CN107248174A - A kind of method for tracking target based on TLD algorithms - Google Patents
A kind of method for tracking target based on TLD algorithms Download PDFInfo
- Publication number
- CN107248174A CN107248174A CN201710340039.1A CN201710340039A CN107248174A CN 107248174 A CN107248174 A CN 107248174A CN 201710340039 A CN201710340039 A CN 201710340039A CN 107248174 A CN107248174 A CN 107248174A
- Authority
- CN
- China
- Prior art keywords
- feature point
- target
- tracking
- image
- module
- 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
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
- G06T7/269—Analysis of motion using gradient-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- 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
- G06T7/248—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving reference images or patches
-
- 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
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Abstract
The present invention relates to a kind of method for tracking target based on TLD algorithms, including：Tracking module, detection module, study module, including：Target area is chosen in initialization from the first two field picture；The tracking module is according to corresponding target area information in information prediction current frame image in target area in previous frame and uses perception hash algorithm to be screened to target area information in the current frame image to determine target following result；The detection module determines object detection results；Target following region is determined according to the target following result and the object detection results；The target following region is inputted to the study module and learnt.The characteristic point of tracking module is screened using hash algorithm is perceived in a kind of method for tracking target based on TLD algorithms of the present invention, it calculates simple, the robustness of algorithm is improved while significantly improving tracking module real-time, the method for tracking target based on TLD algorithms is preferably applied in actual tracking system.
Description
Technical field
The invention belongs to target following technical field, and in particular to a kind of method for tracking target based on TLD algorithms.
Background technology
Target following is all widely used in military and civilian field.With the hair at full speed of modern aerospace technology
Exhibition, various headways and mobility more and more higher, to target following it is also proposed that higher and higher requirement.Target following is handle
Automatically control, image procossing, information science combine, form one kind can from picture signal automatic identification in real time
Target, extraction and future position information, automatically track target Motion Technology.
It is used as a kind of monotrack algorithm detected based on on-line study, TLD (Tracking-Learning-
Detection, abbreviation TLD) track algorithm is of great interest due to its good tracking performance.
The algorithm is mainly made up of tracking module, detection module, three parts of study module.Tracking module uses intermediate value light
Stream method predicts the position of target in the current frame；Detection module has used three layers of cascade classifier, variance grader, random fern
Grader, nearest neighbor classifier, detect the position of target in the current frame；The method that study module utilizes P-N on-line studies,
" remarkable characteristic " of tracking module and the object module and relevant parameter of detection module are constantly updated, so that tracking effect
More stablize, it is robust, reliable.
TLD algorithms preferably resolve tracked target the problems such as the deformation, partial occlusion occurred during tracked.
Existing TLD algorithms, such as number of patent application are a kind of 201610530203.0 method for tracking target based on TLD algorithms
In, tracking module determines characteristic point using uniform sampling site, computationally intensive, and real-time performance of tracking is poor, and Partial Feature point only includes mesh
Target positional information, without representativeness, it is impossible to reliably tracked so that tracking result is easily drifted about.Using equal
Even sampling site is obtained after a large amount of characteristic points, and characteristic point, computationally intensive, poor real are screened using pyramid LK optical flow methods and NCC methods.
Therefore, existing its computation complexity of TLD algorithms is high, computationally intensive, with the need of real system in terms of real-time
Seek survival in larger gap, the real-time for how improving the target tracking algorism of TLD algorithms is still a class for being rich in challenge
Topic.
The content of the invention
In order to solve the above-mentioned problems in the prior art, the invention provides a kind of target based on TLD algorithms with
Track method, including：Tracking module, detection module, study module, the method for tracking target include：
Target area is chosen in initialization from the first two field picture；
The tracking module is believed according to corresponding target area in information prediction current frame image in target area in previous frame
Cease and use perception hash algorithm to be screened to determine target following result to target area information in the current frame image；
The detection module determines object detection results；
Target following region is determined according to the target following result and the object detection results；
The target following region is inputted to the study module and learnt.
In one embodiment of the invention, the tracking module is according to information prediction present frame in target area in previous frame
In image corresponding target area information and using perceive hash algorithm target area information in the current frame image is carried out
Screen to determine that target following result includes：
Fisrt feature point is extracted to the previous frame objective area in image；
Predict the fisrt feature point corresponding second feature point in the current frame image；
The second feature point is screened using the perception hash algorithm and extracts third feature point；
Determine target location as the target following result by the third feature point.
In one embodiment of the invention, fisrt feature point is extracted to the previous frame objective area in image to use
SURF algorithm.
In one embodiment of the invention, the SURF algorithm includes：
Graphical rule space is calculated using previous frame image target area；
Described image metric space is determined into the fisrt feature point by non-maxima suppression.
In one embodiment of the invention, the fisrt feature point is SURF angle points.
In one embodiment of the invention, predict the fisrt feature point corresponding second in the current frame image
Characteristic point uses pyramid LK optical flow methods.
In one embodiment of the invention, the second feature point is screened using the perception hash algorithm and extracts the 3rd
Characteristic point includes：
First Hash fingerprint of image block where calculating the fisrt feature point；
Second Hash fingerprint of image block where calculating the second feature point；
Calculate the first Hash fingerprint of the fisrt feature point and the second Hash fingerprint of the corresponding second feature point
Between Hamming distance；
The intermediate value of the Hamming distance is chosen as threshold value, the second feature point is screened and obtains the third feature point.
In one embodiment of the invention, using the intermediate value of the Hamming distance as threshold value, the second feature is screened
Point, which obtains the third feature point, to be included：Image block by the Hamming distance less than or equal to the threshold value and less than 15 is corresponding
The second feature point retains, and otherwise screens out.
In one embodiment of the invention, the detection module determines that object detection results include：
Global traversal is carried out to the current frame image to obtain the target to be selected that the detection module is determined；
Using the variance grader, random fern grader and nearest neighbor classifier of the detection module to described at least one
Target carry out processing to be selected forms the object detection results.
In one embodiment of the invention, global traversal is carried out to the current frame image and uses sliding window mode.
The embodiment of the present invention is using perception hash algorithm to tracking module in a kind of method for tracking target based on TLD algorithms
Characteristic point screened, it calculates simple, and the robustness of algorithm is improved while significantly improving tracking module real-time, is made
Method for tracking target based on TLD algorithms is preferably applied in actual tracking system.
Brief description of the drawings
Fig. 1 is a kind of schematic flow sheet of the method for tracking target based on TLD algorithms provided in an embodiment of the present invention；
Fig. 2 is former for a kind of tracking module work of method for tracking target based on TLD algorithms provided in an embodiment of the present invention
Manage schematic diagram.
Embodiment
Further detailed description is done to the present invention with reference to specific embodiment, but embodiments of the present invention are not limited to
This.
Embodiment one
Fig. 1 is referred to, Fig. 1 is a kind of flow of the method for tracking target based on TLD algorithms provided in an embodiment of the present invention
Schematic diagram；Wherein, the TLD algorithms include tracking module, detection module, study module, and this method comprises the following steps：
(a) target area is chosen in initialization from the first two field picture；
(b) tracking module is according to corresponding target area in information prediction current frame image in target area in previous frame
Information simultaneously uses perception hash algorithm to be screened to determine target following knot to target area information in the current frame image
Really；
(c) detection module determines object detection results；
(d) target following region is determined according to the target following result and the object detection results；
(e) the target following region is inputted to the study module and learnt.
Alternatively, step (b) includes：
(b1) fisrt feature point is extracted to the previous frame objective area in image；
(b2) fisrt feature point corresponding second feature point in the current frame image is predicted；
(b3) the second feature point is screened using the perception hash algorithm and extracts third feature point；
(b4) determine target location as the target following result by the third feature point.
Alternatively, step (b1) uses SURF algorithm.
Alternatively, the SURF algorithm includes：
(b11) graphical rule space is calculated using previous frame objective area in image；
(b22) described image metric space is determined into the fisrt feature point by non-maxima suppression.
Alternatively, the fisrt feature point is SURF angle points.
Alternatively, step (b2) uses pyramid LK optical flow methods.
Alternatively, step (b3) includes：
(b31) the first Hash fingerprint of image block where calculating each fisrt feature point；
(b32) the second Hash fingerprint of image block where calculating each second feature point；
(b33) calculate all fisrt feature points the first Hash fingerprint and the corresponding second feature point second
Hamming distance between Hash fingerprint；
(b34) intermediate value of all Hamming distances is chosen as threshold value, is screened the second feature point and is obtained described the
Three characteristic points.
Alternatively, step (b34) includes：
The Hamming distance is less than or equal to the threshold value and is less than the 15 corresponding second feature point guarantor of image block
Stay, otherwise screen out.Alternatively, step (c) includes：
(c1) global traversal is carried out to the current frame image to obtain the target to be selected that the detection module is determined；
(c2) using variance grader, random fern grader and the nearest neighbor classifier of the detection module at least one
The target carry out processing to be selected forms the object detection results.
Alternatively, global traversal is carried out to the current frame image using a stroke window mode.
The present embodiment is using spy of the perception hash algorithm to tracking module in a kind of method for tracking target based on TLD algorithms
Levy and screened, it calculates simple, and the robustness of algorithm is improved while significantly improving tracking module real-time, makes to be based on
The method for tracking target of TLD algorithms is preferably applied in actual tracking system.
Embodiment two
Refer to Fig. 2, a kind of tracking module work of the method for tracking target based on TLD algorithms provided in an embodiment of the present invention
Make principle schematic.This tracking module has been substantially carried out improving at following two：
Improved at first：Characteristic point selection mechanism for TLD algorithm keeps track modules is improved；
Reason：Tracking module characteristic point quantity as obtained by uniform sampling site is more in TLD algorithms, causes to calculate quantitative change greatly, with
Track real-time is deteriorated, and Partial Feature point only includes the positional information of target, without representativeness, it is impossible to by reliably with
Track so that tracking result is easily drifted about；
Solution：Scale invariability, rotational invariance, the advantages of calculating speed is fast, we are had based on SURF angle points
Case extracts SURF angle points to target area and replaces uniform sampling site more fast and accurately to complete the matched jamming process in later stage.
The step-by-step procedures for extracting SURF angle points to any input picture is as follows：
The first step：Build black plug (Hessian) matrix
Assuming that input picture is f, a Hessian matrix, Hessian are calculated to image f each pixel f (x, y)
Matrix is defined as follows：
Hessian determinants of a matrix are：
Hessian matrixes can characterize the local curvature put on Two-dimensional Surfaces, and the determinant of a matrix provides the strong of curvature
Degree, it is the pixel (all having higher curvature in multiple directions) with higher local curvature to define angle point.Because characteristic point is needed
Possesses yardstick independence, so before Hessian matrix constructions are carried out, entering first to input picture using different σ yardsticks
Row gaussian filtering, the calculating of Hessian matrixes is carried out to filtered image again.Hessian matrixes after after filtering are represented
It is as follows：
Wherein L (x, σ) represent second order standard gaussian kernel function under σ yardsticks with input picture f point (x, y) place convolution
As a result.
To make SURF algorithm as efficient as possible, gaussian kernel function is replaced using approximate Gaussian core, then it is above-mentioned filtered
Hessian determinant of a matrix approximate representations are：
Det (H)=DxxDyy-(0.9Dxy)2
In formula, DxxRepresent second order local derviation of the pixel under approximate Gaussian core in horizontal direction, DyyRepresent the pixel
Second order local derviation under approximate Gaussian core on vertical direction, DxyRepresent mixing Second Order Partial of the pixel under approximate Gaussian core
Lead.
Second step：Build metric space.
The metric space of image is to be exactly by choosing different scale σ values to above-mentioned approximate Gaussian core, realizing different scale
Template and input picture carry out convolution algorithm.To the template and each pixel of the image after input picture convolution of each yardstick
Point calculates the approximation of its Hessian matrix determinant, the gray value of pixel is replaced with the approximation of determinant, so as to obtain
Response image under the yardstick, different scale σ just obtains multiple response images so as to the metric space of pie graph picture.
3rd step：Non-maxima suppression determines characteristic point.
Compare surrounding when the Hessian matrix determinants approximation of some pixel obtains local maximum, the i.e. pixel
When other put brighter or darker in neighborhood, can determine that the point is exactly a characteristic point.In the metric space of image, by each pixel
Point is compared the approximation of determinant with 26 pixels of its 3-dimensional neighborhood, if the approximation of the determinant of the point is this 26
Maximum or minimum value in point, then it is exactly a characteristic point to assert the point.
Improved at second：For the improvement project of tracking module intermediate value optical flow method；
Reason：Existing TLD algorithms intermediate value optical flow method obtains a large amount of characteristic points by uniform sampling site, then utilizes forward light
Stream predicts the position of these characteristic points in the next frame, finally utilizes the front and rear FB (Forward- gone out to optical flow computation
Backword) error and normalized-cross-correlation function NCC Double Selection mechanism, screen out the larger characteristic point of deviation, with remaining
Characteristic point finally determines target location.Wherein pyramid LK optical flow methods and NCC are and the existing TLD than relatively time-consuming algorithm
In order to calculate the pyramid LK light streams twice of FB errors, cause amount of calculation excessive, real-time is deteriorated.
The characteristic point extracted to target area by SURF algorithm is improved at first, and not only quantity is few, and has been
Possesses the characteristic point of target critical information, carrying out forward light stream to these points predicts behind its position in the next frame, then uses FB
Error and this strong Filtering system screening characteristic points of NCC are not necessarily to, and add amount of calculation.
Solution：The SURF angle points that this programme is extracted to target area in previous frame predict it before to optical flow method
After character pair point in the current frame, the screening of characteristic point directly is carried out by perceiving hash algorithm, this algorithm calculates letter
Single, real-time is good.
It is a kind of similarity measurements quantity algorithm without ginseng to perceive hash algorithm, and it is referred to by comparing the Hash that two pictures are generated
Line carries out similarity measurement.Perceive hash algorithm detailed step as follows：
The first step：Image sampling
The picture of arbitrary size is narrowed down to 8*8 size, 64 pixels, do not keep aspect ratio altogether, need to only be become
Into 8*8 square.The picture difference that different sizes, proportional band come can so be abandoned.
Second step：Calculate average value
Calculate the average gray of 64 pixels.
3rd step：The gray scale of compared pixels
The gray scale of each pixel and average gray are compared, 1 is then designated as more than or equal to average value, otherwise is designated as
0。
4th step：Generate Hash fingerprint
By the comparative result of previous step, combine, the binary string of one 64 is constituted, here it is this image block
Hash fingerprint.The order of combination is unimportant, as long as ensureing all pictures all according to same combination.
To sum up, improved with reference at above-mentioned two, improvement and the improvement to characteristic point Filtering system to extracting feature point mode,
The change of TLD algorithm keeps track modules whole flow process is as follows：
Assuming that former frame is It-1, present frame is It,
(1) mode for extracting SURF characteristic points first with SURF algorithm extracts It-1The fisrt feature of target area in frame
Point, the fisrt feature point is SURF angle points；
(2) the fisrt feature point extracted using pyramid LK optical flow methods forward trace is in present frame ItIn position be correspondence
Second feature point；
(3) to present frame ItIt is middle to track obtained each second feature point and obtain the using perceiving hash algorithm and screen
Three characteristic points；
(4) present frame I is finally determined finally according to third feature point after screeningtThe position of middle target.
Embodiment three
The present embodiment is on the basis of above-described embodiment, with reference to the workflow of TLD algorithms, provides and is calculated using perception Hash
TLD algorithm flows after method, improvement of the emphasis to tracking module is illustrated.
The workflow of TLD algorithms is specific as follows：
S01：Choose the position of target and size in start frame；
In the start frame of video (or sequence of pictures), the rectangle for including target by mouse manually determined one obtains institute
The target location that need to be tracked and size, that is, record coordinates of targets and width is high.
S02：Tracking module extracts SURF angle points using SURF algorithm to previous frame objective area in image, specifically includes：
S021：It regard the target area of former frame as input picture during said extracted SURF characteristic points first；
S022：Convolution is carried out with the corresponding approximate Gaussian core templates of different scale σ and input picture, to the image after convolution
Each pixel utilize this formula det (H)=DxxDyy-(0.9Dxy)2The approximation of Hessian matrix determinants is calculated, is obtained
Response image, and then obtain the graphical rule space that the corresponding response images of different scale σ are constituted.
S023：In the metric space of image, each pixel is compared into determinant with 26 pixels of its 3-dimensional neighborhood
Approximation, if the approximation of the determinant of the point is the maximum or minimum value in this 26 points, assert that the point is exactly
One characteristic point.
S03：Fisrt feature point corresponding second feature in current frame image is predicted using pyramid LK optical flow methods
Point；
S04：The second feature point is screened using the perception hash algorithm and extracts third feature point, is specifically included：
S041：Centered on each SURF angle points extracted by target area in former frame, the pixel of 8*8 in its neighborhood is taken
Image block is constituted, the first Hash fingerprint of image block where calculating each SURF angle points using said sensed hash algorithm.
S042：For each SURF angle point, to predict in current frame image corresponding second feature point in
The heart, takes the pixel of 8*8 in its neighborhood to constitute image block, the second Hash fingerprint of image block where calculating each characteristic point.
S043：Calculating the Hamming distance of the first Hash fingerprint and the second Hash fingerprint, (two character string correspondence positions are not
With the number of character).If Hamming distance is 0, illustrate that two image blocks are much like, Hamming distance is less than 5, then two pictures slightly have
Difference, but it is relatively more similar, and Hamming distance is more than 15, then it is assumed that two image blocks are entirely different.
S044：The intermediate values of all Hamming distances is chosen as threshold value, Hamming distance is less than or equal to intermediate value and less than 15
The corresponding characteristic point of image block is that third feature point retains, and is otherwise screened out.Finally determine to work as finally according to all third feature points
The position of target in previous frame.
S05：The detection module determines object detection results；
S06：Target following region is determined according to the target following result and the object detection results；
S07：The target following region is inputted to the study module and learnt.
Example IV
The present embodiment is on the basis of above-described embodiment, with reference to the workflow of TLD algorithms, provides another using perception
TLD algorithm flows after hash algorithm, improvement of the emphasis to tracking module is illustrated.
S01：Choose the position of target and size in start frame；
S02：Tracking module extracts fisrt feature point to the uniform sampling site of previous frame objective area in image；
S03：Fisrt feature point corresponding second feature in current frame image is predicted using pyramid LK optical flow methods
Point；
S04：The second feature point is screened using the perception hash algorithm and extracts third feature point, is specifically included：
S041：Centered on each fisrt feature point extracted by target area in former frame, the pixel of 8*8 in its neighborhood is taken
Point constitutes image block, the first Hash fingerprint of image block where calculating each fisrt feature point using said sensed hash algorithm.
S042：For each fisrt feature point, using predict in current frame image corresponding second feature point as
Center, takes the pixel of 8*8 in its neighborhood to constitute image block, the second Hash fingerprint of image block where calculating each characteristic point.
S043：Calculating the Hamming distance of the first Hash fingerprint and the second Hash fingerprint, (two character string correspondence positions are not
With the number of character).If Hamming distance is 0, illustrate that two image blocks are much like, Hamming distance is less than 5, then two pictures slightly have
Difference, but it is relatively more similar, and Hamming distance is more than 15, then it is assumed that two image blocks are entirely different.
S044：The intermediate values of all Hamming distances is chosen as threshold value, Hamming distance is less than or equal to intermediate value and less than 15
The corresponding characteristic point of image block is that third feature point retains, and is otherwise screened out.Finally determine to work as finally according to all third feature points
The position of target in previous frame.
S05：The detection module determines object detection results；
S06：Target following region is determined according to the target following result and the object detection results；
S07：The target following region is inputted to the study module and learnt.
Above content is to combine specific preferred embodiment further description made for the present invention, it is impossible to assert
The specific implementation of the present invention is confined to these explanations.For general technical staff of the technical field of the invention,
On the premise of not departing from present inventive concept, some simple deduction or replace can also be made, should all be considered as belonging to the present invention's
Protection domain.
Claims (10)
1. a kind of method for tracking target based on TLD algorithms, including：Tracking module, detection module, study module, its feature exist
In, including：
Target area is chosen in initialization from the first two field picture；
The tracking module according to corresponding target area information in information prediction current frame image in target area in previous frame simultaneously
Perception hash algorithm is used to be screened to determine target following result to target area information in the current frame image；
The detection module determines object detection results；
Target following region is determined according to the target following result and the object detection results；
The target following region is inputted to the study module and learnt.
2. according to the method described in claim 1, it is characterised in that the tracking module is according to target area information in previous frame
Predict current frame image in corresponding target area information and using perceive hash algorithm to target area in the current frame image
Domain information is screened to determine that target following result includes：
Fisrt feature point is extracted to the previous frame objective area in image；
Predict the fisrt feature point corresponding second feature point in the current frame image；
The second feature point is screened using the perception hash algorithm and extracts third feature point；
Determine target location as the target following result by the third feature point.
3. method according to claim 2, it is characterised in that first is extracted to the previous frame objective area in image special
Levy and use SURF algorithm.
4. method according to claim 3, it is characterised in that the SURF algorithm includes：
Graphical rule space is calculated using previous frame image target area；
Described image metric space is determined into the fisrt feature point by non-maxima suppression.
5. method according to claim 3, it is characterised in that the fisrt feature point is SURF angle points.
6. method according to claim 2, it is characterised in that the prediction fisrt feature point is in the current frame image
Corresponding second feature point uses pyramid LK optical flow methods.
7. method according to claim 2, it is characterised in that the second feature is screened using the perception hash algorithm
Point, which extracts third feature point, to be included：
First Hash fingerprint of image block where calculating the fisrt feature point；
Second Hash fingerprint of image block where calculating the second feature point；
Calculate between the first Hash fingerprint of the fisrt feature point and the second Hash fingerprint of the corresponding second feature point
Hamming distance；
The intermediate value of the Hamming distance is chosen as threshold value, the second feature point is screened and obtains the third feature point.
8. method according to claim 7, it is characterised in that using the intermediate value of the Hamming distance as threshold value, screens institute
Stating second feature point and obtaining the third feature point includes：The Hamming distance is less than or equal to the threshold value and is less than 15 figure
As the corresponding second feature point of block retains, otherwise screen out.
9. according to the method described in claim 1, it is characterised in that the detection module determines that object detection results include：
Global traversal is carried out to the current frame image to obtain the target to be selected that the detection module is determined；
Using the variance grader, random fern grader and nearest neighbor classifier of the detection module to be selected at least one described
Target carry out processing forms the object detection results.
10. method according to claim 9, it is characterised in that global traversal is carried out to the current frame image using cunning
Window mode.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710340039.1A CN107248174A (en) | 2017-05-15 | 2017-05-15 | A kind of method for tracking target based on TLD algorithms |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710340039.1A CN107248174A (en) | 2017-05-15 | 2017-05-15 | A kind of method for tracking target based on TLD algorithms |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107248174A true CN107248174A (en) | 2017-10-13 |
Family
ID=60017349
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710340039.1A Pending CN107248174A (en) | 2017-05-15 | 2017-05-15 | A kind of method for tracking target based on TLD algorithms |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107248174A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107967692A (en) * | 2017-11-28 | 2018-04-27 | 西安电子科技大学 | A kind of target following optimization method based on tracking study detection |
| CN108320306A (en) * | 2018-03-06 | 2018-07-24 | 河北新途科技有限公司 | Merge the video target tracking method of TLD and KCF |
| CN108447079A (en) * | 2018-03-12 | 2018-08-24 | 中国计量大学 | A kind of method for tracking target based on TLD algorithm frames |
| CN109146918A (en) * | 2018-06-11 | 2019-01-04 | 西安电子科技大学 | A kind of adaptive related objective localization method based on piecemeal |
| CN109270963A (en) * | 2018-09-29 | 2019-01-25 | 武汉理工大学 | A kind of wheelchair automatic following system and method based on reflective fabric and laser radar |
| CN111383247A (en) * | 2018-12-29 | 2020-07-07 | 北京易讯理想科技有限公司 | Method for enhancing image tracking stability of pyramid LK optical flow algorithm |
| CN112651999A (en) * | 2021-01-19 | 2021-04-13 | 滨州学院 | Unmanned aerial vehicle ground target real-time tracking method based on space-time context perception |
| CN112836684A (en) * | 2021-03-09 | 2021-05-25 | 上海高德威智能交通系统有限公司 | Auxiliary driving-based method, device and equipment for calculating degree change rate of eye scale |
| CN112989100A (en) * | 2019-12-16 | 2021-06-18 | 中国移动通信集团辽宁有限公司 | Indoor positioning method and device based on image fingerprints |
| CN113689467A (en) * | 2021-07-30 | 2021-11-23 | 稿定（厦门）科技有限公司 | Feature point optimization method and system suitable for plane tracking |
| CN113792670A (en) * | 2021-09-16 | 2021-12-14 | 中国电子科技集团公司第五十四研究所 | High-precision airborne target identification method based on adaptive reconstruction |
| CN114066936A (en) * | 2021-11-06 | 2022-02-18 | 中国电子科技集团公司第五十四研究所 | Target reliability tracking method in small target capturing process |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102881024A (en) * | 2012-08-24 | 2013-01-16 | 南京航空航天大学 | Tracking-learning-detection (TLD)-based video object tracking method |
| CN104331901A (en) * | 2014-11-26 | 2015-02-04 | 北京邮电大学 | TLD-based multi-view target tracking device and method |
| CN105139420A (en) * | 2015-08-03 | 2015-12-09 | 山东大学 | Particle filter and perceptual hash-based video target tracking method |
| CN105844669A (en) * | 2016-03-28 | 2016-08-10 | 华中科技大学 | Video target real-time tracking method based on partial Hash features |
| CN106204649A (en) * | 2016-07-05 | 2016-12-07 | 西安电子科技大学 | A kind of method for tracking target based on TLD algorithm |
-
2017
- 2017-05-15 CN CN201710340039.1A patent/CN107248174A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102881024A (en) * | 2012-08-24 | 2013-01-16 | 南京航空航天大学 | Tracking-learning-detection (TLD)-based video object tracking method |
| CN104331901A (en) * | 2014-11-26 | 2015-02-04 | 北京邮电大学 | TLD-based multi-view target tracking device and method |
| CN105139420A (en) * | 2015-08-03 | 2015-12-09 | 山东大学 | Particle filter and perceptual hash-based video target tracking method |
| CN105844669A (en) * | 2016-03-28 | 2016-08-10 | 华中科技大学 | Video target real-time tracking method based on partial Hash features |
| CN106204649A (en) * | 2016-07-05 | 2016-12-07 | 西安电子科技大学 | A kind of method for tracking target based on TLD algorithm |
Non-Patent Citations (3)
| Title |
|---|
| S.N.SHARMA 等: "Multi-object tracking using TLD framework", 《2016 IEEE INTERNATIONAL CONFERENCE ON RECENT TRENDS IN ELECTRONICS, INFORMATION & COMMUNICATION TECHNOLOGY (RTEICT)》 * |
| 李子印 等: "融合图像感知哈希技术的运动目标跟踪", 《中国图象图形学报》 * |
| 李敏敏: "基于TLD模型的目标跟踪方法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107967692A (en) * | 2017-11-28 | 2018-04-27 | 西安电子科技大学 | A kind of target following optimization method based on tracking study detection |
| CN108320306A (en) * | 2018-03-06 | 2018-07-24 | 河北新途科技有限公司 | Merge the video target tracking method of TLD and KCF |
| CN108447079A (en) * | 2018-03-12 | 2018-08-24 | 中国计量大学 | A kind of method for tracking target based on TLD algorithm frames |
| CN109146918A (en) * | 2018-06-11 | 2019-01-04 | 西安电子科技大学 | A kind of adaptive related objective localization method based on piecemeal |
| CN109270963A (en) * | 2018-09-29 | 2019-01-25 | 武汉理工大学 | A kind of wheelchair automatic following system and method based on reflective fabric and laser radar |
| CN111383247A (en) * | 2018-12-29 | 2020-07-07 | 北京易讯理想科技有限公司 | Method for enhancing image tracking stability of pyramid LK optical flow algorithm |
| CN112989100A (en) * | 2019-12-16 | 2021-06-18 | 中国移动通信集团辽宁有限公司 | Indoor positioning method and device based on image fingerprints |
| CN112989100B (en) * | 2019-12-16 | 2023-07-18 | 中国移动通信集团辽宁有限公司 | Indoor positioning method and device based on image fingerprint |
| CN112651999A (en) * | 2021-01-19 | 2021-04-13 | 滨州学院 | Unmanned aerial vehicle ground target real-time tracking method based on space-time context perception |
| CN112836684A (en) * | 2021-03-09 | 2021-05-25 | 上海高德威智能交通系统有限公司 | Auxiliary driving-based method, device and equipment for calculating degree change rate of eye scale |
| CN113689467A (en) * | 2021-07-30 | 2021-11-23 | 稿定（厦门）科技有限公司 | Feature point optimization method and system suitable for plane tracking |
| CN113689467B (en) * | 2021-07-30 | 2022-07-12 | 稿定（厦门）科技有限公司 | Feature point optimization method and system suitable for plane tracking |
| CN113792670A (en) * | 2021-09-16 | 2021-12-14 | 中国电子科技集团公司第五十四研究所 | High-precision airborne target identification method based on adaptive reconstruction |
| CN114066936A (en) * | 2021-11-06 | 2022-02-18 | 中国电子科技集团公司第五十四研究所 | Target reliability tracking method in small target capturing process |
| CN114066936B (en) * | 2021-11-06 | 2023-09-12 | 中国电子科技集团公司第五十四研究所 | Target reliability tracking method in small target capturing process |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107248174A (en) | A kind of method for tracking target based on TLD algorithms | |
| CN108062525B (en) | Deep learning hand detection method based on hand region prediction | |
| CN113065558A (en) | Lightweight small target detection method combined with attention mechanism | |
| Roa'a et al. | Generation of high dynamic range for enhancing the panorama environment | |
| CN109685045B (en) | Moving target video tracking method and system | |
| Chen et al. | Haze removal using radial basis function networks for visibility restoration applications | |
| CN109711268B (en) | Face image screening method and device | |
| CN113591968A (en) | Infrared weak and small target detection method based on asymmetric attention feature fusion | |
| CN111402237B (en) | Video image anomaly detection method and system based on space-time cascade self-encoder | |
| CN111160291B (en) | Human eye detection method based on depth information and CNN | |
| CN110941996A (en) | Target and track augmented reality method and system based on generation of countermeasure network | |
| CN111815528A (en) | Bad weather image classification enhancement method based on convolution model and feature fusion | |
| Fan et al. | Multiscale cross-connected dehazing network with scene depth fusion | |
| Yu et al. | Image and video dehazing using view-based cluster segmentation | |
| CN110414430B (en) | Pedestrian re-identification method and device based on multi-proportion fusion | |
| Xu et al. | COCO-Net: A dual-supervised network with unified ROI-loss for low-resolution ship detection from optical satellite image sequences | |
| Lv et al. | Blind dehazed image quality assessment: a deep CNN-based approach | |
| CN114241372A (en) | Target identification method applied to sector-scan splicing | |
| CN107248175B (en) | TLD target tracking method based on circular projection matching algorithm | |
| WO2014172875A1 (en) | Moving object detection | |
| CN108229281B (en) | Neural network generation method, face detection device and electronic equipment | |
| CN106778822B (en) | Image straight line detection method based on funnel transformation | |
| CN105608674A (en) | Image registration, interpolation and denoising-based image enhancement method | |
| CN110111368B (en) | Human body posture recognition-based similar moving target detection and tracking method | |
| CN111127355A (en) | Method for finely complementing defective light flow graph and application thereof |
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: 20171013 |