CN108765455A - A kind of target tenacious tracking method based on TLD algorithms - Google Patents
A kind of target tenacious tracking method based on TLD algorithms Download PDFInfo
- Publication number
- CN108765455A CN108765455A CN201810506760.8A CN201810506760A CN108765455A CN 108765455 A CN108765455 A CN 108765455A CN 201810506760 A CN201810506760 A CN 201810506760A CN 108765455 A CN108765455 A CN 108765455A
- Authority
- CN
- China
- Prior art keywords
- tracking
- module
- target
- frame
- image block
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The target tenacious tracking method based on TLD algorithms that the invention discloses a kind of, includes the following steps：(1) initialization training is carried out in the start frame of video to be tracked.(2) during tracking, detection module and tracking module work independently：Detection module scanning current frame image obtains image block and passes sequentially through variance grader, merges grader, nearest neighbor classifier, is then clustered to image block；Tracking module is using intermediate value optical flow method by previous frame tracking result prediction present frame target location, and calculate the Euclidean distance D between the two frames target location central point, if D is more than an adaptive threshold value, judge that present frame tracking failure, tracking module do not export any result；(3) module output tracking result is integrated；(4) to current updated target location, new positive negative sample is generated, to update detection module.(4) cycle (2)-(3) is until tracking terminates.This method can improve the stability of tracking to a certain extent.
Description
Technical field
The target tenacious tracking method based on TLD algorithms that the present invention relates to a kind of, feature are carried out to failure detection mechanism
Adaptive threshold value setting, is applied to computer vision, target detection, target following etc., belongs to the mesh in computer vision
Mark tracking field.
Background technology
TLD track algorithms are a kind of single goal long-time track algorithms proposed by Zdenka Kalal.The algorithm by with
Three track module, detection module, study module module compositions.Simple track algorithm be difficult correction tracking drift error and
Will continue to accumulate the error of tracking, once and target disappear from the visual field, tracking be just inevitably generated failure.Simple
Detection algorithm needs a large amount of sample to carry out offline supervised training, may not apply to the tracing task of unknown object, and by
It is established offline in object module, once so if target varies widely, tracking is just easy to generate failure.TLD
Detection algorithm and track algorithm are combined and by learning come real-time update object module.
For target tracking algorism, target is blocked or is disappeared to completely behind the visual field the case where occurring again and passed through
Can often it occur.Therefore can target tracking algorism unsuccessfully make just to tracking caused by due to blocking or disappearing to the visual field completely
True judgement is particularly important：If target tracking algorism cannot make accurate judgment, even when target is not in video,
Target tracking algorism also will continue to update object module.Obvious this contaminated object module can not carry out target effective
Characterization.Therefore TLD algorithms in tracking module add failure detection mechanism to judge whether target disappears or hidden completely
Gear.In the video frame for being determined tracking failure, the study module of TLD algorithms will not carry out model modification to it, in this way it is avoided that
Object module is contaminated.But in TLD algorithms, the threshold value of tracking failure detection mechanism is set to a fixed value,
It is difficult to adapt to various target following scenes.Especially in the case where quickly movement occurs for target, due to tracking failure detection machine
The threshold value of system is less than the movement of target, and failure detection mechanism can judge that tracking failure has occurred in present frame, this is a kind of mistake
Judge.
Invention content
The invention solves technical problems to be：For the failure inspection of TLD algorithms tracker when quickly movement occurs for target
The erroneous judgement of survey mechanism causes the problem of tracking failure, it is proposed that adaptive tracking failure detection mechanism passes through adaptive threshold
Adjustment overcomes the problems, such as the erroneous judgement of original failure detection mechanism.The experiment carried out in open sets of video data shows that this method can
To improve the stability of tracking to a certain extent.
The present invention solve the technical solution that uses of above-mentioned technical problem for：A kind of target tenacious tracking based on TLD algorithms
Method is specified tracking window to form positive negative sample and is initialized to detection module in the start frame of video to be tracked by user
Training.During tracking, detection module and tracking module work independently：Detection module scans current frame image and obtains image block
And it passes sequentially through variance grader, merge grader, nearest neighbor classifier.Tracking module passes through previous frame using intermediate value optical flow method
Tracking prediction present frame target location.It integrates module synthesis detection module and tracking module carries out the output of tracking result.And
To current updated target location, new positive negative sample is generated, to update detection module.
Wherein, in the start frame of video to be tracked, tracking window is specified by user, specified track window of then adjusting the distance
Several windows are chosen in mouthful nearest scanning grid window carry out a series of affine transformation and form initial positive sample, and right
Choosing is searched at random far from specified tracking window obtains initial negative sample.The positive and negative initial sample obtained is used for detection module
Carry out initialization training.
Wherein, it during tracking, after detection module carries out network scanning acquisition image block to current frame image, counts first
The variance of each image block is calculated, the image block that variance is less than some threshold value is received, into merging grader.Pass through several differences
The average posterior probability values that obtain more afterwards of pixel that carry out of basic classification device, the image block more than some threshold value received,
Into nearest neighbor classifier.By carrying out the zero-mean normalized of gray scale to the image block for entering nearest neighbor classifier, with
Image block in object module carries out the normalized similarity calculation of cross-correlation, if similarity is more than some threshold value, judges
Current image block is target area, is otherwise determined as background.
Wherein, during tracking, be utilized N frames before present frame for information about to original tracking failure detection mechanism
Carry out adaptive threshold value setting.Wherein, in the initialization of tracking failure detection mechanism, the tracking of the preceding N frames of entire video
Failure detection threshold value is arranged to a higher value, that is, it is not in tracking failure to give tacit consent in the preceding N frames of video.
Wherein, during tracking, tracking module and detection module independent operating, and use adaptive tracing failure detection
Testing result is finally merged output target following result by mechanism into line trace failure detection with tracking result.In each frame
To current updated target location, new positive negative sample is generated, to update object module and detection module.
Compared with prior art, the beneficial effects of the invention are as follows：
This method can be adaptively adjusted the threshold value of tracking failure detection mechanism, to make target tracking algorism in mesh
When different motion occurs for mark, can unsuccessfully it be made correctly being blocked completely as target or being tracked caused by leaving the visual field
Judge, and then realizes more stable tracking.
Description of the drawings
Point coordinates calculates schematic diagram centered on Fig. 1；
Fig. 2 is to successfully track the variation of frame number in the case of different N values；
Fig. 3 is the comparison diagram of the success rate (Pascal score) of experimental data set；
Specific implementation mode
Opinion specific implementation mode further illustrates the present invention below in conjunction with the accompanying drawings.
Defining the distance between rectangle frame 1 and rectangle frame 2 first is：
Wherein (x1,y1)、(x2,y2) be respectively rectangle frame 1 and rectangle frame 2 center point coordinate.Define two rectangle frames
The distance between for the Euclidean distance between two rectangle frame center point coordinates.
For rectangle frame as shown in Figure 1, center point coordinate (x0,y0) calculation formula be：
Based in the case where target does not occur being blocked completely or leaving the visual field, the movement of target should be continuous
The fact that, it is considered herein that the distance between the rectangle frame of preceding N frames tracking result can reflect mesh in present frame i (i=N+1)
Target movement degree, so as to carry out the adaptive setting of threshold value based on this distance.In order to reduce the shadow of tracker mistake
It rings, the present invention sets the threshold value of present frame using the average value of distance between preceding N frames tracking result rectangle frame.
In video frame i (i > N), the threshold θ of failure detection is trackedfailureIt is defined as shown in formula (3)：
Wherein α is adjustment factor, is defined as：
It is initialized as 1.α is described in detail in extended meeting afterwards.
In video frame i (3≤i≤N), since tracking result is less, in order to avoid track it is unstable caused by mistake, this
Invention shown in the formula (5) in a manner of to tracking the threshold θ of failure detectionfailureIt is configured：
Wherein θc=10 be the acquiescence threshold residual value of TLD algorithms, and α is adjustment factor, is initialized as 1.
In video frame i (1≤i≤2), it can be arranged for reference without data.In view of in actual video frame, target
Movement be continuous, the visual field just will not be blocked or be disappeared to completely to target in the second frame.Based on such a fact, originally
Invention think be not in video frame i (1≤i≤2) caused by due to target is blocked or disappeared to completely the visual field with
Track fails.Therefore, the present invention comes to carry out Initialize installation to tracking failure detection threshold value with formula (6)：
θfailure=10 θc ...(6)
Wherein θc=10 be the acquiescence threshold residual value of TLD algorithms.
Occur tracking failure after, tracking module do not export it is any as a result, study module pause object module update until
Detection module detects successfully in global detection and resets tracking module, and tracking module can just restart.In view of this feelings
The setting of condition, adaptive threshold of the invention on the basis of the setting of the threshold value of former frame, is adjusted after tracking failure occurs for target
The value of whole adjustment factor α is 1.2, i.e., appropriate to increase tracking failure detection threshold value, occurs quickly to move suddenly to avoid because of target
When caused by failure detection mechanism judge by accident.
As previously mentioned, adaptive failure detection threshold value is set as：
Wherein θc=10 be the acquiescence threshold residual value of TLD algorithms, and α ∈ { 1,1.2 } are adjustment factor.
It is tested using Deer data sets, is carried out under the value of { 3,4,5,6,7,8,9,10,11,12,13 } N ∈
Test, to find the setting that a rational value completes the failure detection mechanism of adaptive threshold adjustment.In fact N can take
Value is 1 or 2, i.e., is configured to threshold value using the distance between tracking result rectangle frame of preceding 1 frame or preceding 2 frame.But
There is larger contingency for very few data, therefore not 1 and 2 included in the practical value of N.In addition in view of N takes
Value conference the influence of excessive previous movement result is added, and it is this influence it is not necessarily positive, the value of N > 13 not by
Test is added.
Fig. 2 is the variation that frame number is successfully tracked in the case of different N values.Figure it is seen that with the increasing of N values
Greatly, it successfully tracks frame number and stablizes in N >=6 after rising and no longer change 65 or so.And totalframes is 71 in Deer data sets
Frame.It is the 27th that as N >=6, tracking, which occurs to repeat in the video frame of tracking failure, 28,31,32 frames, and this several frame is exactly entire
Target most motion intense in Deer data sets.Therefore it attempts to increase regulated values of the adjustment factor α after tracking failure occurs, i.e.,
Increase tracking failure detection threshold value when tracking failure occurs, successfully tracking frame number has a subtle promotion, but there are still with
Track fails.This is because inevitably failing caused by the definition of α itself.Because α is defined on generation in initial definition
Higher value is taken to adapt to the larger movement that target in present frame may occur when tracking failure, is also printed in actual test result
This point is demonstrate,proved.
Therefore, present invention selection can make the N values for successfully tracking the convergent minimum of frame number set formula (7).I.e. certainly
Adjustment failure detection threshold value θfailureFor：
Wherein N=6, θc=10,
It should be noted that the training Jing Guo target tracking result in the video frame extremely slowly moved, θfailureIt can quilt
It is trained for a very small value, then emergent quick movement, θfailureIt needs multiframe just and can be conditioned factor alpha to move to
One rational horizontal.
Therefore, the present invention adds the mechanism as shown in formula (9) to enhance the robustness of the above method in the above-mentioned methods：
θfailure=max (θfailure,θc)...(9)
Wherein θc=10.
The improvement of addition formula (9), BlurOwl, BlurBody, Deer of the selection comprising quickly movement (FastMotion),
Jumping, BlurCar2 data set are tested, and test result is as shown in table 1.
As can be seen from Table 1, method proposed by the present invention can significantly increase the tracking stability of TLD algorithms.
The Statistical Comparison result of 1 experimental data set of table
Part of that present invention that are not described in detail belong to the well-known technology of those skilled in the art.
Those of ordinary skill in the art it should be appreciated that more than embodiment be intended merely to illustrate the present invention,
And be not used as limitation of the invention, if in the spirit of the present invention, to embodiment described above variation,
Modification will all be fallen in the range of claims of the present invention.
Claims (5)
1. a kind of target tenacious tracking method based on TLD algorithms, it is characterized in that：In the start frame of video to be tracked, by user
Specified tracking window forms positive negative sample and carries out initialization training to detection module, during tracking, detection module and tracking
Module works independently：Detection module scanning current frame image obtain image block and pass sequentially through variance grader, merge grader,
Nearest neighbor classifier, and to being clustered by the image block of these three graders, tracking module is passed through using intermediate value optical flow method
Previous frame tracking result predicts present frame target location, and calculate present frame target location and previous frame target location central point it
Between Euclidean distance D, if D be more than an adaptive threshold value, judge present frame tracking failure, tracking module does not export any
As a result, integrating the output of module synthesis detection module and tracking module progress tracking result and to current updated target position
It sets, new positive negative sample is generated, to update detection module.
2. a kind of target tenacious tracking method based on TLD algorithms according to claim 1, it is characterized in that：To be tracked
In the start frame of video, tracking window is specified by user, the nearest scanning grid window of specified tracking window of then adjusting the distance
Middle several windows of selection carry out a series of affine transformation and form initial positive sample, and to far from specified tracking window with
Machine searches choosing and obtains initial negative sample, and the positive and negative initial sample obtained is used for carrying out initialization training to detection module.
3. a kind of target tenacious tracking method based on TLD algorithms according to claim 1, it is characterized in that：It was tracking
Cheng Zhong calculates the variance of each image block, variance first after detection module carries out network scanning acquisition image block to current frame image
Image block less than some threshold value is received, and into grader is merged, passes through the picture of basic classification device several different progress
The average posterior probability values that element obtains more afterwards, the image block more than some threshold value is received, into nearest neighbor classifier.Pass through
Image block to entering nearest neighbor classifier carries out the zero-mean normalized of gray scale, is carried out with the image block in object module
The normalized similarity calculation of cross-correlation judges current image block for target area if similarity is more than some threshold value, no
Then it is determined as background.
4. a kind of target tenacious tracking method based on TLD algorithms according to claim 1, it is characterized in that：It was tracking
Carrying out adaptive threshold value to original tracking failure detection mechanism for information about and setting for N frames before present frame is utilized in Cheng Zhong,
Wherein, in the initialization of tracking failure detection mechanism, the tracking failure detection threshold value of the preceding N frames of entire video is arranged to one
A higher value, that is, it is not in tracking failure to give tacit consent in the preceding N frames of video.
5. a kind of target tenacious tracking method based on TLD algorithms according to claim 1, it is characterized in that：It was tracking
Cheng Zhong, tracking module and detection module independent operating, and using adaptive tracing failure detection mechanism into line trace failure detection,
Testing result is finally merged to output target following with tracking result as a result, in each frame to current updated target position
It sets, new positive negative sample is generated, to update object module and detection module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810506760.8A CN108765455B (en) | 2018-05-24 | 2018-05-24 | Target stable tracking method based on TLD algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810506760.8A CN108765455B (en) | 2018-05-24 | 2018-05-24 | Target stable tracking method based on TLD algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108765455A true CN108765455A (en) | 2018-11-06 |
| CN108765455B CN108765455B (en) | 2021-09-21 |
Family
ID=64005313
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810506760.8A Active CN108765455B (en) | 2018-05-24 | 2018-05-24 | Target stable tracking method based on TLD algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108765455B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635657A (en) * | 2018-11-12 | 2019-04-16 | 平安科技（深圳）有限公司 | Method for tracking target, device, equipment and storage medium |
| CN109858526A (en) * | 2019-01-08 | 2019-06-07 | 沈阳理工大学 | Sensor-based multi-target track fusion method in a kind of target following |
| CN109917818A (en) * | 2019-01-31 | 2019-06-21 | 天津大学 | Collaboratively searching based on ground robot contains method |
| CN110472562A (en) * | 2019-08-13 | 2019-11-19 | 新华智云科技有限公司 | Position ball video clip detection method, device, system and storage medium |
| CN111627046A (en) * | 2020-05-15 | 2020-09-04 | 北京百度网讯科技有限公司 | Target part tracking method and device, electronic equipment and readable storage medium |
| CN113243026A (en) * | 2019-10-04 | 2021-08-10 | Sk电信有限公司 | Apparatus and method for high resolution object detection |
| CN113284167A (en) * | 2021-05-28 | 2021-08-20 | 深圳数联天下智能科技有限公司 | Face tracking detection method, device, equipment and medium |
| CN114782496A (en) * | 2022-06-20 | 2022-07-22 | 杭州闪马智擎科技有限公司 | Object tracking method and device, storage medium and electronic device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103871081A (en) * | 2014-03-29 | 2014-06-18 | 湘潭大学 | Method for tracking self-adaptive robust on-line target |
| CN106651909A (en) * | 2016-10-20 | 2017-05-10 | 北京信息科技大学 | Background weighting-based scale and orientation adaptive mean shift method |
| CN107679455A (en) * | 2017-08-29 | 2018-02-09 | 平安科技（深圳）有限公司 | Target tracker, method and computer-readable recording medium |
-
2018
- 2018-05-24 CN CN201810506760.8A patent/CN108765455B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103871081A (en) * | 2014-03-29 | 2014-06-18 | 湘潭大学 | Method for tracking self-adaptive robust on-line target |
| CN106651909A (en) * | 2016-10-20 | 2017-05-10 | 北京信息科技大学 | Background weighting-based scale and orientation adaptive mean shift method |
| CN107679455A (en) * | 2017-08-29 | 2018-02-09 | 平安科技（深圳）有限公司 | Target tracker, method and computer-readable recording medium |
Non-Patent Citations (1)
| Title |
|---|
| ZDENEK KALAL 等: "Tracking-Learning-Detection", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635657A (en) * | 2018-11-12 | 2019-04-16 | 平安科技（深圳）有限公司 | Method for tracking target, device, equipment and storage medium |
| CN109635657B (en) * | 2018-11-12 | 2023-01-06 | 平安科技（深圳）有限公司 | Target tracking method, device, equipment and storage medium |
| CN109858526A (en) * | 2019-01-08 | 2019-06-07 | 沈阳理工大学 | Sensor-based multi-target track fusion method in a kind of target following |
| CN109858526B (en) * | 2019-01-08 | 2023-08-18 | 沈阳理工大学 | Multi-target track fusion method based on sensor in target tracking |
| CN109917818A (en) * | 2019-01-31 | 2019-06-21 | 天津大学 | Collaboratively searching based on ground robot contains method |
| CN109917818B (en) * | 2019-01-31 | 2021-08-13 | 天津大学 | Collaborative search containment method based on ground robot |
| CN110472562A (en) * | 2019-08-13 | 2019-11-19 | 新华智云科技有限公司 | Position ball video clip detection method, device, system and storage medium |
| CN113243026A (en) * | 2019-10-04 | 2021-08-10 | Sk电信有限公司 | Apparatus and method for high resolution object detection |
| CN111627046A (en) * | 2020-05-15 | 2020-09-04 | 北京百度网讯科技有限公司 | Target part tracking method and device, electronic equipment and readable storage medium |
| CN113284167A (en) * | 2021-05-28 | 2021-08-20 | 深圳数联天下智能科技有限公司 | Face tracking detection method, device, equipment and medium |
| CN113284167B (en) * | 2021-05-28 | 2023-03-07 | 深圳数联天下智能科技有限公司 | Face tracking detection method, device, equipment and medium |
| CN114782496A (en) * | 2022-06-20 | 2022-07-22 | 杭州闪马智擎科技有限公司 | Object tracking method and device, storage medium and electronic device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108765455B (en) | 2021-09-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108765455A (en) | A kind of target tenacious tracking method based on TLD algorithms | |
| CN106204649B (en) | A kind of method for tracking target based on TLD algorithm | |
| US7982774B2 (en) | Image processing apparatus and image processing method | |
| CN103971386B (en) | A kind of foreground detection method under dynamic background scene | |
| CN106780557A (en) | A kind of motion target tracking method based on optical flow method and crucial point feature | |
| CN112184759A (en) | Moving target detection and tracking method and system based on video | |
| SanMiguel et al. | Adaptive online performance evaluation of video trackers | |
| Pan et al. | Robust and accurate object tracking under various types of occlusions | |
| CN107169989A (en) | A kind of multi-object tracking method assessed based on data correlation and track | |
| CN111160212A (en) | Improved tracking learning detection system and method based on YOLOv3-Tiny | |
| CN111191535B (en) | Pedestrian detection model construction method based on deep learning and pedestrian detection method | |
| CN110070560A (en) | Movement direction of object recognition methods based on target detection | |
| CN108022254A (en) | A kind of space-time contextual target tracking based on sign point auxiliary | |
| CN110569785A (en) | Face recognition method based on fusion tracking technology | |
| CN108846850B (en) | Target tracking method based on TLD algorithm | |
| CN111161309A (en) | Searching and positioning method for vehicle-mounted video dynamic target | |
| Ali et al. | Moving object detection using self adaptive Gaussian Mixture Model for real time applications | |
| KR101396838B1 (en) | Video stabilization method and system by selection one of various motion models | |
| CN114638808A (en) | Multi-scene video jitter detection method based on video monitoring | |
| WO2020095644A1 (en) | State-change detection device, state-change detection method, and computer-readable recording medium | |
| CN110910428B (en) | Real-time multi-target tracking method based on neural network | |
| Sajid et al. | Background subtraction under sudden illumination change | |
| Kerdvibulvech et al. | Vision-based guitarist fingering tracking using a Bayesian classifier and particle filters | |
| Biswas et al. | Short local trajectory based moving anomaly detection | |
| CN110602487B (en) | Video image jitter detection method based on TSN (time delay network) |
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 |