CN108734109A - A kind of visual target tracking method and system towards image sequence - Google Patents
A kind of visual target tracking method and system towards image sequence Download PDFInfo
- Publication number
- CN108734109A CN108734109A CN201810373435.9A CN201810373435A CN108734109A CN 108734109 A CN108734109 A CN 108734109A CN 201810373435 A CN201810373435 A CN 201810373435A CN 108734109 A CN108734109 A CN 108734109A
- Authority
- CN
- China
- Prior art keywords
- target
- convolution
- regression model
- training
- size
- 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 35
- 230000000007 visual effect Effects 0.000 title claims abstract description 23
- 238000000605 extraction Methods 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000003475 lamination Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 3
- 230000007613 environmental effect Effects 0.000 abstract description 5
- 238000005516 engineering process Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 230000007704 transition Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000002547 anomalous effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of visual target tracking method and system towards image sequence, wherein visual target tracking method include the following steps, utilize given initialisation image and target rectangle frame to be tracked, convolution regression model of the training for target following;Utilize the position for the convolution forecast of regression model target that training obtains;On the basis of target prodiction result, the size of target is further predicted;Position and the size of target are obtained according to tracking, updates convolution regression model.The present invention, it is related to the technologies such as the training of target entirety regression model, the training of target texture regression model, target prodiction, target sizes prediction, trace model update, the interference of various environmental factors in tracking scene can fully be overcome, it realizes the Accurate Prediction to target location and size, there is higher commercial value and research significance.
Description
Technical field
The present invention relates to technical field of computer vision, and in particular to a kind of visual target tracking side towards image sequence
Method and system.
Background technology
In computer vision field, it usually needs using intelligent algorithm automatic identification and analysis video information, to real
Now to the intelligentized control method of equipment.The target tracking algorism of view-based access control model image sequence can make full use of existing towards single width
The algorithm of target detection of image tracks the movement locus of target in video fast and reliablely, is improved to understand and analyzing video
Technical support.
With expanding rapidly for commercial production scale, automation and intelligence degree in industrial products production process also need
It is continuously improved.Such as in video monitoring system, needs intelligent algorithm automatic identification and detect the abnormal thing occurred in video
Part.And visual target tracking algorithm can automatically track each target in video, and the movement locus of target is obtained, to divide
It analyses and understands that the anomalous event in video provides crucial technological means.But traditional visual target tracking algorithm there is also
Following defect:
(1) size of target can not be predicted well, especially when apparent deformation occurs for target, tradition tracking
Algorithm can not accurately predict target sizes, cause to lose target in subsequent tracking, and can not be provided for video analysis understanding can
The bottom-up information leaned on.
(2) under the interference of a variety of environmental factors cannot accurately, reliably carry out target following.
In view of this, being badly in need of being improved existing visual target tracking algorithm, propose that one kind can overcome a variety of rings
Border factor interference, and the visual target tracking algorithm of Accurate Prediction target location and size.
Invention content
The technical problem to be solved by the present invention is to existing visual target tracking algorithm, there are the positions of unpredictable target
It sets and size, and tracking process is easy to be interfered by environment, cannot accurately and reliably carry out the problem of target following.
In order to solve the above-mentioned technical problem, the technical solution adopted in the present invention is to provide a kind of regarding towards image sequence
The technical method for feeling target following, includes the following steps:
Using given initialisation image and target rectangle frame to be tracked, convolution of the training for target following returns mould
Type;
Utilize the position for the convolution forecast of regression model target that training obtains;
On the basis of target prodiction result, the size of target is further predicted;
Position and the size of target are obtained according to tracking, updates convolution regression model.
In the above scheme, the method for training convolutional regression model includes the following steps:
Step 10, feature extraction network of the structure for expressing feature of interest characteristic, the network can be based on arbitrarily being used for
The feature extracting method for expressing target information is realized;
Step 11, the corresponding feature of feature extraction network extraction current input image in step 10 is utilized;
Step 12, structure is realized based on single convolutional network layer, and object-oriented whole convolution regression model, the convolution
Convolution kernel size and the target of layer are in the in the same size of feature space, while the output channel of convolution layer network is 1, the convolutional layer
The output of network can be used for predicting the position of target;
Step 13, based on the feature extracted in step 11, corresponding training label figure is generated, training label figure is according to one
A two-dimensional Gaussian function generates, and peak value has corresponded to the true position of target, utilized gradient descent algorithm iteration optimization step 12
In single convolutional layer;
Step 14, structure is realized based on single convolutional network layer, and the convolution regression model of object-oriented texture, the convolution
Convolution kernel size and the target of layer are in the in the same size of feature space, while the output channel of convolution layer network is 1, the convolutional layer
The output of network can be used for predicting the foreground of target;
Step 15, it based on the feature extracted in step 11, generates corresponding training label and schemes, with a square in the label figure
Shape frame identifies the foreground of target, this utilizes the single convolutional layer in gradient descent algorithm iteration optimization step 12;
Step 16, the initial training of convolution regression model terminates.
In the above scheme, target prodiction method has specifically included following steps:
Step 20, the feature extraction network built in step 10, the corresponding feature of extraction current input image, after being are utilized
Continuous target following is prepared;
Step 21, the object-oriented whole convolution obtained in the characteristics of image input step 12 obtained in step 20 is returned
Return network, the target prodiction figure H (x based on target entirety regression model are calculatedt,yt);
Step 22, the convolution of the object-oriented texture obtained in the characteristics of image input step 14 obtained in step 20 is returned
Return network, the target prospect prognostic chart T (x based on target texture regression model are calculatedt,yt);
Step 23, on the target prospect prognostic chart obtained in step 22 carry out mean filter operation, Filtering Template it is big
It is small in the same size with target, the target prodiction figure F (x based on target prospect are calculatedt,yt);
Step 24, obtained in step 21 and step 23 two kinds of target prodiction figures are superimposed, are obtained final
Target prodiction figure, and according to the corresponding index future position of maximum value in position prediction figure, calculation formula is:
In the above scheme, the target prodiction figure F (x based on target prospect in the step 23t,yt) calculating it is public
Formula is:Wherein wt-1And ht-1It is respectively identified in previous frame tracking
Obtained target sizes, R (xt,yt,wt-1,ht-1) denotation coordination be (xt,yt), size wt-1,ht-1Rectangle frame.T(i,j)
The value in the rectangle frame of the target prospect prognostic chart of target texture regression model corresponding to each pixel is represented, i, j are square
Pixel in shape frame
In the above scheme, target sizes prediction techniques includes the following steps:
Step 30, the feature extraction network built in step 10, the corresponding feature of extraction current input image, after being are utilized
Continuous target following is prepared;
Step 31, the convolution of the object-oriented texture obtained in the characteristics of image input step 14 obtained in step 30 is returned
Return network, the target prospect prognostic chart T (x based on target texture regression model are calculatedt,yt);
Step 32, the position x of current goal is obtainedt,ytAnd size w of the known target in previous framet,htAfterwards, it calculates
Target sizes are wt,htPosterior probability;
Step 33, it reuses the method in step 32 and calculates the corresponding posterior probability of multiple target candidate sizes, selection
The maximum target sizes of posterior probability are as final target sizes predicted value;
Step 34, target sizes prediction terminates.
In the above scheme, target sizes are w in the step 32t,htThe calculation formula of posterior probability be:P(wt,ht
|O,xt,yt,wt-1,ht-1)=P (O | xt,yt,wt,ht)P(wt,ht|wt-1,ht-1), wherein P (O | xt,yt,wt,ht) indicate target
Position and size state be (xt,yt,wt,ht) probability, P (O | xt,yt,wt,ht)P(wt,ht|wt-1,ht-1) indicate adjacent two
Target sizes state transition probability between frame,
P(O|xt,yt,wt,ht)=A (wt,ht)-B(wt,ht), wherein A (wt,ht) indicate candidate target rectangle frame (xt,yt,wt,ht) on
Average criterion foreground probability, B (wt,ht) indicate target rectangle frame (xt,yt,wt,ht) surrounding background area average criterion before
Scape probability.
In the above scheme, update convolution regression model includes the following steps:
Step 40, it according to predicted obtained target location, generates for training object-oriented whole convolution to return
The label figure of model, and utilize the monovolume lamination network parameter in gradient descent method update step 12;
Step 41, it according to predicted obtained target sizes, generates for training the convolution of object-oriented texture to return
The label figure of model, and utilize the monovolume lamination network parameter in gradient descent method update step 14;
Step 42, convolution regression model update terminates.
The visual target tracking system towards image sequence that the present invention also provides a kind of, including:
Training module, convolution regression model of the training for target following;
Target prodiction module utilizes the position for the convolution forecast of regression model target that training obtains;
Target sizes prediction module further predicts the size of target on the basis of target prodiction result;
Update module obtains position and the size of target according to tracking, updates convolution regression model.
Compared with prior art, the present invention relates to the training of target entirety regression model, the training of target texture regression model, mesh
The technologies such as cursor position prediction, target sizes prediction, trace model update, can fully overcome various environmental factors in tracking scene
Interference, realize to the Accurate Prediction of target location and size, there is higher commercial value and research significance.
Description of the drawings
Fig. 1 is the flow diagram of the position of the present invention present invention;
Fig. 2 is the input initial frame image of the present invention;
Fig. 3 is that the convolution regression model of the present invention trains flow diagram;
Fig. 4 is the whole regression model schematic diagram of the present invention;
Fig. 5 is the texture regression model schematic diagram of the present invention;
Fig. 6 is the target prodiction flow diagram of the present invention;
Fig. 7 is the target prodiction figure based on whole regression model of the present invention;
Fig. 8 is the target prospect prognostic chart based on texture regression model of the present invention;
Fig. 9 is the target prodiction figure based on texture regression model of the present invention;
Figure 10 is that the target sizes of the present invention predict flow diagram.
Specific implementation mode
The present invention provides a kind of technical methods of the visual target tracking towards image sequence, can fully overcome tracking
The interference of various environmental factors in scene realizes to the Accurate Prediction of target location and size, have higher commercial value and
Research significance.The present invention is described in detail with specific implementation mode with reference to the accompanying drawings of the specification.
As depicted in figs. 1 and 2, the technical method of a kind of visual target tracking towards image sequence provided by the invention,
Specifically it may include following steps:
Using given initialisation image and target rectangle frame to be tracked, convolution of the training for target following returns mould
Type;
Utilize the position for the convolution forecast of regression model target that training obtains;
On the basis of target prodiction result, the size of target is further predicted;
Position and the size of target are finally obtained according to tracking, update convolution regression model.
Correspondingly, the visual target tracking system towards image sequence that the present invention also provides a kind of, including training module,
Target prodiction module, target sizes prediction module and update module.
Convolution regression model of the training module training for target following;
The position for the convolution forecast of regression model target that target prodiction module is obtained using training;
Target sizes prediction module further predicts the size of target on the basis of target prodiction result;
Update module obtains position and the size of target according to tracking, updates convolution regression model.
The present invention relates to the training of target entirety regression model, the training of target texture regression model, target prodiction, targets
The technologies such as size prediction, trace model update, can fully overcome the interference of various environmental factors in tracking scene, realize to mesh
The Accurate Prediction of cursor position and size has higher commercial value and research significance.
As shown in figure 3, the method for training convolutional regression model specifically includes following steps:
The feature extraction network for expressing feature of interest characteristic is built, which can be based on arbitrary for expressing target
The feature extracting method of information is realized;
Utilize the corresponding feature of feature extraction network extraction current input image;
Structure is realized based on single convolutional network layer, and object-oriented whole convolution regression model, the volume of the convolutional layer
Core size is accumulated with target in the in the same size of feature space, while the output channel of convolution layer network is 1, the convolution layer network
Output can be used for predicting the position of target;
The feature for expressing target information based on said extracted generates corresponding training label figure, training label
Figure is generated according to a two-dimensional Gaussian function, and peak value has corresponded to the true position of target, excellent using gradient descent algorithm iteration
Change single convolutional layer;
Structure is realized based on single convolutional network layer, and the convolution regression model of object-oriented texture, the volume of the convolutional layer
Core size is accumulated with target in the in the same size of feature space, while the output channel of convolution layer network is 1, the convolution layer network
Output can be used for predicting the foreground of target;
The feature for expressing target information based on said extracted generates corresponding training label and schemes, in the label figure
The foreground of target is identified with a rectangle frame, this utilizes the single convolutional layer of gradient descent algorithm iteration optimization;
Convolution regression model initial training terminates.
As shown in Fig. 4~Fig. 9, target prodiction method has specifically included following steps:
Using the feature extraction network for expressing feature of interest characteristic of above-mentioned structure, extraction current input image corresponds to
Feature, prepare for subsequent target following;
Obtained characteristics of image is input to above-mentioned object-oriented whole convolution Recurrent networks, is calculated based on target
The target prodiction figure H (x of whole regression modelt,yt);
Characteristics of image obtained above is input to the convolution Recurrent networks of object-oriented texture, is calculated based on target
The target prospect prognostic chart T (x of texture regression modelt,yt);
Mean filter operation, the size of Filtering Template and the size one of target will be carried out on above-mentioned target prospect prognostic chart
It causes, the target prodiction figure F (x based on target prospect is calculatedt,yt), wherein F (xt,yt) calculation formula be:
Wherein wt-1And ht-1It is respectively identified at the target sizes obtained in previous frame tracking, R (xt,yt,wt-1,ht-1)R(xt,
yt,wt-1,ht-1) denotation coordination be (xt,yt), size wt-1,ht-1Rectangle frame, T (i, j) represent target texture return mould
Value in the rectangle frame of the target prospect prognostic chart of type corresponding to each pixel, i, j are the pixel in rectangle frame;
By two kinds of target prodictions figure H (x obtained abovet,yt) and F (xt,yt) be superimposed, it obtains final
Target prodiction figure, and according to the corresponding index future position of maximum value in position prediction figure, calculation formula is:
As shown in Figure 10, the prediction techniques of target sizes includes the following steps:
Using the feature extraction network for expressing feature of interest characteristic of above-mentioned structure, extraction current input image corresponds to
Feature, prepare for subsequent target following;
Characteristics of image obtained above is input to the convolution Recurrent networks for the object-oriented texture that convolution layer network obtains,
The target prospect prognostic chart T (x based on target texture regression model are calculatedt,yt);
Obtain the position x of current goalt,ytAnd size w of the known target in previous framet,htAfterwards, it is big to calculate target
Small is wt,htPosterior probability P (O | xt,yt,wt,ht), wherein calculation formula is:
P(wt,ht|O,xt,yt,wt-1,ht-1)=P (O | xt,yt,wt,ht)P(wt,ht|wt-1,ht-1), wherein P (O | xt,yt,wt,ht)
Indicate that the position of target and size state are (xt,yt,wt,ht) probability, P (O | xt,yt,wt,ht)P(wt,ht|wt-1,ht-1) indicate
Target sizes state transition probability between adjacent two frame,
P(O|xt,yt,wt,ht)=A (wt,ht)-B(wt,ht), wherein A (wt,ht) indicate candidate target rectangle frame (xt,yt,wt,ht) on
Average criterion foreground probability, B (wt,ht) indicate target rectangle frame (xt,yt,wt,ht) surrounding background area average criterion before
Scape probability;
It reuses the above method and calculates the corresponding posterior probability of multiple target candidate sizes, select posterior probability maximum
Target sizes are as final target sizes predicted value;
Target sizes prediction terminates.
The update of convolution regression model mainly includes the following steps that:
According to predicted obtained target location, the mark for training object-oriented whole convolution regression model is generated
Note figure, and update above-mentioned monovolume lamination network parameter using gradient descent method;
According to predicted obtained target sizes, the mark of the convolution regression model for training object-oriented texture is generated
Note figure, and update above-mentioned monovolume lamination network parameter using gradient descent method;
The update of convolution regression model terminates.
The method of the present invention is needed at given one by algorithm keeps track using continuous sequence of video images as input data
After target rectangle frame, pass through the training of target entirety regression model, the training of target texture regression model, target prodiction, target
Size prediction, trace model update and etc., realize being continuously tracked to the target in image sequence.This method can be in target
Rotate or target be blocked when be accurately tracked by target, while solving traditional visual target tracking algorithm
The problem of being difficult to Accurate Prediction target sizes, can be when target deforms upon, the size of Accurate Prediction target.The present invention simultaneously
The method of proposition has tracking accuracy high, and the speed of service is very fast, it is insensitive to the interference of background environment the features such as, controlled in industry
The occasions such as system, automated production have boundless application prospect.
The invention is not limited in above-mentioned preferred forms, and anyone should learn that is made under the inspiration of the present invention
Structure change, the technical schemes that are same or similar to the present invention are each fallen within protection scope of the present invention.
Claims (8)
1. a kind of technical method of the visual target tracking towards image sequence, which is characterized in that include the following steps:
Utilize given initialisation image and target rectangle frame to be tracked, convolution regression model of the training for target following;
Utilize the position for the convolution forecast of regression model target that training obtains;
On the basis of target prodiction result, the size of target is further predicted;
Position and the size of target are obtained according to tracking, updates convolution regression model.
2. a kind of technical method of visual target tracking towards image sequence according to claim 1, which is characterized in that
The method of training convolutional regression model includes the following steps:
Step 10, feature extraction network of the structure for expressing feature of interest characteristic, the network can be based on arbitrary for expressing
The feature extracting method of target information is realized;
Step 11, the corresponding feature of feature extraction network extraction current input image in step 10 is utilized;
Step 12, structure is realized based on single convolutional network layer, and object-oriented whole convolution regression model, the convolutional layer
Convolution kernel size and target are in the in the same size of feature space, while the output channel of convolution layer network is 1, the convolution layer network
Output can be used for predicting the position of target;
Step 13, based on the feature extracted in step 11, corresponding training label figure is generated, training label figure is according to one two
It ties up Gaussian function to generate, peak value has corresponded to the true position of target, using in gradient descent algorithm iteration optimization step 12
Single convolutional layer;
Step 14, structure is realized based on single convolutional network layer, and the convolution regression model of object-oriented texture, the convolutional layer
Convolution kernel size and target are in the in the same size of feature space, while the output channel of convolution layer network is 1, the convolution layer network
Output can be used for predicting the foreground of target;
Step 15, it based on the feature extracted in step 11, generates corresponding training label and schemes, with a rectangle frame in the label figure
The foreground of target is identified, this utilizes the single convolutional layer in gradient descent algorithm iteration optimization step 12;
Step 16, the initial training of convolution regression model terminates.
3. according to claim, a kind of technical method of visual target tracking towards image sequence described in 2, feature exists
In target prodiction method has specifically included following steps:
Step 20, using the feature extraction network built in step 10, the corresponding feature of extraction current input image is subsequent
Target following is prepared;
Step 21, the object-oriented whole convolution obtained in the characteristics of image input step 12 obtained in step 20 is returned into net
The target prodiction figure H (x based on target entirety regression model are calculated in networkt,yt);
Step 22, the convolution of the object-oriented texture obtained in the characteristics of image input step 14 obtained in step 20 is returned into net
The target prospect prognostic chart T (x based on target texture regression model are calculated in networkt,yt);
Step 23, on the target prospect prognostic chart obtained in step 22 carry out mean filter operation, the size of Filtering Template with
Target it is in the same size, the target prodiction figure F (x based on target prospect are calculatedt,yt);
Step 24, obtained in step 21 and step 23 two kinds of target prodiction figures are superimposed, obtain final mesh
Cursor position prognostic chart, and according to the corresponding index future position of maximum value in position prediction figure, calculation formula is:
4. a kind of technical method of visual target tracking towards image sequence according to claim 3, feature
It is, the target prodiction figure F (x based on target prospect in the step 23t,yt) calculation formula be:Wherein wt-1And ht-1It is respectively identified in previous frame tracking and obtains
Target sizes, R (xt,yt,wt-1,ht-1) denotation coordination be (xt,yt), size wt-1,ht-1Rectangle frame, T (i, j) represent
Value in the rectangle frame of the target prospect prognostic chart of target texture regression model corresponding to each pixel, i, j are rectangle frame
Interior pixel.
5. a kind of technical method of visual target tracking towards image sequence according to claim 2, which is characterized in that
Target sizes prediction technique includes the following steps:
Step 30, using the feature extraction network built in step 10, the corresponding feature of extraction current input image is subsequent
Target following is prepared;
Step 31, the convolution of the object-oriented texture obtained in the characteristics of image input step 14 obtained in step 30 is returned into net
The target prospect prognostic chart T (x based on target texture regression model are calculated in networkt,yt);
Step 32, the position x of current goal is obtainedt,ytAnd size w of the known target in previous framet,htAfterwards, target is calculated
Size is wt,htPosterior probability;
Step 33, it reuses the method in step 32 and calculates the corresponding posterior probability of multiple target candidate sizes, select posteriority
The target sizes of maximum probability are as final target sizes predicted value;
Step 34, target sizes prediction terminates.
6. a kind of technical method of visual target tracking towards image sequence according to claim 5, which is characterized in that
Target sizes are w in the step 32t,htThe calculation formula of posterior probability be:P(wt,ht|O,xt,yt,wt-1,ht-1)=P (O
|xt,yt,wt,ht)P(wt,ht|wt-1,ht-1), wherein P (O | xt,yt,wt,ht) indicate that the position of target and size state are (xt,
yt,wt,ht) probability, P (O | xt,yt,wt,ht)P(wt,ht|wt-1,ht-1) indicate target sizes state transfer between adjacent two frame
Probability,P(O|xt,yt,wt,ht)=A (wt,ht)-
B(wt,ht), wherein A (wt,ht) indicate candidate target rectangle frame (xt,yt,wt,ht) on average criterion foreground probability, B (wt,
ht) indicate target rectangle frame (xt,yt,wt,ht) surrounding background area average criterion foreground probability.
7. a kind of technical method of visual target tracking towards image sequence according to claim 2, which is characterized in that
Update convolution regression model includes the following steps:
Step 40, it according to predicted obtained target location, generates for training object-oriented whole convolution regression model
Label figure, and using gradient descent method update step 12 in monovolume lamination network parameter;
Step 41, according to predicted obtained target sizes, the convolution regression model for training object-oriented texture is generated
Label figure, and using gradient descent method update step 14 in monovolume lamination network parameter;
Step 42, convolution regression model update terminates.
8. a kind of visual target tracking system towards image sequence, which is characterized in that including:
Training module, convolution regression model of the training for target following;
Target prodiction module utilizes the position for the convolution forecast of regression model target that training obtains;
Target sizes prediction module further predicts the size of target on the basis of target prodiction result;
Update module obtains position and the size of target according to tracking, updates convolution regression model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810373435.9A CN108734109B (en) | 2018-04-24 | 2018-04-24 | Visual target tracking method and system for image sequence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810373435.9A CN108734109B (en) | 2018-04-24 | 2018-04-24 | Visual target tracking method and system for image sequence |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108734109A true CN108734109A (en) | 2018-11-02 |
CN108734109B CN108734109B (en) | 2020-11-17 |
Family
ID=63939209
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810373435.9A Active CN108734109B (en) | 2018-04-24 | 2018-04-24 | Visual target tracking method and system for image sequence |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108734109B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109829936A (en) * | 2019-01-29 | 2019-05-31 | 青岛海信网络科技股份有限公司 | A kind of method and apparatus of target tracking |
CN111027586A (en) * | 2019-11-04 | 2020-04-17 | 天津大学 | Target tracking method based on novel response map fusion |
WO2020224479A1 (en) * | 2019-05-06 | 2020-11-12 | 腾讯科技(深圳)有限公司 | Method and apparatus for acquiring positions of target, and computer device and storage medium |
CN112378397A (en) * | 2020-11-02 | 2021-02-19 | 中国兵器工业计算机应用技术研究所 | Unmanned aerial vehicle target tracking method and device and unmanned aerial vehicle |
CN112465859A (en) * | 2019-09-06 | 2021-03-09 | 顺丰科技有限公司 | Method, device, equipment and storage medium for detecting fast moving object |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102881012A (en) * | 2012-09-04 | 2013-01-16 | 上海交通大学 | Vision target tracking method aiming at target scale change |
CN103106667A (en) * | 2013-02-01 | 2013-05-15 | 山东科技大学 | Motion target tracing method towards shielding and scene change |
CN103632382A (en) * | 2013-12-19 | 2014-03-12 | 中国矿业大学(北京) | Compressive sensing-based real-time multi-scale target tracking method |
CN105321188A (en) * | 2014-08-04 | 2016-02-10 | 江南大学 | Foreground probability based target tracking method |
EP3229206A1 (en) * | 2016-04-04 | 2017-10-11 | Xerox Corporation | Deep data association for online multi-class multi-object tracking |
CN107403175A (en) * | 2017-09-21 | 2017-11-28 | 昆明理工大学 | Visual tracking method and Visual Tracking System under a kind of movement background |
US20180032817A1 (en) * | 2016-07-27 | 2018-02-01 | Conduent Business Services, Llc | System and method for detecting potential mugging event via trajectory-based analysis |
-
2018
- 2018-04-24 CN CN201810373435.9A patent/CN108734109B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102881012A (en) * | 2012-09-04 | 2013-01-16 | 上海交通大学 | Vision target tracking method aiming at target scale change |
CN103106667A (en) * | 2013-02-01 | 2013-05-15 | 山东科技大学 | Motion target tracing method towards shielding and scene change |
CN103632382A (en) * | 2013-12-19 | 2014-03-12 | 中国矿业大学(北京) | Compressive sensing-based real-time multi-scale target tracking method |
CN105321188A (en) * | 2014-08-04 | 2016-02-10 | 江南大学 | Foreground probability based target tracking method |
EP3229206A1 (en) * | 2016-04-04 | 2017-10-11 | Xerox Corporation | Deep data association for online multi-class multi-object tracking |
US20180032817A1 (en) * | 2016-07-27 | 2018-02-01 | Conduent Business Services, Llc | System and method for detecting potential mugging event via trajectory-based analysis |
CN107403175A (en) * | 2017-09-21 | 2017-11-28 | 昆明理工大学 | Visual tracking method and Visual Tracking System under a kind of movement background |
Non-Patent Citations (5)
Title |
---|
CHAO MA等: ""Adaptive Correlation Filters with Long-Term and Short-Term Memory for Object Tracking"", 《INTERNATIONAL JOURNAL OF COMPUTER VISION (2018)》 * |
KAI CHEN等: ""Convolutional Regression for Visual Tracking"", 《ARXIV》 * |
MARTIN DANELLJAN等: ""Accurate Scale Estimation for Robust Visual Tracking"", 《BMVC 2014 • COMPUTER SCIENCE》 * |
YIBING SONG等: ""CREST: Convolutional Residual Learning for Visual Tracking"", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
齐飞等: ""基于均值漂移的视觉目标跟踪方法综述"", 《计算机工程》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109829936A (en) * | 2019-01-29 | 2019-05-31 | 青岛海信网络科技股份有限公司 | A kind of method and apparatus of target tracking |
CN109829936B (en) * | 2019-01-29 | 2021-12-24 | 青岛海信网络科技股份有限公司 | Target tracking method and device |
WO2020224479A1 (en) * | 2019-05-06 | 2020-11-12 | 腾讯科技(深圳)有限公司 | Method and apparatus for acquiring positions of target, and computer device and storage medium |
CN112465859A (en) * | 2019-09-06 | 2021-03-09 | 顺丰科技有限公司 | Method, device, equipment and storage medium for detecting fast moving object |
CN111027586A (en) * | 2019-11-04 | 2020-04-17 | 天津大学 | Target tracking method based on novel response map fusion |
CN112378397A (en) * | 2020-11-02 | 2021-02-19 | 中国兵器工业计算机应用技术研究所 | Unmanned aerial vehicle target tracking method and device and unmanned aerial vehicle |
CN112378397B (en) * | 2020-11-02 | 2023-10-10 | 中国兵器工业计算机应用技术研究所 | Unmanned aerial vehicle target tracking method and device and unmanned aerial vehicle |
Also Published As
Publication number | Publication date |
---|---|
CN108734109B (en) | 2020-11-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108734109A (en) | A kind of visual target tracking method and system towards image sequence | |
CN110660082B (en) | Target tracking method based on graph convolution and trajectory convolution network learning | |
CN110929578B (en) | Anti-shielding pedestrian detection method based on attention mechanism | |
CN106157319B (en) | The conspicuousness detection method in region and Pixel-level fusion based on convolutional neural networks | |
CN106408594B (en) | Video multi-target tracking based on more Bernoulli Jacob's Eigen Covariances | |
CN106845621B (en) | Dense population number method of counting and system based on depth convolutional neural networks | |
CN109886085A (en) | People counting method based on deep learning target detection | |
CN108109162B (en) | Multi-scale target tracking method using self-adaptive feature fusion | |
CN112132014B (en) | Target re-identification method and system based on non-supervised pyramid similarity learning | |
CN103259962A (en) | Target tracking method and related device | |
CN107689052A (en) | Visual target tracking method based on multi-model fusion and structuring depth characteristic | |
CN105550678A (en) | Human body motion feature extraction method based on global remarkable edge area | |
CN108133172A (en) | Method, the analysis method of vehicle flowrate and the device that Moving Objects are classified in video | |
CN104992223A (en) | Intensive population estimation method based on deep learning | |
CN103971386A (en) | Method for foreground detection in dynamic background scenario | |
CN104091348A (en) | Multi-target tracking method integrating obvious characteristics and block division templates | |
CN109919053A (en) | A kind of deep learning vehicle parking detection method based on monitor video | |
CN112836640A (en) | Single-camera multi-target pedestrian tracking method | |
CN110298297A (en) | Flame identification method and device | |
CN107944354B (en) | Vehicle detection method based on deep learning | |
CN110084201B (en) | Human body action recognition method based on convolutional neural network of specific target tracking in monitoring scene | |
CN109583355B (en) | People flow counting device and method based on boundary selection | |
CN105654508A (en) | Monitoring video moving target tracking method based on self-adaptive background segmentation and system thereof | |
CN106127812A (en) | A kind of passenger flow statistical method of non-gate area, passenger station based on video monitoring | |
CN107103614B (en) | Dyskinesia detection method based on level independent element coding |
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 |
Effective date of registration: 20231031 Address after: No. 548, 5th Floor, Building 10, No. 28 Linping Avenue, Donghu Street, Linping District, Hangzhou City, Zhejiang Province Patentee after: Hangzhou Tuke Intelligent Information Technology Co.,Ltd. Address before: No. 182, Minzu Avenue, Hongshan District, Wuhan City, Hubei Province Patentee before: SOUTH CENTRAL University FOR NATIONALITIES |
|
TR01 | Transfer of patent right |