CN108288045A - A kind of mobile video live streaming/monitor video acquisition source tagsort method - Google Patents
A kind of mobile video live streaming/monitor video acquisition source tagsort method Download PDFInfo
- Publication number
- CN108288045A CN108288045A CN201810095732.1A CN201810095732A CN108288045A CN 108288045 A CN108288045 A CN 108288045A CN 201810095732 A CN201810095732 A CN 201810095732A CN 108288045 A CN108288045 A CN 108288045A
- Authority
- CN
- China
- Prior art keywords
- video
- similarity
- word frequency
- frequency vector
- live streaming
- 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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/7867—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, title and artist information, manually generated time, location and usage information, user ratings
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Library & Information Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Disclosed in this invention is a kind of mobile video live streaming/monitor video acquisition source tagsort method, and the method includes S1：Word frequency vector is established based on the description information to all videos of presorting, establishes the word frequency vector of each video；S2：According to the word frequency of video vector, the similarity of each video and other videos is calculated using cosine similarity, according to the size of similarity to all video divides into several classes.Video source tagsort method proposed by the present invention, in conjunction with some parameters outside video image, including key factors such as keyword, geographical locations, calculate the similarity degree of mobile video acquisition source acquisition content, and then it is divided into several classes, each type can represent an independent event, can reach more careful, accurately classify to video source.
Description
Technical field
Technical field according to the present invention is the method classified to video, specially a kind of mobile video acquisition source
Tagsort method.
Technical background
With universal and 4G networks comprehensive covering of cell phone terminal, video movement live streaming, video mobile monitor
Equal new medias the Internet, applications are widely used in the production of people, life.In this applications, mobile phone can be used as video and adopt
The stream media service node of rear end is accessed in the source of collection, and video image information is issued to public user.Mobile live streaming/monitor video
Acquisition source can bring the continuous variation of content meaning due to its randomness, but the content in wherein many sources is often directed toward identical society
It can event.
The patent that notification number is CN107180074A disclosed a kind of video classification methods and dress on March 31st, 2017
It sets, the method includes：Obtain video file；The key frame for extracting each camera lens in the video file obtains multiple videos
Frame；For each video frame, the classification of the video frame is determined；Count the duration of the classification and each video frame of all video frame；
According to statistical result, the video file is classified.The video classification methods and device provided through the embodiment of the present invention,
It may be implemented in time to classify to the video on website, improve user experience, saved manpower.In the prior art, it uses
Key frame images processing, goes to judge whether video belongs to same event, there are still deficiencies.Because being directed to same event, same field
Scape, but the shooting angle residing for photographer is different, and the video image finally presented also can be different, will result in key frame not
Together, it is divided into different classifications so as to cause belonging to the video for playing same event, the classification to make mistake is done to video.
Invention content
The purpose of the present invention is to solve in the above problem, existing deficiency provides a kind of mobile video live streaming/monitoring
Video acquisition source tagsort method, the method that can preferably solve visual classification can represent an independent event per one kind.
The technical solution adopted in the present invention is to provide a kind of mobile video live streaming/monitor video acquisition source tagsort side
Method, the method includes S1：Word frequency vector is established based on the description information to all videos of presorting, establishes each video
Word frequency vector；S2：According to the word frequency of video vector, the similarity of each video and other videos is calculated using cosine similarity,
According to the size of similarity to all video divides into several classes.
Preferably, the m-vector that the word frequency vector is made of 0 and 1.
Preferably, the similarity for calculating each video and other videos is specially：Choose the word of any one video
Frequency Vector Groups, using cosine similarity algorithm calculate selected video its to the word frequency Vector Groups of other videos to each other similar
Degree, is sorted out according to the size of similarity, a new word frequency Vector Groups is reselected in remaining word frequency Vector Groups, weight
Step is calculated again, until all word frequency Vector Groups have been chosen, if all videos are divided into Ganlei.
Preferably, its similarity of word frequency Vector Groups with other videos to each other of the video selected by the calculating, by phase
Video like degree more than threshold value 0.7 is classified as one kind with selected video composition.
Preferably, each video corresponds to a geographic position data, based on the geographic position data of video to regarding
Frequency further divides.
Preferably, the step of further division includes：If in any classification in Ganlei, an optional video, root
According to geographic position data, by the video with it is similar in on selected video geographical location at a distance of the video group for being no more than certain distance
At a new set, optionally a video will be geographical with selected video in such according to geographic position data in the set
On position in the video merging to new set for being no more than certain distance, until the video in set has all been chosen.
Preferably, the certain distance is 600-1200 meters.
The present invention compared with prior art, have the advantage that for：
1. technical solution using the present invention may be implemented to carry out classification processing to mobile video acquisition source, user helped to carry out
Big data mining analysis, focus incident tracking and system resource draw the work such as grade distribution.
2. video source tagsort method proposed by the present invention, in conjunction with some parameters outside video image, including keyword,
The key factors such as geographical location, calculate the similarity degree of mobile video acquisition source acquisition content, and then are divided into several classes, each
Type can represent an independent event, can reach more careful, accurately classify to video source.
Description of the drawings
Fig. 1 is the flow chart of the present invention.
Fig. 2 is the schematic process of video source classifying and dividing of the present invention.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, below with reference to drawings and examples,
Invention is further described in detail.
First, the description information based on video establishes word frequency vector.General mobile video source can carry when being established by initiator
For several keywords as label for the description to the monitor event, or brief abstract can be used（It is removable to be divided into word）Description
The related content of live streaming or monitoring, these contents can be used as the important evidence of the social effects of videos such as excavation live streaming/monitoring, this
Method establishes the Vector Groups of characterization video by the key words frequency of occurrences.
S1:It is as follows：
1）Set video source set V={ V of existing online service1, V2... ..., Vm}
2）Set each video ViA corresponding crucial phrase is Pi = { PI1,PI2 ... ...Pij, each set PiIncluding
Word quantity can differ.
3）Take A=P1 ∪ P2 ∪ P3……∪Pm, then set A includes the non-duplicate independent crucial of all video sources
Lemma element, quantity set h, A={ A1, A2, A3... ..., Ah}
4）Set AiFor video ViA corresponding word frequency vector, Ai={Ai1, Ai2, Ai3... ..., Aih}
Wherein vector Aix∈Ai, then its vectorial exploitation method be：
If Ax ∈ Pi, then Aix= 1；
If Ax ∉ Pi, then Aix= 0；
Therefore, the word frequency vector A of video can be obtainedI,AiA m-vector being made of 0 and 1.
In order to make it easy to understand, specific example is as follows, two video sources of same time：
Video source 1, keyword are：The continents Pa world museum, the areas A, food win meeting, rostrum roadshow
Video source 2, keyword are：The continents Pa world museum, the areas B, An Bohui, robot building exhibition section
Then vectorial participle group is A={ continents Pa world museum, the areas A, the areas B, the rich meeting of food, An Bohui, rostrum roadshow, robot building
Exhibition section }, then
The word frequency vector value A of corresponding video source 11={ 1,1,0,1,0,1,0 }
The word frequency vector value A of corresponding video source 22={ 1,0,1,0,1,0,1 }
S2 specific steps：Word frequency vector based on each video that S1 steps obtain, video is calculated using cosine similarity algorithm
The similar characteristic in source, if being classified as Ganlei.Cosine similarity algorithm is to weigh two groups according to the corner dimension of vector space
A kind of effective ways of the similar probability of multi-C vector.Two groups of vector cosine values are found out according to the cosine law, cosine value more connects
It is bordering on 1, then angle is smaller, and similarity degree is higher, and cosine value is closer to 0, then angle is bigger, and similarity degree is lower.Work as angle
Then it is believed that two groups of vectors overlap when being 0, the two is completely similar, and when angle is 90 degree, then vector direction is at a right angle, then can recognize
It is completely dissimilar for the two.
5）From A1, A2, A3... ..., AmRandomly choose a Vector Groups Aj, calculate its to other Vector Groups to each other similar
Degree, computational methods are as follows：
Such as AjWith another vector AyBetween similarity：
6）Selection and AjSimilarity is more than the video source of threshold value 0.7, with AjForm new one kind, it is believed that in the video source in such
Appearance is more close, and maximum probability belongs to the same event of monitoring.
7）A Vector Groups are reselected in remaining Vector Groups, repeat step（5）, until all Vector Groups are selected
It is complete.
8）Finally obtain a classification results, Q={ Q1, Q2……Qw, contain all video sources in V, Q1, Q2……
QW,Respectively represent different classifications.
S3 specific steps：Geographical location information based on video acquisition source further divides the class that S2 steps obtain.It is existing
Many activity names are meant that much like in real field scape, but can not also be determined completely and be directed to same event, therefore also need to borrow
Help judgement precision of other factors raisings to classification.Therefore it is more than that daughter element is more than 1 in the class obtained for S2 steps
Class, the geographical location information that place can be uploaded based on video source further classify to the class that S2 steps generate, will to each other
The closer point of distance is further subdivided into one kind, is final result.
9）If all video sources correspond to a current geographic position data in V（It can be by mobile phone terminal timing acquiring
And it uploads, is described in a manner of longitude and latitude）G={G1, G2,……Gm, then can include several G first in some classification Qx
Element.
10）A classification Qx is selected in Q, calculates the air line distance between arbitrary source node in Qx, the distance between node can
It is identified as Dij；
11）A source G is arbitrarily selected in Qxi, while selecting all in Qx and GiDistance D is no more than threshold value, and threshold value is
1000 meters of all the points, with GiIt is combined into new class Qy, G at this timeiIt can be labeled as Selected；One is selected again in Qy not mark
Know the point Gj for being Selected, while selecting all in Qx and Gj distances D and being no more than threshold value（For example it may be set to 1000 meters）
All the points, be merged into Qy（Ignore if any the point repeated）, Gj can be labeled as Selected at this time, this step of repetition,
Until point all in Qy has been set to Selected；
12) node at this time from gathering the Qy ultimately produced in Qx all removes, and repeats step 11）, until all members of Qx
Element is all finished by selection.
13）Repeat step 10), 11）、12）, until class of the daughter element more than 1 is fully completed screening in all Q.
14）In summary the classification results more segmented finally can be obtained in method：Q’={Q’1, Q '2... ... Q 'z,
Q’1, Q '2... ... Q 'Z,Respectively finer classification is wrapped wherein containing all video sources in V in each classification
The video source contained is regarded as greater probability and is directed toward same event, and the video source in different classifications is believed that not larger pass
Connection property.
The present invention is described by embodiment, but is not limited the invention, with reference to description of the invention, institute
Other variations of disclosed embodiment, are such as readily apparent that the professional person of this field, such variation should belong to
Within the scope of the claims in the present invention limit.
Claims (7)
1. a kind of mobile video live streaming/monitor video acquires source tagsort method, it is characterised in that：The method includes S1：
Word frequency vector is established based on the description information to all videos of presorting, establishes the word frequency vector of each video；S2：According to video
Word frequency vector, the similarity of each video and other videos is calculated using cosine similarity, according to the size of similarity to institute
There is video divide into several classes.
2. mobile video live streaming according to claim 1/monitor video acquires source tagsort method, it is characterised in that：
The m-vector that above-mentioned word frequency vector is made of 0 and 1.
3. mobile video live streaming according to claim 1/monitor video acquires source tagsort method, it is characterised in that：
The similarity of the above-mentioned each video of calculating and other videos is specially：The word frequency Vector Groups for choosing any one video, using remaining
String similarity algorithm calculates selected video its similarity of word frequency Vector Groups with other videos to each other, according to similarity
Size is sorted out, and a new word frequency Vector Groups are reselected in remaining word frequency Vector Groups, compute repeatedly step, until
All word frequency Vector Groups have been chosen, if all videos are divided into Ganlei.
4. mobile video live streaming according to claim 3/monitor video acquires source tagsort method, it is characterised in that：
Its similarity of word frequency Vector Groups with other videos to each other of video selected by above-mentioned calculating, is more than threshold value 0.7 by similarity
Video and selected video composition be classified as one kind.
5. mobile video live streaming according to claim 1/monitor video acquires source tagsort method, it is characterised in that：
Each video corresponds to a geographic position data, is further divided to video based on the geographic position data of video.
6. mobile video live streaming according to claim 5/monitor video acquires source tagsort method, it is characterised in that：
The step of further division includes：If in any classification in Ganlei, an optional video, according to geographic position data,
By the video with it is similar in on selected video geographical location at a distance of be no more than certain distance video form a new set,
An alternative video in the set, according to geographic position data, by such on selected video geographical location at a distance of not surpassing
It crosses in the video merging to new set of certain distance, until the video in set has all been chosen.
7. mobile video live streaming according to claim 6/monitor video acquires source tagsort method, it is characterised in that：
The certain distance is 600-1200 meters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810095732.1A CN108288045B (en) | 2018-01-31 | 2018-01-31 | Mobile video live broadcast/monitoring video acquisition source feature classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810095732.1A CN108288045B (en) | 2018-01-31 | 2018-01-31 | Mobile video live broadcast/monitoring video acquisition source feature classification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108288045A true CN108288045A (en) | 2018-07-17 |
| CN108288045B CN108288045B (en) | 2020-11-24 |
Family
ID=62836039
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810095732.1A Active CN108288045B (en) | 2018-01-31 | 2018-01-31 | Mobile video live broadcast/monitoring video acquisition source feature classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108288045B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117255212A (en) * | 2023-11-20 | 2023-12-19 | 北京泰伯科技有限公司 | Remote emergency live broadcast control method and related equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101419614A (en) * | 2008-12-03 | 2009-04-29 | 深圳市迅雷网络技术有限公司 | Video resource clustering method and device |
| CN103488689A (en) * | 2013-09-02 | 2014-01-01 | 新浪网技术(中国)有限公司 | Mail classification method and mail classification system based on clustering |
| CN103678702A (en) * | 2013-12-30 | 2014-03-26 | 优视科技有限公司 | Video duplicate removal method and device |
| CN103838835A (en) * | 2014-02-25 | 2014-06-04 | 中国科学院自动化研究所 | Network sensitive video detection method |
| CN104182421A (en) * | 2013-05-27 | 2014-12-03 | 华东师范大学 | Video clustering method and detecting method |
| CN104794215A (en) * | 2015-04-27 | 2015-07-22 | 北京工业大学 | Fast recursive clustering method suitable for large-scale data |
| CN104834686A (en) * | 2015-04-17 | 2015-08-12 | 中国科学院信息工程研究所 | Video recommendation method based on hybrid semantic matrix |
-
2018
- 2018-01-31 CN CN201810095732.1A patent/CN108288045B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101419614A (en) * | 2008-12-03 | 2009-04-29 | 深圳市迅雷网络技术有限公司 | Video resource clustering method and device |
| CN104182421A (en) * | 2013-05-27 | 2014-12-03 | 华东师范大学 | Video clustering method and detecting method |
| CN103488689A (en) * | 2013-09-02 | 2014-01-01 | 新浪网技术(中国)有限公司 | Mail classification method and mail classification system based on clustering |
| CN103678702A (en) * | 2013-12-30 | 2014-03-26 | 优视科技有限公司 | Video duplicate removal method and device |
| CN103838835A (en) * | 2014-02-25 | 2014-06-04 | 中国科学院自动化研究所 | Network sensitive video detection method |
| CN104834686A (en) * | 2015-04-17 | 2015-08-12 | 中国科学院信息工程研究所 | Video recommendation method based on hybrid semantic matrix |
| CN104794215A (en) * | 2015-04-27 | 2015-07-22 | 北京工业大学 | Fast recursive clustering method suitable for large-scale data |
Non-Patent Citations (4)
| Title |
|---|
| 刘璐 等: ""基于文本扩展模型的网络视频聚类方法"", 《智能系统学报》 * |
| 艾丽丽: ""基于文本挖掘的视频资源分类研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 陈翰: ""新闻视频的语义提取与自动分类技术研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 黄海琦: ""基于多模态策略的网络视频聚类方法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117255212A (en) * | 2023-11-20 | 2023-12-19 | 北京泰伯科技有限公司 | Remote emergency live broadcast control method and related equipment |
| CN117255212B (en) * | 2023-11-20 | 2024-01-26 | 北京泰伯科技有限公司 | Remote emergency live broadcast control method and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108288045B (en) | 2020-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guo et al. | The emergence of visual crowdsensing: Challenges and opportunities | |
| CN104754413B (en) | Method and apparatus for identifying television signals and recommending information based on image search | |
| US20160034712A1 (en) | System and method for event-related content discovery, curation, and presentation | |
| CN106203498B (en) | Urban scene garbage detection method and system | |
| CN110796098B (en) | Method, device, equipment and storage medium for training and auditing content auditing model | |
| US20160110381A1 (en) | Methods and systems for social media-based profiling of entity location by associating entities and venues with geo-tagged short electronic messages | |
| CN103365936A (en) | Video recommendation system and method thereof | |
| US9275460B2 (en) | Reference orientations for viewing panoramic images | |
| CN106575280B (en) | System and method for analyzing user-associated images to produce non-user generated labels and utilizing the generated labels | |
| Giridhar et al. | On localizing urban events with instagram | |
| CN205263812U (en) | Distributing type face identification orbit searching system | |
| US20170039264A1 (en) | Area modeling by geographic photo label analysis | |
| CN105631404B (en) | The method and device that photo is clustered | |
| CN103699679A (en) | Method and equipment for retrieving information of target objects | |
| US20240037142A1 (en) | Systems and methods for filtering of computer vision generated tags using natural language processing | |
| WO2020052275A1 (en) | Image processing method and apparatus, terminal device, server and system | |
| CN103077236A (en) | System and method for realizing video knowledge acquisition and marking function of portable-type device | |
| CN103577400A (en) | Location information providing method and system | |
| CN105354252A (en) | Information processing method and apparatus | |
| CN110019827B (en) | Corpus generation method, apparatus, device and computer storage medium | |
| CN111445442A (en) | Crowd counting method and device based on neural network, server and storage medium | |
| CN114708426A (en) | Target detection method, model training method, device, equipment and storage medium | |
| CN106250396B (en) | Automatic image label generation system and method | |
| CN108288045A (en) | A kind of mobile video live streaming/monitor video acquisition source tagsort method | |
| CN113920585A (en) | Behavior recognition method and device, equipment and storage medium |
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 |