CN109800757A - A kind of video text method for tracing based on layout constraint - Google Patents
A kind of video text method for tracing based on layout constraint Download PDFInfo
- Publication number
- CN109800757A CN109800757A CN201910006843.5A CN201910006843A CN109800757A CN 109800757 A CN109800757 A CN 109800757A CN 201910006843 A CN201910006843 A CN 201910006843A CN 109800757 A CN109800757 A CN 109800757A
- Authority
- CN
- China
- Prior art keywords
- text
- frame
- track
- formula
- result
- 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
Abstract
It is tracked to solve the mobile lower more texts of significantly camera, the invention proposes a kind of video text method for tracing based on layout constraint.The input of this method is the text detection of video and video frame as a result, output is the trace information after text tracking.Firstly, carrying out the initialization of text track by the testing result of initial video frame, then the testing result of the text track of previous frame and present frame is sent into method for tracing of the invention to the update for carrying out text track.The core that text track updates is that the character area for detecting present frame corresponds to existing text track, which can be considered as a kind of Data Matching problem.The present invention designs a new Data Matching cost function in response to this problem, obtains best matching result by solution cost function.By repeating track renewal process until processing terminate for video, text tracking result is finally obtained.The present invention introduces layout constraint in Data Matching cost function, carries out text tracking by the overall appearance structure between character area, it is possible to prevente effectively from because camera Large Amplitude Motion leads to error tracking as a result, having preferably tracking effect.
Description
Technical field
The present invention relates to the text method for tracing in a kind of field of video processing, especially natural scene shooting video.
Background technique
Text in video includes high-layer semantic information, usually closely related with video content.Therefore, video text mentions
It takes and being played an important role in application of many based on Media Analysis, such as is blind person's auxiliary system, driving assistance system, autonomous
Mobile robot etc..The extraction of video text generally includes text detection and text tracking, and text detection is completed to video frame figure
The positioning of text target as in, text tracking complete that identical character area is mapped in consecutive image sequence.Video
In text usually there is time redundancy characteristic, i.e. text can be deposited disappear over time in video.Using the characteristic, lead to
Crossing text tracer technique can be improved the stability and precision of video text detection.In addition, text tracking can also be video point
Analysis provides other relevant informations, such as: the time point of text appearing and subsiding, text are in certain time in video time sequence
Motion profile etc..Some real time processing systems can also improve system using the time redundancy characteristic of text in video and handle
Speed.It can be seen that text tracer technique is played an important role in based on video analysis application.
Existing video text method for tracing can not handle the tracing problem of more texts when camera significantly moves well.
Because text usually will not individually occur, but occur with packed form in natural scene.These texts often have identical
Size, length-width ratio and color characteristic, and the feature that most tracing algorithms extracts can not distinguish these texts well, can make
Text can not be correctly tracked at the matching of mistake.As camera is when significantly moving, such case can be more serious.
Based on the above issues, the invention proposes a kind of video text method for tracing based on layout constraint is big to solve
More texts tracking under amplitude camera is mobile.
Summary of the invention
In order to solve more text tracing problems under significantly camera is mobile, the invention proposes one kind to be based on layout constraint
Video text method for tracing.Method flow of the invention is as shown in Fig. 1.The input of this method is video and video single frames figure
The text detection of picture is as a result, output is the track of each character area in video, i.e., spatial information (position in each frame
Coordinate and width are high).Firstly, carrying out the initialization of text track by the word area detection result of initial video frame, then will
The text track of previous frame and the word area detection result of present frame are sent into progress text track in method for tracing of the invention
Update, repeat the process until video processing complete, finally obtain text tracking result.Text track update core be by
The character area that present frame detects corresponds to existing text track, which can be considered as a kind of Data Matching problem.This
Invention designs a new Data Matching cost function in response to this problem, obtains best match knot by solution cost function
Fruit.The present invention introduces layout constraint in Data Matching cost function, carries out text by the overall appearance structure between character area
Word tracking, it is possible to prevente effectively from because camera Large Amplitude Motion leads to error tracking as a result, having preferably tracking effect.This hair
Bright detail is as follows.
1, design data matching cost function
The character area that text track is included in present frame is defined first.I-th of text in setting video t frame text track
The state in block domain isWhereinFor the abscissa and vertical seat of character area central point
Mark,For the region horizontal and vertical movement speed in the picture,For the width and height of character area,For
(present invention extracts RGB color histogram to the character area color characteristic, and there are 16 characteristic areas in each channel, and three channels are total
48).The state of character area is set as in all text tracks of t frameWherein i ∈ Nt, NtIt indicates in t frame
The number of character area.
For every two character area, need to establish the correlation of its position and speed, this correlation can regard
For a kind of structural constraint, structural constraint can be by shown in formula (1):
WhereinThe structural constraint for indicating character area i and j, can be expressed as all constraints of iThen all character areas are constrained in t frame
Tracking task is matched to word area detection result on existing text track, if p-th of text in t frame
Area detection result information is For the centre coordinate of word area detection result,For the width and height of word area detection result.Then the collection of all word area detection results is combined into t frameWherein p ∈ Mt, MtFor the number for detecting character area.
Two-value symbol a is used in the present inventioni,pIt indicates text track and word area detection result match condition, works as text
Track i matches with word area detection result p, then ai,p=1, otherwise ai,p=0.For in t-1 frame text track with
Word area detection in t frame is as a result, its Data Matching situation can be described by formula (2):
A=argminC (St-1,Rt-1,Dt) (2)
Wherein A={ ai,p|i∈Nt-1,p∈Mt, a text track is at most matched to a character area in this method
Testing result, C (St-1,Rt-1,Dr) indicate that text track and all possible pairing of word area detection result are gathered.And it is best
Matching result be set minimum value argminC (St-1,Rt-1,Dt)。
In successive frame, too big variation will not occur for mutual distance between the text with same background.When camera motion,
Text and other texts should keep similar appearance.In text tracking, this method considers the similar of text in consecutive frame simultaneously
Property and associated system other text appearances similitude, similitude, that is, cloth of appearance is laid out in this consecutive frame around text
Office's constraint.Cost function C (S based on layout constraintt-1,Rt-1,Dt) as shown in formula (3):
WhereinWithIndicate that the frame text track t-1 and t frame detect character area
Cost value is distinguished, using area size ratio and Duplication calculate cost in the present invention, as shown in formula (4) and (5):
In formulaWithRespectively indicate t-1 frame text track i and t frame word area detection
As a result the width and height of p,Indicate t-1 frame text track i and t frame word area detection result
The minimum extraneous area for surrounding frame overlapping of p region,It indicates to merge area.
In formula (3)Indicate t frame word area detection result p with the structural constraint of t-1 frameEstimation rangeExternal appearance characteristic and the similitude of corresponding t-1 frame text track j external appearance characteristic, calculation formula such as (6)
(7) shown in:
H in formulab(s) RGB color normalization histogram feature is indicated, B is feature sum, and b is index,Include
The center point coordinate of estimation range position and the width of estimation range are high.
2, cost function optimization and solution
It is calculated to simplify, the present invention uses the matching of formula (8) and formula (9) constrained trajectory and testing result, when not
When meeting condition, it is considered as ai,p=0.Formula (8) and (9) are as follows:
S in formulaaAnd sbThe state between two character areas is indicated, when interregional distance and excessive relative velocity, then it is assumed that
The two not can be carried out matching.τ value is 10 in the present invention.
Finally, all pairing character area cost values can be calculated according to formula (2), and obtain a Nt-1×Mt's
Similarity matrix.Use " the Kuhn H W.The Hungarian method for the assignment problem of document 1
[J] .Naval Research Logistics, 1955, the method that (1-2): 83-97. " is proposed can calculate best match knot
Fruit.It as a result is the matrix of a 2 × Q, which is the matching of text track index number and word area detection result call number
Matrix, wherein Q is matching number.Using the matching matrix, existing text track new space letter in the current frame can be updated
Breath (position coordinates and width are high), i.e. the character area tracking of completion present frame.Such as: t-1 frame has 3 articles of texts tracked
Track, t frame have 3 character areas detected, and the matching matrix obtained after inventive algorithm calculates is such as shown in (10):
The matrix first row indicates text track index number, and secondary series indicates word area detection result call number.Its table
Show the 1st article of text track it is corresponding be the 2nd character area detected, the 2nd article of text track be corresponding to be the 1st and detects
Character area, corresponding the 3rd article of text track is the 3rd character area detected.According to the matching matrix, by t frame
3 character area corresponding coordinates detected and the high information of width replace the spatial information in corresponding text track, complete t frame
The update of text track.
3, beneficial effect
The present invention can accurately track the text track in video when extensive camera is mobile.The present invention uses text
Tracking field well known data library Minetto is tested.Minetto database includes 5 sections of scene text videos, video frame point altogether
Resolution is 640 × 480.In test phase, the text detection result of video and every frame video image is input to the present invention and is chased after
In track algorithm, algorithm exports the track of each character area in video, i.e. spatial information (position of the character area in every frame
Coordinate and width are high).Become by calculating multi-target tracking accuracy rate (MOTP), multi-target tracking accuracy (MOTA) and track index
Change number (IDS) three known evaluation indexes to measure the validity of inventive algorithm.With document 2 " Pei W Y, Yang C,
Meng L Y,et al.Scene Video Text Tracking With Graph Matching[J].IEEE Access,
Method is compared in 2018,6:19419-19426. ", and the video text tracking proposed by the present invention based on layout constraint exists
Performance greatly improves on Minetto database, specifically: MOTP index improves 6%, MOTA and improves 19%, IDS promotion
One times.
Detailed description of the invention
Fig. 1 is the video text method for tracing flow chart based on layout constraint.
Specific embodiment
Referring to Fig.1, specific step is as follows for the video text method for tracing proposed by the present invention based on layout constraint:
Step 1: input video and text detection result
The present invention is built upon in video text testing result.Text detection can be divided on line to be detected under detection and line.It is right
In detecting on line, input video first then frame by frame or frame-skipping detects text will test in the result input present invention and carry out text
Then tracking carries out next frame text detection again, repeat the process until video processing is completed.It is defeated first for being detected under line
Enter video, then carries out text detection until video processing completion, the detection structure of video and each frame is finally input to this
Text tracking is carried out in invention.Method for tracing proposed by the present invention can be applied on line simultaneously and detect under detection and line.
Step 2: text Track Initiation
Track Initiation is carried out to the testing result of video first frame, the character area that each is detected is considered as one
New text track is simultaneously indexed number, then calculates the state S of all character areast, speed (u in statet,vt) just
Beginning turns to (0,0), t=1.And the structural constraint R between character area two-by-two is calculated according to formula (1)t, t=1.Simultaneously according to about
Beam formula (8) and (9) remove undesirable structural constraint, and record remaining structure constrains R1With text track state S1。
Step 3: text track updates
The text track more new stage passes through existing text track in the word area detection result and t-1 frame of t frame
It is matched, the word area detection result additional space information (position coordinates and width are high) that will match to replaces corresponding text rail
The spatial information of character area in mark.Stage input is the text track state S of t-1 framet-1, structural constraint Rt-1With t
Frame word area detection result Dt, export as updated text trajectory range information.
Step 3.1: Data Matching
The word area detection result of text track and t frame to t-1 frame is matched, and forms N altogethert-1×Mt
A combinations of pairs.Then a N is obtained using the cost value that formula (3) calculate all pairingst-1×MtSimilarity matrix.Make
Judged with constraint is carried out first with formula (8) before formula (3), when being unsatisfactory for condition, the calculating of formula (3) is skipped, by this
Pairing cost value is set as 999.Use " the Kuhn H W.The Hungarian method for the assignment of document 1
Problem [J] .Naval Research Logistics, 1955, the method that (1-2): 83-97. " is proposed can calculate most
Good matching result.It as a result is the matrix of a 2 × Q, which is that text track index number and word area detection result index
Number matching matrix, wherein Q be matching number.
Step 3.1: update is matched to track
If text track matches with current character area detection result, " the Links I K F.An of document 3 is utilized
The Kalman filtering algorithm proposed in Introduction to the Kalman Filter [J] .1995. " uses character area
Testing resultTo in track stateInto
Row updates, while updating the normalization color histogram of character areaObtain new state St。
Step 3.2: update does not match track
Existing text detection algorithm often will appear detection leakage phenomenon, and text track is caused to fail to be matched to character area inspection
Survey result.Updated text track state S is utilized at this timetWith the structural constraint R of t-1 framet-1To the text not being matched to
Predicted that formula is as follows using formula (11) in track:
Wherein NrFor the number for being matched to text track, (x, y) is character area regional center coordinate, and (Δ x, Δ y) are
Centre coordinate distance difference in structural constraint.The regional center coordinate of prediction is replaced to the old seat not being matched in text track
Mark, and replacement number is recorded, when replacing number greater than 3, it is considered as text track disappearance, the track is deleted from text track
Information.
Step 3.3: initializing new track
If t frame word area detection result p fails and any text path matching, then it is assumed that new text rail occur
Mark establishes new track stateAnd the new track state is added in existing track.
Step 3.4: the constraint of more new literacy track configuration
The structural constraint between text track two-by-two is calculated according to formula (1).It is removed simultaneously according to constraint formulations (8) and (9)
Undesirable structural constraint, record remaining structure constrain Rt。
Step 3.5: output updates track
The updated text rail spatial information of t frame (position coordinates and width are high) is exported, records and updates all tracks and deposit
Number living.
Step 4: output character trace information
Step 3 is repeated until video processing is completed, according to the time redundancy characteristic of text in video, i.e., text is in video
In often exist one section after can just disappear.The present invention filters non-legible region using the time redundancy characteristic, when depositing for track
When number living is less than or equal to 15 frame, judges the non-legible region in this track, delete the text trace information.By filtration treatment, most
Remaining text trace information is exported eventually.
Claims (1)
1. the video text method for tracing based on layout constraint, core essentially consists in two parts: Data Matching cost function,
Cost function optimization and solution.
(1) Data Matching cost function:
The character area that text track is included in present frame is defined first.I-th of literal field in setting video t frame text track
The state in domain isWhereinFor the abscissa and ordinate of character area central point,For the region horizontal and vertical movement speed in the picture,For the width and height of character area,For this article
(present invention extracts RGB color histogram to word region color feature, and there are 16 characteristic areas in each channel, three channels totally 48
It is a).The state of character area is set as in all text tracks of t frameWherein i ∈ Nt, NtIndicate t frame Chinese
The number in block domain.
For every two character area, need to establish the correlation of its position and speed, this correlation can be considered as one
Kind structural constraint, structural constraint can be by shown in formula (1):
WhereinThe structural constraint for indicating character area i and j, can be expressed as all constraints of iThen all character areas are constrained in t frame
Tracking task is matched to word area detection result on existing text track, if p-th of character area in t frame
Testing result information is For the centre coordinate of word area detection result,For the width and height of word area detection result.Then the collection of all word area detection results is combined into t frameWherein p ∈ Mt, MtFor the number for detecting character area.
Two-value symbol a is used in the present inventioni,pText track and word area detection result match condition are indicated, as text track i
Match with word area detection result p, then ai,p=1, otherwise ai,p=0.For in t-1 frame text track and t frame
In word area detection as a result, its Data Matching situation can be described by formula (2):
A=argminC (St-1,Rt-1,Dt) (2)
Wherein A={ ai,p|i∈Nt-1,p∈Mt, a text track is at most matched to a word area detection knot in this method
Fruit, C (St-1,Rt-1,Dt) indicate that text track and all possible pairing of word area detection result are gathered.And optimal matching
It as a result is set minimum value argminC (St-1,Rt-1,Dt)。
In successive frame, too big variation will not occur for mutual distance between the text with same background.When camera motion, text
Similar appearance should be kept with other texts.In text tracking, this method consider in consecutive frame simultaneously the similitude of text with
The similitude of other text appearances of associated system, the similitude for being laid out appearance in this consecutive frame around text are laid out about
Beam.Cost function C (S based on layout constraintt-1,Rt-1,Dt) as shown in formula (3):
WhereinWithIndicate that the frame text track t-1 and t frame detect the difference of character area
Cost value, using area size ratio and Duplication calculate cost in the present invention, as shown in formula (4) and (5):
In formulaWithRespectively indicate t-1 frame text track i and t frame word area detection result p
Width and height,Indicate t-1 frame text track i and t frame word area detection result p region
The minimum extraneous area for surrounding frame overlapping,It indicates to merge area.
In formula (3)Indicate t frame word area detection result p with the structural constraint of t-1 frameIn advance
Survey regionExternal appearance characteristic and the similitude of corresponding t-1 frame text track j external appearance characteristic, calculation formula such as (6) and (7)
It is shown:
H in formulab(s) RGB color normalization histogram feature is indicated, B is feature sum, and b is index,Include Target area
The center point coordinate of domain position and the width of estimation range are high.
(2) cost function optimization and solution:
It is calculated to simplify, the present invention uses the matching of formula (8) and formula (9) constrained trajectory and testing result, when being unsatisfactory for
When condition, it is considered as ai,p=0.Formula (8) and (9) are as follows:
S in formulaaAnd sbThe state between two character areas is indicated, when interregional distance and excessive relative velocity, then it is assumed that the two
Matching is not can be carried out.τ value is 10 in the present invention.
Finally, all pairing character area cost values can be calculated according to formula (2), and obtain a Nt-1×MtSimilarity
Matrix.Use " the Kuhn H W.The Hungarian method for the assignment problem [J] of document 1
.Naval Research Logistics, 1955, the method that (1-2): 83-97. " is proposed can calculate best matching result.
It as a result is the matrix of a 2 × Q, which is the matching square of text track index number and word area detection result call number
Battle array, wherein Q is matching number.Using the matching matrix, existing text track new spatial information in the current frame can be updated
(position coordinates and width are high), i.e. the character area tracking of completion present frame.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910006843.5A CN109800757B (en) | 2019-01-04 | 2019-01-04 | Video character tracking method based on layout constraint |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910006843.5A CN109800757B (en) | 2019-01-04 | 2019-01-04 | Video character tracking method based on layout constraint |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109800757A true CN109800757A (en) | 2019-05-24 |
| CN109800757B CN109800757B (en) | 2022-04-19 |
Family
ID=66558550
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910006843.5A Active CN109800757B (en) | 2019-01-04 | 2019-01-04 | Video character tracking method based on layout constraint |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109800757B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021164479A1 (en) * | 2020-02-21 | 2021-08-26 | 华为技术有限公司 | Video text tracking method and electronic device |
| CN114463376A (en) * | 2021-12-24 | 2022-05-10 | 北京达佳互联信息技术有限公司 | Video character tracking method and device, electronic equipment and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101276416A (en) * | 2008-03-10 | 2008-10-01 | 北京航空航天大学 | Text tracking and multi-frame reinforcing method in video |
| TW201039149A (en) * | 2009-04-17 | 2010-11-01 | Yu-Chieh Wu | Robust algorithms for video text information extraction and question-answer retrieval |
| CN104244073A (en) * | 2014-09-26 | 2014-12-24 | 北京大学 | Automatic detecting and recognizing method of scroll captions in videos |
| WO2015165524A1 (en) * | 2014-04-30 | 2015-11-05 | Longsand Limited | Extracting text from video |
| CN107545210A (en) * | 2016-06-27 | 2018-01-05 | 北京新岸线网络技术有限公司 | A kind of method of video text extraction |
| CN108052941A (en) * | 2017-12-19 | 2018-05-18 | 北京奇艺世纪科技有限公司 | A kind of news caption tracking and device |
| CN108229476A (en) * | 2018-01-08 | 2018-06-29 | 北京奇艺世纪科技有限公司 | Title area detection method and system |
| CN108256493A (en) * | 2018-01-26 | 2018-07-06 | 中国电子科技集团公司第三十八研究所 | A kind of traffic scene character identification system and recognition methods based on Vehicular video |
| CN108363981A (en) * | 2018-02-28 | 2018-08-03 | 北京奇艺世纪科技有限公司 | A kind of title detection method and device |
| CN108694393A (en) * | 2018-05-30 | 2018-10-23 | 深圳市思迪信息技术股份有限公司 | A kind of certificate image text area extraction method based on depth convolution |
-
2019
- 2019-01-04 CN CN201910006843.5A patent/CN109800757B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101276416A (en) * | 2008-03-10 | 2008-10-01 | 北京航空航天大学 | Text tracking and multi-frame reinforcing method in video |
| TW201039149A (en) * | 2009-04-17 | 2010-11-01 | Yu-Chieh Wu | Robust algorithms for video text information extraction and question-answer retrieval |
| WO2015165524A1 (en) * | 2014-04-30 | 2015-11-05 | Longsand Limited | Extracting text from video |
| CN104244073A (en) * | 2014-09-26 | 2014-12-24 | 北京大学 | Automatic detecting and recognizing method of scroll captions in videos |
| CN107545210A (en) * | 2016-06-27 | 2018-01-05 | 北京新岸线网络技术有限公司 | A kind of method of video text extraction |
| CN108052941A (en) * | 2017-12-19 | 2018-05-18 | 北京奇艺世纪科技有限公司 | A kind of news caption tracking and device |
| CN108229476A (en) * | 2018-01-08 | 2018-06-29 | 北京奇艺世纪科技有限公司 | Title area detection method and system |
| CN108256493A (en) * | 2018-01-26 | 2018-07-06 | 中国电子科技集团公司第三十八研究所 | A kind of traffic scene character identification system and recognition methods based on Vehicular video |
| CN108363981A (en) * | 2018-02-28 | 2018-08-03 | 北京奇艺世纪科技有限公司 | A kind of title detection method and device |
| CN108694393A (en) * | 2018-05-30 | 2018-10-23 | 深圳市思迪信息技术股份有限公司 | A kind of certificate image text area extraction method based on depth convolution |
Non-Patent Citations (4)
| Title |
|---|
| LIANG WU等: "A New Technique for Multi-Oriented Scene Text Line Detection and Tracking in Video", 《IEEE TRANSACTIONS ON MULTIMEDIA》 * |
| XU-CHENG YIN等: "Text Detection, Tracking and Recognition in Video: A Comprehensive Survey", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| 刁月华: "网络视频字幕提取识别系统的设计与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 密聪杰等: "基于多帧图像的视频文字跟踪和分割算法", 《计算机研究与发展》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021164479A1 (en) * | 2020-02-21 | 2021-08-26 | 华为技术有限公司 | Video text tracking method and electronic device |
| CN114463376A (en) * | 2021-12-24 | 2022-05-10 | 北京达佳互联信息技术有限公司 | Video character tracking method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109800757B (en) | 2022-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106096577B (en) | A kind of target tracking method in camera distribution map | |
| CN109522854B (en) | Pedestrian traffic statistical method based on deep learning and multi-target tracking | |
| CN106097391B (en) | A kind of multi-object tracking method of the identification auxiliary based on deep neural network | |
| CN109919981A (en) | A kind of multi-object tracking method of the multiple features fusion based on Kalman filtering auxiliary | |
| CN109816689A (en) | A kind of motion target tracking method that multilayer convolution feature adaptively merges | |
| CN110490901A (en) | The pedestrian detection tracking of anti-attitudes vibration | |
| CN109949340A (en) | Target scale adaptive tracking method based on OpenCV | |
| CN105940430B (en) | Personnel's method of counting and its device | |
| CN107424171A (en) | A kind of anti-shelter target tracking based on piecemeal | |
| CN108665485A (en) | A kind of method for tracking target merged with twin convolutional network based on correlation filtering | |
| CN107784663A (en) | Correlation filtering tracking and device based on depth information | |
| CN110390292A (en) | Based on the remote sensing video frequency vehicle object detecting and tracking method for dynamically associating model | |
| CN105654516B (en) | Satellite image based on target conspicuousness is to ground weak moving target detection method | |
| CN109974721A (en) | A kind of vision winding detection method and device based on high-precision map | |
| CN104992453A (en) | Target tracking method under complicated background based on extreme learning machine | |
| CN108830171A (en) | A kind of Intelligent logistics warehouse guide line visible detection method based on deep learning | |
| CN110490905A (en) | A kind of method for tracking target based on YOLOv3 and DSST algorithm | |
| CN107622507B (en) | Air target tracking method based on deep learning | |
| CN109800757A (en) | A kind of video text method for tracing based on layout constraint | |
| CN103853794B (en) | Pedestrian retrieval method based on part association | |
| CN110443829A (en) | It is a kind of that track algorithm is blocked based on motion feature and the anti-of similarity feature | |
| CN109816693A (en) | Anti- based on multimodal response blocks correlation filtering tracking and systems/devices | |
| CN109297489A (en) | A kind of indoor navigation method based on user characteristics, electronic equipment and storage medium | |
| CN109740609A (en) | A kind of gauge detection method and device | |
| CN106204633A (en) | A kind of student trace method and apparatus based on computer vision |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Feng Xiaoyi Inventor after: Song Zhendong Inventor after: Wang Xihan Inventor after: Jiang Xiaoyue Inventor after: Xia Zhaoqiang Inventor after: Peng Jinye Inventor after: Xie Hongmei Inventor after: Li Huifang Inventor after: He Guiqing Inventor before: Feng Xiaoyi Inventor before: Wang Xihan Inventor before: Jiang Xiaoyue Inventor before: Xia Zhaoqiang Inventor before: Peng Jinye Inventor before: Xie Hongmei Inventor before: Li Huifang Inventor before: He Guiqing Inventor before: Song Zhendong |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |