CN109887001A - Method for tracking target, device, computer equipment and storage medium - Google Patents
Method for tracking target, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN109887001A CN109887001A CN201910099560.XA CN201910099560A CN109887001A CN 109887001 A CN109887001 A CN 109887001A CN 201910099560 A CN201910099560 A CN 201910099560A CN 109887001 A CN109887001 A CN 109887001A
- Authority
- CN
- China
- Prior art keywords
- video frame
- target object
- current video
- target
- model
- 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.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
This application involves a kind of method for tracking target, this method comprises: acquisition includes the current video frame image of target object, extract the characteristics of image in current video frame image, using characteristics of image as the input of position model, obtain the output valve of position model, it determines that the target object in current video frame image whether there is according to output valve to block, when determining in the presence of blocking, using the corresponding position model of a upper video frame images as the corresponding position model of current video frame image, when there is no when blocking for determination, target position of the target object in current video frame image is determined according to output valve, then position model is updated, obtain position model corresponding with current video frame.The method increase the accuracy of tracking and improve the efficiency of target following.Furthermore, it is also proposed that a kind of target tracker, computer equipment and storage medium.
Description
Technical field
The present invention relates to field of computer technology, more particularly, to a kind of method for tracking target, device, computer equipment and
Storage medium.
Background technique
The ultimate challenge that target following at present faces is occlusion issue.When blocking, it is easy to occur tracking to lose
The phenomenon that mistake.Traditional target tracking algorism all can not when processing target is blocked well out active target the problem of, when
When blocking or tracking accuracy is low or calculating speed is slow, tracks low efficiency.
Summary of the invention
Based on this, it is necessary in view of the above-mentioned problems, providing a kind of tracking accuracy height and high-efficient target following side
Method, device, computer equipment and storage medium.
In a first aspect, the embodiment of the present invention provides a kind of method for tracking target, which comprises
Acquisition includes the current video frame image of target object, and the image extracted in the current video frame image is special
Sign;
Using described image feature as the input of position model, the position model is according to mesh in a upper video frame images
What the location updating of mark object obtained;
The output valve for obtaining the position model determines the target object in current video frame image according to the output valve
With the presence or absence of blocking;
It is corresponding using the corresponding position model of a upper video frame images as current video frame image when determining in the presence of blocking
Position model;
When determining there is no when blocking, target of the target object in current video frame image is determined according to the output valve
Position;
The position model is updated according to the target object corresponding target position, is obtained and current video frame
Corresponding position model.
Second aspect, the embodiment of the present invention provide a kind of target tracker, and described device includes:
First extraction module, for obtain include target object current video frame image, extract the current video
Characteristics of image in frame image;
First input module, for using described image feature as the input of position model, the position model to be basis
The location updating of target object obtains in a upper video frame images;
First blocks determining module, for obtaining the output valve of the position model, is determined according to the output valve current
Target object in video frame images, which whether there is, to be blocked;
Block processing module, for when determine exist block when, using the corresponding position model of a upper video frame images as
The corresponding position model of current video frame image;
Position determination module, for determining target object current according to the output valve when determining there is no when blocking
Target position in video frame images;
First update module, for being carried out more according to the corresponding target position of the target object to the position model
Newly, position model corresponding with current video frame is obtained.
The third aspect, the embodiment of the present invention provide a kind of computer equipment, including memory and processor, the memory
It is stored with computer program, when the computer program is executed by the processor, so that the processor executes following steps:
Acquisition includes the current video frame image of target object, and the image extracted in the current video frame image is special
Sign;
Using described image feature as the input of position model, the position model is according to mesh in a upper video frame images
What the location updating of mark object obtained;
The output valve for obtaining the position model determines the target object in current video frame image according to the output valve
With the presence or absence of blocking;
It is corresponding using the corresponding position model of a upper video frame images as current video frame image when determining in the presence of blocking
Position model;
When determining there is no when blocking, target of the target object in current video frame image is determined according to the output valve
Position;
The position model is updated according to the target object corresponding target position, is obtained and current video frame
Corresponding position model.
Fourth aspect, the embodiment of the present invention provide a kind of computer readable storage medium, are stored with computer program, described
When computer program is executed by processor, so that the processor executes following steps:
Acquisition includes the current video frame image of target object, and the image extracted in the current video frame image is special
Sign;
Using described image feature as the input of position model, the position model is according to mesh in a upper video frame images
What the location updating of mark object obtained;
The output valve for obtaining the position model determines the target object in current video frame image according to the output valve
With the presence or absence of blocking;
It is corresponding using the corresponding position model of a upper video frame images as current video frame image when determining in the presence of blocking
Position model;
When determining there is no when blocking, target of the target object in current video frame image is determined according to the output valve
Position;
The position model is updated according to the target object corresponding target position, is obtained and current video frame
Corresponding position model.
Above-mentioned method for tracking target, device, computer equipment and storage medium mention after getting current video frame image
The characteristics of image in current video frame image is taken, then using characteristics of image as the input of position model, position model is basis
The location updating of target object obtains in a upper video frame images, obtains the output valve of position model, according to output valve head
First judge that the target object in current video frame image with the presence or absence of blocking, if there is blocking, does not then update position model, if
There is no blocking, then target position of the target object in current video frame image is determined according to output valve, then target object
Target position position model is updated, obtain the corresponding position model of current video frame.Lead in the method for tracking target
It crosses to detect whether to exist and block, only there is no when blocking, just position model is updated, is significantly reduced due to blocking
With homologue interference and target object deformation quantity it is big caused by track and lose, tracking accuracy is high, and due to being only in place
It joined shadowing during setting model modification, not additional time loss is tracked, mentioned with not influencing real-time
The high efficiency of target following.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
The structure shown according to these attached drawings obtains other attached drawings.
Fig. 1 is the flow chart of method for tracking target in one embodiment;
Fig. 2 is the flow chart of method for tracking target in another embodiment;
Fig. 3 is the schematic diagram that various sizes of image block is extracted in one embodiment;
Fig. 4 is the schematic diagram for accordingly being exported image block characteristics input Scale Model in one embodiment;
Fig. 5 is the flow diagram of target following in one embodiment;
Fig. 6 A is the schematic diagram of the Gaussian response figure before blocking occur in one embodiment;
Fig. 6 B is the schematic diagram of the Gaussian response figure after blocking occur in one embodiment;
Fig. 7 is the structural block diagram of target tracker in one embodiment;
Fig. 8 is the structural block diagram of target tracker in another embodiment;
Fig. 9 is the internal structure chart of computer equipment in one embodiment.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention 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 shown in Figure 1, proposing a kind of method for tracking target, which both can be used for terminal, can also be with
Applied to server, illustrated for being applied to terminal in the present embodiment, specifically includes the following steps:
Step 102, acquisition includes the current video frame image of target object, extracts the image in current video frame image
Feature.
Wherein, target object can be personage, be also possible to animal, plant, can also be object etc., i.e., target object can
With the customized setting of demand according to tracking.Current video frame image is the video image of current target object to be detected, currently
Video frame images, which can be directly to shoot using camera, obtains image, is also possible to the video image prestored.
In order to position the position of target object in current video frame image, need to carry out feature to current video frame image to mention
It takes.In one embodiment, feature extraction can use HOG feature extraction.HOG(Histogram of Oriented
Gradient, histograms of oriented gradients) be characterized in it is a kind of in computer vision and image procossing be used to carry out object detection
Feature Descriptor, by calculating the gradient orientation histogram with statistical picture regional area come constitutive characteristic.In another implementation
In example, feature extraction can also be using LBP feature, Haar feature etc., LBP (Local Binary Pattern, local binary mould
Formula) it is a kind of operator for describing image local textural characteristics.Haar feature includes edge feature, linear character, center spy
It seeks peace diagonal line feature.In one embodiment, terminal is mentioned from current video frame image according to the preset size for extracting frame
A series of image block characteristics are taken out, convenient for the subsequent position for determining target object according to image block characteristics.
Step 104, using characteristics of image as the input of position model, position model is according to mesh in a upper video frame images
What the location updating of mark object obtained.
Wherein, position model is to be trained update according to the position of target object in a upper video frame images to obtain.
Using the feature of the image block where target object as Positive training sample, using the feature of other image blocks as negative training sample pair
Position model is trained, and the prediction of target object position is carried out using the position model that training obtains.Since target object exists
Form in different video images is variation, so needing constantly to obtain target object according to detection to update position mould
What the mode that type, i.e. position model are utilized in line training obtained.It predicts to work as using the corresponding position model of a upper video frame
The position of target object in preceding video frame, by can determine whether mesh according to output using characteristics of image as the input of position model
Mark the position of object.
Position model is carried out according to the position of target object in the first video frame images (initial video frame image)
Training obtains, and by artificially selecting the target object to be tracked in the first video frame images, determines target object in the first view
Position in frequency frame image, then extracts a series of image block characteristics, using the feature of the image block where target object as
Positive sample, other image block characteristics are trained as negative sample, obtain position model corresponding with the first video frame images, so
It goes to predict the position of the target object in next video frame images using the position model afterwards.For example, in initial video frame image
In select some personage.Then the corresponding mark of the image block where the personage is set as 1, other do not include the figure for having the personage
As the mark of block is set as 0, by being trained some column image block characteristics extracted to position model as training sample, obtain
To position model corresponding with initial video frame image, detected the position model as target object in next video frame images
Model.
Step 106, the output valve for obtaining position model determines the target object in current video frame image according to output valve
With the presence or absence of blocking, when in the presence of blocking, 108 being entered step, when there is no blocking, entering step 110.
Wherein, after characteristics of image is input to position model, the output valve of position model is obtained, according to output valve head
It first determines that the target object in current video frame image whether there is to block.When in the presence of blocking, the update of time-out position model,
Using the position model of a upper video frame images as the position model of current video frame image, when there is no when blocking, then first
The position that target object is determined according to output valve, then updates position model.
In one embodiment, position model is realized using position filtering device, and corresponding output valve is Gaussian function
Several responses finds optimal filter by training to get position model is arrived.By the way that the characteristics of image extracted is input to
Position model obtains corresponding output valve.The target object in current video frame image can be calculated according to obtained output valve
With the presence or absence of blocking, when there is shelter to occur, the concussion degree of the response of Gaussian function becomes larger, and shakes threshold value by default,
When concussion degree is greater than concussion threshold value, then it is determined to have and blocks.When existing, when blocking, then holding position model is constant, when not
When in the presence of blocking, then need to update position model.
Step 108, using the corresponding position model of a upper video frame images as the corresponding position mould of current video frame image
Type.
Wherein, when interfering there are shelter or homologue target object, the update of stop position model is conducive to drop
It is low block caused by tracking lose and as target deformation quantity it is big caused by track lose.For example, at one to target person
In the scene being tracked, during tracking, target person has temporarily been blocked in the event of shelter, then has been needed temporarily
The update of stop position model, otherwise will appear misrecognition and identification less than caused pursuing missing, so needing at this time
Using the corresponding position model of a upper video frame images as the corresponding position model of current video frame image.
Step 110, target position of the target object in current video frame image is determined according to output valve.
Wherein, when judgement, which is not present, blocks, then determined target object current according to the output valve in position model
Target position in video frame images.Then the position of target object is determined according to output valve.In one embodiment, multiple
Position corresponding to maximum value in output valve is target position of the target object in current video frame image.
Step 112, position model is updated according to target object corresponding target position, is obtained and current video frame
Corresponding position model.
Wherein, target object is being determined behind the target position in current video frame image, according to the target position pair
Position model carries out re -training update, obtains position model corresponding with current video frame.Specifically, according to target object pair
Other, using the feature of the image block where target object as positive sample, are not included target object by the target position answered
Image block characteristics are updated training to position model as negative sample, obtain updated position model.Above-mentioned target following
Method increases the judgement for the mechanism of blocking, when target object is blocked, the update of stop position model advantageously reduce due to
It blocks bring tracking to lose, improves the accuracy of tracking, and the judgement for blocking mechanism is the mistake in position model modification
It is added in journey, does not bring additional time loss, do not influenced the real-time of tracking, improve the efficiency of target following.
Above-mentioned method for tracking target, after getting current video frame image, the image extracted in current video frame image is special
Sign, then using characteristics of image as the input of position model, position model is according to target object in a upper video frame images
What location updating obtained, the output valve of position model is obtained, the mesh in current video frame image is first determined whether according to the output valve
Mark object if there is blocking, does not then update position model, blocks if it does not exist with the presence or absence of blocking, then true according to output valve
Set the goal target position of the object in current video frame image, and then the target position of target object carries out more position model
Newly, the corresponding position model of current video frame is obtained.It is blocked in the method for tracking target by detecting whether to exist, is not only deposited
When blocking, just position model is updated, is significantly reduced due to blocking and homologue interference and target object shape
It tracks and loses caused by variable is big, tracking accuracy is high, and due to being only that joined screening during the model modification of position
Gear judgement, not additional time loss are tracked, improve the efficiency of target following with not influencing real-time.
As shown in Fig. 2, in one embodiment, being carried out more according to the corresponding target position of target object to position model
Newly, after obtaining position model corresponding with current video frame, further includes:
Step 114, the image block of the target position interception different scale according to target object in current video frame image.
Wherein, determining target object behind the target position in current video frame image, it is also necessary to determine target pair
As the size in current video frame image.In one embodiment, after the target position of target object has been determined, with
Centered on the middle of image block where the target object, intercept the picture of different scale, thus obtained it is a series of not
With the image block of scale.As shown in figure 3, in one embodiment, to extract the schematic diagram of various sizes of image block, in figure, mesh
Marking outside object includes that various sizes of outline border respectively represents various sizes of image block.
Step 116, feature extraction is carried out to each image block and obtains image block characteristics corresponding with each image block.
Wherein, after having obtained various sizes of image block, the extraction of image block characteristics is carried out for each image block, than
Such as, its Feature Descriptor is sought for each image block (Patch), that is, proposes corresponding image block characteristics.The extraction of feature can be with
Using HOG feature, other features can also be used, for example, LBP feature.
Step 118, using image block characteristics as the input of Scale Model, Scale Model is according in a upper video frame images
What the size of target object updated.
Wherein, using the image block characteristics of each image block of extraction as the input of Scale Model, Scale Model is basis
The size of target object updates in a upper video frame images, since the size of target object is the process of a gradual change,
For example, when an object from as far as it is close when, the display of object is increasing, so want real-time update Scale Model, so as to
It predicts that obtained target size is more accurate, while being also beneficial to improve the accuracy of target following.
Step 120, the output valve for obtaining Scale Model determines the matched mesh of target object according to the output valve of Scale Model
Dimensioning.
Wherein, output valve corresponding with each image block characteristics is obtained.It in one embodiment, will be in multiple output valves
Maximum value is as the matched image block of target object, using the size of the image block as the matched target size of target object.Such as
It is in one embodiment, by the f1 of a series of images block feature of extraction, f2, f3 ..., fn is as Scale Model shown in Fig. 4
Input, obtain accordingly exporting g1, g2, g3 ..., the schematic diagram of gn.
Step 122, Scale Model is updated according to target object matched target size, is obtained and current video frame
Corresponding Scale Model.
Wherein, after the corresponding target size of target object has been determined, by the feature pair of the corresponding image block of target size
The mark answered is set as 1, and the feature of other image blocks is set as 0, is then updated training to Scale Model, obtains updated ruler
Spend model, i.e., Scale Model corresponding with current video frame.It is tracked using the estimation that Scale Model carries out size to target object,
Be conducive to improve the accuracy of target following.
Above-mentioned method for tracking target is each responsible for carrying out target positioning and to mesh using position model and Scale Model
Mark carries out the estimation of size, is conducive to the accuracy for further increasing target following through the cooperation between the two, and in position model
During update, the judgement for the mechanism of blocking joined, significantly reduce due to blocking and homologue interference and target pair
It tracks and loses caused by pictograph variable is big.
In one embodiment, after the output valve for obtaining the Scale Model, further includes: according to the scale
The output valve of model determines that the target object in current video frame image whether there is and blocks;It, will be upper when determining in the presence of blocking
The corresponding Scale Model of one video frame images is as the corresponding Scale Model of current video frame image;It is blocked when determining to be not present
When, then the step of determining the target object matched target size into the output valve according to the Scale Model.
Wherein, bring influence is blocked in order to further decrease, target pair is judged according to the output valve of Scale Model again
As when in the presence of blocking, then not updating the Scale Model with the presence or absence of blocking, when there is no when blocking, it is determined that target object
Then size in current video frame image is convenient for the update of model.In above-mentioned method for tracking target, not only in position
It joined shadowing in the update of model, and also joined shadowing during Scale Model, be conducive to improve
The accuracy of shadowing, and then the accuracy of target following is improved, and shadowing mechanism is all the process in model modification
Middle addition, not additional time loss improves tracking performance.
As shown in figure 5, in one embodiment, the flow diagram of target following.Firstly, obtaining the first video frame figure
Picture obtains position model and Scale Model according to the training of the positions and dimensions for the target object selected in the first video frame images,
Then next video frame images are obtained as current video frame image, each image block characteristics pair are obtained according to position model respectively
The output valve answered is blocked if it exists according to output valve it is first determined whether in the presence of blocking, then holding position model is constant.Then
Continue to obtain next video frame images as current video frame image, repeats the above process the tracking for carrying out target object.If no
In the presence of blocking, then the location updating position model of the target object determined according to output valve then should be centered on target position
The feature for extracting the image block of multiple and different sizes obtains corresponding output using image block characteristics as the input of Scale Model
Value, then judges whether there is and blocks, block if it exists, then do not update Scale Model, continues to obtain next video frame images work
For current video frame image, the tracking for carrying out target object is repeated the above process.It blocks, is then determined according to output valve if it does not exist
The size of target object updates Scale Model.Then proceed to obtain next frame video image as in the repetition of current video frame image
State the tracking that process carries out target object.
In one embodiment, described to determine whether the target object in current video frame image is deposited according to the output valve
It is blocking, comprising: when the concussion degree for determining corresponding Gaussian response figure according to the output valve is greater than preset concussion threshold value,
Then determine that the target object in current video frame image exists to block.
Wherein, the calculating of degree of concussion can be the number by calculating unstable location point to obtain, and be also possible to pass through
The difference between maximum value and the average value of other values is calculated to obtain.When concussion degree is greater than preset concussion threshold value, then determine
Target object in current video frame image, which exists, to be blocked.As shown in fig. 6, Fig. 6 A is to occur before blocking in one embodiment
Gaussian response figure schematic diagram, Fig. 6 B is the schematic diagram of the Gaussian response figure after occurring blocking, it is evident that the maximum before blocking
Than more prominent, the maximum value after blocking is obviously reduced value.
In one embodiment, described to determine whether the target object in current video frame image is deposited according to the output valve
It is blocking, comprising: obtain the average value of the maximum output value and remaining output valve in multiple output valves;When the maximum output value
When difference between the average value is less than preset threshold, then determines that the target object in current video frame image exists and hide
Gear.
Wherein, for not blocking the maximum value of the case where, the corresponding Gaussian response figure of target object are bigger,
The corresponding Gauss response value of his image block is all smaller, and the two gap is obvious, and when blocking, maximum value and minimum value
Gap before reduces.So the corresponding output valve of each image block characteristics can be calculated separately, maximum output value is then obtained,
Remaining output valve is averagely obtained into average value, by the difference between maximum output value and average value to determine whether depositing
It is blocking, when difference between the two is less than preset threshold, is illustrating that the shock range of Gaussian response figure is big, target object exists
It blocks.
In one embodiment, described that the position model is carried out more according to the target object corresponding target position
Newly, position model corresponding with current video frame image is obtained, comprising: extract multiple images from the current video frame image
The feature of block is as training sample, using the feature of the corresponding image block in the target position as positive sample, by other image blocks
Feature the position model is trained as negative sample;Will the obtained position model of training as with current video frame figure
As corresponding position model.
Wherein, target object is obtained in current video frame behind target position by identification, it will be corresponding to target position
Image block is as positive sample, using the feature of other image blocks as negative sample, is updated training to position model, obtains and work as
The corresponding position model of preceding video frame images, in order to according to the position model to the target object in next video frame images into
Line trace identification.
In one embodiment, position model is realized using filter.A series of figure is extracted from current video frame
The desired output of image block where target object is labeled as 1, by the phase of other image blocks as training sample by picture block feature
Output is hoped to be labeled as 0, corresponding filter response (output valve) is Gaussian function gj, optimal filter H is found by trainingt,
Corresponding formula can be expressed as follows:
Wherein, htIndicate position filtering device, fjIndicate the training sample of input, gjIndicate the response of Gaussian function, ε table
Show penalty values, fi,gi,htIt is the matrix of M × N.Ht, Fj, GjIt is f respectivelyi,gi,htTable after carrying out discrete Fourier transform
Show.Indicate HtConjugation.
Its penalty values ε minimum can be obtained:Wherein, l indicates a certain of feature
Dimension,Indicate GjConjugation, wherein η is to learn
Habit rate.
Optimal filter H is being calculatedt, using the image block characteristics Z extracted as input, with position filtering device HtInto
Row operation, the calculation formula for obtaining response y are expressed as follows:Wherein,Indicate HtConjugation, F { } indicate
Corresponding function.Using the corresponding position of the maximum value in obtained response as the target position of target object.
In one embodiment, described that target of the target object in current video frame image is determined according to the output valve
Position, comprising: the maximum value in multiple output valves acquired;Exist using the corresponding position of the maximum value as target object
Target position in current video frame image.
Wherein, using the maximum value in output valve as the target position of target object.
As shown in fig. 7, in one embodiment it is proposed that a kind of target tracker, the device include:
First extraction module 702, for obtain include target object current video frame image, extract and described work as forward sight
Characteristics of image in frequency frame image;
First input module 704, for using described image feature as the input of position model, the position model to be root
It is obtained according to the location updating of target object in a upper video frame images;
First blocks determining module 706, for obtaining the output valve of the position model, is worked as according to output valve determination
Target object in preceding video frame images, which whether there is, to be blocked;
Processing module 708 is blocked, for when determining in the presence of blocking, the corresponding position model of a upper video frame images to be made
For the corresponding position model of current video frame image;
Position determination module 710, for determining that target object is being worked as according to the output valve when determining there is no when blocking
Target position in preceding video frame images;
First update module 712, for being carried out according to the corresponding target position of the target object to the position model
It updates, obtains position model corresponding with current video frame.
As shown in figure 8, in one embodiment, above-mentioned target tracker further include:
Interception module 714, it is different for being intercepted according to target position of the target object in current video frame image
The image block of scale;
Second extraction module 716, it is corresponding with each image block for being obtained to the progress feature extraction of each described image block
Image block characteristics;
Second input module 718, for using described image block feature as the input of Scale Model, the Scale Model is
It is updated according to the size of target object in a upper video frame images;
Size determining module 720, for obtaining the output valve of the Scale Model, according to the output valve of the Scale Model
Determine the matched target size of the target object;
Second update module 722, for being carried out according to the matched target size of the target object to the Scale Model
It updates, obtains Scale Model corresponding with current video frame.
In one embodiment, above-mentioned target tracker further include: second blocks determining module, for according to the ruler
The output valve of degree model determines that the target object in current video frame image whether there is and blocks.It, will when determining in the presence of blocking
The corresponding Scale Model of upper video frame images is as the corresponding Scale Model of current video frame image, when determining that there is no block
When, then notify size determining module to determine the matched target size of the target object according to the output valve of the Scale Model.
In one embodiment, described first block determining module be also used to it is corresponding high when being determined according to the output valve
When the concussion degree of this response diagram is greater than preset concussion threshold value, then determines that the target object in current video frame image exists and hide
Gear.
In one embodiment, it described first blocks determining module and is also used to obtain maximum output value in multiple output valves
With the average value of remaining output valve;When the difference between the maximum output value and the average value is less than preset threshold, then
Determine that the target object in current video frame image exists to block.
In one embodiment, first update module is also used to extract multiple figures from the current video frame image
As the feature of block is as training sample, using the feature of the corresponding image block in the target position as positive sample, by other images
The feature of block is trained the position model as negative sample;Will the obtained position model of training as with current video frame
The corresponding position model of image.
In one embodiment, the maximum value in multiple output valves that position determination module is also used to acquire;By institute
State target position of the corresponding position of maximum value as target object in current video frame image.
Fig. 9 shows the internal structure chart of computer equipment in one embodiment.The computer equipment can be terminal,
It can be server.As shown in figure 9, the computer equipment includes processor, memory and the network connected by system bus
Interface.Wherein, memory includes non-volatile memory medium and built-in storage.The non-volatile memory medium of the computer equipment
It is stored with operating system, can also be stored with computer program, when which is executed by processor, may make that processor is real
Existing method for tracking target.Computer program can also be stored in the built-in storage, it, can when which is executed by processor
So that processor performance objective tracking.Network interface with external for being communicated.It will be understood by those skilled in the art that
Structure shown in Fig. 9, only the block diagram of part-structure relevant to application scheme, is not constituted to application scheme institute
The restriction for the computer equipment being applied thereon, specific computer equipment may include than more or fewer portions as shown in the figure
Part perhaps combines certain components or with different component layouts.
In one embodiment, method for tracking target provided by the present application can be implemented as a kind of shape of computer program
Formula, computer program can be run in computer equipment as shown in Figure 9.Composition can be stored in the memory of computer equipment should
Each process template of target tracker.For example, the first extraction module 702, the first input module 704, first blocking determination
Module 706 blocks processing module 708, position determination module 710 and the first update module 712.
A kind of computer equipment, including memory and processor, the memory are stored with computer program, the calculating
When machine program is executed by the processor, so that the processor executes following steps: acquisition includes the current of target object
Video frame images extract the characteristics of image in the current video frame image;Using described image feature as the defeated of position model
Enter, the position model is obtained according to the location updating of target object in a upper video frame images;Obtain the position mould
The output valve of type determines that the target object in current video frame image whether there is according to the output valve and blocks;When determination is deposited
When blocking, using the corresponding position model of a upper video frame images as the corresponding position model of current video frame image;When true
Determine to determine target position of the target object in current video frame image according to the output valve there is no when blocking;According to institute
It states the corresponding target position of target object to be updated the position model, obtains position corresponding with current video frame mould
Type.
In one embodiment, the position model is carried out according to the target object corresponding target position described
It updates, after obtaining position model corresponding with current video frame, when the processor is executed by the computer program, also uses
In execution following steps: according to the image of target position interception different scale of the target object in current video frame image
Block;Feature extraction is carried out to each described image block and obtains image block characteristics corresponding with each image block;By described image block
Input of the feature as Scale Model, the Scale Model are updated according to the size of target object in a upper video frame images
It arrives;The output valve for obtaining the Scale Model determines that the target object is matched according to the output valve of the Scale Model
Target size;The Scale Model is updated according to the target object matched target size, is obtained and current video
The corresponding Scale Model of frame.
In one embodiment, after the output valve for obtaining the Scale Model, the processor is by the meter
When calculation machine program executes, it is also used to execute following steps: current video frame image is determined according to the output valve of the Scale Model
In target object with the presence or absence of blocking;When determine exist block when, using the corresponding Scale Model of a upper video frame images as
The corresponding Scale Model of current video frame image;It is described according to the Scale Model there is no when blocking, then entering when determining
Output valve determines the step of target object matched target size.
In one embodiment, described to determine whether the target object in current video frame image is deposited according to the output valve
It is blocking, comprising: when the concussion degree for determining corresponding Gaussian response figure according to the output valve is greater than preset concussion threshold value,
Then determine that the target object in current video frame image exists to block.
In one embodiment, described to determine whether the target object in current video frame image is deposited according to the output valve
It is blocking, comprising: obtain the average value of the maximum output value and remaining output valve in multiple output valves;When the maximum output value
When difference between the average value is less than preset threshold, then determines that the target object in current video frame image exists and hide
Gear.
In one embodiment, described that the position model is carried out more according to the target object corresponding target position
Newly, position model corresponding with current video frame image is obtained, comprising: extract multiple images from the current video frame image
The feature of block is as training sample, using the feature of the corresponding image block in the target position as positive sample, by other image blocks
Feature the position model is trained as negative sample;Will the obtained position model of training as with current video frame figure
As corresponding position model.
In one embodiment, described that target of the target object in current video frame image is determined according to the output valve
Position, comprising: the maximum value in multiple output valves acquired;Exist using the corresponding position of the maximum value as target object
Target position in current video frame image.
A kind of computer readable storage medium is stored with computer program, when the computer program is executed by processor,
So that the processor executes following steps: acquisition includes the current video frame image of target object, and extraction is described to work as forward sight
Characteristics of image in frequency frame image;Using described image feature as the input of position model, the position model is according to upper one
The location updating of target object obtains in video frame images;The output valve for obtaining the position model, according to the output valve
It determines that the target object in current video frame image whether there is to block;When determining in the presence of blocking, by a upper video frame images
Corresponding position model is as the corresponding position model of current video frame image;When determination, which is not present, blocks, according to described defeated
It is worth the target position for determining target object in current video frame image out;According to the corresponding target position pair of the target object
The position model is updated, and obtains position model corresponding with current video frame.
In one embodiment, the position model is carried out according to the target object corresponding target position described
It updates, after obtaining position model corresponding with current video frame, when the processor is executed by the computer program, also uses
In execution following steps: according to the image of target position interception different scale of the target object in current video frame image
Block;Feature extraction is carried out to each described image block and obtains image block characteristics corresponding with each image block;By described image block
Input of the feature as Scale Model, the Scale Model are updated according to the size of target object in a upper video frame images
It arrives;The output valve for obtaining the Scale Model determines that the target object is matched according to the output valve of the Scale Model
Target size;The Scale Model is updated according to the target object matched target size, is obtained and current video
The corresponding Scale Model of frame.
In one embodiment, after the output valve for obtaining the Scale Model, the processor is by the meter
When calculation machine program executes, it is also used to execute following steps: current video frame image is determined according to the output valve of the Scale Model
In target object with the presence or absence of blocking;When determine exist block when, using the corresponding Scale Model of a upper video frame images as
The corresponding Scale Model of current video frame image;It is described according to the Scale Model there is no when blocking, then entering when determining
Output valve determines the step of target object matched target size.
In one embodiment, described to determine whether the target object in current video frame image is deposited according to the output valve
It is blocking, comprising: when the concussion degree for determining corresponding Gaussian response figure according to the output valve is greater than preset concussion threshold value,
Then determine that the target object in current video frame image exists to block.
In one embodiment, described to determine whether the target object in current video frame image is deposited according to the output valve
It is blocking, comprising: obtain the average value of the maximum output value and remaining output valve in multiple output valves;When the maximum output value
When difference between the average value is less than preset threshold, then determines that the target object in current video frame image exists and hide
Gear.
In one embodiment, described that the position model is carried out more according to the target object corresponding target position
Newly, position model corresponding with current video frame image is obtained, comprising: extract multiple images from the current video frame image
The feature of block is as training sample, using the feature of the corresponding image block in the target position as positive sample, by other image blocks
Feature the position model is trained as negative sample;Will the obtained position model of training as with current video frame figure
As corresponding position model.
In one embodiment, described that target of the target object in current video frame image is determined according to the output valve
Position, comprising: the maximum value in multiple output valves acquired;Exist using the corresponding position of the maximum value as target object
Target position in current video frame image.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a non-volatile computer and can be read
In storage medium, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, provided herein
Each embodiment used in any reference to memory, storage, database or other media, may each comprise non-volatile
And/or volatile memory.Nonvolatile memory may include that read-only memory (ROM), programming ROM (PROM), electricity can be compiled
Journey ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include random access memory
(RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms, such as static state RAM
(SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhanced SDRAM
(ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) directly RAM (RDRAM), straight
Connect memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of above embodiments can be combined arbitrarily, for simplicity of description, not to above-described embodiment
In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance
Shield all should be considered as described in this specification.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
The limitation to the application the scope of the patents therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the concept of this application, various modifications and improvements can be made, these belong to the guarantor of the application
Protect range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (10)
1. a kind of method for tracking target, which is characterized in that the described method includes:
Acquisition includes the current video frame image of target object, extracts the characteristics of image in the current video frame image;
Using described image feature as the input of position model, the position model is according to target pair in a upper video frame images
What the location updating of elephant obtained;
Whether the output valve for obtaining the position model determines the target object in current video frame image according to the output valve
In the presence of blocking;
When determining in the presence of blocking, using the corresponding position model of a upper video frame images as the corresponding position of current video frame image
Set model;
When determining there is no when blocking, target position of the target object in current video frame image is determined according to the output valve
It sets;
The position model is updated according to the target object corresponding target position, is obtained corresponding with current video frame
Position model.
2. the method according to claim 1, wherein described according to the corresponding target position of the target object
The position model is updated, after obtaining position model corresponding with current video frame, further includes:
According to the image block of target position interception different scale of the target object in current video frame image;
Feature extraction is carried out to each described image block and obtains image block characteristics corresponding with each image block;
Using described image block feature as the input of Scale Model, the Scale Model is according to target in a upper video frame images
What the size of object updated;
The output valve for obtaining the Scale Model determines the matched mesh of the target object according to the output valve of the Scale Model
Dimensioning;
The Scale Model is updated according to the target object matched target size, is obtained corresponding with current video frame
Scale Model.
3. according to the method described in claim 2, it is characterized in that, after the output valve for obtaining the Scale Model,
Further include:
It determines that the target object in current video frame image whether there is according to the output valve of the Scale Model to block;
When determining in the presence of blocking, using the corresponding Scale Model of a upper video frame images as the corresponding ruler of current video frame image
Spend model;
When determining there is no when blocking, then the target object matching is determined into the output valve according to the Scale Model
Target size the step of.
4. the method according to claim 1, wherein described determine current video frame image according to the output valve
In target object with the presence or absence of blocking, comprising:
When the concussion degree for determining corresponding Gaussian response figure according to the output valve is greater than preset concussion threshold value, then determine to work as
Target object in preceding video frame images, which exists, to be blocked.
5. the method according to claim 1, wherein described determine current video frame image according to the output valve
In target object with the presence or absence of blocking, comprising:
Obtain the average value of the maximum output value and remaining output valve in multiple output valves;
When the difference between the maximum output value and the average value is less than preset threshold, then current video frame image is determined
In target object exist block.
6. the method according to claim 1, wherein described according to the corresponding target position pair of the target object
The position model is updated, and obtains position model corresponding with current video frame image, comprising:
The feature of multiple images block is extracted from the current video frame image as training sample, and the target position is corresponded to
Image block feature as positive sample, the position model is trained using the feature of other image blocks as negative sample;
The position model that training is obtained is as position model corresponding with current video frame image.
7. the method according to claim 1, wherein described determine target object current according to the output valve
Target position in video frame images, comprising:
The maximum value in multiple output valves acquired;
Target position using the corresponding position of the maximum value as target object in current video frame image.
8. a kind of target tracker, which is characterized in that described device includes:
First extraction module, for obtain include target object current video frame image, extract the current video frame figure
Characteristics of image as in;
First input module, for using described image feature as the input of position model, the position model to be according to upper one
The location updating of target object obtains in video frame images;
First blocks determining module, for obtaining the output valve of the position model, determines current video according to the output valve
Target object in frame image, which whether there is, to be blocked;
Block processing module, for when determine exist block when, will the corresponding position model of a upper video frame images as currently
The corresponding position model of video frame images;
Position determination module, for determining target object in current video according to the output valve when determining there is no when blocking
Target position in frame image;
First update module is obtained for being updated according to the corresponding target position of the target object to the position model
To position model corresponding with current video frame.
9. a kind of computer equipment, including memory and processor, the memory is stored with computer program, the computer
When program is executed by the processor, so that the processor executes the step such as any one of claims 1 to 7 the method
Suddenly.
10. a kind of computer readable storage medium is stored with computer program, when the computer program is executed by processor,
So that the processor is executed such as the step of any one of claims 1 to 7 the method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910099560.XA CN109887001A (en) | 2019-01-31 | 2019-01-31 | Method for tracking target, device, computer equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910099560.XA CN109887001A (en) | 2019-01-31 | 2019-01-31 | Method for tracking target, device, computer equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109887001A true CN109887001A (en) | 2019-06-14 |
Family
ID=66927488
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910099560.XA Pending CN109887001A (en) | 2019-01-31 | 2019-01-31 | Method for tracking target, device, computer equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109887001A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110533690A (en) * | 2019-08-08 | 2019-12-03 | 河海大学 | The core correlation filtering Method for Underwater Target Tracking of Fusion Features and learning rate optimization |
CN111598925A (en) * | 2020-05-15 | 2020-08-28 | 武汉卓目科技有限公司 | Visual target tracking method and device based on ECO algorithm and region growth segmentation |
CN112967399A (en) * | 2021-03-31 | 2021-06-15 | 东莞中国科学院云计算产业技术创新与育成中心 | Three-dimensional time sequence image generation method and device, computer equipment and storage medium |
CN113190909A (en) * | 2021-05-21 | 2021-07-30 | 杭州群核信息技术有限公司 | Method, device and storage medium for determining position reasonability of target object |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009085233A2 (en) * | 2007-12-21 | 2009-07-09 | 21Ct, Inc. | System and method for visually tracking with occlusions |
CN107644430A (en) * | 2017-07-27 | 2018-01-30 | 孙战里 | Target following based on self-adaptive features fusion |
CN107680119A (en) * | 2017-09-05 | 2018-02-09 | 燕山大学 | A kind of track algorithm based on space-time context fusion multiple features and scale filter |
US20180047173A1 (en) * | 2016-08-12 | 2018-02-15 | Qualcomm Incorporated | Methods and systems of performing content-adaptive object tracking in video analytics |
CN108596951A (en) * | 2018-03-30 | 2018-09-28 | 西安电子科技大学 | A kind of method for tracking target of fusion feature |
CN108830879A (en) * | 2018-05-29 | 2018-11-16 | 上海大学 | A kind of unmanned boat sea correlation filtering method for tracking target suitable for blocking scene |
-
2019
- 2019-01-31 CN CN201910099560.XA patent/CN109887001A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009085233A2 (en) * | 2007-12-21 | 2009-07-09 | 21Ct, Inc. | System and method for visually tracking with occlusions |
US20180047173A1 (en) * | 2016-08-12 | 2018-02-15 | Qualcomm Incorporated | Methods and systems of performing content-adaptive object tracking in video analytics |
CN107644430A (en) * | 2017-07-27 | 2018-01-30 | 孙战里 | Target following based on self-adaptive features fusion |
CN107680119A (en) * | 2017-09-05 | 2018-02-09 | 燕山大学 | A kind of track algorithm based on space-time context fusion multiple features and scale filter |
CN108596951A (en) * | 2018-03-30 | 2018-09-28 | 西安电子科技大学 | A kind of method for tracking target of fusion feature |
CN108830879A (en) * | 2018-05-29 | 2018-11-16 | 上海大学 | A kind of unmanned boat sea correlation filtering method for tracking target suitable for blocking scene |
Non-Patent Citations (2)
Title |
---|
MENGMENG WANG等: "Large margin object tracking with circulant feature maps", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
顾培婷等: "抗遮挡的相关滤波目标跟踪算法", 《华侨大学学报(自然科学版)》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110533690A (en) * | 2019-08-08 | 2019-12-03 | 河海大学 | The core correlation filtering Method for Underwater Target Tracking of Fusion Features and learning rate optimization |
CN110533690B (en) * | 2019-08-08 | 2022-02-11 | 河海大学 | Nuclear correlation filtering underwater target tracking method based on feature fusion and learning rate optimization |
CN111598925A (en) * | 2020-05-15 | 2020-08-28 | 武汉卓目科技有限公司 | Visual target tracking method and device based on ECO algorithm and region growth segmentation |
CN111598925B (en) * | 2020-05-15 | 2023-10-17 | 武汉卓目科技有限公司 | Visual target tracking method and device based on ECO algorithm and region growing segmentation |
CN112967399A (en) * | 2021-03-31 | 2021-06-15 | 东莞中国科学院云计算产业技术创新与育成中心 | Three-dimensional time sequence image generation method and device, computer equipment and storage medium |
CN113190909A (en) * | 2021-05-21 | 2021-07-30 | 杭州群核信息技术有限公司 | Method, device and storage medium for determining position reasonability of target object |
CN113190909B (en) * | 2021-05-21 | 2023-02-24 | 杭州群核信息技术有限公司 | Method, device and storage medium for determining position reasonability of target object |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109887001A (en) | Method for tracking target, device, computer equipment and storage medium | |
CN108550161B (en) | Scale self-adaptive kernel-dependent filtering rapid target tracking method | |
CN109033955B (en) | Face tracking method and system | |
CN107452015B (en) | Target tracking system with re-detection mechanism | |
CN108961308B (en) | Residual error depth characteristic target tracking method for drift detection | |
CN108198201A (en) | A kind of multi-object tracking method, terminal device and storage medium | |
CN109035295B (en) | Multi-target tracking method, device, computer equipment and storage medium | |
CN109711332B (en) | Regression algorithm-based face tracking method and application | |
CN105488468A (en) | Method and device for positioning target area | |
CN113191180B (en) | Target tracking method, device, electronic equipment and storage medium | |
CN106709938A (en) | Multi-target tracking method based on improved TLD (tracking-learning-detected) | |
CN111027347A (en) | Video identification method and device and computer equipment | |
CN113327272A (en) | Robustness long-time tracking method based on correlation filtering | |
CN111950498A (en) | Lane line detection method and device based on end-to-end instance segmentation | |
CN108898057A (en) | Track method, apparatus, computer equipment and the storage medium of target detection | |
CN110660077A (en) | Multi-scale target tracking method fusing multiple features | |
CN113763424B (en) | Real-time intelligent target detection method and system based on embedded platform | |
CN113379789B (en) | Moving target tracking method in complex environment | |
CN110472607A (en) | A kind of ship tracking method and system | |
CN110334703B (en) | Ship detection and identification method in day and night image | |
CN109727268A (en) | Method for tracking target, device, computer equipment and storage medium | |
US20230342951A1 (en) | Object tracking apparatus, object tracking method, and program | |
CN111815677A (en) | Target tracking method and device, terminal equipment and readable storage medium | |
CN112488062A (en) | Image identification method, device, equipment and medium | |
CN110866484B (en) | Driver face detection method, computer device and computer readable storage medium |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190614 |