CN109978908A - A kind of quick method for tracking and positioning of single goal adapting to large scale deformation - Google Patents
A kind of quick method for tracking and positioning of single goal adapting to large scale deformation Download PDFInfo
- Publication number
- CN109978908A CN109978908A CN201910219613.7A CN201910219613A CN109978908A CN 109978908 A CN109978908 A CN 109978908A CN 201910219613 A CN201910219613 A CN 201910219613A CN 109978908 A CN109978908 A CN 109978908A
- Authority
- CN
- China
- Prior art keywords
- target
- coordinate
- parameter
- template
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/223—Analysis of motion using block-matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Closed-Circuit Television Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to it is a kind of adapt to large scale deformation but the quick method for tracking and positioning of target, the following steps are included: template matching, determine the coordinate and size of target template image, cyclic convolution is done with current frame image and target template image and acquires target response matrix with Ridge Regression Modeling Method, as rough coordinates of targets at response matrix maximum value；Motion detection, before and after frames images match obtain visual angle and shake parameter, and acquire difference diagram；It counts difference diagram and obtains motion detection output coordinate；The rough coordinates of targets is corrected with the motion detection output coordinate, obtains precision target coordinate；Template renewal updates target template with the precision target coordinate, updates ridge regression parameter；The present invention can rapidly and accurately track set goal position, and suitable for the scene that target shape has large scale to change, there is good robustness, can be applied in the scenes such as shooting auto-focusing, monitor video target lock-on.
Description
Technical field
The invention belongs to video signal treatment technique fields, and in particular to it is a kind of adapt to large scale deformation but target it is quick
Method for tracking and positioning.
Background technique
Target following is an important research direction in computer vision.It is in human-computer interaction, machine recognition and artificial
The fields such as intelligence have a wide range of applications.
In target tracking domain, the variation of target shape is always a problem.Existing method is built in first frame
Vertical standard form, does template matching in rear frame with this and acquires target position.But once deformation occurs for target during tracking
(such as shooting angle changes, the overturning etc. of human body), template matching is likely to fail.
And the variation of shape necessarily implies that violent change has occurred in pixel at target place, therefore, is very suitable for moving
Detection is to detect target.
Summary of the invention
After deformation occurs in order to solve target existing in the prior art, the technical issues of template matching fails, the present invention
Provide following technical scheme:
A kind of quick method for tracking and positioning of single goal adapting to large scale deformation, comprising the following steps:
Step 1, template matching determine the coordinate and size of target template image, with current frame image and target template figure
As doing cyclic convolution and acquiring target response matrix with Ridge Regression Modeling Method, as rough coordinates of targets at response matrix maximum value；
Step 2, motion detection,
Step 2.1, before and after frames images match obtain visual angle and shake parameter, and acquire difference diagram；
Step 2.2, statistics difference diagram obtain motion detection output coordinate；
Step 2.3 corrects the rough coordinates of targets with the motion detection output coordinate, obtains precision target coordinate；
Step 3, template renewal update target template with the precision target coordinate, update ridge regression parameter.
As a further explanation of the present invention, in the step 1 current frame image and target template image cyclic convolution
Matrix is,
If target template image is z, current frame image img, in above formula | | img | | it is the norm of img, | | z | | for z's
Norm, FFT and IFFT respectively represent the two-dimensional fast fourier transform and inverse transformation of image,For the conjugation of FFT (z),
σ is the standard deviation of Gauss kernel method；
The target response matrix is,
R=IFFT (FFT (K) FFT (γ))
γ is ridge regression parameter matrix in above formula, and R is target response matrix, and wherein the position at top is rough mesh
Coordinate is marked, [x is expressed ast, yt]。
As a further explanation of the present invention, the prediction technique of shaking parameter is in the step 2.1,
θ=[α, β, ε] is to shake parameter in above formula, and α is the parameter to teetertotter, and β is the parameter of double swerve, and ε is mirror
Head zooming parameter, Ic are current frame image, and Ip is previous frame image, and Z (Ip, ε) is by present frame with the figure after parameter ε scaling
Picture；
The difference diagram of before and after frames image is,
D is difference diagram in above formula.
As a further explanation of the present invention, motion detection output coordinate is in the step 2.2,
If target sizes are [sizex, sizey], x in above formulamdFor the x coordinate position of moving target, ymdFor moving target
Y-coordinate position, D (i, j) is value of the difference diagram at coordinate [i, j]；
D in above formulasumIt is the sum of the weight of entire difference diagram window.
As a further explanation of the present invention, precision target coordinate is in the step 2.3,
ρ is weight coefficient, [x in above formulafinal, yfinal] it is accurate template coordinate.
As a further explanation of the present invention, updated target template is in the step 3,
In above formulaFor updated target template, δ is weight coefficient, and updated ridge regression parameter is,
γ=(K+ λ I)-1S
γ is updated ridge regression parameter in above formula, and I is and the equirotal unit matrix of K, I and regularization coefficient λ
Regularization is played the role of in cooperation, and S is normal response matrix.
Compared with prior art, what the present invention obtained has the beneficial effect that
The present invention is matched by before and after frames eliminates visual angle shaking, and does frame difference to detect moving target, and the two result is added
Weight average can rapidly and accurately track set goal position, and suitable for the scene that target shape has large scale to change, have
Good robustness can be applied in the scenes such as shooting auto-focusing, monitor video target lock-on.
The present invention is described in further details below with reference to accompanying drawings and embodiments.
Detailed description of the invention
Fig. 1 is the system block diagram of this method.
Fig. 2 is target response matrix R.
Fig. 3 is the present frame effect image of frame-to-frame differences method.
Fig. 4 is the previous frame effect image of frame-to-frame differences method.
Fig. 5 is the difference diagram between the present frame and previous frame of frame-to-frame differences method.
When Fig. 6 is target mobile and shaking with camera lens, frame-to-frame differences method present frame effect picture.
When Fig. 7 is target mobile and shaking with camera lens, frame-to-frame differences method previous frame effect picture.
When Fig. 8 is target mobile and shaking with camera lens, the difference diagram between frame-to-frame differences method present frame and previous frame.
Fig. 9 is present frame effect picture one after correcting shaking parameter by this method.
Figure 10 is previous frame effect picture one after correcting shaking parameter by this method.
Figure 11 is between the present frame and previous frame directly acquired with frame-to-frame differences method after being corrected by this method and shaking parameter
Difference diagram one.
Figure 12 is after being corrected by this method and shaking parameter, then by the correction of camera lens jitter parameter, then uses frame-to-frame differences
The difference diagram one between present frame and previous frame that method acquires.
Figure 13 is present frame effect picture two after correcting shaking parameter by this method.
Figure 14 is previous frame effect picture two after correcting shaking parameter by this method.
Figure 15 is between the present frame and previous frame directly acquired with frame-to-frame differences method after being corrected by this method and shaking parameter
Difference diagram two.
Figure 16 is after being corrected by this method and shaking parameter, then by the correction of camera lens jitter parameter, then uses frame-to-frame differences
The difference diagram two between present frame and previous frame that method acquires.
Figure 17 is normal response matrix S.
Figure 18 is tracking target position effect picture one.
Figure 19 is Figure 18 through KCF method treated tracking effect figure.
Figure 20 is Figure 18 through SAMF method treated tracking effect figure.
Figure 21 is that Figure 18 tracks effect picture after the processing of DSST method.
Figure 22 is Figure 18 through Staple method treated tracking effect figure.
Figure 23 is the tracking effect figure after Figure 18 treated by the present method.
Figure 24 is tracking target position effect picture two.
Figure 25 is Figure 24 through KCF method treated tracking effect figure.
Figure 26 is Figure 24 through SAMF method treated tracking effect figure.
Figure 27 is that Figure 24 tracks effect picture after the processing of DSST method.
Figure 28 is Figure 24 through Staple method treated tracking effect figure.
Figure 29 is the tracking effect figure after Figure 24 treated by the present method.
Specific embodiment
Reach the technical means and efficacy that predetermined purpose is taken for the present invention is further explained, below in conjunction with attached drawing and reality
Example is applied to a specific embodiment of the invention, structure feature and its effect, detailed description are as follows.
To solve technical problem of the existing technology, technical solution provided in this embodiment is broadly divided into template matching rank
Three section, motion detection stage and template renewal stage parts, as shown in Figure 1, being needed in first frame artificially to setting the goal in head
The position of frame and size, and then original image is cut, obtain the standard drawing centered on tracking target.
One, the template matching stage,
If target template image is that z, current frame image img are rolled up with frequency domain multiplication product instead of time domain for acceleration technique
It accumulates, then the convolution matrix K of current frame image and target template image are as follows:
In formula (1) | | img | | it is the norm of img, | | z | | it is the norm of z, FFT and IFFT respectively represent the two dimension of image
Fast Fourier Transform (FFT) and inverse transformation,For the conjugation of FFT (z), σ is the standard deviation of Gauss kernel method；Then target response
Matrix is,
R=IFFT (FFT (K) FFT (γ)) (2)
γ is ridge regression parameter matrix in formula (2).As shown in Fig. 2, R is target response matrix.Target response matrix R is
One with image response matrix of the same size, the size of each coordinate upper value represents that change position related to target in R
Degree, the degree of correlation the big, target more is possible in the coordinate points, therefore target position is that maximum value institute in matrix R is in place
It sets.By traversing R matrix, where finding its maximum value, its coordinate is recorded, [x is expressed ast, yt], the coordinate is to assist hereinafter
Middle method for testing motion.
Two, motion detection stage
Frame-to-frame differences method is usually used in detecting moving target, and basic principle is the Europe for asking adjacent two inter frame image two-by-two between pixel
Formula distance and obtain difference diagram, the value of each point represents the variation degree of present frame and previous frame in difference diagram, movement
Target then appears on the big point of variation degree.Its effect is as shown in figure 3, figure 4 and figure 5.Wherein Fig. 3 and 4 is respectively current
Frame image and previous frame image, Fig. 5 are difference diagram between the two.From figure it can clearly be seen that when camera is static, frame
Differences method can more accurately obtain moving target position.But once two frames not only have the movement of target, with camera lens
It shakes, a large amount of background pixels also change.At this time in the output difference component of frame-to-frame differences method can not accurate recognition go out moving target
Place, as shown in Fig. 6, Fig. 7, Fig. 8.
Therefore this method predicts shaking parameter θ=[α, β, the ε] at video current shooting visual angle by before and after frames matching first,
Frame-to-frame differences method is carried out again after camera lens shakes correction.The prediction technique of the shaking parameter are as follows:
θ is to shake parameter in formula (3), and α is the parameter to teetertotter, and β is the parameter of double swerve, and ε is lens zoom ginseng
Number, Ic are current frame image, and Ip is previous frame image, and Z (Ic, ε) is by present frame with the image after parameter ε scaling；
Shaking parameter θ with visual angle is reference, seeks the Euclidean distance of adjacent two figures after correction, can obtain between two frame of front and back
Difference diagram D,
D is that lens shake the difference diagram after parameter correction in formula (4), and effect is as shown in Fig. 9 to Figure 16.Wherein Fig. 9 and
Figure 13 is current frame image, and Figure 10 and Figure 14 are previous frame images.Figure 11 and Figure 15 is that two field pictures are directly acquired with frame difference method
Difference diagram.As shown, due to the shaking of camera lens, background pixel is also in movement in image, so being also difficult in difference diagram
The place of lock motion target.But in the correction for passing through camera lens jitter parameter, difference diagram such as Figure 12 and Figure 16 of two frame of front and back
It is shown, it is clear that background information is inhibited, and the position where moving target shows bigger difference value, and effect is obvious.
On difference diagram, [x is exported with the rough coordinates of targets in template matching staget, yt] centered on, taking a size is mould
Twice of plate of retrieval window is weighted and averaged each pixel coordinate, is transported using the value of difference diagram as weight in the window
Moving-target position, if moving target size is [sizex, sizey], then motion detection output coordinate is,
X in formula (5)mdFor the x coordinate position of moving target, ymdFor the y-coordinate position of moving target, D (i, j) is difference
Value D of the figure at coordinate [i, j]sumIt is the sum of the weight of entire difference diagram window；
Finally the output of template matching and the output of motion detection are weighted and averaged by we,
ρ is weight coefficient, [x in above formulafinal, yfinal] it is accurate template coordinate.
Three, the template renewal stage,
In present frame calculation template position [xfinal, yfinal] after, centered on this coordinate, with target sizes [sizex,
sizey] it is scale, it reacquires present frame target template z ', present frame template z ' and upper frame template z and is weighted and averaged to obtain new mould
PlateThe tracing task that new template will replace old template to participate in next frame:
δ is that weight coefficient recalculates response matrix K to update ridge with formula (1) after obtaining new template in formula (8)
Regression parameter γ,
γ=(K+ λ I)-1S (9)
α is updated ridge regression parameter in formula (9), and I is and the equirotal unit matrix of K, I and regularization coefficient λ
Regularization is played the role of in cooperation, and S is normal response matrix.Wherein λ value is the positive real number of a very little.S value is and image
In the same size, the minimum dimensional Gaussian nuclear shape of standard deviation, shape is as shown in figure 17.
There are in the target scale significantly video scene of deformation, this method has absolute predominance, with advanced track side
Method is compared, and this method robustness is higher, can complete the tracing task that other methods cannot be completed.Its effect shows as shown in Figure 6.
To sum up, this method be mainly characterized in that: 1) predict and eliminate camera lens shake by the matching of before and after frames image.
Then frame difference detection is carried out to moving target；2) motion detection algorithm and correlation filtering is combined to track.Using motion detection method
The result of track algorithm is corrected, better tracking effect is obtained.3) algorithm can be in the quick motion deformation of target
Complete more complicated tracing task.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
Specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs, exist
Under the premise of not departing from present inventive concept, a number of simple deductions or replacements can also be made, all shall be regarded as belonging to of the invention
Protection scope.
Claims (6)
1. a kind of quick method for tracking and positioning of single goal for adapting to large scale deformation, which comprises the following steps:
Step 1, template matching determine the coordinate and size of target template image, make of current frame image and target template image
Cyclic convolution simultaneously acquires target response matrix with Ridge Regression Modeling Method, as rough coordinates of targets at response matrix maximum value；
Step 2, motion detection,
Step 2.1, before and after frames images match obtain visual angle and shake parameter, and acquire difference diagram；
Step 2.2, statistics difference diagram obtain motion detection output coordinate；
Step 2.3 corrects the rough coordinates of targets with the motion detection output coordinate, obtains precision target coordinate；
Step 3, template renewal update target template with the precision target coordinate, update ridge regression parameter.
2. according to the method described in claim 1, it is characterized by: current frame image and target template image in the step 1
Convolution matrix be,
If target template image is z, current frame image img, in above formula | | img | | it is the norm of img, | | z | | it is the model of z
Number, FFT and IFFT respectively represent the two-dimensional fast fourier transform and inverse transformation of image,For the conjugation of FFT (z), σ
For the standard deviation of Gauss kernel method；
The target response matrix is,
R=IFFT (FFT (K) FFT (γ))
γ is ridge regression parameter matrix in above formula, and R is target response matrix, and wherein the position at top is that rough target is sat
Mark, is expressed as [xt, yt]。
3. according to the method described in claim 2, it is characterized by: in the step 2.1 shake parameter prediction technique be,
θ=[α, β, ε] is to shake parameter in above formula, and α is the parameter to teetertotter, and β is the parameter of double swerve, and ε is camera lens contracting
Parameter is put, Ic is current frame image, and Ip is previous frame image, and Z (Ic, ε) is by present frame with the image after parameter ε scaling；
The difference diagram of before and after frames image is,
D is difference diagram in above formula.
4. according to the method described in claim 3, it is characterized by: in the step 2.2 motion detection output coordinate be,
If target sizes are [sizex, sizey], x in above formulamdFor the x coordinate position of moving target, ymdIt is sat for the y of moving target
Cursor position, D (i, j) are value of the difference diagram at coordinate [i, j]；
D in above formulasumIt is the sum of the weight of entire difference diagram window.
5. according to the method described in claim 4, it is characterized by: in the step 2.3 precision target coordinate be,
ρ is weight coefficient, [x in above formulafinal, yfinal] it is accurate template coordinate.
6. according to the method described in claim 5, it is characterized by: in the step 3 updated target template be,
In above formulaFor updated target template, δ is weight coefficient, and updated ridge regression parameter is,
γ=(K+ λ I)-1S
γ is updated ridge regression parameter in above formula, and I is to cooperate with the equirotal unit matrix of K, I and regularization coefficient λ
Play the role of regularization, S is normal response matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910219613.7A CN109978908B (en) | 2019-03-21 | 2019-03-21 | Single-target rapid tracking and positioning method suitable for large-scale deformation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910219613.7A CN109978908B (en) | 2019-03-21 | 2019-03-21 | Single-target rapid tracking and positioning method suitable for large-scale deformation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109978908A true CN109978908A (en) | 2019-07-05 |
| CN109978908B CN109978908B (en) | 2023-04-28 |
Family
ID=67080039
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910219613.7A Active CN109978908B (en) | 2019-03-21 | 2019-03-21 | Single-target rapid tracking and positioning method suitable for large-scale deformation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109978908B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110555405A (en) * | 2019-08-30 | 2019-12-10 | 北京迈格威科技有限公司 | Target tracking method and device, storage medium and electronic equipment |
| CN110660090A (en) * | 2019-09-29 | 2020-01-07 | Oppo广东移动通信有限公司 | Subject detection method and apparatus, electronic device, and computer-readable storage medium |
| CN115631359A (en) * | 2022-11-17 | 2023-01-20 | 诡谷子人工智能科技(深圳)有限公司 | Image data processing method and device for machine vision recognition |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056158A (en) * | 2016-06-03 | 2016-10-26 | 西安电子科技大学 | Template image global searching method based on mask matrix and fast Fourier transform |
| US20180053294A1 (en) * | 2016-08-19 | 2018-02-22 | Sony Corporation | Video processing system and method for deformation insensitive tracking of objects in a sequence of image frames |
| CN108230367A (en) * | 2017-12-21 | 2018-06-29 | 西安电子科技大学 | A kind of quick method for tracking and positioning to set objective in greyscale video |
| CN108876816A (en) * | 2018-05-31 | 2018-11-23 | 西安电子科技大学 | Method for tracking target based on adaptive targets response |
| CN108961308A (en) * | 2018-06-01 | 2018-12-07 | 南京信息工程大学 | A kind of residual error depth characteristic method for tracking target of drift detection |
-
2019
- 2019-03-21 CN CN201910219613.7A patent/CN109978908B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056158A (en) * | 2016-06-03 | 2016-10-26 | 西安电子科技大学 | Template image global searching method based on mask matrix and fast Fourier transform |
| US20180053294A1 (en) * | 2016-08-19 | 2018-02-22 | Sony Corporation | Video processing system and method for deformation insensitive tracking of objects in a sequence of image frames |
| CN108230367A (en) * | 2017-12-21 | 2018-06-29 | 西安电子科技大学 | A kind of quick method for tracking and positioning to set objective in greyscale video |
| CN108876816A (en) * | 2018-05-31 | 2018-11-23 | 西安电子科技大学 | Method for tracking target based on adaptive targets response |
| CN108961308A (en) * | 2018-06-01 | 2018-12-07 | 南京信息工程大学 | A kind of residual error depth characteristic method for tracking target of drift detection |
Non-Patent Citations (2)
| Title |
|---|
| 库涛等: "尺度目标的频域核回归跟踪研究", 《空军工程大学学报(自然科学版)》 * |
| 曲巨宝等: "运动图像快速跟踪技术研究", 《重庆师范大学学报(自然科学版)》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110555405A (en) * | 2019-08-30 | 2019-12-10 | 北京迈格威科技有限公司 | Target tracking method and device, storage medium and electronic equipment |
| CN110555405B (en) * | 2019-08-30 | 2022-05-06 | 北京迈格威科技有限公司 | Target tracking method and device, storage medium and electronic equipment |
| CN110660090A (en) * | 2019-09-29 | 2020-01-07 | Oppo广东移动通信有限公司 | Subject detection method and apparatus, electronic device, and computer-readable storage medium |
| CN110660090B (en) * | 2019-09-29 | 2022-10-25 | Oppo广东移动通信有限公司 | Subject detection method and apparatus, electronic device, and computer-readable storage medium |
| US11538175B2 (en) | 2019-09-29 | 2022-12-27 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for detecting subject, electronic device, and computer readable storage medium |
| CN115631359A (en) * | 2022-11-17 | 2023-01-20 | 诡谷子人工智能科技(深圳)有限公司 | Image data processing method and device for machine vision recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109978908B (en) | 2023-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105245841B (en) | A kind of panoramic video monitoring system based on CUDA | |
| CN110796010B (en) | Video image stabilizing method combining optical flow method and Kalman filtering | |
| CN111951325B (en) | Pose tracking method, pose tracking device and electronic equipment | |
| CN106991690B (en) | A kind of video sequence synchronous method based on moving target timing information | |
| WO2020253618A1 (en) | Video jitter detection method and device | |
| CN101383899A (en) | Video image stabilizing method for space based platform hovering | |
| CN105809626A (en) | Self-adaption light compensation video image splicing method | |
| CN109978908A (en) | A kind of quick method for tracking and positioning of single goal adapting to large scale deformation | |
| CN111383252B (en) | Multi-camera target tracking method, system, device and storage medium | |
| CN109711241B (en) | Object detection method and device and electronic equipment | |
| CN110555377B (en) | Pedestrian detection and tracking method based on fish eye camera overlooking shooting | |
| Wang et al. | Video stabilization: A comprehensive survey | |
| CN106713740A (en) | Positioning tracking camera shooting method and system | |
| CN110827321A (en) | Multi-camera cooperative active target tracking method based on three-dimensional information | |
| CN103500454A (en) | Method for extracting moving target of shaking video | |
| Wan et al. | Drone image stitching using local mesh-based bundle adjustment and shape-preserving transform | |
| CN113971684B (en) | Real-time robust target tracking method based on KCF and SURF features | |
| CN116579920A (en) | Image stitching method and system based on heterogeneous multimode panoramic stereoscopic imaging system | |
| Dardelet et al. | An event-by-event feature detection and tracking invariant to motion direction and velocity | |
| Li et al. | Deep online video stabilization using imu sensors | |
| CN112508998A (en) | Visual target alignment method based on global motion | |
| CN117152206A (en) | Multi-target long-term tracking method for unmanned aerial vehicle | |
| Chen | Sports sequence images based on convolutional neural network | |
| Chui et al. | Event-Based Feature Tracking in Continuous Time with Sliding Window Optimization | |
| Murakami et al. | Athlete 3D pose estimation from a monocular TV sports video using pre-trained temporal convolutional networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |