CN117152203A - Target tracking method and device based on multi-core correlation filter and electronic equipment - Google Patents
Target tracking method and device based on multi-core correlation filter and electronic equipment Download PDFInfo
- Publication number
- CN117152203A CN117152203A CN202310946698.5A CN202310946698A CN117152203A CN 117152203 A CN117152203 A CN 117152203A CN 202310946698 A CN202310946698 A CN 202310946698A CN 117152203 A CN117152203 A CN 117152203A
- Authority
- CN
- China
- Prior art keywords
- target
- frame image
- image
- sample
- core
- 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
- 238000000034 method Methods 0.000 title claims abstract description 73
- 230000004044 response Effects 0.000 claims abstract description 119
- 238000012360 testing method Methods 0.000 claims abstract description 90
- 239000000523 sample Substances 0.000 claims description 130
- 230000006870 function Effects 0.000 claims description 55
- 125000004122 cyclic group Chemical group 0.000 claims description 20
- 238000011156 evaluation Methods 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 18
- 238000000605 extraction Methods 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 11
- 230000001133 acceleration Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 9
- 239000013074 reference sample Substances 0.000 claims description 9
- 238000012952 Resampling Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000004891 communication Methods 0.000 description 6
- 239000000284 extract Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 206010000117 Abnormal behaviour Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/10—Image enhancement or restoration using non-spatial domain filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a target tracking method and device based on a multi-core correlation filter and electronic equipment, and relates to the technical field of computer vision, wherein the method comprises the following steps: traversing the image sequence, and extracting features from the current frame image traversed currently to obtain a test sample; the current frame image comprises a target object; performing correlation operation on the test sample based on the multi-core correlation filter to obtain a response graph, and determining the position corresponding to the maximum response value in the response graph as the target position of the target object in the current frame image; updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for carrying out correlation operation on the test sample corresponding to the next frame of image so as to determine the target position of the target object in the next frame of image. The invention can improve the accuracy of target tracking.
Description
Technical Field
The present invention relates to the field of computer vision technologies, and in particular, to a target tracking method and apparatus based on a multi-core correlation filter, and an electronic device.
Background
In the existing target tracking method, more and more computer vision workers begin to research and utilize the related filtering theory to track the target.
In the related art, feature extraction is generally performed on a target object in each frame of image of an image sequence, correlation operation is performed on the extracted features based on a trained filter for the features extracted from each frame of image, a response chart is obtained, a position corresponding to a maximum response value in the response chart is determined as a position of the target object in the frame of image, and the position of the target object in each frame of image can be obtained according to the same method, so that tracking of the target is realized.
However, in the related art, the correlation calculation is performed on the features extracted from each frame of image based on the filter trained at one time, which reduces the accuracy of target tracking.
Disclosure of Invention
Aiming at the problems existing in the prior art, the embodiment of the invention provides a target tracking method and device based on a multi-core correlation filter and electronic equipment.
The invention provides a target tracking method based on a multi-core correlation filter, which comprises the following steps:
traversing the image sequence, and extracting features from the current frame image traversed currently to obtain a test sample; the current frame image comprises a target object;
Performing correlation operation on the test sample based on a multi-core correlation filter to obtain a response graph, and determining a position corresponding to a maximum response value in the response graph as a target position of the target object in the current frame image;
updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
According to the target tracking method based on the multi-core correlation filter provided by the invention, before traversing the image sequence, the method further comprises the following steps:
acquiring a first frame image in the image sequence; the first frame image comprises a target area; the target area comprises the target object;
performing cyclic shift on the first frame image to obtain at least two shift images;
determining the characteristics of the target area in the shift image as a virtual sample; the features include color features and HOG features;
optimizing an objective function based on the virtual sample to obtain a sample coefficient and a kernel coefficient corresponding to the condition that the objective function is the minimum value;
The multi-core correlation filter is determined based on the sample coefficients and the kernel coefficients.
According to the target tracking method based on the multi-core correlation filter provided by the invention, the method extracts the characteristics from the current frame image traversed currently to obtain the test sample, and comprises the following steps:
extracting color features from the current frame image based on a formula (1), extracting HOG features from the current frame image based on a formula (2), and determining the color features and the HOG features as the test sample;
wherein,representing a color feature extraction kernel for characterizing the color features of the C-th color channel, C representing the number of color channels, and>test sample z representing the j-th frame j Color value on the c-th color channel, and->Representing the color value of the ith virtual sample on the c-th color channel, +.>Representing HOG feature extraction kernel function for characterizing test sample z j And HOG features of the ith virtual sample.
According to the target tracking method based on the multi-core correlation filter provided by the invention, the multi-core correlation filter is used for carrying out correlation operation on the test sample to obtain a response diagram, and the target tracking method comprises the following steps:
for each virtual sample, performing correlation operation on the test sample based on a formula (3) to obtain a response value corresponding to the correlation between the virtual sample and the test sample;
Determining the response map based on equation (4);
wherein y is j (z) test sample z representing the ith virtual sample and the jth frame j The response value corresponding to the correlation between M represents the number of kernel functions, d m Represents the kernel coefficient corresponding to the mth kernel function, l represents the size of the cyclic shift matrix P, and alpha i For the sample coefficient corresponding to the i-th virtual sample,test sample z representing the ith virtual sample and the jth frame j Correlation between; y (z) represents a response map, +.>Is to->As a cyclic matrix of the first row, representing the mth kernel function k m In test sample z and located sample transformed by cyclic shift matrix P>Results of the evaluation therebetween.
According to the target tracking method based on the multi-core correlation filter provided by the invention, the determining of the position corresponding to the maximum response value in the response map as the target position of the target object in the current frame image comprises the following steps:
performing acceleration calculation on the response map based on a formula (5) to obtain an accelerated response map, and determining a position corresponding to a maximum response value in the accelerated response map as a target position of the target object in the current frame image;
wherein y is 1 (z) represents the response after acceleration, and by element-wise multiplication, Representing a one-dimensional Fourier transform, ">Representing the corresponding conjugate transformation->Representing inverse Fourier transform, y 1 The element with the maximum response value in (z) is accepted as the current frame image +.>Target position of the target object.
According to the target tracking method based on the multi-core correlation filter provided by the invention, the multi-core correlation filter is updated based on the target position, and the updated multi-core correlation filter is obtained, and the target tracking method comprises the following steps:
performing scale space search on the current frame image based on the target position to obtain an optimal target scale;
and updating the multi-core correlation filter based on the characteristics corresponding to the optimal target scale and the optimal target scale in the current frame image to obtain the updated multi-core correlation filter.
According to the target tracking method based on the multi-core correlation filter provided by the invention, the method for searching the scale space of the current frame image based on the target position to obtain the optimal target scale comprises the following steps:
extracting a patch area with a preset scale from the current frame image;
resampling the patch area based on the scale corresponding to the previous frame of image to obtain a reference sample;
Determining a target response map based on the reference sample and the accelerated response map;
calculating peak value and side lobe ratio of the target response graph, and determining an evaluation function based on the peak value and the side lobe ratio;
and optimizing the evaluation function based on a golden section method to obtain the optimal target scale.
According to the target tracking method based on the multi-core correlation filter provided by the invention, the method for updating the multi-core correlation filter based on the characteristics corresponding to the optimal target scale and the optimal target scale in the current frame image to obtain the updated multi-core correlation filter comprises the following steps:
updating a positioned sample corresponding to a previous frame of image based on a formula (6) to obtain an updated positioned sample, and updating the multi-core correlation filter based on the updated positioned sample to obtain the updated multi-core correlation filter;
wherein,representing the color value of the updated located sample in the mth color channel, +.>Representing the color value, η, of the located sample at the mth color channel m Indicates learning rate (I/O)>Representing patch areas corresponding to optimal target scales s * Representing the optimal target scale >Representing the scale corresponding to the target object in the previous frame of image,/->Representing the current frame image being extracted in the mth color channelTaken patch area,/>Represents the center position of the current frame image, +.>Representing the current frame image, λ represents the regularization parameters.
The invention also provides a target tracking device based on the multi-core correlation filter, which comprises:
the extraction unit is used for traversing the image sequence, extracting features from the current frame image traversed currently, and obtaining a test sample; the current frame image comprises a target object;
the operation unit is used for carrying out correlation operation on the test sample based on a multi-core correlation filter to obtain a response graph, and determining the position corresponding to the maximum response value in the response graph as the target position of the target object in the current frame image;
the updating unit is used for updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
The invention also provides electronic equipment, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the target tracking method based on the multi-core correlation filter when executing the program.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a multi-core correlation filter based target tracking method as described in any of the above.
The invention also provides a computer program product comprising a computer program which, when executed by a processor, implements a multi-core correlation filter based object tracking method as described in any one of the above.
According to the target tracking method, the target tracking device and the electronic equipment based on the multi-core correlation filter, the image sequence is traversed, the characteristics are extracted from the current frame image which is traversed currently, the test sample is obtained, the correlation operation is carried out on the test sample based on the multi-core correlation filter, the response graph is obtained, the position corresponding to the maximum response value in the response graph is determined to be the target position of the target object in the current frame image, the multi-core correlation filter is updated based on the target position, the correlation operation is carried out on the test sample corresponding to the next frame image based on the updated multi-core correlation filter, and the target position of the target object in the next frame image is determined until the image traversal in the image sequence is finished. As can be seen, when the correlation operation is performed on the test sample corresponding to each frame of image, the multi-core correlation filter is updated based on the target position of the target object in the previous frame of image, so that the multi-core correlation filter after each update is matched with the real-time characteristic of the target object, and the accuracy of target tracking can be improved.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a target tracking method based on a multi-core correlation filter provided by the invention;
FIG. 2 is a second flow chart of the target tracking method based on the multi-core correlation filter provided by the invention;
FIG. 3 is a third flow chart of the target tracking method based on the multi-core correlation filter provided by the invention;
FIG. 4 is a schematic diagram of a target tracking device based on a multi-core correlation filter according to the present invention;
fig. 5 is a schematic diagram of the physical structure of the electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The following describes a target tracking method based on a multi-core correlation filter according to the present invention with reference to fig. 1 to 3. The execution subject of the embodiment of the invention can be electronic equipment such as a terminal, a computer, a server or a server cluster, or can be a target tracking device based on a multi-core correlation filter, which is arranged in the electronic equipment, and the target tracking device based on the multi-core correlation filter can be realized by software, hardware or a combination of the two.
Fig. 1 is one of flow diagrams of a target tracking method based on a multi-core correlation filter according to an embodiment of the present invention, as shown in fig. 1, the target tracking method based on a multi-core correlation filter includes the following steps:
step 101, traversing an image sequence, and extracting features from a current frame image traversed currently to obtain a test sample; the current frame image comprises a target object.
The target object is an object to be tracked by a user, and can be a vehicle, a person or an object, and the like.
For example, collecting multi-frame images with time sequence collected in the fields of video monitoring, automatic driving, intelligent home and the like, wherein the multi-frame images form an image sequence, the image sequence is used as a resource required for tracking a target object, the current frame image can be a second frame image in the image sequence, color features and direction gradient histogram (Histogram of Oriented Gradient, HOG) features in the current frame image are extracted, and the color features and the HOG features in the current frame image are used as test samples. For each frame of image there will be multiple virtual samples, which are generated by cyclic shifts.
Step 102, performing correlation operation on the test sample based on a multi-core correlation filter to obtain a response diagram, and determining a position corresponding to a maximum response value in the response diagram as a target position of the target object in the current frame image.
Illustratively, the multi-core correlation filter is an algorithm for tracking a target in real time. The target tracking of the multi-core correlation filter can be seen as a regression process. In target tracking, it is necessary to predict the position of the target object in the current frame image, which can be regarded as a regression problem. By training the multi-core correlation filter, a relationship between the characteristics of the target object in the frame image and the position of the target object in the frame image can be learned, and the position of the target object in the frame image can be predicted using the relationship. And extracting the characteristics of the current frame image from each frame image to obtain a test sample, and calculating the correlation between the test sample and each virtual sample, namely the similarity. By calculating the correlation, a response map can be obtained in which each position corresponds to the likelihood that the target object is at that position. Then, the position having the largest response value is selected as the target position of the target object in the current frame image. The multi-core correlation filter can evaluate the similarity between the virtual sample and the test sample by utilizing the color feature and the HOG feature simultaneously. The color feature may capture color information of a target object in the current frame image and the HOG feature may capture shape information and texture information of the target object in the current frame image. Continuous tracking of the target object is achieved by determining the position of the target object in each frame of image.
Step 103, updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
For example, when determining the target position of the target object in the current frame image, the multi-core correlation filter may be updated based on the target position, so that the updated multi-core correlation filter may be synchronized with the real-time change of the appearance of the target object, for example, the target object is a vehicle, the appearance of the target object may be a driving angle of the vehicle on a road, or the like. And aiming at the next frame image in the image sequence, extracting color features and HOG features from the next frame image to obtain a test sample corresponding to the next frame image, carrying out correlation operation on the test sample based on the updated multi-core correlation filter to obtain a response image corresponding to the next frame image, determining the position corresponding to the maximum response value in the response image corresponding to the next frame image as the target position of the target object in the next frame image, updating the updated multi-core correlation filter again based on the target position of the target object in the next frame image, and repeating the steps until all the frame images in the image sequence are traversed, and continuously tracking the target object based on the target position of the target object in each frame image determined after the traversing is completed.
It should be noted that, in the field of video monitoring, a target object that can be tracked may be a vehicle, a target object, or the like; in the automatic driving field, the target tracking can predict the action track of surrounding targets to make decisions and plan; in the field of smart home, people in the home can be tracked in real time, and abnormal behaviors or invasion can be detected; tracking the position and behavior of the pet, whether the pet leaves a designated area, needs feeding, and the like, and timely sending a notification to the user.
According to the target tracking method based on the multi-core correlation filter, an image sequence is traversed, features are extracted from a current frame image which is traversed currently to obtain a test sample, correlation operation is carried out on the test sample based on the multi-core correlation filter to obtain a response chart, the position corresponding to the maximum response value in the response chart is determined to be the target position of a target object in the current frame image, the multi-core correlation filter is updated based on the target position, correlation operation is carried out on the test sample corresponding to a next frame image based on the updated multi-core correlation filter to determine the target position of the target object in the next frame image until image traversal in the image sequence is completed. As can be seen, when the correlation operation is performed on the test sample corresponding to each frame of image, the multi-core correlation filter is updated based on the target position of the target object in the previous frame of image, so that the multi-core correlation filter after each update is matched with the real-time characteristic of the target object, and the accuracy of target tracking can be improved.
In an embodiment, fig. 2 is a second flowchart of the target tracking method based on a multi-core correlation filter according to the embodiment of the present invention, as shown in fig. 2, before the step 101, the target tracking method based on a multi-core correlation filter further includes the following steps:
104, acquiring a first frame image in the image sequence; the first frame image comprises a target area, and the target area comprises a target object.
For example, a first frame image in the image sequence is acquired, and the user can frame a target region including the target object in the first frame image based on the requirement, and extract color features and HOG features of the target region.
And 105, performing cyclic shift on the first frame image to obtain at least two shift images.
For example, when the first frame image is acquired, the first frame image may be sequentially shifted based on cyclic shift, and in the shifting process, the frame-selected target area always exists, and each obtained shifted image also includes the target area.
Step 106, determining the characteristics of the target area in the shift image as a virtual sample; the features include color features and HOG features.
For each shift image, for example, the color feature and the HOG feature of the target area in the shift image may be extracted, and the color feature and the HOG feature of the target area are used as virtual samples, where one shift image corresponds to one virtual sample, and then multiple shift images obtain multiple virtual samples.
And 107, optimizing an objective function based on the virtual sample to obtain a sample coefficient and a kernel coefficient corresponding to the condition that the objective function is the minimum value.
Wherein the sample coefficient α represents the importance of each virtual sample in the response, and the kernel coefficient d represents the weight of the mth kernel function.
Illustratively, the training objective of the multi-kernel correlation filter is to find an objective function f (x) such that in the virtual sample x i And virtual sample x i Regression target y of (2) i The above minimum mean square error can be expressed by the following equation (7):
where l denotes the size of the cyclic shift matrix P, i.e. the number of virtual samples, f is located in a bounded convex subset in the regenerated kernel Hilbert space defined by a positive definite kernel function k (·, ·), λ++0, denotes the regularization parameter, k denotes the k-order norm, and according to the regenerated kernel theorem, the solution f of the Gibbs Tikhonov regularization problem * Can be represented by the following formula (8):
wherein α= (α) 0 ,α 1 ,…,α l-1 ) T K represents a positive semi-definite kernel matrix, and the element of K is K (x i X). The problem becomes a problem of solving α, which can be expressed by the following equation (9):
wherein y= (y) 0 ,y 1 ,…,y l-1 ) T ，Representing the domain.
X is the number i Representing the virtual samples in the derivation process,representing a weighted average from the virtual samples.
Research has shown that using multiple kernel functions rather than a single kernel function can increase the degree of discrimination of the algorithm, given the underlying kernel function k m M=1, 2, …, M, the usual method is to add k (x i X) is represented as a convex combination of basis kernel functions, i.e. k (x) i ,x)＝d T k(x i ,x)，k(x i ,x)＝(k 1 (x i ,x),k 2 (x i ,x),…,k M (x i ,x)) T ，k(x i X) represents the kernel function k at the located sample x i And the similarity evaluation result between the current samples x, wherein the positioned samples are the results of the weighted operation of the previous frame image through the filter, and the current samples are virtual samples in the virtual data set of the current frame image. Performing similarity evaluation by using kernel function through the virtual sample determined by the previous frame image and the virtual sample in the virtual data set of the current frame image, and d= (d) 1 ,d 2 ,…,d M ) T And (2) andthe following formula (10) can thus be obtained:
wherein K is m Is the m-th basic kernel matrix, K m The elements of (2) areSubstituting K in equation (10) into equation (9) and introducing the pair d m The additional constraint of the sum can obtain an objective function F (alpha, d) of the Tikhonov regularization problem of a multi-core version, and F (alpha, d) can be expressed by the following formula (11):
wherein λ=10 -3 ，v＝10 -2 To ensure that all the combining coefficients are positive, we use in equation (11)Rather than d m I.e. +.>It is noted that formula (11) is equivalent to the third term +.>As a constraint, v is a constrained multi-core optimization problem of lagrangian lagranger multiplier, and its optimal solution can be expressed as the following formula (12):
wherein,
it should be noted that equation (11) is also a Lagrangian function, v is a Lagrangian multiplier, but notAnd performing second power. In this case, min α,d F (alpha, d) is d 2 Given α. Since the optimal solution of the linear programming problem is always at the vertex of the linearly viable area, optimal d *2 Must be a unit vector, which means that the combination of multiple kernel functions will be discarded, leaving only one kernel function. This situation is not consistent with our goal of exploring multiple kernel functions simultaneously to improve tracking performance. Thus, the pair +_in formula (11)>To make the second power of the lawThe equation (12) is a representation of the optimal solution of the objective function, and thus the sample coefficient α and the kernel coefficient d in the case where the objective function is the minimum value can be obtained.
Step 108, determining the multi-core correlation filter based on the sample coefficients and the kernel coefficients.
For example, when the sample coefficient α and the kernel coefficient d in the case where the objective function is the minimum value are obtained, the response calculation of the learned multi-core correlation filter may be expressed by the following equation (13):
wherein w is c Weight coefficient representing color feature, w h The weighting coefficients representing the HOG features, for adjusting the color features and the HOG feature contribution in the response calculation,color feature representing target area, +.>HOG features representing the target region, y (z) being the Gaussian regression label corresponding to the target region, i.e. the response value, y j J in (z) is the index of the current frame image, y j (z) is a response value corresponding to the j-th frame image.
In this embodiment, the virtual sample includes a color feature and a HOG feature, and the combination of the color feature and the HOG feature can characterize the target object, so that accuracy of a sample coefficient and a kernel coefficient obtained by final training is improved, that is, accuracy and robustness of target tracking by the multi-core correlation filter determined based on the sample coefficient and the kernel coefficient are improved.
In an embodiment, the extracting the features from the current frame image in the step 101 to obtain the test sample may be implemented specifically by the following ways:
Extracting color features from the current frame image based on a formula (1), extracting HOG features from the current frame image based on a formula (2), and determining the color features and the HOG features as the test sample;
wherein,representing a color feature extraction kernel for characterizing the color features of the C-th color channel, C representing the number of color channels, and>test sample z representing the j-th frame j Color value on the c-th color channel, and->Representing the color value of the ith virtual sample on the c-th color channel, +.>Representing a HOG feature extraction kernel for characterizing test samples z of the jth frame j And HOG features of the ith virtual sample.
Note that the color feature extraction method may include a color histogram or a color moment. For the color histogram, the histogram similarity may be used as a kernel function, and the above formula (1) is a histogram intersection kernel function, or other kernel functions may be used, which is not limited in the present invention.
Note that, the HOG feature extraction is used for object detection and identification in the image, the above formula (2) is a linear kernel function, and other kernel functions may also be used, which is not limited in the present invention.
In an embodiment, in the step 102, the correlation operation is performed on the test sample based on the multi-core correlation filter to obtain a response chart, which may be specifically implemented by the following ways:
For each virtual sample, performing correlation operation on the test sample based on a formula (3) to obtain a response value corresponding to the correlation between the virtual sample and the test sample;
determining the response map based on equation (4);
wherein y is j (z) test sample z representing the ith virtual sample and the jth frame j The response value corresponding to the correlation between M represents the number of kernel functions, d m Represents the kernel coefficient corresponding to the mth kernel function, l represents the size of the cyclic shift matrix P, and alpha i For the sample coefficient corresponding to the i-th virtual sample,test sample z representing the ith virtual sample and the jth frame j Correlation between; y (z) represents a response map, +.>Is to->As a cyclic matrix of the first row, representing the mth kernel function k m After the test sample z and the cyclic shift matrix P are transformedBit sample->Results of evaluation between, α is α i Is represented as a vector, +.>Is->A set, denoted as a vector. The located samples are the result of the filter weighting operation of the previous frame image. For example, correlation operation is performed between the test sample and each virtual sample by using the sample coefficient α and the kernel coefficient d in the learned multi-core correlation filter. The virtual sample set is test sample z j Generated by cyclic shift matrix P, i.e. z j ＝P j z. For each virtual sample->Correlation operations may be performed using equation (14) above.
In an embodiment, in the step 102, the determining the position corresponding to the maximum response value in the response map as the target position of the target object in the current frame image may be specifically implemented by the following manner:
performing acceleration calculation on the response map based on a formula (5) to obtain an accelerated response map, and determining a position corresponding to a maximum response value in the accelerated response map as a target position of the target object in the current frame image;
wherein y is 1 (z) represents the response after acceleration, and by element-wise multiplication,representing a one-dimensional Fourier transform, ">Representing the corresponding conjugate transformation->Representing inverse Fourier transform, y 1 The element with the maximum response value in (z) is accepted as the current frame image +.>The target position of the target object is calculated using a fast fourier transform acceleration.
In this embodiment, the cyclic matrix representation core is used to convert the calculation of the response graph into matrix multiplication operation, so that the purpose of quickly calculating the response graph can be achieved, and the efficiency of target tracking is further improved.
In an embodiment, in step 103, the multi-core correlation filter is updated based on the target position, so as to obtain an updated multi-core correlation filter, which may be specifically implemented by the following ways:
performing scale space search on the current frame image based on the target position to obtain an optimal target scale; and updating the multi-core correlation filter based on the characteristics corresponding to the optimal target scale and the optimal target scale in the current frame image to obtain the updated multi-core correlation filter.
When determining the target position of the target object in the current frame image, the scale space search is performed on the current frame image based on the target position, the optimal target size is efficiently found by using an optimization method, then the color features and the HOG features corresponding to the optimal target scale in the current frame image are extracted, and the multi-core correlation filter is updated based on the color features and the HOG features corresponding to the optimal target scale in the optimal target scale and the current frame image, so that the updated multi-core correlation filter is obtained.
In this embodiment, a scale space search is performed on a current frame image based on a target position to obtain an optimal target scale, and the multi-core correlation filter is updated based on the optimal target scale and color features and HOG features corresponding to the optimal target scale in the current frame image, so that the accuracy of the updated multi-core correlation filter is higher.
In an embodiment, fig. 3 is a third flow chart of the target tracking method based on the multi-core correlation filter according to the embodiment of the present invention, as shown in fig. 3, and the foregoing scale space search is performed on the current frame image based on the target position to obtain an optimal target scale, which may be specifically implemented by the following steps:
step 301, extracting a patch area with a preset scale from the current frame image.
And 302, resampling the patch area based on the scale corresponding to the previous frame of image to obtain a reference sample.
Step 303, determining a target response graph based on the reference sample and the accelerated response graph.
Step 304, calculating peak value and side lobe ratio of the target response diagram, and determining an evaluation function based on the peak value and the side lobe ratio.
And 305, optimizing the evaluation function based on a golden section method to obtain the optimal target scale.
Illustratively, an evaluation function ρ(s) is defined to evaluate the quality of the target scale, and for ρ(s), a patch region of a predetermined scale needs to be extracted from the current frame imageWherein (1)>Is the central position, s is the preset scale; the corresponding scale of the last frame of image is used +. >For patch area->Resampling is carried out to be used as a reference sample; generating a response graph y (·) using the baseline sample and equation (5) above) And calculating the peak-to-side lobe ratio (PSR) of the response graph y (·) to obtain an evaluation function rho(s). It can be derived that ρ(s) typically has a dominant modality, representing the optimal target scale. To find the optimal target scale s * The golden section method (0.618 method) is used to optimize ρ(s), which represents a function of the target scale estimation, and peak-to-side lobe ratio (PSR) is used to estimate the response of the target.
The iterative process of the golden section method comprises the following steps:
a) The search interval [0.9,1.1] is initialized as a range of values for the target scale s, where 0.9 and 1.1 are scaling factors relative to the length and width of the target bounding box in the previous frame of image.
b) In each iteration, the search interval is narrowed to 0.618 times the previous round.
c) When the length of the search interval is less than 1 pixel, the iteration is stopped.
d) Returning to the final optimal target scale s * 。
The specific formula for target scale evaluation can be expressed by the following formula (14):
wherein,representing +.>The preset scale extracted from the Chinese medicinal materials is s, and the central position is +.>Is (are) patch area->Representing resampling of the patch area to the corresponding scale of the previous frame image +. >Y (·) is a response plot generated using the baseline sample and equation (5) above.
The patch area extraction process comprises the following steps:
given a current frame imageCenter position +.>And a preset scale s, defining the patch area size according to the preset scale s to +.>For the center, from the current frame image +.>And extracting the patch area.
Let the patch area be of size w×h, the extraction of the patch area can be expressed by the following formula (15):
wherein, x indicating the central positionIs of the abscissa, iota y Representing the center position +.>Is defined by the vertical coordinate of (c). Thus, a patch area can be obtained>For the evaluation and tracking process of the target dimensions.
In this embodiment, in the scale estimation stage, the golden section method is adopted to determine the optimal target scale of the target object, and compared with the traditional exhaustion method, the method has lower calculation complexity and higher efficiency.
In an embodiment, the updating the multi-core correlation filter based on the features corresponding to the optimal target scale and the optimal target scale in the current frame image to obtain the updated multi-core correlation filter may be specifically implemented by the following manner:
updating a positioned sample corresponding to a previous frame of image based on a formula (6) to obtain an updated positioned sample, and updating the multi-core correlation filter based on the updated positioned sample to obtain the updated multi-core correlation filter;
Wherein the positioned sample corresponding to the previous frame image is the test sample corresponding to the previous frame image,representing the color value of the updated located sample in the m-th color channel, which is the weighted sum of the updated located sample and the color value of the m-th color channel of the target area of the current frame image, ">Representing the color value of the located sample in the mth color channel, the test sample needs to be scaled quickly to the optimal target scale s by the power law of dolla r * ，η m Indicates learning rate (I/O)>Representing patch areas corresponding to optimal target scales s * Representing the optimal target scale>Representing the scale corresponding to the target object in the previous frame of image,/->Representing the patch area extracted in the mth color channel of the current frame image,/for the current frame image>Represents the center position of the current frame image, +.>Representing the current frame image, λ representing the regularization parameters,
it should be noted that the power law algorithm of dolla r is a method for scaling image features, and the core idea is to construct an image pyramid and scale features. Firstly, the power law algorithm of Doll ar constructs an image pyramid, a series of images with different scales are obtained by scaling an original image for a plurality of times, the images can be obtained by different scaling factors, the scaling factor used in the invention is 0.618, for each scale of image, the power law algorithm of Doll ar extracts color features and hog features, and then the color features and hog features have consistent sizes and proportions at different scales by scaling the features.
In this embodiment, the power law of Doll ar is used to accelerate feature scaling, so that the efficiency and accuracy of the algorithm are further improved, and the high computational cost of readjusting the image block and extracting the features thereof in each iteration is avoided.
The target tracking device based on the multi-core correlation filter provided by the invention is described below, and the target tracking device based on the multi-core correlation filter described below and the target tracking method based on the multi-core correlation filter described above can be referred to correspondingly.
Fig. 4 is a schematic structural diagram of a target tracking device based on a multi-core correlation filter according to an embodiment of the present invention, and as shown in fig. 4, the target tracking device 400 based on a multi-core correlation filter includes an extracting unit 401, an operation unit 402, and an updating unit 403; wherein:
an extracting unit 401, configured to traverse the image sequence, and extract features from the currently traversed current frame image, so as to obtain a test sample; the current frame image comprises a target object;
an operation unit 402, configured to perform a correlation operation on the test sample based on a multi-core correlation filter, obtain a response chart, and determine a position corresponding to a maximum response value in the response chart as a target position of the target object in the current frame image;
An updating unit 403, configured to update the multi-core related filter based on the target location, to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
According to the target tracking device based on the multi-core correlation filter, an image sequence is traversed, features are extracted from a current frame image which is traversed currently to obtain a test sample, correlation operation is carried out on the test sample based on the multi-core correlation filter to obtain a response chart, the position corresponding to the maximum response value in the response chart is determined to be the target position of a target object in the current frame image, the multi-core correlation filter is updated based on the target position, correlation operation is carried out on the test sample corresponding to a next frame image based on the updated multi-core correlation filter to determine the target position of the target object in the next frame image until image traversal in the image sequence is completed. As can be seen, when the correlation operation is performed on the test sample corresponding to each frame of image, the multi-core correlation filter is updated based on the target position of the target object in the previous frame of image, so that the multi-core correlation filter after each update is matched with the real-time characteristic of the target object, and the accuracy of target tracking can be improved.
Based on any of the above embodiments, the target tracking device 400 based on a multi-core correlation filter further includes:
an acquisition unit configured to acquire a first frame image in the image sequence; the first frame image comprises a target area; the target area comprises the target object;
the mobile unit is used for circularly shifting the first frame image to obtain at least two shifted images;
a first determining unit configured to determine a feature of the target region in the shift image as a virtual sample; the features include color features and HOG features;
the optimizing unit is used for optimizing the objective function based on the virtual sample to obtain a sample coefficient and a kernel coefficient corresponding to the condition that the objective function is the minimum value;
and the second determining unit is used for determining the multi-core correlation filter based on the sample coefficient and the kernel coefficient.
Based on any of the above embodiments, the extracting unit 401 is specifically configured to:
extracting color features from the current frame image based on a formula (1), extracting HOG features from the current frame image based on a formula (2), and determining the color features and the HOG features as the test sample;
Wherein,representing a color feature extraction kernel for characterizing the color features of the C-th color channel, C representing the number of color channels, and>test sample z representing the j-th frame j Color value on the c-th color channel, and->Representing the color value of the ith virtual sample on the c-th color channel, +.>Representing a HOG feature extraction kernel for characterizing test samples z of the jth frame j And HOG features of the ith virtual sample.
Based on any of the above embodiments, the operation unit 402 is specifically configured to:
for each virtual sample, performing correlation operation on the test sample based on a formula (3) to obtain a response value corresponding to the correlation between the virtual sample and the test sample;
determining the response map based on equation (4);
wherein y is j (z) test sample z representing the ith virtual sample and the jth frame j The response value corresponding to the correlation between M represents the number of kernel functions, d m Represents the kernel coefficient corresponding to the mth kernel function, l represents the size of the cyclic shift matrix P, and alpha i For the sample coefficient corresponding to the i-th virtual sample,test sample z representing the ith virtual sample and the jth frame j Correlation between; y (z) represents a response map, +.>Is to->As a cyclic matrix of the first row, Representing the mth kernel function k m In test sample z and located sample transformed by cyclic shift matrix P>Results of the evaluation therebetween.
Based on any of the above embodiments, the operation unit 402 is further specifically configured to:
performing acceleration calculation on the response map based on a formula (5) to obtain an accelerated response map, and determining a position corresponding to a maximum response value in the accelerated response map as a target position of the target object in the current frame image;
wherein y is 1 (z) represents the response after acceleration, and by element-wise multiplication,representing a one-dimensional Fourier transform, ">Representing the corresponding conjugate transformation->Representing inverse Fourier transform, y 1 The element with the maximum response value in (z) is accepted as the current frame image +.>Target position of the target object.
Based on any of the above embodiments, the updating unit 403 is specifically configured to:
performing scale space search on the current frame image based on the target position to obtain an optimal target scale;
and updating the multi-core correlation filter based on the characteristics corresponding to the optimal target scale and the optimal target scale in the current frame image to obtain the updated multi-core correlation filter.
Based on any of the above embodiments, the updating unit 403 is further specifically configured to:
Extracting a patch area with a preset scale from the current frame image;
resampling the patch area based on the scale corresponding to the previous frame of image to obtain a reference sample;
determining a target response map based on the reference sample and the accelerated response map;
calculating peak value and side lobe ratio of the target response graph, and determining an evaluation function based on the peak value and the side lobe ratio;
and optimizing the evaluation function based on a golden section method to obtain the optimal target scale.
Based on any of the above embodiments, the updating unit 403 is further specifically configured to:
updating a positioned sample corresponding to a previous frame of image based on a formula (6) to obtain an updated positioned sample, and updating the multi-core correlation filter based on the updated positioned sample to obtain the updated multi-core correlation filter;
wherein,representing the color value of the updated located sample in the mth color channel, +.>Representing the color value, η, of the located sample at the mth color channel m Representing learning rate，/>Representing patch areas corresponding to optimal target scales s * Representing the optimal target scale>Representing the scale corresponding to the target object in the previous frame of image,/- >Representing the patch area extracted in the mth color channel of the current frame image,/for the current frame image>Represents the center position of the current frame image, +.>Representing the current frame image, λ represents the regularization parameters.
Fig. 5 is a schematic physical structure of an electronic device according to an embodiment of the present invention, as shown in fig. 5, the electronic device may include: processor 510, communication interface (Communications Interface) 520, memory 530, and communication bus 540, wherein processor 510, communication interface 520, memory 530 complete communication with each other through communication bus 540. Processor 510 may invoke logic instructions in memory 530 to perform a multi-core correlation filter based target tracking method comprising: traversing the image sequence, and extracting features from the current frame image traversed currently to obtain a test sample; the current frame image comprises a target object;
performing correlation operation on the test sample based on a multi-core correlation filter to obtain a response graph, and determining a position corresponding to a maximum response value in the response graph as a target position of the target object in the current frame image;
updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
Further, the logic instructions in the memory 530 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product, where the computer program product includes a computer program, where the computer program can be stored on a non-transitory computer readable storage medium, and when the computer program is executed by a processor, the computer can execute the target tracking method based on the multi-core correlation filter provided by the above methods, and the method includes: traversing the image sequence, and extracting features from the current frame image traversed currently to obtain a test sample; the current frame image comprises a target object;
Performing correlation operation on the test sample based on a multi-core correlation filter to obtain a response graph, and determining a position corresponding to a maximum response value in the response graph as a target position of the target object in the current frame image;
updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the method for target tracking based on a multi-core correlation filter provided by the above methods, the method comprising: traversing the image sequence, and extracting features from the current frame image traversed currently to obtain a test sample; the current frame image comprises a target object;
performing correlation operation on the test sample based on a multi-core correlation filter to obtain a response graph, and determining a position corresponding to a maximum response value in the response graph as a target position of the target object in the current frame image;
Updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. The target tracking method based on the multi-core correlation filter is characterized by comprising the following steps of:
traversing the image sequence, and extracting features from the current frame image traversed currently to obtain a test sample; the current frame image comprises a target object;
performing correlation operation on the test sample based on a multi-core correlation filter to obtain a response graph, and determining a position corresponding to a maximum response value in the response graph as a target position of the target object in the current frame image;
updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
2. The multi-kernel correlation filter based object tracking method of claim 1, wherein prior to traversing the sequence of images, the method further comprises:
acquiring a first frame image in the image sequence; the first frame image comprises a target area; the target area comprises the target object;
performing cyclic shift on the first frame image to obtain at least two shift images;
determining the characteristics of the target area in the shift image as a virtual sample; the features include color features and HOG features;
optimizing an objective function based on the virtual sample to obtain a sample coefficient and a kernel coefficient corresponding to the condition that the objective function is the minimum value;
the multi-core correlation filter is determined based on the sample coefficients and the kernel coefficients.
3. The method for target tracking based on the multi-core correlation filter according to claim 1, wherein the extracting features from the current frame image of the current traversal to obtain the test sample comprises:
extracting color features from the current frame image based on a formula (1), extracting HOG features from the current frame image based on a formula (2), and determining the color features and the HOG features as the test sample;
Wherein,representing a color feature extraction kernel for characterizing the color features of the C-th color channel, C representing the number of color channels, and>test sample z representing the j-th frame j Color value on the c-th color channel, and->Representing the color value of the ith virtual sample on the c-th color channel, +.>Representing HOG feature extraction kernel function for characterizing test sample z j And HOG features of the ith virtual sample.
4. The target tracking method based on the multi-core correlation filter according to claim 3, wherein the performing correlation operation on the test sample based on the multi-core correlation filter to obtain a response chart includes:
for each virtual sample, performing correlation operation on the test sample based on a formula (3) to obtain a response value corresponding to the correlation between the virtual sample and the test sample;
determining the response map based on equation (4);
wherein y is j (z) test sample z representing the ith virtual sample and the jth frame j The response value corresponding to the correlation between M represents the number of kernel functions, d m Represents the kernel coefficient corresponding to the mth kernel function, l represents the size of the cyclic shift matrix P, and alpha i For the sample coefficient corresponding to the i-th virtual sample, Test sample z representing the ith virtual sample and the jth frame j Correlation between; y (z) represents a response map, +.>Is to->As a cyclic matrix of the first row,representing the mth kernel function k m In test sample z and located sample transformed by cyclic shift matrix P>Results of the evaluation therebetween.
5. The method for target tracking based on a multi-core correlation filter according to claim 4, wherein determining the position corresponding to the maximum response value in the response map as the target position of the target object in the current frame image comprises:
performing acceleration calculation on the response map based on a formula (5) to obtain an accelerated response map, and determining a position corresponding to a maximum response value in the accelerated response map as a target position of the target object in the current frame image;
wherein y is 1 (z) represents the response after acceleration, and by element-wise multiplication,representing a one-dimensional fourier transform,representing the corresponding conjugate transformation->Representing inverse Fourier transform, y 1 The element taking the maximum response value in (z) is accepted as the target position of the target object in the current frame image J.
6. The method for tracking a target based on a multi-core correlation filter according to claim 5, wherein updating the multi-core correlation filter based on the target position to obtain an updated multi-core correlation filter comprises:
Performing scale space search on the current frame image based on the target position to obtain an optimal target scale;
and updating the multi-core correlation filter based on the characteristics corresponding to the optimal target scale and the optimal target scale in the current frame image to obtain the updated multi-core correlation filter.
7. The method for target tracking based on a multi-core correlation filter according to claim 6, wherein the performing a scale space search on the current frame image based on the target position to obtain an optimal target scale comprises:
extracting a patch area with a preset scale from the current frame image;
resampling the patch area based on the scale corresponding to the previous frame of image to obtain a reference sample;
determining a target response map based on the reference sample and the accelerated response map;
calculating peak value and side lobe ratio of the target response graph, and determining an evaluation function based on the peak value and the side lobe ratio;
and optimizing the evaluation function based on a golden section method to obtain the optimal target scale.
8. The method for tracking a target based on a multi-core correlation filter according to claim 7, wherein updating the multi-core correlation filter based on the characteristics corresponding to the optimal target scale and the optimal target scale in the current frame image to obtain the updated multi-core correlation filter comprises:
Updating a positioned sample corresponding to a previous frame of image based on a formula (6) to obtain an updated positioned sample, and updating the multi-core correlation filter based on the updated positioned sample to obtain the updated multi-core correlation filter;
wherein,representing the color value of the updated located sample in the mth color channel, +.>Representing the color value, η, of the located sample at the mth color channel m Indicates learning rate (I/O)>Representing patch areas corresponding to optimal target scales s * Representing the optimal target scale>Representing the scale corresponding to the target object in the previous frame of image,/->Representing the patch area extracted in the mth color channel of the current frame image,/for the current frame image>Represents the center position of the current frame image, +.>Representing the currentFrame image, λ represents regularization parameters.
9. A target tracking device based on a multi-core correlation filter, comprising:
the extraction unit is used for traversing the image sequence, extracting features from the current frame image traversed currently, and obtaining a test sample; the current frame image comprises a target object;
the operation unit is used for carrying out correlation operation on the test sample based on a multi-core correlation filter to obtain a response graph, and determining the position corresponding to the maximum response value in the response graph as the target position of the target object in the current frame image;
The updating unit is used for updating the multi-core related filter based on the target position to obtain an updated multi-core related filter; the updated multi-core correlation filter is used for performing correlation operation on a test sample corresponding to a next frame of image so as to determine the target position of the target object in the next frame of image.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the multi-core correlation filter-based object tracking method of any one of claims 1 to 8 when the program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310946698.5A CN117152203A (en) | 2023-07-28 | 2023-07-28 | Target tracking method and device based on multi-core correlation filter and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310946698.5A CN117152203A (en) | 2023-07-28 | 2023-07-28 | Target tracking method and device based on multi-core correlation filter and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117152203A true CN117152203A (en) | 2023-12-01 |
Family
ID=88883223
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310946698.5A Pending CN117152203A (en) | 2023-07-28 | 2023-07-28 | Target tracking method and device based on multi-core correlation filter and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117152203A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117768351A (en) * | 2024-02-22 | 2024-03-26 | 鹏城实验室 | interference evaluation method and related equipment of Internet of vehicles system |
-
2023
- 2023-07-28 CN CN202310946698.5A patent/CN117152203A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117768351A (en) * | 2024-02-22 | 2024-03-26 | 鹏城实验室 | interference evaluation method and related equipment of Internet of vehicles system |
| CN117768351B (en) * | 2024-02-22 | 2024-05-07 | 鹏城实验室 | Interference evaluation method and related equipment of Internet of vehicles system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112800876B (en) | Super-spherical feature embedding method and system for re-identification | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| CN109919241B (en) | Hyperspectral unknown class target detection method based on probability model and deep learning | |
| CN112215119B (en) | Small target identification method, device and medium based on super-resolution reconstruction | |
| CN111650453A (en) | Power equipment diagnosis method and system based on windowing characteristic Hilbert imaging | |
| CN109284668B (en) | Pedestrian re-identification method based on distance regularization projection and dictionary learning | |
| CN111091147B (en) | Image classification method, device and equipment | |
| CN113111716B (en) | Remote sensing image semiautomatic labeling method and device based on deep learning | |
| CN114627447A (en) | Road vehicle tracking method and system based on attention mechanism and multi-target tracking | |
| CN117152203A (en) | Target tracking method and device based on multi-core correlation filter and electronic equipment | |
| Yang et al. | Visual tracking with long-short term based correlation filter | |
| CN114749342A (en) | Method, device and medium for identifying coating defects of lithium battery pole piece | |
| CN111640138A (en) | Target tracking method, device, equipment and storage medium | |
| CN112329784A (en) | Correlation filtering tracking method based on space-time perception and multimodal response | |
| Li et al. | On uncertainty estimation in active learning for image segmentation | |
| CN115527050A (en) | Image feature matching method, computer device and readable storage medium | |
| CN113392937A (en) | 3D point cloud data classification method and related device thereof | |
| CN111950362A (en) | Golden monkey face image identification method, device, equipment and storage medium | |
| Feng et al. | A novel saliency detection method for wild animal monitoring images with WMSN | |
| CN113076876B (en) | Face spoofing detection method and system based on three-dimensional structure supervision and confidence weighting | |
| CN109241981B (en) | Feature detection method based on sparse coding | |
| CN113673465A (en) | Image detection method, device, equipment and readable storage medium | |
| CN110765898B (en) | Method and device for determining object and key point thereof in image | |
| CN116758419A (en) | Multi-scale target detection method, device and equipment for remote sensing image | |
| CN107808164B (en) | Texture image feature selection method based on firework algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |