CN108665485A - A kind of method for tracking target merged with twin convolutional network based on correlation filtering - Google Patents
A kind of method for tracking target merged with twin convolutional network based on correlation filtering Download PDFInfo
- Publication number
- CN108665485A CN108665485A CN201810342324.1A CN201810342324A CN108665485A CN 108665485 A CN108665485 A CN 108665485A CN 201810342324 A CN201810342324 A CN 201810342324A CN 108665485 A CN108665485 A CN 108665485A
- Authority
- CN
- China
- Prior art keywords
- target
- frame image
- convolutional network
- video sequence
- present
- 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
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/262—Analysis of motion using transform domain methods, e.g. Fourier domain methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/269—Analysis of motion using gradient-based methods
-
- 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
-
- 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
- G06T2207/20048—Transform domain processing
- G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]
Abstract
The invention discloses a kind of method for tracking target merged with twin convolutional network based on correlation filtering, including：Using the target signature of (t 1) frame image of the first convolutional network extraction known target position, the search characteristics figure of t frame images is extracted using the second convolutional network；Fast Fourier Transform (FFT) is carried out to the target signature of (t 1) frame image and obtains the target area of (t 1) frame image, correlation filtering is carried out to the search characteristics figure of t frame images and obtains the region of search of t frame images, calculate the cross correlation between the region of search and the target area of (t 1) frame image of t frame images, the target score figure of t frame images is obtained, the target location of t frame images is obtained according to the target score figure of t frame images；And then the target location of each frame image in video sequence is obtained, realize the target following to video sequence.The present invention can overcome illumination, block, and the influence of posture and scale carries out real-time modeling method.
Description
Technical field
The invention belongs to the crossing domain of computer vision, depth convolutional network and pattern-recognition, more particularly, to
A kind of method for tracking target merged with twin convolutional network based on correlation filtering.
Background technology
Target following has very important status in computer vision, however due to the complexity of natural scene, mesh
It marks to the sensibility of illumination variation, tracks the requirement to real-time and robustness, and block, the factors such as posture and dimensional variation
Presence so that tracking problem is still highly difficult.Traditional method for tracking target, can not the feature abundant to Objective extraction make
Target strict differences and background are susceptible to tracking drift phenomenon, therefore can not track target for a long time；With deep learning
It rising, existing general convolutional neural networks can effectively extract the abundant feature of target, but network parameter is excessive, if
It to track online, it is virtually impossible to meet the requirement of real-time performance, Practical Project utility value is limited.
The high performance calculating device such as raising and GPU due to hardware performance is popularized, and the real-time of tracking is no longer difficult
To overcome the problems, such as, effective target appearance model is only vital during tracking.The essence of target following is one
The process of a similarity measurement has naturally excellent due to the special construction of twin convolutional network in terms of similarity measurement
Gesture, and there is convolutional coding structure, abundant feature can be extracted for target following.Pure is used based on twin convolutional network
Off-line training, it is online to track, although real-time can be met the requirements in high performance computation equipment, not in line target mould
The problems such as dynamic of plate updates, and is difficult to overcome illumination, blocks, posture and scale.
Invention content
For the disadvantages described above or Improvement requirement of the prior art, the present invention provides one kind based on correlation filtering and twin volume
The method for tracking target of the product network integration, thus solving the prior art does not have the dynamic of online target template to update, and is difficult to
Overcome illumination, block, the technical issues of posture and scale.
To achieve the above object, the present invention provides a kind of target merged with twin convolutional network based on correlation filtering with
Track method, the twin convolutional network are 2 identical first convolutional networks and the second convolutional network, the target following
Method includes：
(1) target signature for utilizing (t-1) frame image of the first convolutional network extraction known target position utilizes the
Two convolutional networks extract the search characteristics figure of t frame images；
(2) Fast Fourier Transform (FFT) is carried out to the target signature of (t-1) frame image and obtains the mesh of (t-1) frame image
Region is marked, carrying out correlation filtering to the search characteristics figure of t frame images obtains the region of search of t frame images, calculates t frame figures
Cross correlation between the region of search and the target area of (t-1) frame image of picture, obtains the target score of t frame images
Figure, the target location of t frame images is obtained according to the target score figure of t frame images；
Wherein, the 1st frame image in video sequence is demarcated when t is 2 in t >=2, executes step (1)-(2) and obtains
To the target location of the 2nd frame image, when t is 3, executes step (1)-(2) and obtain the target location of the 3rd frame image, with such
It pushes away, obtains the target location of each frame image in video sequence, realize the target following to video sequence.
Further, the correlation filtering in step (2) includes：
Smothing filtering is carried out using the search characteristics figure of cosine window function or sine-window function pair t frame images,
Then Fast Fourier Transform (FFT) is used from spatial transform to frequency domain, to obtain the search characteristics figure of the t frame images after smothing filtering
To the region of search of t frame images.
Further, the first convolutional network and the second convolutional network include five convolutional layers, five convolutional layers
There are one down-sampling pond layers respectively after preceding two layers of convolution.
Further, twin convolutional network is trained convolutional network, and the training method of the twin convolutional network is：
Collecting sample video sequence is utilized to being marked per the target location of frame sample image in Sample video sequence
Sample video sequence training convolutional network after label is joined in training with the minimum objective optimization network of logarithm loss function
Number, obtains trained convolutional network.
Further, logarithm loss function is：
L (y, v)=log (1+exp (- yv))
Wherein, v is the confidence score of the target location of sample image, and y is the label of the target location of sample image, l (y,
V) it is error amount.
Further, method for tracking target further includes：
When t is 2, the 1st frame image in video sequence is demarcated, step (1)-(2) is executed and obtains the 1st frame image
Region of search and the target area of the 2nd frame image between cross correlation pass through and minimize logarithm loss using cross correlation
Function backpropagation updates the network parameter of twin convolutional network.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect：
(1) the method for the present invention combines twin convolutional network using correlation filtering, and target following is improved using correlation filtering
Real-time feature-rich and accurate measured similarity side are extracted, so that the method for the present invention can using twin convolutional network
It effectively to extract the feature-rich of target, while being realized using correlation filtering and smoothly being updated in line template, reach efficient reality
When target following.
(2) the logarithm loss function that the present invention uses accelerates the training speed of network, is effectively prevented in training process
There is gradient disappearance or gradient disperse.Can it is accurate, robust, carry out target following in real time.The present invention uses cosine window
Function or sine-window function carry out smothing filtering, eliminate the noise that Fourier transformation generates on the image.Using quick
Convolution operation inside spatial domain can be become the operation of the dot product inside frequency domain by Fourier transformation, be significantly reduced calculation amount.
Description of the drawings
Fig. 1 is the flow chart of method for tracking target provided in an embodiment of the present invention；
Fig. 2 is the flow chart of method for tracking target in detail provided in an embodiment of the present invention；
Fig. 3 is the flow chart of correlation filtering provided in an embodiment of the present invention；
Fig. 4 (a) is the first frame image of the first video sequence provided in an embodiment of the present invention；
Fig. 4 (b) is the first frame image of the second video sequence provided in an embodiment of the present invention；
Fig. 4 (c) is the first frame image of third video sequence provided in an embodiment of the present invention；
Fig. 4 (d) is the first frame image of the 4th video sequence provided in an embodiment of the present invention；
Fig. 4 (e) is the first frame image of the 5th video sequence provided in an embodiment of the present invention；
Fig. 4 (f) is the first frame image of the 6th video sequence provided in an embodiment of the present invention；
Fig. 5 (a1) is provided in an embodiment of the present invention using the progress target following of the first video sequence of the method for the present invention pair
50th frame image；
Fig. 5 (a2) is provided in an embodiment of the present invention using the progress target following of the first video sequence of the method for the present invention pair
100th frame image；
Fig. 5 (a3) is provided in an embodiment of the present invention using the progress target following of the first video sequence of the method for the present invention pair
150th frame image；
Fig. 5 (b1) is provided in an embodiment of the present invention using the progress target following of the second video sequence of the method for the present invention pair
50th frame image；
Fig. 5 (b2) is provided in an embodiment of the present invention using the progress target following of the second video sequence of the method for the present invention pair
100th frame image；
Fig. 5 (b3) is provided in an embodiment of the present invention using the progress target following of the second video sequence of the method for the present invention pair
150th frame image；
Fig. 5 (c1) be it is provided in an embodiment of the present invention using the method for the present invention to third video sequence carry out target following
50th frame image；
Fig. 5 (c2) be it is provided in an embodiment of the present invention using the method for the present invention to third video sequence carry out target following
100th frame image；
Fig. 5 (c3) be it is provided in an embodiment of the present invention using the method for the present invention to third video sequence carry out target following
150th frame image；
Fig. 5 (d1) is provided in an embodiment of the present invention using the progress target following of the 4th video sequence of the method for the present invention pair
50th frame image；
Fig. 5 (d2) is provided in an embodiment of the present invention using the progress target following of the 4th video sequence of the method for the present invention pair
100th frame image；
Fig. 5 (d3) is provided in an embodiment of the present invention using the progress target following of the 4th video sequence of the method for the present invention pair
150th frame image；
Fig. 5 (e1) is provided in an embodiment of the present invention using the progress target following of the 5th video sequence of the method for the present invention pair
50th frame image；
Fig. 5 (e2) is provided in an embodiment of the present invention using the progress target following of the 5th video sequence of the method for the present invention pair
100th frame image；
Fig. 5 (e3) is provided in an embodiment of the present invention using the progress target following of the 5th video sequence of the method for the present invention pair
150th frame image；
Fig. 5 (f1) is provided in an embodiment of the present invention using the progress target following of the 6th video sequence of the method for the present invention pair
50th frame image；
Fig. 5 (f2) is provided in an embodiment of the present invention using the progress target following of the 6th video sequence of the method for the present invention pair
100th frame image；
Fig. 5 (f3) is provided in an embodiment of the present invention using the progress target following of the 6th video sequence of the method for the present invention pair
150th frame image.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, right
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 in addition, technical characteristic involved in the various embodiments of the present invention described below
It does not constitute a conflict with each other and can be combined with each other.
As shown in Figure 1, a kind of method for tracking target merged with twin convolutional network based on correlation filtering, the twin volume
Product network is 2 identical first convolutional networks and the second convolutional network, the method for tracking target include：
(1) target signature for utilizing (t-1) frame image of the first convolutional network extraction known target position utilizes the
Two convolutional networks extract the search characteristics figure of t frame images；
(2) Fast Fourier Transform (FFT) is carried out to the target signature of (t-1) frame image and obtains the mesh of (t-1) frame image
Region is marked, carrying out correlation filtering to the search characteristics figure of t frame images obtains the region of search of t frame images, calculates t frame figures
Cross correlation between the region of search and the target area of (t-1) frame image of picture, obtains the target score of t frame images
Figure, the target location of t frame images is obtained according to the target score figure of t frame images；
Wherein, the 1st frame image in video sequence is demarcated when t is 2 in t >=2, executes step (1)-(2) and obtains
To the target location of the 2nd frame image, when t is 3, executes step (1)-(2) and obtain the target location of the 3rd frame image, with such
It pushes away, obtains the target location of each frame image in video sequence, realize the target following to video sequence.
In detail, as shown in Fig. 2, a kind of method for tracking target merged with twin convolutional network based on correlation filtering, institute
State twin convolutional network includes for 2 identical first convolutional networks and the second convolutional network, the method for tracking target：
Utilize the extensive visual identity challenge matches of ImageNet (ILSVRC, ImageNet, Large Scale Visual
Recognition Challenge) in video database as Sample video sequence, in Sample video sequence per frame sample
The target location of image is marked, and is damaged with logarithm in training using the Sample video sequence training convolutional network after label
The minimum objective optimization network parameter of function is lost, trained convolutional network is obtained.
(1) target signature for utilizing (t-1) frame image of the first convolutional network extraction known target position utilizes the
Two convolutional networks extract the search characteristics figure of t frame images；
(2) Fast Fourier Transform (FFT) is carried out to the target signature of (t-1) frame image and obtains the mesh of (t-1) frame image
Region is marked, carrying out correlation filtering to the search characteristics figure of t frame images obtains the region of search of t frame images, calculates t frame figures
Cross correlation between the region of search and the target area of (t-1) frame image of picture, obtains the target score of t frame images
Figure, the target location of t frame images is obtained according to the target score figure of t frame images；
Wherein, the 1st frame image in video sequence is demarcated when t is 2 in t >=2, executes step (1)-(2) and obtains
To the target location of the 2nd frame image, when t is 3, executes step (1)-(2) and obtain the target location of the 3rd frame image, with such
It pushes away, obtains the target location of each frame image in video sequence, realize the target following to video sequence.
As shown in figure 3, being carried out using the search characteristics figure of cosine window function or sine-window function pair t frame images
Then smothing filtering uses Fast Fourier Transform (FFT) by the search characteristics figure of the t frame images after smothing filtering from spatial transform
To frequency domain, the region of search of t frame images is obtained.
When t is 2, the 1st frame image in video sequence is demarcated, step (1)-(2) is executed and obtains the 1st frame image
Region of search and the target area of the 2nd frame image between cross correlation pass through and minimize logarithm loss using cross correlation
Function backpropagation updates the network parameter of twin convolutional network, obtains target score figure (score as in figure is accordingly schemed),
Accordingly, it predicts the target frame of the 2nd frame image, then carries out Fast Fourier Transform (FFT), slided using the 1st frame image of calibration
Averaging model updates, and the target location of the 2nd obtained frame image is used to calculate the template of correlation as the 3rd frame.
Fig. 4 (a) is the first frame image of the first video sequence provided in an embodiment of the present invention；Fig. 4 (b) is implementation of the present invention
The first frame image for the second video sequence that example provides；Fig. 4 (c) is the first of third video sequence provided in an embodiment of the present invention
Frame image；Fig. 4 (d) is the first frame image of the 4th video sequence provided in an embodiment of the present invention；Fig. 4 (e) is implementation of the present invention
The first frame image for the 5th video sequence that example provides；Fig. 4 (f) is the first of the 6th video sequence provided in an embodiment of the present invention
Frame image；The position of target and size are wherein calibrated to the input for being used as convolutional network.
Fig. 5 (a1) is provided in an embodiment of the present invention using the progress target following of the first video sequence of the method for the present invention pair
50th frame image；Fig. 5 (a2) be it is provided in an embodiment of the present invention using the first video sequence of the method for the present invention pair carry out target with
100th frame image of track；Fig. 5 (a3) is provided in an embodiment of the present invention using the first video sequence of the method for the present invention pair progress mesh
Mark the 150th frame image of tracking；As can be seen that method for tracking target proposed by the present invention can effectively trace into appearance deformation
Target.
Fig. 5 (b1) is provided in an embodiment of the present invention using the progress target following of the second video sequence of the method for the present invention pair
50th frame image；Fig. 5 (b2) be it is provided in an embodiment of the present invention using the second video sequence of the method for the present invention pair carry out target with
100th frame image of track；Fig. 5 (b3) is provided in an embodiment of the present invention using the second video sequence of the method for the present invention pair progress mesh
Mark the 150th frame image of tracking；As can be seen that method for tracking target proposed by the present invention can move mould effective against target
Paste.
Fig. 5 (c1) be it is provided in an embodiment of the present invention using the method for the present invention to third video sequence carry out target following
50th frame image；Fig. 5 (c2) be it is provided in an embodiment of the present invention using the method for the present invention to third video sequence carry out target with
100th frame image of track；Fig. 5 (c3) is that use the method for the present invention provided in an embodiment of the present invention carries out mesh to third video sequence
Mark the 150th frame image of tracking；As can be seen that method for tracking target proposed by the present invention can be dry effective against similar background
It disturbs.
Fig. 5 (d1) is provided in an embodiment of the present invention using the progress target following of the 4th video sequence of the method for the present invention pair
50th frame image；Fig. 5 (d2) be it is provided in an embodiment of the present invention using the 4th video sequence of the method for the present invention pair carry out target with
100th frame image of track；Fig. 5 (d3) is provided in an embodiment of the present invention using the 4th video sequence of the method for the present invention pair progress mesh
Mark the 150th frame image of tracking；As can be seen that method for tracking target proposed by the present invention can effectively trace into quick movement
Target.
Fig. 5 (e1) is provided in an embodiment of the present invention using the progress target following of the 5th video sequence of the method for the present invention pair
50th frame image；Fig. 5 (e2) be it is provided in an embodiment of the present invention using the 5th video sequence of the method for the present invention pair carry out target with
100th frame image of track；Fig. 5 (e3) is provided in an embodiment of the present invention using the 5th video sequence of the method for the present invention pair progress mesh
Mark the 150th frame image of tracking；As can be seen that method for tracking target proposed by the present invention can change effective against target scale
And illumination variation.
Fig. 5 (f1) is provided in an embodiment of the present invention using the progress target following of the 6th video sequence of the method for the present invention pair
50th frame image；Fig. 5 (f2) be it is provided in an embodiment of the present invention using the 6th video sequence of the method for the present invention pair carry out target with
100th frame image of track；Fig. 5 (f3) is provided in an embodiment of the present invention using the 6th video sequence of the method for the present invention pair progress mesh
Mark the 150th frame image of tracking.As can be seen that method for tracking target proposed by the present invention can be blocked effective against target.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, all within the spirits and principles of the present invention made by all any modification, equivalent and improvement etc., should all include
Within protection scope of the present invention.
Claims (6)
1. a kind of method for tracking target merged with twin convolutional network based on correlation filtering, which is characterized in that the twin volume
Product network is 2 identical first convolutional networks and the second convolutional network, the method for tracking target include：
(1) target signature for utilizing (t-1) frame image of the first convolutional network extraction known target position, utilizes volume Two
The search characteristics figure of product network extraction t frame images；
(2) Fast Fourier Transform (FFT) is carried out to the target signature of (t-1) frame image and obtains the target area of (t-1) frame image
Domain carries out correlation filtering to the search characteristics figure of t frame images and obtains the region of search of t frame images, calculates t frame images
Cross correlation between region of search and the target area of (t-1) frame image obtains the target score figure of t frame images, root
The target location of t frame images is obtained according to the target score figure of t frame images；
Wherein, the 1st frame image in video sequence is demarcated when t is 2 in t >=2, executes step (1)-(2) and obtains the 2nd
The target location of frame image executes step (1)-(2) and obtains the target location of the 3rd frame image when t is 3, and so on, it obtains
To the target location of each frame image in video sequence, the target following to video sequence is realized.
2. a kind of method for tracking target merged with twin convolutional network based on correlation filtering as described in claim 1, special
Sign is that the correlation filtering in the step (2) includes：
Smothing filtering is carried out using the search characteristics figure of cosine window function or sine-window function pair t frame images, then
Using Fast Fourier Transform (FFT) by the search characteristics figure of the t frame images after smothing filtering from spatial transform to frequency domain, obtain t
The region of search of frame image.
3. a kind of method for tracking target merged with twin convolutional network based on correlation filtering as claimed in claim 1 or 2,
It being characterized in that, first convolutional network and the second convolutional network include five convolutional layers, and preceding the two of five convolutional layers
There are one down-sampling pond layers respectively after layer convolution.
4. a kind of method for tracking target merged with twin convolutional network based on correlation filtering as claimed in claim 1 or 2,
It is characterized in that, the twin convolutional network is trained convolutional network, and the training method of the twin convolutional network is：
Collecting sample video sequence utilizes label to being marked per the target location of frame sample image in Sample video sequence
Sample video sequence training convolutional network afterwards, with the minimum objective optimization network parameter of logarithm loss function, is obtained in training
To trained convolutional network.
5. a kind of method for tracking target merged with twin convolutional network based on correlation filtering as claimed in claim 4, special
Sign is that the logarithm loss function is：
L (y, v)=log (1+exp (- yv))
Wherein, v is the confidence score of the target location of sample image, and y is the label of the target location of sample image, and l (y, v) is
Error amount.
6. a kind of method for tracking target merged with twin convolutional network based on correlation filtering as claimed in claim 4, special
Sign is that the method for tracking target further includes：
When t is 2, the 1st frame image in video sequence is demarcated, step (1)-(2) is executed and obtains searching for the 1st frame image
Cross correlation between rope region and the target area of the 2nd frame image, using cross correlation, by minimizing logarithm loss function
Backpropagation updates the network parameter of twin convolutional network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810342324.1A CN108665485B (en) | 2018-04-16 | 2018-04-16 | Target tracking method based on relevant filtering and twin convolution network fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810342324.1A CN108665485B (en) | 2018-04-16 | 2018-04-16 | Target tracking method based on relevant filtering and twin convolution network fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108665485A true CN108665485A (en) | 2018-10-16 |
| CN108665485B CN108665485B (en) | 2021-07-02 |
Family
ID=63783613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810342324.1A Expired - Fee Related CN108665485B (en) | 2018-04-16 | 2018-04-16 | Target tracking method based on relevant filtering and twin convolution network fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108665485B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543559A (en) * | 2018-10-31 | 2019-03-29 | 东南大学 | Method for tracking target and system based on twin network and movement selection mechanism |
| CN109598684A (en) * | 2018-11-21 | 2019-04-09 | 华南理工大学 | In conjunction with the correlation filtering tracking of twin network |
| CN109712171A (en) * | 2018-12-28 | 2019-05-03 | 上海极链网络科技有限公司 | A kind of Target Tracking System and method for tracking target based on correlation filter |
| CN110210551A (en) * | 2019-05-28 | 2019-09-06 | 北京工业大学 | A kind of visual target tracking method based on adaptive main body sensitivity |
| CN110309835A (en) * | 2019-06-27 | 2019-10-08 | 中国人民解放军战略支援部队信息工程大学 | A kind of image local feature extracting method and device |
| CN110415271A (en) * | 2019-06-28 | 2019-11-05 | 武汉大学 | One kind fighting twin network target tracking method based on the multifarious generation of appearance |
| CN110473231A (en) * | 2019-08-20 | 2019-11-19 | 南京航空航天大学 | A kind of method for tracking target of the twin full convolutional network with anticipation formula study more new strategy |
| CN110807793A (en) * | 2019-09-29 | 2020-02-18 | 南京大学 | Target tracking method based on twin network |
| CN111260688A (en) * | 2020-01-13 | 2020-06-09 | 深圳大学 | Twin double-path target tracking method |
| CN111340850A (en) * | 2020-03-20 | 2020-06-26 | 军事科学院系统工程研究院系统总体研究所 | Ground target tracking method of unmanned aerial vehicle based on twin network and central logic loss |
| CN111415373A (en) * | 2020-03-20 | 2020-07-14 | 北京以萨技术股份有限公司 | Target tracking and segmenting method, system and medium based on twin convolutional network |
| CN112686957A (en) * | 2019-10-18 | 2021-04-20 | 北京华航无线电测量研究所 | Quick calibration method for sequence image |
| CN113592899A (en) * | 2021-05-28 | 2021-11-02 | 北京理工大学重庆创新中心 | Method for extracting correlated filtering target tracking depth features |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184778A (en) * | 2015-08-25 | 2015-12-23 | 广州视源电子科技股份有限公司 | Detection method and apparatus |
| CN106650630A (en) * | 2016-11-11 | 2017-05-10 | 纳恩博（北京）科技有限公司 | Target tracking method and electronic equipment |
| US20170132472A1 (en) * | 2015-11-05 | 2017-05-11 | Qualcomm Incorporated | Generic mapping for tracking target object in video sequence |
| CN107452025A (en) * | 2017-08-18 | 2017-12-08 | 成都通甲优博科技有限责任公司 | Method for tracking target, device and electronic equipment |
-
2018
- 2018-04-16 CN CN201810342324.1A patent/CN108665485B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184778A (en) * | 2015-08-25 | 2015-12-23 | 广州视源电子科技股份有限公司 | Detection method and apparatus |
| US20170132472A1 (en) * | 2015-11-05 | 2017-05-11 | Qualcomm Incorporated | Generic mapping for tracking target object in video sequence |
| CN106650630A (en) * | 2016-11-11 | 2017-05-10 | 纳恩博（北京）科技有限公司 | Target tracking method and electronic equipment |
| CN107452025A (en) * | 2017-08-18 | 2017-12-08 | 成都通甲优博科技有限责任公司 | Method for tracking target, device and electronic equipment |
Non-Patent Citations (2)
| Title |
|---|
| JANGHOON CHOI等: "Deep Meta Learning for Real-Time Visual Tracking based on Target-Specific Feature Space", 《ARXIV》 * |
| QIANG WANG等: "DCFNet: discriminant correlation filters network for visual tracking", 《ARXIV》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543559A (en) * | 2018-10-31 | 2019-03-29 | 东南大学 | Method for tracking target and system based on twin network and movement selection mechanism |
| CN109543559B (en) * | 2018-10-31 | 2021-12-28 | 东南大学 | Target tracking method and system based on twin network and action selection mechanism |
| CN109598684A (en) * | 2018-11-21 | 2019-04-09 | 华南理工大学 | In conjunction with the correlation filtering tracking of twin network |
| CN109712171A (en) * | 2018-12-28 | 2019-05-03 | 上海极链网络科技有限公司 | A kind of Target Tracking System and method for tracking target based on correlation filter |
| CN109712171B (en) * | 2018-12-28 | 2023-09-01 | 厦门瑞利特信息科技有限公司 | Target tracking system and target tracking method based on correlation filter |
| CN110210551B (en) * | 2019-05-28 | 2021-07-30 | 北京工业大学 | Visual target tracking method based on adaptive subject sensitivity |
| CN110210551A (en) * | 2019-05-28 | 2019-09-06 | 北京工业大学 | A kind of visual target tracking method based on adaptive main body sensitivity |
| CN110309835A (en) * | 2019-06-27 | 2019-10-08 | 中国人民解放军战略支援部队信息工程大学 | A kind of image local feature extracting method and device |
| CN110309835B (en) * | 2019-06-27 | 2021-10-15 | 中国人民解放军战略支援部队信息工程大学 | Image local feature extraction method and device |
| CN110415271B (en) * | 2019-06-28 | 2022-06-07 | 武汉大学 | Appearance diversity-based method for tracking generation twin-resisting network target |
| CN110415271A (en) * | 2019-06-28 | 2019-11-05 | 武汉大学 | One kind fighting twin network target tracking method based on the multifarious generation of appearance |
| CN110473231A (en) * | 2019-08-20 | 2019-11-19 | 南京航空航天大学 | A kind of method for tracking target of the twin full convolutional network with anticipation formula study more new strategy |
| CN110807793A (en) * | 2019-09-29 | 2020-02-18 | 南京大学 | Target tracking method based on twin network |
| CN110807793B (en) * | 2019-09-29 | 2022-04-22 | 南京大学 | Target tracking method based on twin network |
| CN112686957A (en) * | 2019-10-18 | 2021-04-20 | 北京华航无线电测量研究所 | Quick calibration method for sequence image |
| CN111260688A (en) * | 2020-01-13 | 2020-06-09 | 深圳大学 | Twin double-path target tracking method |
| CN111340850A (en) * | 2020-03-20 | 2020-06-26 | 军事科学院系统工程研究院系统总体研究所 | Ground target tracking method of unmanned aerial vehicle based on twin network and central logic loss |
| CN111415373A (en) * | 2020-03-20 | 2020-07-14 | 北京以萨技术股份有限公司 | Target tracking and segmenting method, system and medium based on twin convolutional network |
| CN113592899A (en) * | 2021-05-28 | 2021-11-02 | 北京理工大学重庆创新中心 | Method for extracting correlated filtering target tracking depth features |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108665485B (en) | 2021-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108665485A (en) | A kind of method for tracking target merged with twin convolutional network based on correlation filtering | |
| CN108319972B (en) | End-to-end difference network learning method for image semantic segmentation | |
| CN104573731B (en) | Fast target detection method based on convolutional neural networks | |
| CN112184752A (en) | Video target tracking method based on pyramid convolution | |
| CN106355602B (en) | A kind of Multi-target position tracking video frequency monitoring method | |
| CN109191491A (en) | The method for tracking target and system of the twin network of full convolution based on multilayer feature fusion | |
| CN110276264B (en) | Crowd density estimation method based on foreground segmentation graph | |
| CN110135500A (en) | Method for tracking target under a kind of more scenes based on adaptive depth characteristic filter | |
| CN107423760A (en) | Based on pre-segmentation and the deep learning object detection method returned | |
| CN109949340A (en) | Target scale adaptive tracking method based on OpenCV | |
| CN108090918A (en) | A kind of Real-time Human Face Tracking based on the twin network of the full convolution of depth | |
| CN110210551A (en) | A kind of visual target tracking method based on adaptive main body sensitivity | |
| CN107103326A (en) | The collaboration conspicuousness detection method clustered based on super-pixel | |
| CN108665481A (en) | Multilayer depth characteristic fusion it is adaptive resist block infrared object tracking method | |
| CN103886619B (en) | A kind of method for tracking target merging multiple dimensioned super-pixel | |
| CN108346159A (en) | A kind of visual target tracking method based on tracking-study-detection | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN106530340B (en) | A kind of specified object tracking | |
| CN110473231B (en) | Target tracking method of twin full convolution network with prejudging type learning updating strategy | |
| CN110097575B (en) | Target tracking method based on local features and scale pool | |
| CN109299701A (en) | Expand the face age estimation method that more ethnic group features cooperate with selection based on GAN | |
| CN106599994A (en) | Sight line estimation method based on depth regression network | |
| CN108288282A (en) | A kind of adaptive features select method for tracking target based on convolutional neural networks | |
| CN104392223A (en) | Method for recognizing human postures in two-dimensional video images | |
| CN110827262B (en) | Weak and small target detection method based on continuous limited frame infrared image |
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 | ||
| 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: 20210702 |