CN106651918A - Method for extracting foreground under shaking background - Google Patents
Method for extracting foreground under shaking background Download PDFInfo
- Publication number
- CN106651918A CN106651918A CN201710083910.4A CN201710083910A CN106651918A CN 106651918 A CN106651918 A CN 106651918A CN 201710083910 A CN201710083910 A CN 201710083910A CN 106651918 A CN106651918 A CN 106651918A
- Authority
- CN
- China
- Prior art keywords
- image
- motion vector
- frame
- field picture
- gravity
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Abstract
The invention relates to a method for extracting a foreground under a shaking background. The method comprises the following steps: S1) initializing the size of an image slide block, the step length of the slide block movement and the position of the slide block; S2) reading front and rear frames of images from a video and acquiring binary images be_frame and af_frame; S3) extracting the image slide block on the same position in the be_frame and af_frame, calculating a gravity center difference of two extracting results and acquiring motion vectors of front and rear frames on the current position of the image slide block; S4) moving the image slide block according to the step length and repeating the step S3) till completing the extraction for the whole image; S5) calculating an average motion vector of the front and rear frames according to all the acquired motion vectors of the front and rear frames; S6) reversely translating the rear frame image according to the acquired average motion vector, thereby acquiring a relieved shaking error image; S7) utilizing a Gaussian mixture model to extract the foreground for the front and rear frames of images after the shaking is adjusted. Compared with the prior art, the method provided by the invention has the advantages of wide application scope and the like.
Description
Technical field
The present invention relates to a kind of foreground extracting method, more particularly, to the foreground extracting method under a kind of shake background.
Background technology
In order to improve the quality and level of city security, picture pick-up device has been applied to almost all of public place, but
Video acquired in substantial amounts of picture pick-up device but cannot also comprehensively, intelligently, accurately to it be analyzed at present.If utilized
It is neither actual also uneconomical that manpower carrys out checking monitoring video.And study and find, people's its attention after viewing monitoring video 20 minutes
Power will drop to unacceptable degree, and the unconventional phenomenon occurred in image can be turned a blind eye to.And one can adapt to
The video foreground extraction algorithm of complex scene, just automatically can extract to the prospect in video, it is also possible to by what is extracted
Prospect carries out intelligent analysis as pattern-recognition and the input of the system of motion analysis to video.Therefore, video foreground is extracted
Development be conducive to social security system improve and public safety level lifting.
In perhaps multi-disciplinary research, also it is no lack of the figure of foreground extraction technology：Doctor can after some drugses under clothes for patients
So that medicine is followed the trail of using foreground extraction technology in trace in patient body, whether checking medicine accurately arrives at focus and occurs
Effect；The scholar of animal behavior research may not necessarily look-out target individual in a long time behavior, but utilize intelligence
Video analytic system is replacing.Militarily, foreground extraction can also be used for catching the behavior of common-denominator target with target tracking technology,
To strengthen the defense, aid in attack.
When the equipment for shooting video is in almost static, gauss hybrid models relatively accurately can carry foreground object
Take out, but for the video that there is shake lacks adaptability.
The content of the invention
The purpose of the present invention is provided under a kind of shake background for the defect for overcoming above-mentioned prior art to exist
Foreground extracting method.
The purpose of the present invention can be achieved through the following technical solutions：
A kind of foreground extracting method under shake background, including：
Step S1：The position of the size, the step-length of slide block movement and slide block of initialisation image slide block；
Step S2：From video read before and after two field picture, and obtain prior image frame binary image be_frame and after
The binary image af_frame of two field picture；
Step S3：Same position of the image slider in the binary image of prior image frame and the binary image of rear two field picture
Extraction is put, the center of gravity for calculating two extraction results is poor, the motion vector of frame before and after obtaining under present image slide position；
Step S4：Image slider is moved by step-length, repeat step S3, until completing the extraction of whole image；
Step S5：According to the average motion vector of frame before and after all of motion vector computation of frame in front and back for obtaining；
Step S6：Rear two field picture is reversely translated according to the average motion vector for obtaining, the jitter error figure being mitigated
Picture；
Step S7：Two field picture carries out foreground extraction before and after after being shaken to adjustment using gauss hybrid models.
Step S6 is specifically included：
Step S61：Whether current average motion vector is judged more than or equal to 2, if if it has, then execution step S62
It is no, then execution step S64；
Step S62：The size of image slider is amplified into current twice, repeat step S3 to S5 obtains another average
Motion vector, and using the mean value of this average motion vector and former average motion vector as final average motion vector, and hold
Row step S63；
Step S63：Rear two field picture is reversely translated according to final average motion vector, the jitter error figure being mitigated
Picture；
Step S64：The binary image of the binary image of prior image frame and rear two field picture is amplified into current 10
Times, repeat repeat step S3 to S5, another average motion vector is obtained, and using this average motion vector as final mean motion
Vector, and execution step S65；
Step S65：Rear two field picture is amplified into current 10 times, according to final average motion vector to amplification after rear frame
Image is reversely translated, and is then contracted to life size, the jitter error image being mitigated.
The original dimension of described image slide block is 50 pixel × 50 pixels, and step-length is 25 pixels.
The motion vector of frame is specially before and after described：
Wherein：CxbeThe x coordinate of the center of gravity of part, Cy are extracted by image slider for prior image framebeSchemed for prior image frame
As slide block extracts the y-coordinate of the center of gravity of part, CxafThe x coordinate of the center of gravity of part, Cy are extracted by image slider for rear two field pictureaf
The y-coordinate of the center of gravity of part is extracted by image slider for rear two field picture.
Barycentric coodinates are specially：
Wherein：Cx is the x coordinate of center of gravity, and Cy is the y-coordinate of center of gravity, and W is that image extracts all of part by image slider
The pixel value sum of pixel, wiFor the pixel value of pixel i, xiFor the x coordinate of pixel i, yiFor the y-coordinate of pixel i, m is
Image is extracted the pixel total number of part by image slider.
Compared with prior art, the present invention has advantages below：
1) coloured image is converted into binary image, you can accelerate arithmetic speed, the memory space of program can be reduced again,
On this basis, by the way of image slider slides and extracts, it is easy to iterative algorithm to automatically complete the extraction to whole image,
Mean value is calculated by several times to be conducive to amplifying difference, sensitiveness is improved, and then improves shake rectification effect.
2) if there is larger difference in the molecule of image block center of gravity, can be with the size of adjusting tile so that calculate center of gravity
Denominator value it is larger, the change on molecule is then weakened, makes the calculating of center of gravity more accurate, finally by present frame according to average fortune
Dynamic vector does reversely translation, the jitter error image being mitigated, if the skew of flating generation is less, by frame figure in front and back
As amplifying N times, according to center of gravity calculation motion vector, then reversely translation is done to present frame according to average motion vector, finally will
The present frame for obtaining reduces N times, the jitter error being mitigated.
Description of the drawings
Fig. 1 is the key step schematic flow sheet of the inventive method；
Fig. 2 is the previous frame image of an example；
Fig. 3 is the latter two field picture of an example；
Fig. 4 (a) is the image of non-shake video 1 in experiment；
Fig. 4 (b) is the foreground image that non-shake video 1 is extracted by gauss hybrid models method in experiment；
Fig. 4 (c) is the foreground image that non-shake video 1 is extracted by the application method in experiment；
Fig. 5 (a) is the image that video 2 is shaken in experiment；
Fig. 5 (b) is that the foreground image that video 2 is extracted by gauss hybrid models method is shaken in experiment；
Fig. 5 (c) is that the foreground image that video 2 is extracted by the application method is shaken in experiment；
Fig. 6 (a) is the image that video 3 is shaken in experiment；
Fig. 6 (b) is that the foreground image that video 3 is extracted by gauss hybrid models method is shaken in experiment；
Fig. 6 (c) is that the foreground image that video 3 is extracted by the application method is shaken in experiment.
Specific embodiment
Below in conjunction with the accompanying drawings the present invention is described in detail with specific embodiment.The present embodiment is with technical solution of the present invention
Premised on implemented, give detailed embodiment and specific operating process, but protection scope of the present invention is not limited to
Following embodiments.
In the shooting of video, if video camera there occurs motion, position of the actionless object in picture in practice
Putting will occur change in location, and its position has changed to (x2, y2) by (x1, y1), and motion vector is：
The direct result that the shake of video camera is caused is that the object in video produces identical displacement, and block motion estimation method will
Present frame is divided into the fritter of formed objects, then block most like therewith is searched in former frame to each fritter, contrast
Coordinate between two blocks draws the displacement of each image block.The misalignment for considering all image blocks is just obtained in that shooting sets
Standby motion conditions, draw the motion vector of background.
Block motion estimation algorithm based on square.Square is a numerical characteristic of stochastic variable.To stochastic variable X, it
K rank squares computing formula be E (X-EX)k, wherein EX is the expectation of the stochastic variable.Algorithm can be described using n ranks square
Image block estimates the motion vector of image block, in the application, the fortune of block is carried out as the first moment of image block using center of gravity square
It is dynamic to estimate.
As shown in Figures 2 and 3, it is respectively the image of same position in the two field picture of t-1 moment and t in a video
Block.Due to the shake of video camera so that the image in block there occurs translation (open circle there occurs shifting), the center of gravity of image block
Change therewith.On the premise of previous frame removal pixel and rear frame immigration pixel difference is ignored, it is believed that the mobile arrow of image block
Amount is exactly the motion vector of image.
A kind of foreground extracting method under shake background, as shown in figure 1, including：
Step S1：The position of the size, the step-length of slide block movement and slide block of initialisation image slide block, wherein image slider
Original dimension be 50 pixel × 50 pixels, step-length be 25 pixels；
Step S2：From video read before and after two field picture, and obtain prior image frame binary image be_frame and after
The binary image af_frame of two field picture；
Step S3：Same position of the image slider in the binary image of prior image frame and the binary image of rear two field picture
Extraction is put, the center of gravity for calculating two extraction results is poor, the motion vector of frame before and after obtaining under present image slide position；
Step S4：Image slider is moved by step-length, repeat step S3, until completing the extraction of whole image；
Step S5：According to the average motion vector of frame before and after all of motion vector computation of frame in front and back for obtaining,
In front and back the motion vector of frame is specially：
Wherein：CxbeThe x coordinate of the center of gravity of part, Cy are extracted by image slider for prior image framebeSchemed for prior image frame
As slide block extracts the y-coordinate of the center of gravity of part, CxafThe x coordinate of the center of gravity of part, Cy are extracted by image slider for rear two field pictureaf
The y-coordinate of the center of gravity of part is extracted by image slider for rear two field picture；
Step S6：Rear two field picture is reversely translated according to the average motion vector for obtaining, the jitter error figure being mitigated
Picture, specifically includes：
Step S61：Whether current average motion vector is judged more than or equal to 2, if if it has, then execution step S62
It is no, then execution step S64；
Step S62：The size of image slider is amplified into current twice, repeat step S3 to S5 obtains another average
Motion vector, and using the mean value of this average motion vector and former average motion vector as final average motion vector, and hold
Row step S63；
Step S63：Rear two field picture is reversely translated according to final average motion vector, the jitter error figure being mitigated
Picture；
Step S64：The binary image of the binary image of prior image frame and rear two field picture is amplified into current 10
Times, repeat repeat step S3 to S5, another average motion vector is obtained, and using this average motion vector as final mean motion
Vector, and execution step S65；
Step S65：Rear two field picture is amplified into current 10 times, according to final average motion vector to amplification after rear frame
Image is reversely translated, and is then contracted to life size, the jitter error image being mitigated.
Barycentric coodinates are specially：
Wherein：Cx is the x coordinate of center of gravity, and Cy is the y-coordinate of center of gravity, and W is that image extracts all of part by image slider
The pixel value sum of pixel, wiFor the pixel value of pixel i, xiFor the x coordinate of pixel i, yiFor the y-coordinate of pixel i, m is
Image is extracted the pixel total number of part by image slider.
In the block analysis of video sequential, if ignoring previous frame removes the difference that pixel is moved into rear frame, it is believed that
The mobile vector of block center of gravity is exactly the motion vector of block.But in most cases, removing and move into can deposit between pixel
In difference, there may come a time when there is larger difference, in order to reduce the impact of this species diversity as far as possible, employ in the application following two
Measure：
The value that coloured image describes pixel using RGB color is often larger, when the size of image block is than larger
When, the operand for calculating block center of gravity is larger, so coloured image is converted into edge binary map by the application using Canny algorithms
Picture, you can accelerate arithmetic speed, the memory space of program can be reduced again.
(1) motion vector computation of larger two field picture in front and back is shaken
But the comparison of block or of problems is directly carried out to bianry image, because removal pixel cannot be determined and moved
Enter the relation of pixel：If it is black to remove pixel, the pixel of immigration is white, then the center of gravity of block will be moved down, otherwise
Then move in the center of gravity of block.Employ in the experiment of this paper and be extracted the black matrix binary image at edge to two field picture to carry out block
Relatively.So, the pixel that previous frame is removed and rear frame is moved into is essentially black color dots, obtains block centre-of-gravity motion vector.When segmentation
When image block is less, if the molecule of image block center of gravity occurs larger difference, the calculating of center of gravity can occur larger error, in order to drop
Low error, can be with the size of adjusting tile so that the denominator value for calculating center of gravity is larger, and the change on molecule is then weakened, makes
The calculating of center of gravity is more accurate.Finally present frame is done into reversely translation, the jitter error being mitigated according to average motion vector
Image.
(2) motion vector computation of less two field picture in front and back is shaken
The translation of image can only be carried out in pixel scale, if the skew of flating generation is less, such as side-play amount
Less than a pixel, now center of gravity calculation skew then produces error.Therefore, when current method calculating centre-of gravity shift is 0 or 1, then
N times will be amplified by two field picture in front and back, according to center of gravity calculation motion vector, then do reversely flat to present frame according to average motion vector
Move, finally the present frame for obtaining is reduced into N times, the jitter error being mitigated.
Step S7：Two field picture carries out foreground extraction before and after after being shaken to adjustment using gauss hybrid models, using Gauss
It is foreground extraction prior art quite ripe at present that mixed model carries out foreground extraction, and the application is no longer described in detail, substantially introduced
It is as follows：
Given a collection of observed data X={ x1, x2..., xN, this batch data is generated altogether by M single Gauss model, but
Concrete certain data xiBelong to which single Gauss model, each ratio α of single Gauss model in mixed modelj, mathematic expectaion μj
With covariance CjAll unknown, these mix from the sample data of different Gaussian Profiles, just become gauss hybrid models.It is high
The probability density function of this mixed model is：
Wherein
Order
All parameters of gauss hybrid models are estimated by sample set X
The probability density function of sample X is
Gauss hybrid models are one background model of video extraction.When a new pixel is read, can one by one with it is known
Single Gauss model matched, the order of matching is matched from low to high according to the priority of each model.If sent out during matching
Now the pixel matches with certain single Gauss model and is considered as the pixel and belongs to background dot, and the list is updated using the pixel
The parameters of Gauss model.Illustrate that it is foreground point if pixel is not belonging to any single Gauss model.
Experiment
For the foreground extraction under complex background, herein using gauss hybrid models (prior art) and the application method two
Plant algorithm.For the adaptability to video scene of com-parison and analysis algorithm, voluntarily recording 20 groups of videos carries out foreground extraction experiment.
Wherein there are 5 groups to be non-shake video, 15 groups is shake video.Each group of video carries out foreground extraction using two kinds of algorithms respectively.
The following is 3 groups in 20 groups of videos.
Three frames that every group of sample picture takes video are shown, and each two field picture shows using gauss hybrid models and this Shen
Please method extraction result.The part irised out with red line in figure is that gauss hybrid models extract result presence substantially with context of methods
The region of difference.
Video shown in Fig. 4 (a) (b) (c) is non-shake video, as can be seen from Figure Gaussian Mixture mould when non-shake
The extraction result of type and context of methods does not have significant difference.Fig. 5 (a) (b) (c) and Fig. 6 (a) (b) (c) show shake video
Foreground extraction result.As seen from the figure, the application algorithm shows the extraction result unanimous on the whole with gauss hybrid models, but
There is less wrong prospect than gauss hybrid models.This shows that the application algorithm can carried out to a certain degree just to two field picture
True moving step sizes so that the prospect of error extraction is reduced.
20 video bags of experiment are containing about 100～200 frames or so.This two classes video is therefrom chosen at random respectively 10 frames and
20 frames.Table 1 represents the frame of the result better than gauss hybrid models of this paper algorithms extraction prospect in the frame that each video is randomly selected
Number and ratio.Wherein V1 to V5 is non-shake video, and remaining is shake video.
The foreground extraction Comparative result of table 1
In non-shake video, the performance of this algorithm is consistent with gauss hybrid models, therefore improved frame number ratio is always
0.In the video of shake, if extract frame be just shake more violent so improvement effect can be obvious, if taken out
The dither frame amplitude very little got, improved space also can accordingly reduce.The data of comprehensive upper table, this method is under complex scene
The improvement amplitude of foreground extraction is about 21%.
Above it is demonstrated experimentally that this algorithm compares gauss hybrid models has to the foreground extraction of the video that there is jitter phenomenon
Well adapting to property.
Claims (5)
1. it is a kind of shake background under foreground extracting method, it is characterised in that include：
Step S1：The position of the size, the step-length of slide block movement and slide block of initialisation image slide block；
Step S2：Two field picture before and after reading from video, and obtain the binary image be_frame and rear frame figure of prior image frame
The binary image af_frame of picture；
Step S3：Same position of the image slider in the binary image of prior image frame and the binary image of rear two field picture is carried
Take, the center of gravity for calculating two extraction results is poor, the motion vector of frame before and after obtaining under present image slide position；
Step S4：Image slider is moved by step-length, repeat step S3, until completing the extraction of whole image；
Step S5：According to the average motion vector of frame before and after all of motion vector computation of frame in front and back for obtaining；
Step S6：Rear two field picture is reversely translated according to the average motion vector for obtaining, the jitter error image being mitigated；
Step S7：Two field picture carries out foreground extraction before and after after being shaken to adjustment using gauss hybrid models.
2. the foreground extracting method under a kind of shake background according to claim 1, it is characterised in that step S6 tool
Body includes：
Step S61：Whether current average motion vector is judged more than or equal to 2, if it has, then execution step S62 is if it has not, then
Execution step S64；
Step S62：The size of image slider is amplified into current twice, repeat step S3 to S5 obtains another mean motion
Vector, and using the mean value of this average motion vector and former average motion vector as final average motion vector, and perform step
Rapid S63；
Step S63：Rear two field picture is reversely translated according to final average motion vector, the jitter error image being mitigated；
Step S64：The binary image of the binary image of prior image frame and rear two field picture is amplified into current 10 times, weight
Multiple repeat step S3 to S5, obtains another average motion vector, and using this average motion vector as final average motion vector,
And execution step S65；
Step S65：Rear two field picture is amplified into current 10 times, according to final average motion vector to amplification after rear two field picture
Reversely translated, be then contracted to life size, the jitter error image being mitigated.
3. it is according to claim 1 and 2 it is a kind of shake background under foreground extracting method, it is characterised in that described image
The original dimension of slide block is 50 pixel × 50 pixels, and step-length is 25 pixels.
4. the foreground extracting method under a kind of shake background according to claim 1, it is characterised in that frame before and after described
Motion vector is specially：
Wherein：CxbeThe x coordinate of the center of gravity of part, Cy are extracted by image slider for prior image framebeIt is prior image frame by image slider
Extract the y-coordinate of the center of gravity of part, CxafThe x coordinate of the center of gravity of part, Cy are extracted by image slider for rear two field pictureafFor rear frame
Image is extracted the y-coordinate of the center of gravity of part by image slider.
5. it is according to claim 1 it is a kind of shake background under foreground extracting method, it is characterised in that barycentric coodinates are concrete
For：
Wherein：Cx is the x coordinate of center of gravity, and Cy is the y-coordinate of center of gravity, and W is all pixels that image is extracted part by image slider
Pixel value sum, wiFor the pixel value of pixel i, xiFor the x coordinate of pixel i, yiFor the y-coordinate of pixel i, m is image
The pixel total number of part is extracted by image slider.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710083910.4A CN106651918B (en) | 2017-02-16 | 2017-02-16 | Foreground extraction method under shaking background |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710083910.4A CN106651918B (en) | 2017-02-16 | 2017-02-16 | Foreground extraction method under shaking background |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106651918A true CN106651918A (en) | 2017-05-10 |
| CN106651918B CN106651918B (en) | 2020-01-31 |
Family
ID=58846281
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710083910.4A Active CN106651918B (en) | 2017-02-16 | 2017-02-16 | Foreground extraction method under shaking background |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106651918B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109697689A (en) * | 2017-10-23 | 2019-04-30 | 北京京东尚科信息技术有限公司 | Storage medium, electronic equipment, image synthesizing method and device |
| CN109724992A (en) * | 2018-07-23 | 2019-05-07 | 永康市柴迪贸易有限公司 | Cabinet for TV cleannes analytical mechanism |
| CN110458820A (en) * | 2019-08-06 | 2019-11-15 | 腾讯科技（深圳）有限公司 | A kind of multimedia messages method for implantation, device, equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1211872A (en) * | 1997-06-04 | 1999-03-24 | 株式会社日立制作所 | Image signal system converter and TV set |
| CN1647113A (en) * | 2002-04-11 | 2005-07-27 | 皇家飞利浦电子股份有限公司 | Motion estimation unit and method of estimating a motion vector |
| CN1921628A (en) * | 2005-08-23 | 2007-02-28 | 松下电器产业株式会社 | Motion vector detection apparatus and motion vector detection method |
| CN101090456A (en) * | 2006-06-14 | 2007-12-19 | 索尼株式会社 | Image processing device and method, image pickup device and method |
| US8325810B2 (en) * | 2002-06-19 | 2012-12-04 | Stmicroelectronics S.R.L. | Motion estimation method and stabilization method for an image sequence |
| CN104410855A (en) * | 2014-11-05 | 2015-03-11 | 广州中国科学院先进技术研究所 | Jitter detection method of monitoring video |
-
2017
- 2017-02-16 CN CN201710083910.4A patent/CN106651918B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1211872A (en) * | 1997-06-04 | 1999-03-24 | 株式会社日立制作所 | Image signal system converter and TV set |
| CN1647113A (en) * | 2002-04-11 | 2005-07-27 | 皇家飞利浦电子股份有限公司 | Motion estimation unit and method of estimating a motion vector |
| US8325810B2 (en) * | 2002-06-19 | 2012-12-04 | Stmicroelectronics S.R.L. | Motion estimation method and stabilization method for an image sequence |
| CN1921628A (en) * | 2005-08-23 | 2007-02-28 | 松下电器产业株式会社 | Motion vector detection apparatus and motion vector detection method |
| CN101090456A (en) * | 2006-06-14 | 2007-12-19 | 索尼株式会社 | Image processing device and method, image pickup device and method |
| CN104410855A (en) * | 2014-11-05 | 2015-03-11 | 广州中国科学院先进技术研究所 | Jitter detection method of monitoring video |
Non-Patent Citations (2)
| Title |
|---|
| 李尊民: "《电视图像自动跟踪的基本原理》", 30 September 1998 * |
| 胡彦婷: "一种基于视觉感知的非线性幅型比变换方法", 《中国图象图形学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109697689A (en) * | 2017-10-23 | 2019-04-30 | 北京京东尚科信息技术有限公司 | Storage medium, electronic equipment, image synthesizing method and device |
| CN109697689B (en) * | 2017-10-23 | 2023-09-01 | 北京京东尚科信息技术有限公司 | Storage medium, electronic device, video synthesis method and device |
| CN109724992A (en) * | 2018-07-23 | 2019-05-07 | 永康市柴迪贸易有限公司 | Cabinet for TV cleannes analytical mechanism |
| CN110458820A (en) * | 2019-08-06 | 2019-11-15 | 腾讯科技（深圳）有限公司 | A kind of multimedia messages method for implantation, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106651918B (en) | 2020-01-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020238560A1 (en) | Video target tracking method and apparatus, computer device and storage medium | |
| CN107358623B (en) | Relevant filtering tracking method based on significance detection and robustness scale estimation | |
| EP2864933B1 (en) | Method, apparatus and computer program product for human-face features extraction | |
| CN111861925B (en) | Image rain removing method based on attention mechanism and door control circulation unit | |
| CN111127308B (en) | Mirror image feature rearrangement restoration method for single sample face recognition under partial shielding | |
| CN108109162B (en) | Multi-scale target tracking method using self-adaptive feature fusion | |
| CN112184759A (en) | Moving target detection and tracking method and system based on video | |
| CN102156995A (en) | Video movement foreground dividing method in moving camera | |
| CN104751484B (en) | A kind of moving target detecting method and the detecting system for realizing moving target detecting method | |
| CN109919002B (en) | Yellow stop line identification method and device, computer equipment and storage medium | |
| CN112581540B (en) | Camera calibration method based on human body posture estimation in large scene | |
| CN110334589A (en) | A kind of action identification method of the high timing 3D neural network based on empty convolution | |
| CN107862680B (en) | Target tracking optimization method based on correlation filter | |
| CN110827304B (en) | Traditional Chinese medicine tongue image positioning method and system based on deep convolution network and level set method | |
| CN106651918A (en) | Method for extracting foreground under shaking background | |
| CN102314591B (en) | Method and equipment for detecting static foreground object | |
| CN105095898B (en) | A kind of targeted compression cognitive method towards real-time vision system | |
| CN114170570A (en) | Pedestrian detection method and system suitable for crowded scene | |
| CN101184235B (en) | Method and apparatus for implementing background image extraction from moving image | |
| Yang et al. | Background extraction from video sequences via motion-assisted matrix completion | |
| CN109784215B (en) | In-vivo detection method and system based on improved optical flow method | |
| CN113989556A (en) | Small sample medical image classification method and system | |
| Chen | Moving object detection based on background extraction | |
| Li et al. | CDMY: A lightweight object detection model based on coordinate attention | |
| CN111145221A (en) | Target tracking algorithm based on multi-layer depth feature extraction |
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 |