CN102298604A - Video event detection method based on multi-media analysis - Google Patents
Video event detection method based on multi-media analysis Download PDFInfo
- Publication number
- CN102298604A CN102298604A CN2011101404009A CN201110140400A CN102298604A CN 102298604 A CN102298604 A CN 102298604A CN 2011101404009 A CN2011101404009 A CN 2011101404009A CN 201110140400 A CN201110140400 A CN 201110140400A CN 102298604 A CN102298604 A CN 102298604A
- Authority
- CN
- China
- Prior art keywords
- video
- event
- incident
- model
- video data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 27
- 238000000034 method Methods 0.000 claims abstract description 43
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 13
- 241001269238 Data Species 0.000 claims description 5
- 238000013459 approach Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 abstract description 4
- 238000002372 labelling Methods 0.000 abstract 2
- 230000006870 function Effects 0.000 description 8
- 238000012544 monitoring process Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000012706 support-vector machine Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 206010057315 Daydreaming Diseases 0.000 description 1
- 208000019914 Mental Fatigue Diseases 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 230000002547 anomalous effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 235000012054 meals Nutrition 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000035484 reaction time Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a video event detection method based on multi-media analysis, which comprises the following steps: analyzing a video by utilizing text analysis, so as to obtain a small amount of automatic labeling video data; searching an engine through a network video by utilizing a plurality of keywords, so as to obtain a large amount of video data related to the event; training by utilizing a small amount of automatic labeling video data and a large amount of video data related to the event based on a semi-supervision multi-instance learning algorithm of a graph model; constructing a math descriptive model of the event by utilizing the similarity standard of event instances and a positive-negative attributive standard of an event package; solving a large amount of video data related to the event by adopting a restricted convex-concave process method; learning by utilizing a local similarity measurement learning method and space distribution characteristics of a sample, so as to obtain effective similarity; realizing event identification and positioning according to the obtained event model from learning; and carrying out semantic analysis on the video content according to the event model, so as to obtain the position information of the event in the video.
Description
Technical field
The invention belongs to mode identification technology, relate to the Video Events detection method of analyzing based on multimedia.
Background technology
Refer to use a computer based on the Video Events detection technique of multimedia analysis and from video, carry out computing and analysis by computational algorithm software, extract the useful information in the video, finish a technology of this information extraction and understanding, custom holds water, be exactly extraction and the understanding of computing machine to " content " of video, the most key technology of this process is to adopt computing machine that vision signal is analyzed, to extract some the specific incidents that take place in the video scene or the specific behavior of monitoring objective.
Continuous development along with video processing technique, and the needs of international anti-terrorism situation, the intelligent video analysis technology becomes a research focus gradually, developed comparatively ripe product abroad, and be successfully applied to all kinds of supervisory systems, realize detecting automatically and automatic warning function, and at home, each producer and research unit are also in the active research exploitation, and new homemade intelligent video analysis technical products also will occur on market very soon.
The intelligent video analysis The Application of Technology at first is in the monitoring field, real active computer is monitoring and " checking " video in real time, when occurring, remind suspicious event the monitor staff to carry out the artificial affirmation second time again, in this way, can filter out the normal video more than 90%, need not the monitor staff stare at any time and watch, monitor staff's work efficiency is improved greatly, not only make the monitor staff people take into account several roads simultaneously, tens tunnel videos on roads up to a hundred even, and can effectively overcome the intrinsic mental fatigue of monitor staff, physilogical characteristics such as absent minded, what is more important, make for the monitoring of anomalous event by the intelligent video analysis technology and can accomplish to monitor in real time Realtime Alerts, shorten the reaction time when noting abnormalities incident greatly, make original supervisory system be unlikely to become a kind of " video recording inquiry system ".Because have so many significantly advantages, protection and monitor field is applied to this technology in the actual engineering system gradually.
Event detection is the key of various problems such as video analysis and understanding, more and more is subjected to paying attention to widely.At present, (physical culture, news, film, monitoring) has diverse ways for different video source, but lacking a unified method analyzes different video source.The incident Detection Algorithm of many existence relies on the description feature and the domain knowledge of video, and the learning method training pattern of supervision.Because different domain knowledge and insufficient training samples, it is difficult therefore setting up a general framework.
Summary of the invention
In view of this, the semi-supervised learning method that the objective of the invention is to propose a novelty is analyzed the different video data.
For reaching described purpose, the technical scheme that the present invention is based on the Video Events detection method of multimedia analysis comprises the steps:
Step S1: utilize text analyzing, film, physical culture, news video are analyzed, obtain a spot of automatic mark video data;
Step S2: utilize a plurality of keywords,, obtain the relevant a large amount of video data of incident by the Internet video search engine;
Step S3: based on semi-supervised many learn-by-example algorithms of graph model, utilize a spot of automatic mark video data a large amount of video data relevant to train with incident, and utilize the similarity criterion of incident example and the positive and negative attribute criterion of incident bag, construct the mathematics description model of incident, mathematics description model based on this incident, take constrained convex-concave process approach, a large amount of video data that incident is relevant is found the solution;
Step S4: local similar degree tolerance learning method, utilize the spatial characteristics of sample, study obtains effective similarity, the identification of event model realization event and the location that obtain according to study.
Step S5: according to event model, the content of video is carried out semantic analysis, obtain the positional information of incident in the video.
Preferably,, obtain a spot of training data that label is arranged automatically, thereby reduce the cost of artificial mark sample by utilizing text analyzing.
Preferably,, obtain a large amount of no label datas, increase the quantity of information of DATA DISTRIBUTION, also reduce the cost of artificial mark sample by the Internet video search engine.
Preferably, by the semi-supervised many learn-by-example algorithms based on graph model, merging effectively has label data and no label data, obtains event model preferably thereby can train.
Preferably, by local similar degree tolerance learning method, can learn out effective measuring similarity according to class target local distribution relation, thereby improve the correctness of event detection.
Preferably, by event model, the positional information of locating events effectively, thus the high-level semantic of realizing video content is understood.
Beneficial effect of the present invention: the semi-supervised learning method that the present invention proposes a novelty is analyzed the different video data.The inventive method is based on the semi-supervised many learn-by-example algorithms of graph model, and this algorithm utilizes a spot of the have sample of label and a large amount of weak exemplar training event models.Wherein, the sample that label is arranged is to obtain by text analyzing, and no exemplar is to obtain from the internet by keyword.By utilizing semi-supervised learning, from minute excavating the information of no exemplar, thereby solved the problem of lack of training samples.By utilizing no exemplar, improved the accuracy of event detection.By utilizing many learn-by-examples, solved the problem of incident location in the video, thereby can realize index, retrieval and the semantic understanding of video effectively.By utilizing the method for local similar degree tolerance study, solve the incorrect problem of measuring similarity, thereby made up graph model effectively, improved the correctness of event detection.
Description of drawings
Fig. 1 is an incident Detection Algorithm study block diagram of the present invention;
Fig. 2 is the local similar degree study among the present invention;
Fig. 3 is the movie database among the present invention;
Fig. 4 is the sport database among the present invention;
Fig. 5 is the news database among the present invention;
Fig. 6 is visual signature and the audio frequency characteristics among the present invention;
Fig. 7 is the incident positioning result example among the present invention.
Embodiment
For making the purpose, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in more detail.
The technical scheme that the present invention is based on the Video Events detection method of multimedia analysis comprises the steps: that (1) utilizes text analyzing, and film, physical culture, news video are analyzed, and obtains a spot of mark video data; (2) obtain the relevant multitude of video data of incident as no exemplar, increased the quantity of information of DATA DISTRIBUTION; (3) based on semi-supervised many learn-by-example algorithms of graph model, utilize a spot of automatic mark video data multitude of video data relevant to train, obtain event model with incident; (4) event model that obtains based on study, the identification of realization event and detection.By utilizing text analyzing and Internet video search engine, can obtain a spot of the have training data of label and a large amount of no label datas automatically, thereby reduce the cost of artificial mark sample, and increased the quantity of information of DATA DISTRIBUTION, thereby provide the foundation for the training of behavior model.By semi-supervised many learn-by-example algorithms based on graph model, utilize a spot of automatic mark video data a large amount of video data relevant to train with incident, and utilize the similarity criterion of incident example and the positive and negative attribute criterion of incident bag, construct the mathematics description model of incident, mathematics description model based on this incident, take constrained convex-concave process approach, a large amount of video data that incident is relevant is found the solution; Based on the method for local similar degree tolerance study, can utilize the spatial characteristics of sample, thereby study obtains effective similarity, has improved the correctness of event detection.
The present invention is based on the semi-supervised many learn-by-example algorithms of graph model and utilizes a spot of the have sample of label and a large amount of weak exemplar training event models, the sample that label is arranged is to obtain by text analyzing, and weak exemplar is to obtain from the internet by keyword; This algorithm the important point is the weight that how to obtain the limit of graph model, and for addressing this problem, the present invention has adopted many examples to induce similarity mechanism.We are by the algorithm that three kinds of different video source tests propose, and experimental result shows that the method that we propose is effective.The video data collection method is characterized in that obtaining automatically a spot of the have training data of label and a large amount of network datas.Thereby reduce the quantity of artificial mark sample, and improved the training effect of event model.Based on semi-supervised many learn-by-example algorithms of graph model, this part comprises how making up effective event model and solving model how.When making up event model, we consider the similarity criterion of incident example, the positive and negative attribute criterion of incident bag, thus obtain the mathematics description model of incident.Based on this model, we have taked effective constrained convex-concave process approach to find the solution, thereby our method can be used on large-scale database.
The purpose of the present invention's research is the position of orienting interested incident in the video sequence.Therefore, the present invention proposes the research framework of a novelty, it comprises five parts: (1) has label data to collect, and these data obtain by text analyzing; (2) no label data is collected, and these data obtain by network search engines; (3) based on many learn-by-examples, study local similar tolerance; (4) based on semi-supervised many learn-by-example algorithms of graph model, this part comprises the foundation of graph model, how to make up effective figure and solving model how; (5) event detection, identification of event model realization event and location that this part obtains according to study.Fig. 1 illustrates incident Detection Algorithm study block diagram, and for five above-mentioned parts, detailed content is as follows:
(1), label data is arranged
Obtain a spot of mark sample according to text analyzing.For example for film video, we can adopt the captions and the drama of film; Can adopt the literal of network direct broadcasting for sports video; Can use closed caption for news video.
(2), no label data
Utilize a plurality of keyword searches, obtain a large amount of event related data by network.For example we can use Google, and YouTube etc. are by a plurality of keyword retrievals, thereby the video data that the incident that obtains is correlated with has increased the quantity of information of DATA DISTRIBUTION as no exemplar.
(3), local similar tolerance
Because existing method for measuring similarity, as arest neighbors, Gauss is apart from waiting space structure of having ignored sample, so we utilize the local space of sample, the similarity measurement of a part of study.Fig. 2 has shown local similar degree learning method, and based on the local classifiers that this study obtains, we can define similarity measurement effectively.
(4), based on semi-supervised many learn-by-example algorithms of graph model
In order to obtain effective event model according to a spot of mark sample and a large amount of related datas, the many learn-by-examples algorithm based on the semi-supervised learning of graph model is proposed, the detailed content of this algorithm is as follows:
According to the similarity principle, similar incident example should have identical class mark, therefore obtains objective function E
1(f):
Wherein, i and j are the index of incident example, w
IjBe the similarity of incident example i and j, f
iBe the class mark of incident example i, f
jBe the class mark of incident example j, d
iAnd d
jBe respectively the capable and j of the i of w capable do not comprise diagonal element and.
In addition, for negative incident bag (negative video), we require its all incident examples (cutting apart for one of video) to bear, that is to say, the class mark of all incident examples all should be near 0, so we have following constraint:
Wherein, L
1Be the number of positive incident bag, | L| is the number of all incident bags, and b is the index of bag, and j is the index of incident example, x
bPresentation of events bag b.
Again according to positive incident bag (positive video) as long as it is positive requirement that there is an incident example (cutting apart of video) the inside, we obtain following penalty term:
Wherein the implication of scalar such as formula 1 are identical with 2.
According to these three, we obtain following event model:
Event model hereto, we have following two problems: the measuring similarity matrix w how (1) obtains; (2) how to optimize and find the solution; Study obtains local classifiers based on the local similar degree, and it is as follows that we define the similarity measurement matrix:
Mi-SVM wherein
s(x
B, j) be the sorter mi-SVM that obtains for incident example s study mark to the incident example of incident bag b.I
sBe the feature of incident example, r
sIt is the radius of local space.Based on formula (6), we just can obtain the similarity matrix in the formula (4).Because formula (4) is non-convex function, so objective function (4) can not be with the optimization method for solving of existing convex function.And objective function (4) can be regarded a convex function as and adds a concave function, therefore can find the solution with the concavo-convex process of iteration.We just can obtain the class mark of each incident example based on this method for solving, mark according to the class of each incident example and just can obtain the classification of incident bag and the position of orienting interested incident.
(5), event detection:
This part introduces the result of event detection in detail.To divide the following aspects introduction: (1) database; (2) feature description; (3) different similarity recognition results; (4) different learning strategy recognition results; (5) incident positioning result.
(1) database:
We test on three databases, comprise as follows:
Movie database such as Fig. 3 comprise and making a phone call, and fight, and applaud, and run, and drive six incidents such as kiss.
Sport database such as Fig. 4 comprise the basketball foul, the basketball free ball, and basketball is run, basketball shooting, football foul, six incidents such as soccer goal.
News database such as Fig. 5 comprise and having a meal, parade, four incidents such as device of dancing and play music.
(2) feature description:
Feature description is based on video content and carries out event detection, and we use visual signature and audio frequency characteristics, as shown in Figure 6.
(3) different similarity recognition results;
We have compared with the accurate side of different measuring similarities, to the result of event recognition.We have compared traditional k nearest neighbor, Gauss's distance, the method for sparse tolerance and our part tolerance.The result is shown in following table (1): we get method and are better than other method as can be seen from the table, and the result has improved general 7 percentage points.
Table (1) different similarity recognition results
(4) different learning strategy recognition results:
We have compared with different learning strategies, to the result of event recognition.We have compared support vector machine, the support vector machine of many learn-by-examples and the method that we propose.The result is shown in following table (2): our learning method is better than other two kinds of methods as can be seen from the table, and the result has improved general 4 percentage points.The reason that experimental result improves is the position and the related data training event model that utilized a large amount of no label of our method with the method locating events of many learn-by-examples.
Table (2) different learning method recognition results
(5) incident positioning result.
We have compared the result who locatees about incident with the support vector machine method of many learn-by-examples and our method.The result is shown in following table (3): our learning method is better than the support vector machine method of many learn-by-examples as can be seen from the table.The reason that experimental result improves is that our method has utilized the related data of a large amount of no labels to train event model.
Table (3) incident location comparative result
In addition, the result of some incident location is illustrated in fig. 7 shown below, and the red rectangle frame among the figure is the incident positioned area, and changing plan has provided kiss, applauds the incident annotation results that basketball is freely shot and paraded.As can be seen from the results, our method is effective.
The above; only be the embodiment among the present invention, but protection scope of the present invention is not limited thereto, anyly is familiar with the people of this technology in the disclosed technical scope of the present invention; conversion or the replacement expected can be understood, all of the present invention comprising within the scope should be encompassed in.
Claims (6)
1. a Video Events detection method of analyzing based on multimedia is characterized in that, the step that described Video Events detects comprises:
Step S1: utilize text analyzing, film, physical culture, news video are analyzed, obtain a spot of automatic mark video data;
Step S2: utilize a plurality of keywords,, obtain the relevant a large amount of video data of incident by the Internet video search engine;
Step S3: based on semi-supervised many learn-by-example algorithms of graph model, utilize a spot of automatic mark video data a large amount of video data relevant to train with incident, and utilize the similarity criterion of incident example and the positive and negative attribute criterion of incident bag, construct the mathematics description model of incident, mathematics description model based on this incident, take constrained convex-concave process approach, a large amount of video data that incident is relevant is found the solution;
Step S4: local similar degree tolerance learning method, utilize the spatial characteristics of sample, study obtains effective similarity, the identification of event model realization event and the location that obtain according to study.
Step S5: according to event model, the content of video is carried out semantic analysis, obtain the positional information of incident in the video.
2. the Video Events detection method of analyzing based on multimedia according to claim 1 is characterized in that, by utilizing text analyzing, obtains a spot of training data that label is arranged automatically, thereby reduces the cost of artificial mark sample.
3. the Video Events detection method of analyzing based on multimedia according to claim 1 is characterized in that, by the Internet video search engine, obtains a large amount of no label datas, increases the quantity of information of DATA DISTRIBUTION, also reduces the cost of artificial mark sample.
4. the Video Events detection method of analyzing based on multimedia according to claim 1, it is characterized in that, by the semi-supervised many learn-by-example algorithms based on graph model, merging effectively has label data and no label data, obtains event model preferably thereby can train.
5. the Video Events detection method of analyzing based on multimedia according to claim 1, it is characterized in that,, can concern according to class target local distribution by local similar degree tolerance learning method, learn out effective measuring similarity, thereby improve the correctness of event detection.
6. the Video Events detection method of analyzing based on multimedia according to claim 1 is characterized in that, by event model, and the positional information of locating events effectively, thus the high-level semantic of realizing video content is understood.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011101404009A CN102298604A (en) | 2011-05-27 | 2011-05-27 | Video event detection method based on multi-media analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011101404009A CN102298604A (en) | 2011-05-27 | 2011-05-27 | Video event detection method based on multi-media analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102298604A true CN102298604A (en) | 2011-12-28 |
Family
ID=45359024
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011101404009A Pending CN102298604A (en) | 2011-05-27 | 2011-05-27 | Video event detection method based on multi-media analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102298604A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103943107B (en) * | 2014-04-03 | 2017-04-05 | 北京大学深圳研究生院 | A kind of audio frequency and video keyword recognition method based on Decision-level fusion |
| CN107707931A (en) * | 2016-08-08 | 2018-02-16 | 阿里巴巴集团控股有限公司 | Generated according to video data and explain data, data synthesis method and device, electronic equipment |
| CN108846852A (en) * | 2018-04-11 | 2018-11-20 | 杭州电子科技大学 | Monitor video accident detection method based on more examples and time series |
| CN109002463A (en) * | 2018-06-05 | 2018-12-14 | 国网辽宁省电力有限公司信息通信分公司 | A kind of Method for text detection based on depth measure model |
| CN110580557A (en) * | 2018-06-08 | 2019-12-17 | 上海博泰悦臻网络技术服务有限公司 | Intelligent mobile terminal and fragment time management and utilization method thereof based on artificial intelligence |
| CN112989120A (en) * | 2021-05-13 | 2021-06-18 | 广东众聚人工智能科技有限公司 | Video clip query system and video clip query method |
| CN113646800A (en) * | 2018-09-27 | 2021-11-12 | 株式会社OPTiM | Object condition determination system, object condition determination method, and program |
| CN113711619A (en) * | 2020-03-20 | 2021-11-26 | 华为技术有限公司 | Multimedia data storage method, apparatus, device, storage medium and program product |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101339660A (en) * | 2007-07-05 | 2009-01-07 | 韩庆军 | Sports video frequency content analysis method and device |
-
2011
- 2011-05-27 CN CN2011101404009A patent/CN102298604A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101339660A (en) * | 2007-07-05 | 2009-01-07 | 韩庆军 | Sports video frequency content analysis method and device |
Non-Patent Citations (2)
| Title |
|---|
| 《MM'10 Proceedings of international conference on Multimedia》 20101029 Tianzhu Zhang et al A Generic Framework for Event Detectionin Various Video Domains 第103-112页 1-6 , * |
| TIANZHU ZHANG ET AL: "A Generic Framework for Event Detectionin Various Video Domains", 《MM’10 PROCEEDINGS OF INTERNATIONAL CONFERENCE ON MULTIMEDIA》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103943107B (en) * | 2014-04-03 | 2017-04-05 | 北京大学深圳研究生院 | A kind of audio frequency and video keyword recognition method based on Decision-level fusion |
| CN107707931A (en) * | 2016-08-08 | 2018-02-16 | 阿里巴巴集团控股有限公司 | Generated according to video data and explain data, data synthesis method and device, electronic equipment |
| CN108846852A (en) * | 2018-04-11 | 2018-11-20 | 杭州电子科技大学 | Monitor video accident detection method based on more examples and time series |
| CN108846852B (en) * | 2018-04-11 | 2022-03-08 | 杭州电子科技大学 | Monitoring video abnormal event detection method based on multiple examples and time sequence |
| CN109002463A (en) * | 2018-06-05 | 2018-12-14 | 国网辽宁省电力有限公司信息通信分公司 | A kind of Method for text detection based on depth measure model |
| CN110580557A (en) * | 2018-06-08 | 2019-12-17 | 上海博泰悦臻网络技术服务有限公司 | Intelligent mobile terminal and fragment time management and utilization method thereof based on artificial intelligence |
| CN113646800A (en) * | 2018-09-27 | 2021-11-12 | 株式会社OPTiM | Object condition determination system, object condition determination method, and program |
| CN113711619A (en) * | 2020-03-20 | 2021-11-26 | 华为技术有限公司 | Multimedia data storage method, apparatus, device, storage medium and program product |
| CN112989120A (en) * | 2021-05-13 | 2021-06-18 | 广东众聚人工智能科技有限公司 | Video clip query system and video clip query method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102298604A (en) | Video event detection method based on multi-media analysis | |
| US11023523B2 (en) | Video content retrieval system | |
| CN101894276B (en) | Training method of human action recognition and recognition method | |
| US11256741B2 (en) | Video tagging system and method | |
| Zhou et al. | Conceptlearner: Discovering visual concepts from weakly labeled image collections | |
| CN102034096B (en) | Video event recognition method based on top-down motion attention mechanism | |
| CN103020303B (en) | Based on the historical events extraction of internet cross-media terrestrial reference and the searching method of picture concerned | |
| Wang et al. | Video event detection using motion relativity and feature selection | |
| CN102314916A (en) | Video processing method and system | |
| CN111325153B (en) | Student behavior feature intelligent analysis method based on multidimensional data | |
| Zhang et al. | A generic framework for event detection in various video domains | |
| Li et al. | Technique of image retrieval based on multi-label image annotation | |
| Wu et al. | Novelty detection for cross-lingual news stories with visual duplicates and speech transcripts | |
| Li et al. | Social negative bootstrapping for visual categorization | |
| Rautiainen et al. | Cluster-temporal browsing of large news video databases | |
| Zhang et al. | Integration of visual temporal information and textual distribution information for news web video event mining | |
| Avgerinakis et al. | ITI-CERTH participation in TRECVID 2018. | |
| Can et al. | Modeling concept dependencies for event detection | |
| Guo | [Retracted] Intelligent Sports Video Classification Based on Deep Neural Network (DNN) Algorithm and Transfer Learning | |
| CN102043788A (en) | Video search system based on content comparison | |
| Cai et al. | Semantic entity detection by integrating CRF and SVM | |
| Zhang et al. | Adaptive association rule mining for web video event classification | |
| CN114862492A (en) | Method and system for recommending online shopping mall commodities | |
| Zhang et al. | A novel collaborative optimization framework for web video event mining based on the combination of inaccurate visual similarity detection information and sparse textual information | |
| Othman et al. | Features discovery for web classification using support vector machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20111228 |