CN112396065A - Scale-adaptive target tracking method and system based on correlation filtering - Google Patents
Scale-adaptive target tracking method and system based on correlation filtering Download PDFInfo
- Publication number
- CN112396065A CN112396065A CN202011120598.XA CN202011120598A CN112396065A CN 112396065 A CN112396065 A CN 112396065A CN 202011120598 A CN202011120598 A CN 202011120598A CN 112396065 A CN112396065 A CN 112396065A
- Authority
- CN
- China
- Prior art keywords
- calculation
- image
- gradient
- target tracking
- fhog
- 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 32
- 238000001914 filtration Methods 0.000 title claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims abstract description 30
- 230000004044 response Effects 0.000 claims abstract description 18
- 230000008569 process Effects 0.000 claims abstract description 9
- 230000009191 jumping Effects 0.000 claims abstract description 4
- 238000004364 calculation method Methods 0.000 claims description 41
- 238000010606 normalization Methods 0.000 claims description 16
- 230000017105 transposition Effects 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 10
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 4
- 230000001133 acceleration Effects 0.000 claims 1
- 238000012545 processing Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The scale self-adaptive target tracking method and system based on the correlation filtering comprises the following steps: (1) inputting a sequence of images; (2) judging whether the current frame is the first frame of the sequence; (3) initializing a target tracking rectangular frame, and jumping to the step (10); (4) extracting image blocks with changed scales; (5) extracting FHOG characteristics; (6) calculating a cross-correlation matrix and a maximum response value; (7) judging whether to process image blocks under all scales; (8) calculating the maximum response value under all scales; (9) updating the target rectangular frame; (10) extracting images in the search rectangular frame; (11) extracting FHOG characteristics; (12) calculating an autocorrelation matrix; (13) updating parameters of the tracking model; (14) judging whether all the images in the image sequence are loaded or not; (15) and (6) ending.
Description
Technical Field
The invention relates to the technical field of medical image processing, in particular to a scale self-adaptive target tracking method based on relevant filtering and a scale self-adaptive target tracking system based on relevant filtering, which can adapt to a target tracking scene with a changed scale.
Background
With the recent rise of computer vision technology, dynamic object tracking has become a popular topic. Target tracking technology plays an important role in many fields, such as autopilot, scene monitoring, traffic control, unmanned aerial vehicle reconnaissance, and so on.
Before 2010, the field of target tracking mainly uses some classical methods, such as Meanshift, Kalman filtering, and the like. In 2010, the relevant filter technology was introduced into the field for the first time. In recent years, more excellent tracking algorithms have emerged. Henriques et al in KCF (Kernelized Correlation filters) algorithm, utilize the property that the circulant matrix can be diagonalized in Fourier space, transform the operation of the matrix into the dot product of the vector, have greatly promoted the accuracy and speed that the algorithm runs.
However, the conventional KCF algorithm cannot adapt to the size change of the target during operation, which often results in the target being lost and affects the tracking effect. Besides, the KCF algorithm involves a large number of independent pixel-level calculations, and the traditional serial CPU-based programming method limits the execution efficiency of the algorithm.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a scale self-adaptive target tracking method based on relevant filtering, which can adapt to a target tracking scene with a changed scale.
The technical scheme of the invention is as follows: the scale-adaptive target tracking method based on the correlation filtering comprises the following steps:
(1) inputting a sequence of images;
(2) judging whether the current frame is the first frame of the sequence: if yes, executing the step (3); if not, executing step (4);
(3) initializing a target tracking rectangular frame, and jumping to the step (10);
(4) extracting image blocks with changed scales;
(5) extracting FHOG characteristics;
(6) calculating a cross-correlation matrix and a maximum response value;
(7) judging whether to process image blocks under all scales, if so, executing the step (8), and otherwise, executing the step (4);
(8) calculating the maximum response value under all scales;
(9) updating the target rectangular frame;
(10) extracting images in the search rectangular frame;
(11) extracting FHOG characteristics;
(12) calculating an autocorrelation matrix;
(13) updating parameters of the tracking model;
(14) judging whether all the images in the image sequence are loaded: if yes, executing step (15); if not, executing the step (1);
(15) and (6) ending.
The method extracts the image blocks with changed scales, performs FHOG feature extraction, calculates a cross-correlation matrix and a maximum response value, calculates the maximum response value under all scales if processing the image blocks under all scales, performs FHOG feature extraction, calculates an autocorrelation matrix, and updates parameters of a tracking model, so that the improved method can adapt to a target tracking scene with changed scales.
Also provided is a scale-adaptive target tracking system based on correlation filtering, comprising:
an input module configured to input a sequence of images;
an initialization module configured to initialize a target tracking rectangular box;
the extraction module is configured to extract the image block after the scale change;
the single image FHOG characteristic module is configured to extract FHOG characteristics, calculate a self-correlation matrix and update parameters of a tracking model;
and the all-image FHOG characteristic module is configured to extract FHOG characteristics of the image blocks under all scales, calculate cross-correlation matrixes and maximum response values of the FHOG characteristics, calculate the maximum response values under all scales through comparison, and update the target rectangular frame.
Drawings
Fig. 1 is a flowchart of a scale-adaptive target tracking method based on correlation filtering according to the present invention.
Figure 2 is a flow chart of FHOG feature extraction according to the invention.
FIG. 3 is a diagram of thread block, thread and image correspondence for FHOG feature extraction and autocorrelation matrix computation according to the present invention. The image size is W × H, the original image is divided into Wt × Ht blocks, each Block is divided into 16 × 16 threads, and each thread corresponds to one pixel point in the image.
Detailed Description
As shown in fig. 1, the scale-adaptive target tracking method based on correlation filtering includes the following steps:
(1) inputting a sequence of images;
(2) judging whether the current frame is the first frame of the sequence: if yes, executing the step (3); if not, executing step (4);
(3) initializing a target tracking rectangular frame, and jumping to the step (10);
(4) extracting image blocks with changed scales;
(5) extracting FHOG characteristics;
(6) calculating a cross-correlation matrix and a maximum response value;
(7) judging whether to process image blocks under all scales, if so, executing the step (8), and otherwise, executing the step (4);
(8) calculating the maximum response value under all scales;
(9) updating the target rectangular frame;
(10) extracting images in the search rectangular frame;
(11) extracting FHOG characteristics;
(12) calculating an autocorrelation matrix;
(13) updating parameters of the tracking model;
(14) judging whether all the images in the image sequence are loaded: if yes, executing step (15); if not, executing the step (1);
(15) and (6) ending.
The method extracts the image blocks with changed scales, performs FHOG feature extraction, calculates the cross-correlation matrix and the maximum response value, calculates the maximum response value under all scales if processing the image blocks under all scales, performs FHOG feature extraction, calculates the autocorrelation matrix, and updates the parameters of the tracking model, so that the improved method can adapt to the target tracking scene with changed scales.
Preferably, as shown in fig. 2, in the steps (5) and (11), the extracting of FHOG features includes: image transposition, gray level normalization, gradient value calculation, discretization of gradient values, gradient histogram calculation, gradient energy calculation and normalization, and finally 31-dimensional features are obtained.
Preferably, in the steps (5) and (11), image transposition, gray level normalization, gradient calculation, discretization gradient, gradient histogram calculation, gradient energy calculation, normalization and feature transposition are all processes of pixel-by-pixel calculation, and are all accelerated by using CUDA. Each thread processes one pixel, and the correspondence of thread blocks, threads and images is shown in fig. 3.
Preferably, in the step (12), the autocorrelation matrix is calculated, the kernel function is a gaussian kernel, and the calculation formula of the gaussian kernel is formula (1)
Where σ is the standard deviation of the Gaussian kernel, is a constant,
is a matrix dot product operation, F-1Representing a discrete inverse fourier transform.
Preferably, in the calculation of the gaussian kernel, the norm of L2 and the dot product of the matrix are calculated pixel by pixel, and in the design of CUDA, each thread corresponds to the calculation of one pixel. If the size of the processed image is W × H, the correspondence between the thread block, the thread, and the image is obtained.
It will be understood by those skilled in the art that all or part of the steps in the method of the above embodiments may be implemented by a program to instruct associated hardware, the program may be stored in a computer readable storage medium, and when executed, the program includes the steps of the method of the above embodiments, and the storage medium may be: ROM/RAM, magnetic disks, optical disks, memory cards, and the like. Therefore, corresponding to the method of the present invention, the present invention also includes a scale-adaptive target tracking system based on correlation filtering, which is generally expressed in the form of functional blocks corresponding to the steps of the method. The system comprises:
an input module configured to input a sequence of images;
an initialization module configured to initialize a target tracking rectangular box;
the extraction module is configured to extract the image block after the scale change;
the single image FHOG characteristic module is configured to extract FHOG characteristics, calculate a self-correlation matrix and update parameters of a tracking model;
and the all-image FHOG characteristic module is configured to extract FHOG characteristics of the image blocks under all scales, calculate cross-correlation matrixes and maximum response values of the FHOG characteristics, calculate the maximum response values under all scales through comparison, and update the target rectangular frame.
Preferably, in the single-image FHOG feature module and the full-image FHOG feature module, the extracting of FHOG features includes: image transposition, gray level normalization, gradient value calculation, discretization of gradient values, gradient histogram calculation, gradient energy calculation and normalization, and finally 31-dimensional features are obtained.
Preferably, in the single image FHOG feature module and the all-image FHOG feature module, image transposition, gray scale normalization, gradient calculation, discretization, gradient histogram calculation, gradient energy calculation, normalization, and feature transposition are processes of pixel-by-pixel calculation, and are accelerated by using CUDA.
Preferably, in the single-image FHOG feature module, an autocorrelation matrix is calculated, the kernel function is a Gaussian kernel, and the calculation formula of the Gaussian kernel is formula (1)
Where σ is the standard deviation of the Gaussian kernel, is a constant,
is a matrix dot product operation, F-1Representing a discrete inverse fourier transform.
Preferably, in the calculation of the gaussian kernel, the norm of L2 and the dot product of the matrix are calculated pixel by pixel, and in the design of CUDA, each thread corresponds to the calculation of one pixel.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, and all simple modifications, equivalent variations and modifications made to the above embodiment according to the technical spirit of the present invention still fall within the protection scope of the technical solution of the present invention.
Claims (10)
1. The scale self-adaptive target tracking method based on the correlation filtering is characterized by comprising the following steps: which comprises the following steps:
(1) inputting a sequence of images;
(2) judging whether the current frame is the first frame of the sequence: if yes, executing the step (3); if not, executing step (4);
(3) initializing a target tracking rectangular frame, and jumping to the step (10);
(4) extracting image blocks with changed scales;
(5) extracting FHOG characteristics;
(6) calculating a cross-correlation matrix and a maximum response value;
(7) judging whether to process image blocks under all scales, if so, executing the step (8), and otherwise, executing the step (4);
(8) calculating the maximum response value under all scales;
(9) updating the target rectangular frame;
(10) extracting images in the search rectangular frame;
(11) extracting FHOG characteristics;
(12) calculating an autocorrelation matrix;
(13) updating parameters of the tracking model;
(14) judging whether all the images in the image sequence are loaded: if yes, executing step (15); if not, executing the step (1);
(15) and (6) ending.
2. The scale-adaptive target tracking method based on correlation filtering according to claim 1, wherein: in the steps (5) and (11), the extracting of the FHOG characteristic comprises: image transposition, gray level normalization, gradient value calculation, discretization of gradient values, gradient histogram calculation, gradient energy calculation and normalization, and finally 31-dimensional features are obtained.
3. The scale-adaptive target tracking method based on correlation filtering according to claim 2, wherein: in the steps (5) and (11), image transposition, gray level normalization, gradient calculation, discretization gradient, gradient histogram calculation, gradient energy calculation, normalization and feature transposition are all processes of pixel-by-pixel calculation, and are accelerated by using CUDA.
4. The scale-adaptive target tracking method based on correlation filtering according to claim 3, wherein: in the step (12), an autocorrelation matrix is calculated, the kernel function is a Gaussian kernel, and the calculation formula of the Gaussian kernel is a formula (1)
Where σ is the standard deviation of the Gaussian kernel, is a constant,
is a matrix dot product operation, F-1Representing a discrete inverse fourier transform.
5. The scale-adaptive target tracking method based on correlation filtering according to claim 4, wherein: in the calculation of the gaussian kernel, the norm of L2 and the dot product of the matrix are calculated pixel by pixel, and each thread corresponds to the calculation of one pixel when the CUDA is designed.
6. The scale self-adaptive target tracking system based on the correlation filtering is characterized in that: it includes:
an input module configured to input a sequence of images;
an initialization module configured to initialize a target tracking rectangular box;
the extraction module is configured to extract the image block after the scale change;
the single image FHOG characteristic module is configured to extract FHOG characteristics, calculate an autocorrelation matrix and update parameters of a tracking model;
and the all-image FHOG characteristic module is configured to extract FHOG characteristics of the image blocks under all scales, calculate cross-correlation matrixes and maximum response values of the FHOG characteristics, calculate the maximum response values under all scales through comparison, and then update the target rectangular frame.
7. The correlation filtering based scale-adaptive target tracking system of claim 6, wherein: in the single-image FHOG characteristic module and the all-image FHOG characteristic module, the extraction of FHOG characteristics comprises the following steps: image transposition, gray level normalization, gradient value calculation, discretization of gradient values, gradient histogram calculation, gradient energy calculation and normalization, and finally 31-dimensional features are obtained.
8. The correlation filtering based scale-adaptive target tracking system of claim 7, wherein: in the single image FHOG characteristic module and the all-image FHOG characteristic module, image transposition, gray level normalization, gradient calculation, discretization gradient, gradient histogram calculation, gradient energy calculation, normalization and characteristic transposition are all processes of pixel-by-pixel calculation, and CUDA is used for acceleration.
9. The correlation filtering based scale-adaptive target tracking system of claim 8, wherein: in the single image FHOG characteristic module, an autocorrelation matrix is calculated, a kernel function is a Gaussian kernel, and a calculation formula of the Gaussian kernel is a formula (1)
Where σ is the standard deviation of the Gaussian kernel, is a constant,
is a matrix dot product operation, F-1Representing a discrete inverse fourier transform.
10. The correlation filtering based scale-adaptive target tracking system of claim 9, wherein: in the calculation of the gaussian kernel, the norm of L2 and the dot product of the matrix are calculated pixel by pixel, and each thread corresponds to the calculation of one pixel when the CUDA is designed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011120598.XA CN112396065A (en) | 2020-10-19 | 2020-10-19 | Scale-adaptive target tracking method and system based on correlation filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011120598.XA CN112396065A (en) | 2020-10-19 | 2020-10-19 | Scale-adaptive target tracking method and system based on correlation filtering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112396065A true CN112396065A (en) | 2021-02-23 |
Family
ID=74596826
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011120598.XA Pending CN112396065A (en) | 2020-10-19 | 2020-10-19 | Scale-adaptive target tracking method and system based on correlation filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112396065A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113658216A (en) * | 2021-06-24 | 2021-11-16 | 北京理工大学 | Remote sensing target tracking method based on multi-stage self-adaptive KCF and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106991689A (en) * | 2017-04-05 | 2017-07-28 | 西安电子科技大学 | Method for tracking target and GPU based on FHOG and color characteristic accelerate |
| CN109584277A (en) * | 2018-12-07 | 2019-04-05 | 上海应用技术大学 | A kind of nuclear phase pass filter tracking method based on binary search |
| CN110895820A (en) * | 2019-03-14 | 2020-03-20 | 河南理工大学 | KCF-based scale self-adaptive target tracking method |
-
2020
- 2020-10-19 CN CN202011120598.XA patent/CN112396065A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106991689A (en) * | 2017-04-05 | 2017-07-28 | 西安电子科技大学 | Method for tracking target and GPU based on FHOG and color characteristic accelerate |
| CN109584277A (en) * | 2018-12-07 | 2019-04-05 | 上海应用技术大学 | A kind of nuclear phase pass filter tracking method based on binary search |
| CN110895820A (en) * | 2019-03-14 | 2020-03-20 | 河南理工大学 | KCF-based scale self-adaptive target tracking method |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113658216A (en) * | 2021-06-24 | 2021-11-16 | 北京理工大学 | Remote sensing target tracking method based on multi-stage self-adaptive KCF and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3971772A1 (en) | Model training method and apparatus, and terminal and storage medium | |
| CN109035319B (en) | Monocular image depth estimation method, monocular image depth estimation device, monocular image depth estimation apparatus, monocular image depth estimation program, and storage medium | |
| CN108427927B (en) | Object re-recognition method and apparatus, electronic device, program, and storage medium | |
| CN109543548A (en) | A kind of face identification method, device and storage medium | |
| CN108875931B (en) | Neural network training and image processing method, device and system | |
| CN109685830B (en) | Target tracking method, device and equipment and computer storage medium | |
| CN110569844B (en) | Ship recognition method and system based on deep learning | |
| CN108021869A (en) | A kind of convolutional neural networks tracking of combination gaussian kernel function | |
| CN108875655A (en) | A kind of real-time target video tracing method and system based on multiple features | |
| CN111815668B (en) | Target tracking method, electronic device and storage medium | |
| WO2017168462A1 (en) | An image processing device, an image processing method, and computer-readable recording medium | |
| EP3121788A1 (en) | Image feature estimation method and device | |
| CN112396065A (en) | Scale-adaptive target tracking method and system based on correlation filtering | |
| CN115620022A (en) | Object detection method, device, equipment and storage medium | |
| CN115424288A (en) | Visual Transformer self-supervision learning method and system based on multi-dimensional relation modeling | |
| CN112801020B (en) | Pedestrian re-identification method and system based on background graying | |
| CN113128427A (en) | Face recognition method and device, computer readable storage medium and terminal equipment | |
| CN111652080A (en) | Target tracking method and device based on RGB-D image | |
| CN109543556B (en) | Motion recognition method, motion recognition apparatus, medium, and device | |
| CN108229320B (en) | Frame selection method and device, electronic device, program and medium | |
| US11989927B2 (en) | Apparatus and method for detecting keypoint based on deep learning using information change across receptive fields | |
| KR101853211B1 (en) | Complexity Reduction of SIFT for Video based on Frame Difference in the Mobile GPU environment | |
| CN112288748B (en) | Semantic segmentation network training and image semantic segmentation method and device | |
| WO2021056531A1 (en) | Face gender recognition method, face gender classifier training method and device | |
| CN117893766B (en) | Object detection segmentation scheme |
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: 20210223 |