CN108765470A - One kind being directed to the improved KCF track algorithms of target occlusion - Google Patents
One kind being directed to the improved KCF track algorithms of target occlusion Download PDFInfo
- Publication number
- CN108765470A CN108765470A CN201810624972.6A CN201810624972A CN108765470A CN 108765470 A CN108765470 A CN 108765470A CN 201810624972 A CN201810624972 A CN 201810624972A CN 108765470 A CN108765470 A CN 108765470A
- Authority
- CN
- China
- Prior art keywords
- target
- tracking
- algorithm
- algorithms
- occlusion
- 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
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- 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
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- 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
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- 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
- G06T2207/20024—Filtering details
-
- 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
- G06T2207/20048—Transform domain processing
- G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]
-
- 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 invention belongs to technical field of image processing, and in particular to one kind being directed to the improved KCF track algorithms of target occlusion.By the improvement to KCF track algorithms, anti-the blocking property of former algorithm is promoted, the tracking accuracy and success rate of algorithm are improved.Target occlusion determination method, tracking failure determination method is added in the algorithm, and template matches target relocates algorithm, promotes anti-the blocking property of original KCF track algorithms.Using Benchmark data sets and the trolley video of oneself shooting, innovatory algorithm is verified.By experimental result it is found that inventive algorithm is for video sequence is blocked has stronger robustness.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to one kind being directed to the improved KCF track algorithms of target occlusion.
By the improvement to KCF track algorithms, anti-the blocking property of former algorithm is promoted, the tracking accuracy and success rate of algorithm are improved.
Background technology
Target following is one important research direction of visual field, with social progress with science and technology development, target with
Track technology has been widely used in the key areas such as Activity recognition, traffic monitoring, navigational guidance, weaponry.In recent years, mesh
Mark tracking technique achieves huge progress and development, however due to the factor of various complexity in reality, for example quickly move
Object, complicated background, the deformation of object and serious occlusion issue can handle all applied fields without a kind of algorithm
Scape, so target following technology is still a project that is challenging and being studied extensively.
Bolme etc. proposes MOSSE algorithms, which opens the gate of correlation filter, proposes with filter correlation
Form come obtain output response, and then obtain peak response at position be it is desirable that tracking target's center position, calculate
Method only need a target area sample image be used for training objective display model, using discrete Fourier transform by target with
Similarity between all candidate regions is transformed into frequency domain, greatly improves the tracking velocity of algorithm.Henriques etc. is proposed
Detecting and tracking (CSK) algorithm of loop structure, the algorithm carry out cyclic shift to training sample, approximate can regard as to the close of target
Collection sampling, is trained grader to obtain a large amount of training samples.In addition, using the behaviour of cyclic shift to candidate image block
Make, constructs detection process of a large amount of candidate image blocks region for grader.The bright spot of algorithm is, the training and inspection of grader
Survey process is realized under frequency domain and is quickly calculated, greatly improve the processing of track algorithm all using discrete Fourier transform
Speed.Danelljan etc. proposes adaptive color attribute vision tracking (CN) algorithm, base of the algorithm in CSK track algorithms
Color characteristic is added on plinth, and uses self-adaptive reduced-dimensions technology, and the color feature vector of 11 dimensions is dropped into 2 dimensions, is realized certainly
Adapt to color tracking.Zhang etc. proposes that space-time context tracks (STC) algorithm, which uses depth spatio-temporal context information,
Background information around target is added in the training of convolution filter model, for weakening shadow of the partial occlusion to target
It rings.Henriques etc. proposes core correlation filter (KCF) track algorithm on the basis of (CSK), uses the direction Hough (HOG)
Histogram of gradients feature replaces original grey value characteristics, and core correlation filter is expanded to multichannel by single channel, is significantly carried
The performance of tracking is risen.But the anti-ability of blocking of tradition KCF track algorithms is poor, is easy to generate when tracking target and being blocked
The phenomenon that tracking drift or tracking failure, and target can not be given for change again after tracking failure, so proposing a kind of improved
Algorithm promotes anti-the blocking property of KCF.
Invention content
In view of the deficiencies of the prior art, the present invention proposes one kind and being directed to the improved KCF track algorithms of target occlusion,
The algorithm adds on traditional KCF track algorithms blocks the detection algorithm that decision algorithm fails with target following to promote original
Anti- the blocking property of algorithm.
To achieve the above object, the present invention adopts the following technical scheme that:
For the improved KCF track algorithms of target occlusion, it includes,
S1, tracking Target Acquisition, initialize tracker, read video information, with video first frame image and tracking mesh
Frame initialized target tracker is marked, a large amount of sample, training grader are obtained using the method for circular matrix;
S2, judge to track whether target is blocked, if tracking target is blocked, goes to step S3, otherwise go to step
Rapid S6;
S3, judge whether seriously to block, if it is, going to step S4, otherwise go to step S5;
S4, judge whether tracking failure, if tracking failure, repositions tracking target, and initialize tracker, and
Step S6 is gone to, step S5 is otherwise gone to;
S5, stop classifier training;
S6, continue to track target.
The technical program further optimizes, and judges that target is using using the peak value of response of grader in the step S2
It is no to be blocked, when the peak value of response of grader is not less than preset value a, does not block, otherwise block.
The technical program further optimizes, when the peak value of response of grader is big less than preset value a in the step S3
When preset value b, it is determined as partial occlusion, goes to step S5;When the peak value of response of grader is not more than preset value b, then sentence
It is set to and seriously blocks, goes to step S4.
The technical program further optimizes, and whether the step S4 tracks failure using Pasteur's range estimation, that is, uses
Pasteur's distance of present frame and the grey level histogram of previous frame target following frame judges tracking result.
The technical program further optimizes, and the step S4 repositions tracking target using template matching algorithm.
The technical program further optimizes, and the template matching algorithm is difference of two squares matching method.
It is different from the prior art, above-mentioned technical proposal has the following advantages that:
A the peak value of response for) introducing grader judges whether target is blocked, if detecting that target is partly hidden
Gear stops the training to grader.If target is seriously blocked, former frame and present frame target following frame intensity histogram are used
Whether Pasteur's range estimation of figure tracks failure.
B) if Pasteur's distance is more than the threshold value of setting, it is determined as tracking failure, is given for change again using template matching algorithm
Track target.After target is given for change again, KCF trackers are initialized, continue effectively to track target.
Description of the drawings
Fig. 1 is for the improved KCF track algorithms flow chart of target occlusion;
Fig. 2 be KCF track algorithms (black dotted lines) with improve track algorithm (solid black lines) target be at least partially obscured with
And the peak response curve graph seriously blocked;
Fig. 3 be KCF track algorithms (black dotted lines) with improve track algorithm (solid black lines) target be at least partially obscured with
And the Pasteur's distance Curve figure seriously blocked
Fig. 4 is KCF algorithms (black dotted lines) and innovatory algorithm (solid black lines) center error curve diagram;
Fig. 5 is KCF innovatory algorithms and original KCF algorithms in Coke, FaceOcc2, SUV, Jogging, plate armour vehicle model, 5
Experience blocks the tracking result that several frames are representative on data set in various degree.
Specific implementation mode
For the technology contents of technical solution, construction feature, the objects and the effects are described in detail, below in conjunction with specific reality
It applies example and attached drawing is coordinated to be explained in detail.
Referring to Fig. 1, the present invention proposes that one kind being directed to the improved KCF track algorithms of target occlusion, it includes the following steps,
S1, tracking Target Acquisition, initialize tracker, read video information, with video first frame image and tracking mesh
Frame initialized target tracker is marked, a large amount of sample, training grader are obtained using the method for circular matrix;
S2, judge to track whether target is blocked, if tracking target is blocked, goes to step S3, otherwise go to step
Rapid S6;
S3, judge whether seriously to block, if it is, going to step S4, otherwise go to step S5;
S4, judge whether tracking failure, if tracking failure, repositions tracking target, and initialize tracker, and
Step S6 is gone to, step S5 is otherwise gone to;
S5, stop classifier training;
S6, continue to track target, obtain video information.
The tracker of KCF algorithms is a kind of tracker of the dense sampling based on detection, using core ridge regression grader as
Core carries out shifting function in target area using the method for circular matrix, forms a large amount of sample to train grader.Pass through
Kernel function calculates the similarity of selection region and target area, and the region of peak response is as new tracking target.In addition should
Algorithm carries out ingenious transformation to training sample, makes it have the characteristic of circular matrix, using discrete Fourier transform diagonalization, subtracts
The storage and calculating of few several orders of magnitude, significantly improve the speed of service of algorithm, are quickly and effectively examined to target to reach
Survey the purpose with tracking.
A present invention preferably embodiment, for the improved KCF track algorithms of target occlusion, it includes the following steps,
S1, tracking Target Acquisition, initialize tracker, read video information, with video first frame image and tracking mesh
Frame initialized target tracker is marked, a large amount of sample, training grader are obtained using the method for circular matrix.
Fast target detects, and region of the grader corresponding to the element of region of interest domain response maximum value is to track mesh
Mark.The method that circular matrix is still used after obtaining target obtains great amount of samples, continues to be trained grader.
S2, judge to track whether target is blocked, if tracking target is blocked, goes to step S3, otherwise go to step
Rapid S6.
The judgement for blocking algorithm judges different coverage extents using the peak value of response of grader.Track target
When being blocked, still it is unreasonable with the sample training grader of mistake, target occlusion determination method is added, can subtract
Few training of the error sample to grader, to the success rate of boosting algorithm tracking.
Traditional KCF track algorithms, when tracking target is obscured by an object, the image in target frame becomes shelter, still
Algorithm still uses the sample training grader of mistake, so this is unreasonable.In improved KCF track algorithms, KCF algorithms are utilized
In principle the peak response of grader as whether the Judging index blocked, grader response it is bigger explanation more may be
Target, response is smaller to illustrate that target is blocked.Grader peak response numerical value is exported and is saved as the form of text document,
It is analyzed with actual tracking process in conjunction with the peak response numerical value of output.It is found by many experiments, it can be by setting two
Different threshold values determines degree that target is blocked.Set preset value a=0.32 and preset value b=0.21, specific judgement side
Method is as shown in table 1:
Table 1 blocks determination method
Decision condition | Judge result |
Peak response>=0.32 | It does not block |
0.21<Peak response<0.32 | Partial occlusion |
Peak response<=0.21 | Seriously block |
When detecting that target is at least partially obscured, stop the training to grader.When detecting that target is seriously blocked,
Judge to track whether target tracks failure using Pasteur's distance.
S3, judge whether seriously to block, if it is, going to step S4, otherwise go to step S5.
S4, judge to track whether target tracks failure using Pasteur's distance, if tracking failure, repositions tracking
Target, and tracker is initialized, and step S6 is gone to, otherwise go to step S5.
Tracking unsuccessfully judges, using Pasteur's distance of present frame and the grey level histogram of previous frame target following frame to tracking
As a result judged, if detecting that target following fails, tracking target is given for change using the method for template matches, is used in combination
The target frame newly obtained reinitializes tracker with present frame, it is therefore an objective to continue subsequent tracking.
Tracking failure determination method
Pasteur's distance in statistics, measurement be two discrete or continuous probability distribution similitude, Pasteur's coefficient can
It is measured for the correlation to two groups of samples, the separability between class is often measured in classification, is generally used for two
Specific things model carries out the calculating of similarity, often applies the detection and tracking in target.
According to statistical concept, former frame is counted respectively and is used for doing histogram with present frame target following frame grey level histogram
The similarity-rough set of figure, manner of comparison use Pasteur's distance of normal distribution comparison, and value is bigger to illustrate image difference in target frame
Bigger, tracking result is bad, and value is smaller to illustrate that image difference is smaller in target frame, and tracking result is good.By Pasteur's distance values
The form for exporting and saving as text document is analyzed in conjunction with Pasteur's distance values and the actual tracking result of output.Pass through
Many experiments are found, can judge the tracking result of target by given threshold, and the threshold value set is 0.25.
It is using Pasteur's range estimation target after detecting target by serious block according to target occlusion decision algorithm
No tracking failure, specific determination method are as shown in table 2:
2 tracking result determination method of table
Decision condition | Judge result |
Pasteur's distance>=0.25 | Tracking failure |
Pasteur's distance<0.25 | It tracks successfully |
If target following fails, tracking target is given for change again using the method for template matches, and initialize KCF
Algorithm keeps track device continues effectively to track target.
Template matching algorithm
Template matches are exactly to be found in piece image and the most like region of template image, this method principle simple computation
Speed is fast, can be applied to target identification, the multiple fields such as target following.
Algorithm is realized:Over an input image according to from left to right, direction from top to bottom is traversed target template, is found
The similarity of each band of position and template image obtains best match coordinate points, finally according to match point and template image
Rectangle frame marks matching area.
For the tracking environmental of the embodiment, difference of two squares matching method has been selected after tested.
It is realized by the method for template matches in innovatory algorithm and tracking target is re-recognized and positioned.
S5, stop classifier training.
S6, continue to track target, fail with tracking if target occlusion is not detected, the method by constructing kernel function,
Target is effectively tracked using Hough feature.
Experimental result and analysis
In order to verify the validity of this algorithm, the video sequence that target is at least partially obscured in Benchmark data sets is chosen
Coke, FaceOcc2, Woman, video sequence SUV, Jogging and the armoring vehicle model of oneself shooting that target is seriously blocked
Video is tested, and is compared with traditional KCF track algorithms.
Experimental situation and parameter
Experiment porch Visual Studio 2012 configure opencv-2.4.9, and all experiments are in Intel (R) Core
(TM)i3-4310
CPU is completed on the configuration computer of dominant frequency 3.4GHz, 4GB memory.
Peak response data analysis
Fig. 2 be KCF track algorithms (black dotted lines) with improve track algorithm (solid black lines) target be at least partially obscured with
And the peak response curve graph seriously blocked.
As shown in the figure:Just start target not to be blocked, grader response is bigger, and curve is without too big wave
It is dynamic.When blocking, grader response can drastically decline, and big fluctuation occurs in curve, for this phenomenon, innovatory algorithm
Different coverage extents is distinguished by setting two different threshold values.It seriously blocks in video sequence, solid black lines occur steep
Situation is risen, is because innovatory algorithm successfully gives the target with losing for change, and initializes tracker, so peak response numerical value can dash forward
So become larger.
Pasteur's range data is analyzed
Fig. 3 be KCF track algorithms (black dotted lines) with improve track algorithm (solid black lines) target be at least partially obscured with
And the Pasteur's distance Curve figure seriously blocked.
As shown in the figure:In the case of just beginning target following is successful, Pasteur's distance values are smaller, when target is seriously hidden
When gear leads to tracking failure, big fluctuation occurs for Pasteur's distance values meeting rapid increase, curve.For this phenomenon, by setting
Threshold value is determined to judge the result of tracking.It seriously blocks in video sequence, the situation that skyrockets occur in solid black lines, and are because of innovatory algorithm
Target with losing successfully is given for change, target frame is become correctly tracking target from the sample of mistake, so Pasteur's distance values meeting
Become larger suddenly.
Performance Evaluation
In order to assess the performance for improving track algorithm, service precision, success rate, three indexs of processing speed are as judge mark
It is accurate.
Precision
One index for being widely used in evaluation precision is center error, i.e. realistic objective tracking box and manually mark mesh
Mark the distance between center.
Fig. 4 is KCF algorithms (black dotted lines), innovatory algorithm (solid black lines) center error curve diagram.
Table 3 is the average central error amount of different video sequence, and numerical value is smaller, and to represent tracking accuracy higher.From Fig. 4
In and list data find out that the center error of innovatory algorithm is respectively less than original KCF track algorithms, illustrates that innovatory algorithm has
Higher tracking accuracy.
3 average position error of table
Success rate
Table 4 is the tracking success rate of each video sequence, and numerical value is bigger to illustrate that tracking result is better.As can be seen from the table,
The tracking success rate of innovatory algorithm is higher than original KCF algorithms, illustrates that innovatory algorithm has better tracking result.
Table 4 tracks success rate
Video sequence | Former KCF algorithms | Innovatory algorithm |
Coke | 0.153 | 0.579 |
FaceOcc2 | 0.724 | 0.776 |
Woman | 0.884 | 0.905 |
SUV | 0.332 | 0.452 |
Jogging | 0.101 | 0.475 |
Processing speed
Table 5 shows that the requirement of real-time can be met by improving the processing speed of KCF track algorithms, have certain reality
Using.
5 algorithm process speed of table
Video sequence | Image type | Image pixel | Processing speed (FPS) |
Coke | It is colored | 640×480 | 23.6 |
FaceOcc2 | Gray scale | 320×240 | 21.1 |
Woman | It is colored | 352×288 | 29.9 |
SUV | Gray scale | 320×240 | 31.2 |
Jogging | It is colored | 352×288 | 29.6 |
Armoring vehicle model | It is colored | 1280×720 | 17.3 |
Qualitative comparison
(a) of Fig. 5, (b), (c), (d) and (e) respectively illustrate innovatory algorithm and original KCF algorithms Coke,
FaceOcc2, SUV, Jogging, armoring vehicle model, 5 experience block in various degree several frames on data set it is representative with
Track result.Lastrow picture is shown as the tracking result of the original KCF algorithms of the first behavior in wherein Fig. 5, and next line picture is shown
For the tracking result of the second behavior innovatory algorithm.
In Coke, FaceOcc2 data set, since target is at least partially obscured, there is showing for tracking drift in former KCF algorithms
As target frame becomes large in size.Innovatory algorithm lives through after blocking several times, remains able to be accurately positioned and arrives target.SUV,
Jogging, the trolley video data oneself shot are concentrated, and target is seriously blocked, and former KCF track algorithms are in target by blocking
Behind region, there is the phenomenon that tracking failure, hereafter successfully can not be positioned and be tracked again.Innovatory algorithm is in SUV data sets
The 578th frame, the 112nd frame of Jogging data sets, shoot trolley data set 235 frames, target can be given for change again
And it effectively tracks.It can be seen from the above result that this paper algorithms show good robustness for blocking video sequence.
For traditional KCF target tracking algorisms, since target is blocked, there is tracking drift or the phenomenon that fails, and with
The problem of target can not be given for change again after track failure, the present invention propose a kind of for the improved KCF tracking calculation of target following
Method.Target occlusion determination method, tracking failure determination method is added in the algorithm, and template matches target relocates algorithm, promoted former
Anti- the blocking property of KCF track algorithms.Using Benchmark data sets and oneself shooting trolley video, to innovatory algorithm into
Row verification.By experimental result it is found that inventive algorithm is for video sequence is blocked has stronger robustness.
It should be noted that herein, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, are deposited without necessarily requiring or implying between these entities or operation
In any actual relationship or order or sequence.Moreover, the terms "include", "comprise" or its any other variant are intended to
Non-exclusive inclusion, so that process, method, article or terminal device including a series of elements include not only those
Element, but also include other elements that are not explicitly listed, or further include for this process, method, article or end
The intrinsic element of end equipment.In the absence of more restrictions, being limited by sentence " including ... " or " including ... "
Element, it is not excluded that there is also other elements in process, method, article or the terminal device including the element.This
Outside, herein, " being more than ", " being less than ", " being more than " etc. are interpreted as not including this number;" more than ", " following ", " within " etc. understandings
It includes this number to be.
Although the various embodiments described above are described, once a person skilled in the art knows basic wounds
The property made concept, then additional changes and modifications can be made to these embodiments, so example the above is only the implementation of the present invention,
It is not intended to limit the scope of patent protection of the present invention, it is every to utilize equivalent structure made by description of the invention and accompanying drawing content
Or equivalent process transformation, it is applied directly or indirectly in other relevant technical fields, the patent for being similarly included in the present invention
Within protection domain.
Claims (6)
1. one kind being directed to the improved KCF track algorithms of target occlusion, it is characterised in that:It includes,
S1, tracking Target Acquisition, initialize tracker, read video information, with video first frame image and tracking target frame
Initialized target tracker obtains a large amount of sample, training grader using the method for circular matrix;
S2, judge to track whether target is blocked, if tracking target is blocked, goes to step S3, otherwise go to step S6;
S3, judge whether seriously to block, if it is, going to step S4, otherwise go to step S5;
S4, judge whether tracking failure, if tracking failure, repositions tracking target, and initialize tracker, and go to
Otherwise step S6 goes to step S5;
S5, stop classifier training;
S6, continue to track target.
2. as described in claim 1 a kind of for the improved KCF track algorithms of target occlusion, it is characterised in that:The step
Using judging whether target is blocked using the peak value of response of grader in S2, when the peak value of response of grader is not less than preset value
It when a, does not block, otherwise blocks.
3. as claimed in claim 2 a kind of for the improved KCF track algorithms of target occlusion, it is characterised in that:The step
In S3 when the peak value of response of grader, which is less than preset value a, is more than preset value b, it is determined as partial occlusion, goes to step S5;When point
When the peak value of response of class device is not more than preset value b, then it is judged to seriously blocking, goes to step S4.
4. as described in claim 1 a kind of for the improved KCF track algorithms of target occlusion, it is characterised in that:The step
Whether S4 tracks failure using Pasteur's range estimation, that is, uses bar of present frame and the grey level histogram of previous frame target following frame
Family name's distance judges tracking result.
5. as described in claim 1 a kind of for the improved KCF track algorithms of target occlusion, it is characterised in that:The step
S4 repositions tracking target using template matching algorithm.
6. as claimed in claim 5 a kind of for the improved KCF track algorithms of target occlusion, it is characterised in that:The template
Matching algorithm is difference of two squares matching method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810624972.6A CN108765470A (en) | 2018-06-17 | 2018-06-17 | One kind being directed to the improved KCF track algorithms of target occlusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810624972.6A CN108765470A (en) | 2018-06-17 | 2018-06-17 | One kind being directed to the improved KCF track algorithms of target occlusion |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108765470A true CN108765470A (en) | 2018-11-06 |
Family
ID=63978392
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810624972.6A Pending CN108765470A (en) | 2018-06-17 | 2018-06-17 | One kind being directed to the improved KCF track algorithms of target occlusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108765470A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109993769A (en) * | 2019-03-07 | 2019-07-09 | 安徽创世科技股份有限公司 | A kind of multiple-target system of deep learning SSD algorithm combination KCF algorithm |
CN110189365A (en) * | 2019-05-24 | 2019-08-30 | 上海交通大学 | It is anti-to block correlation filtering tracking |
CN110400347A (en) * | 2019-06-25 | 2019-11-01 | 哈尔滨工程大学 | A kind of method for tracking target for judging to block and target relocates |
CN110555866A (en) * | 2019-08-07 | 2019-12-10 | 北京首贝科技发展有限公司 | Infrared target tracking method for improving KCF feature descriptor |
CN111340846A (en) * | 2020-02-25 | 2020-06-26 | 重庆邮电大学 | Multi-feature fusion anti-occlusion target tracking method |
CN111354022A (en) * | 2020-02-20 | 2020-06-30 | 中科星图股份有限公司 | Target tracking method and system based on kernel correlation filtering |
CN111462186A (en) * | 2020-04-03 | 2020-07-28 | 天津理工大学 | Infrared target detection and tracking integrated algorithm based on extension immunity |
CN111476825A (en) * | 2020-03-10 | 2020-07-31 | 重庆邮电大学 | Anti-occlusion target tracking method based on multi-example learning and kernel correlation filter |
CN112633105A (en) * | 2020-12-15 | 2021-04-09 | 重庆电子工程职业学院 | Target tracking and counting system and method |
CN113344964A (en) * | 2021-06-23 | 2021-09-03 | 江苏三恒科技股份有限公司 | Image processing-based mine robot rockfall monitoring and early warning method |
WO2021139787A3 (en) * | 2020-01-09 | 2023-03-02 | 影石创新科技股份有限公司 | Target tracking method for panoramic video, readable storage medium, and computer device |
CN116109975A (en) * | 2023-02-08 | 2023-05-12 | 广州宝立科技有限公司 | Power grid safety operation monitoring image processing method and intelligent video monitoring system |
CN116228817A (en) * | 2023-03-10 | 2023-06-06 | 东南大学 | Real-time anti-occlusion anti-jitter single target tracking method based on correlation filtering |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7813528B2 (en) * | 2007-04-05 | 2010-10-12 | Mitsubishi Electric Research Laboratories, Inc. | Method for detecting objects left-behind in a scene |
CN106204638A (en) * | 2016-06-29 | 2016-12-07 | 西安电子科技大学 | A kind of based on dimension self-adaption with the method for tracking target of taking photo by plane blocking process |
CN107657630A (en) * | 2017-07-21 | 2018-02-02 | 南京邮电大学 | A kind of modified anti-shelter target tracking based on KCF |
CN108010067A (en) * | 2017-12-25 | 2018-05-08 | 北京航空航天大学 | A kind of visual target tracking method based on combination determination strategy |
-
2018
- 2018-06-17 CN CN201810624972.6A patent/CN108765470A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7813528B2 (en) * | 2007-04-05 | 2010-10-12 | Mitsubishi Electric Research Laboratories, Inc. | Method for detecting objects left-behind in a scene |
CN106204638A (en) * | 2016-06-29 | 2016-12-07 | 西安电子科技大学 | A kind of based on dimension self-adaption with the method for tracking target of taking photo by plane blocking process |
CN107657630A (en) * | 2017-07-21 | 2018-02-02 | 南京邮电大学 | A kind of modified anti-shelter target tracking based on KCF |
CN108010067A (en) * | 2017-12-25 | 2018-05-08 | 北京航空航天大学 | A kind of visual target tracking method based on combination determination strategy |
Non-Patent Citations (1)
Title |
---|
范曲: ""复杂场景下运动目标的视觉跟踪方法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109993769B (en) * | 2019-03-07 | 2022-09-13 | 安徽创世科技股份有限公司 | Multi-target tracking system combining deep learning SSD algorithm with KCF algorithm |
CN109993769A (en) * | 2019-03-07 | 2019-07-09 | 安徽创世科技股份有限公司 | A kind of multiple-target system of deep learning SSD algorithm combination KCF algorithm |
CN110189365A (en) * | 2019-05-24 | 2019-08-30 | 上海交通大学 | It is anti-to block correlation filtering tracking |
CN110189365B (en) * | 2019-05-24 | 2023-04-07 | 上海交通大学 | Anti-occlusion correlation filtering tracking method |
CN110400347A (en) * | 2019-06-25 | 2019-11-01 | 哈尔滨工程大学 | A kind of method for tracking target for judging to block and target relocates |
CN110400347B (en) * | 2019-06-25 | 2022-10-28 | 哈尔滨工程大学 | Target tracking method for judging occlusion and target relocation |
CN110555866A (en) * | 2019-08-07 | 2019-12-10 | 北京首贝科技发展有限公司 | Infrared target tracking method for improving KCF feature descriptor |
WO2021139787A3 (en) * | 2020-01-09 | 2023-03-02 | 影石创新科技股份有限公司 | Target tracking method for panoramic video, readable storage medium, and computer device |
CN111354022A (en) * | 2020-02-20 | 2020-06-30 | 中科星图股份有限公司 | Target tracking method and system based on kernel correlation filtering |
CN111354022B (en) * | 2020-02-20 | 2023-08-22 | 中科星图股份有限公司 | Target Tracking Method and System Based on Kernel Correlation Filtering |
CN111340846A (en) * | 2020-02-25 | 2020-06-26 | 重庆邮电大学 | Multi-feature fusion anti-occlusion target tracking method |
CN111340846B (en) * | 2020-02-25 | 2023-02-17 | 重庆邮电大学 | Multi-feature fusion anti-occlusion target tracking method |
CN111476825B (en) * | 2020-03-10 | 2022-08-26 | 重庆邮电大学 | Anti-occlusion target tracking method based on multi-example learning and kernel correlation filter |
CN111476825A (en) * | 2020-03-10 | 2020-07-31 | 重庆邮电大学 | Anti-occlusion target tracking method based on multi-example learning and kernel correlation filter |
CN111462186B (en) * | 2020-04-03 | 2022-04-15 | 天津理工大学 | Infrared target detection and tracking integrated algorithm based on extension immunity |
CN111462186A (en) * | 2020-04-03 | 2020-07-28 | 天津理工大学 | Infrared target detection and tracking integrated algorithm based on extension immunity |
CN112633105A (en) * | 2020-12-15 | 2021-04-09 | 重庆电子工程职业学院 | Target tracking and counting system and method |
CN113344964A (en) * | 2021-06-23 | 2021-09-03 | 江苏三恒科技股份有限公司 | Image processing-based mine robot rockfall monitoring and early warning method |
CN113344964B (en) * | 2021-06-23 | 2024-02-23 | 江苏三恒科技股份有限公司 | Mine robot falling stone monitoring and early warning method based on image processing |
CN116109975A (en) * | 2023-02-08 | 2023-05-12 | 广州宝立科技有限公司 | Power grid safety operation monitoring image processing method and intelligent video monitoring system |
CN116109975B (en) * | 2023-02-08 | 2023-10-20 | 广州宝立科技有限公司 | Power grid safety operation monitoring image processing method and intelligent video monitoring system |
CN116228817A (en) * | 2023-03-10 | 2023-06-06 | 东南大学 | Real-time anti-occlusion anti-jitter single target tracking method based on correlation filtering |
CN116228817B (en) * | 2023-03-10 | 2023-10-03 | 东南大学 | Real-time anti-occlusion anti-jitter single target tracking method based on correlation filtering |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108765470A (en) | One kind being directed to the improved KCF track algorithms of target occlusion | |
CN108447080B (en) | Target tracking method, system and storage medium based on hierarchical data association and convolutional neural network | |
US8340420B2 (en) | Method for recognizing objects in images | |
CN104574445B (en) | A kind of method for tracking target | |
CN102496016B (en) | Infrared target detection method based on space-time cooperation framework | |
CN109785366B (en) | Related filtering target tracking method for shielding | |
CN108961308B (en) | Residual error depth characteristic target tracking method for drift detection | |
CN110728697A (en) | Infrared dim target detection tracking method based on convolutional neural network | |
CN109871763A (en) | A kind of specific objective tracking based on YOLO | |
CN108921873A (en) | The online multi-object tracking method of Markovian decision of filtering optimization is closed based on nuclear phase | |
CN104794729B (en) | SAR image change detection based on conspicuousness guiding | |
CN110555870B (en) | DCF tracking confidence evaluation and classifier updating method based on neural network | |
CN103455820A (en) | Method and system for detecting and tracking vehicle based on machine vision technology | |
CN104680545B (en) | There is the detection method of well-marked target in optical imagery | |
CN110991397B (en) | Travel direction determining method and related equipment | |
CN103500453B (en) | Based on the SAR image salient region detection method of Gamma distribution and neighborhood information | |
CN104751185A (en) | SAR image change detection method based on mean shift genetic clustering | |
CN107679469A (en) | A kind of non-maxima suppression method based on deep learning | |
Zhou et al. | Autonomous detection of crop rows based on adaptive multi-ROI in maize fields. | |
CN103198482A (en) | Remote sensing image change detection method based on difference chart fuzzy membership degree integration | |
CN104680536A (en) | Method for detecting SAR image change by utilizing improved non-local average algorithm | |
CN104537384B (en) | A kind of SAR target discrimination methods of combination likelihood ratio judgement | |
CN110766657A (en) | Laser interference image quality evaluation method | |
CN102148919B (en) | Method and system for detecting balls | |
CN114419140A (en) | Positioning algorithm for light spot center of track laser measuring device |
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 |