570B (en) | A kind of angular-point detection method based on non-causal fractional order gradient operator | |
| CN105740945A (en) | People counting method based on video analysis | |
| CN109001757A (en) | A kind of parking space intelligent detection method based on 2D laser radar | |
| CN105956544B (en) | A method of it extracts the remote sensing image road intersection based on structure index feature | |
| CN106295639A (en) | A kind of virtual reality terminal and the extracting method of target image and device | |
| CN108154150B (en) | Significance detection method based on background prior | |
| CN112164093A (en) | Automatic person tracking method based on edge features and related filtering | |
| CN112329764A (en) | Infrared dim target detection method based on TV-L1 model | |
| WO2012122588A1 (en) | Extraction processes | |
| Zhang et al. | Image sequence segmentation using 3-D structure tensor and curve evolution | |
| CN104978558B (en) | The recognition methods of target and device | |
| JP4020982B2 (en) | Moving image processing device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190924 Termination date: 20210117 |