CN104463864B - Multistage parallel key frame cloud extracting method and system - Google Patents
Multistage parallel key frame cloud extracting method and system Download PDFInfo
- Publication number
- CN104463864B CN104463864B CN201410731007.0A CN201410731007A CN104463864B CN 104463864 B CN104463864 B CN 104463864B CN 201410731007 A CN201410731007 A CN 201410731007A CN 104463864 B CN104463864 B CN 104463864B
- Authority
- CN
- China
- Prior art keywords
- key frame
- frame
- video
- parallel
- camera lens
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
-
- 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
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
- G06V20/47—Detecting features for summarising video content
-
- 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
- G06V20/49—Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes
Abstract
The present invention discloses a kind of multistage parallel key frame cloud extracting method and system.The key frame cloud extraction system includes:Video input module is configured to the video input interface of extraction key frame;Video lens divide module, are configured to carry out shot segmentation to the video of input, to carry out key-frame extraction parallel processing;Multistage parallel key frame cloud processing module is configured to video lens through multistage parallel strategy, and by extracting frame feature vector parallel, parallel clustering operates, and generates each camera lens key frame group;Key frame output module, the key frame group for being configured to generate each camera lens are combined, and finally export the key frame group of entire video.There is higher efficiency when the multistage parallel key frame cloud extracting method and system of the present invention are to handling extensive key-frame extraction, meanwhile, there is good scalability and reliability, disclosure satisfy that the demand of extensive key-frame extraction.
Description
Technical field
The present invention relates to key frame cloud extractive techniques, more particularly, are related to a kind of multistage parallel key frame cloud extraction side
Method and multistage parallel key frame cloud extraction system.
Background technology
Key frame of video refers to most important, the representational one or more image that can be represented in camera lens.Key frame
Extraction can greatly reduce the treating capacity of video data, by the extensive concern of researcher.
Cloud computing seamless can expand to large-scale cluster, and can tolerate the error code of part of nodes, or even very big
Failure, which occurs, for part of nodes will not influence the correct operation of program, therefore cloud computing has preferable scalability and stabilization
Property.
The main method of existing key-frame extraction includes:Method based on shot boundary is closed based on motion analysis extraction
Key frame extracts key frame etc. based on image information extraction key frame, based on cluster.Most of research is concentrated mainly on Video Key
In terms of the extraction accuracy of frame, but the extraction of key frame of video is related to many digital operations, and tool video frame quantity is more, serial video
Key Frame Extraction can time-consuming, and single machine processing capacity is limited, therefore it is necessary to study a kind of multistage parallel key frame cloud extraction sides
Method and cloud extraction system.
Invention content
In order to solve the above-mentioned problems of the prior art, the purpose of the present invention is to provide a kind of multistage parallel key frames
Cloud extraction system, wherein the multistage parallel key frame cloud extraction system includes:Video input module is configured to extraction and closes
The video input interface of key frame;Video lens divide module, are configured to carry out shot segmentation to the video of input, to carry out
Key-frame extraction parallel processing;Multistage parallel key frame cloud processing module, is configured to pass through multistage parallel plan to video lens
Slightly, by extracting frame feature vector, parallel clustering operation parallel, each camera lens key frame group is generated;Key frame output module, by structure
It makes and is combined for the key frame group generated to each camera lens, finally export the key frame group of entire video.
Another object of the present invention, which also resides in, provides a kind of multistage parallel key frame cloud extracting method, wherein the multistage
Key frame cloud extracting method includes parallel:Receiving need to extract the video of key frame;Shot segmentation is carried out to the video;To described
Camera lens carries out multistage parallel key frame cloud extraction operation;The camera lens key frame group extracted is combined, final video is exported
Key frame group.
Further, the multistage parallel key frame cloud extraction operation includes parallel extraction video frame feature vector, basis
Video frame feature vector parallel clustering operates.
Further, the camera lens key frame group combination, including final cluster generates the final key frame of video.
Further, the video feature vector cluster operation can be k-means cluster, fuzzy C-means clustering or other
It is suitble to the characteristics of image clustering algorithm of parallel processing.
The multistage parallel key frame cloud extracting method and cloud extraction system of the present invention can greatly improve key-frame extraction efficiency,
There is good scalability and reliability simultaneously.
Description of the drawings
Fig. 1 is multistage parallel key frame cloud extraction system schematic diagram according to an embodiment of the invention.
Fig. 2 is the flow chart of multistage parallel key frame cloud extraction dispatching method according to an embodiment of the invention.
Fig. 3 is that multistage parallel key frame cloud according to an embodiment of the invention extracts certain example operation figure.
Specific implementation mode
The embodiment of the present invention is described in detail now, the example is illustrated in the accompanying drawings, wherein identical label
Always show same parts.Embodiment is described to explain the present invention below with reference to the accompanying drawings.In the accompanying drawings, in order to clear
For the sake of clear, the thickness of layer and region can be exaggerated.In the following description, in order to avoid known features and/or function need not
The present inventive concept caused by detailed description wanted is obscured, and can omit known features and/or the unnecessary of function is retouched in detail
It states.
Fig. 1 is multistage parallel key frame cloud extraction system schematic diagram according to an embodiment of the invention.
Referring to Fig.1, multistage parallel key frame cloud extraction system according to an embodiment of the invention includes:Video input module
10, it is configured to the video input interface of extraction key frame;Video lens divide module 20, be configured to the video of input into
Row shot segmentation, to carry out key-frame extraction parallel processing;Multistage parallel key frame cloud processing module 30, is configured to regarding
By multistage parallel strategy, by extracting frame feature vector parallel, parallel clustering operates frequency camera lens, generates each camera lens key frame
Group;Key frame output module 40, the key frame group for being configured to generate each camera lens are combined, and finally export entire video
Key frame group.
In addition, multistage parallel key frame cloud extraction operation includes parallel extraction video frame feature vector, according to video frame spy
Levy Vector Parallel cluster operation.
The video feature vector cluster operation can be k-means clusters, fuzzy C-means clustering or other suitable parallel places
The characteristics of image clustering algorithm of reason.
Correspondingly, the present invention also provides a kind of multistage parallel key frame cloud extracting method, Fig. 2 is specifically please referred to,
It is the flow chart of multistage parallel key frame cloud extracting method according to an embodiment of the invention.
With reference to Fig. 2, multistage parallel key frame cloud extracting method according to an embodiment of the invention includes:S1, receiving need to be into
The video of row Key Frame Extraction;S2, video lens segmentation is carried out to video;S3, key frame of video is extracted to camera lens multistage parallel;
S4, output is combined to camera lens key frame group.
In the present embodiment, after video lens segmentation, video frame feature vector can be extracted parallel, can be regarded parallel later
Frequency feature vector clusters operate, and cluster can be the image of k-means clusters, fuzzy C-means clustering or other suitable parallel processings
Feature clustering algorithm.To be to extract video frame Vector Processing and the parallel k-means cluster extraction key frames that carry out parallel below
Example is next, and the present invention will be described.Wherein, K-means algorithms are the very typical clustering algorithms based on distance, are made using distance
For the evaluation index of similitude, that is, think that the distance of two objects is closer, similarity is bigger.And FuzzycMeans Clustering
Major techniques one of of the algorithm fuzzy c-means algorithm (FCMA) as unsupervised machine learning is with fuzzy
Theory establishes the uncertainty description of sample generic to the method for important data analysis and modeling, can more objectively reflect
Real world.
Specifically, with reference to Fig. 3, if after S1, S2 step, video has been divided into k camera lens(K >=1, k ∈ Z).
S3 is constructed in step:
S31:The map operations of level-one camera lens, this operation is realized is distributed to level-one cloud computing platform node by the camera lens of segmentation
In, node described herein is dual role, specially:It is the angles Datanode and TaskTasker in first order cloud platform
Color is Namenode and Jobtracker roles in the cloud computing platform node of the second level.In S301 operations, map construction of function
Input data record<key,value>For<Camera lens ID, video lens position>, function operation is:Camera lens is copied to first
In grade cloud computing platform node, to carry out the map operations of the second level.
S32:Map operation pretreatments in the second level are made to the camera lens being distributed on node.By shot segmentation at one group of video frame
Image.Camera lens 1 includes video frame number N1, camera lens 2 include video frame number N2, camera lens K include video frame number Nk。N1、N2、… NkRefer to
Each camera lens N1、N2、…、NkThe practical video frame number for being included(N1、N2、…Nk≥1, N1、N2、…Nk∈Z).
S33:Carry out second level map operations parallel to each camera lens.The second level map operations, are each frame being distributed to
It is parallel to extract video frame feature vector in next stage cloud computing platform Tasktracker nodes.In the S33 operations, map letters
Number construction input data record<key,value>For<Frame ID, frame picture position>, function operation is to carry out feature to video frame
Vector extraction, exports result<key,value>To form be<Frame ID, frame feature vector>.
S34:Obtained result is operated to S33 and carries out third level map operations.The S34 operations, are to generate S33
's<Frame ID, frame feature vector>Record group is parallel into the cluster operation that racks.Specifically it can be described as:Each data frame is completed to initially
The distance of frame cluster centre calculates, and re-flags its new cluster classification belonged to, what input was generated by S33<Frame ID, frame
Feature vector>All records and last round of iteration(Or initial clustering)Cluster centre.Each map functions read in cluster centre
File is described, each of map function pairs input record point calculates its nearest class center, and does the label of new category.Map functions
Input data record<key,value>For<Frame ID, frame feature vector>;Export intermediate result<key,value>Form be<
Frame generic, frame feature vector>.
Above-mentioned camera lens 1, camera lens 2 ..., the initial cluster center number of camera lens k be described as m1、m2、…、mk(m1、m2、…、
mk>=1, m1、m2、…、mk∈Z)Its value is determining according to each lens data frame sum and certain rule, to a certain extent, m1、
m2、…、mkAlso crucial frame number of each camera lens by generation is represented.
Above-mentioned initial cluster center is configured to:According to m1、m2、…、mkValue, respectively from camera lens 1, camera lens 2 ..., camera lens
M is randomly selected in k1、m2、…、mkA sign feature vector as camera lens 1, camera lens 2 ..., the initial cluster center of camera lens k.
The distance of above-mentioned data frame to cluster centre calculates, and can be described as Euclidean distance, mahalanobis distance etc..
The map functions that S34 is constructed can be described as:
void map(Object key, Text value, Context context)
{
Distance of the calculating frame to each cluster centre;
More above-mentioned distance;
Sum up in the point that of distance recently away from the class belonging to class center frame;
It will<Frame generic, frame feature vector>Intermediate file is written;
}
The above-mentioned map stages can carry out shuffle operations, complete the packet sequencing of results of intermediate calculations.
S35:According to the output of S34, cluster centre is updated, is used for next round map-reduce.The S35 operations, it is defeated
Enter data<key,value>To form be<Cluster category IDs, { record attribute vector set }>;The identical record of all key(I.e.
The record of identical cluster centre category IDs)A reduce task will be given.S35 operations are described as:The identical points of cumulative key
Number and it is each record component sum, seek the mean value of each component, obtain new cluster centre.S35 operation output results<key,
value>To form be<Cluster category IDs, mean vector>.S35 operating process can be described as:
Void reduce (Text key, Iterable<Text> values, Context context)
{
For (all records identical for key)
{
Seek the mean value of each attribute;
}
It will<Cluster category IDs, mean vector>Destination file is written;
}
S36:Iterative process.S36 steps can be described as:Output to S34 is as a result, judge whether the cluster has restrained.Tool
Body can be described as:The cluster centre that more last round of map-reduce is obtained is at a distance from epicycle map-reduce cluster centres.
If distance is less than given threshold values, algorithm terminates.It is no it, then the cluster centre of epicycle is replaced to last round of cluster centre, and
Start the map-reduce operations of a new round.
S4:Output is combined to camera lens key frame group.The final each camera lens key frame of video group poly- to S3 steps carries out group
Close output key frame of video.
In conclusion multistage parallel key frame cloud extracting method according to an embodiment of the invention and cloud extraction system, tool
There is the characteristics of multistage parallel operation, key-frame extraction efficiency can be greatly improved.Meanwhile system is based on cloud computing platform, therefore have
Preferable scalability and stability.
Although being particularly shown and describing the present invention, those skilled in the art with reference to its exemplary embodiment
It should be understood that in the case where not departing from the spirit and scope of the present invention defined by claim, form can be carried out to it
With the various changes in details.
Claims (2)
1. a kind of multistage parallel key frame cloud extracting method, which is characterized in that the multistage parallel key frame cloud extracting method packet
Include following steps:
S1. receiving need to carry out the video of Key Frame Extraction;
S2. video lens segmentation is carried out to video;
S3. key frame of video is extracted to camera lens multistage parallel;Specially:
S31:The map operations of level-one camera lens, the camera lens of segmentation is distributed in level-one cloud computing platform node by realization, described herein
Node is dual role;
S32:Map operation pretreatments in the second level are made to the camera lens being distributed on node, by shot segmentation at one group of video frame images;
S33:Carry out second level map operations parallel to each camera lens, the second level map operations are each frame being distributed to next
It is parallel to extract video frame feature vector in grade cloud computing platform Tasktracker nodes;
S34:Obtained result is operated to S33 and carries out third level map operations, is to generate S33<Frame ID, frame feature to
Amount>Record group is parallel into the cluster operation that racks;
S35:According to the output of S34, cluster centre is updated, is used for next round map-reduce;
S36:Iterative process, the output to S34 is as a result, judge whether the cluster has restrained;
S4. output is combined to camera lens key frame group.
2. multistage parallel key frame cloud extracting method according to claim 1, which is characterized in that the video feature vector
Cluster operation is k-means clusters or fuzzy C-means clustering or other are suitble in the characteristics of image clustering algorithm of parallel processings
It is one or more.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410731007.0A CN104463864B (en) | 2014-12-05 | 2014-12-05 | Multistage parallel key frame cloud extracting method and system |
PCT/CN2015/092349 WO2016086731A1 (en) | 2014-12-05 | 2015-10-21 | Multi-level parallel key frame cloud extraction method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410731007.0A CN104463864B (en) | 2014-12-05 | 2014-12-05 | Multistage parallel key frame cloud extracting method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104463864A CN104463864A (en) | 2015-03-25 |
CN104463864B true CN104463864B (en) | 2018-08-14 |
Family
ID=52909846
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410731007.0A Active CN104463864B (en) | 2014-12-05 | 2014-12-05 | Multistage parallel key frame cloud extracting method and system |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN104463864B (en) |
WO (1) | WO2016086731A1 (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104463864B (en) * | 2014-12-05 | 2018-08-14 | 华南师范大学 | Multistage parallel key frame cloud extracting method and system |
CN108921773A (en) * | 2018-07-04 | 2018-11-30 | 百度在线网络技术(北京)有限公司 | Human body tracking processing method, device, equipment and system |
CN110889857A (en) * | 2019-11-15 | 2020-03-17 | 北京邮电大学 | Mobile Web real-time video frame segmentation method and system |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1719909A (en) * | 2005-07-15 | 2006-01-11 | 复旦大学 | Method for measuring audio-video frequency content change |
CN103064935A (en) * | 2012-12-24 | 2013-04-24 | 深圳先进技术研究院 | System and method for multimedia data parallel processing |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7016540B1 (en) * | 1999-11-24 | 2006-03-21 | Nec Corporation | Method and system for segmentation, classification, and summarization of video images |
CN101296373B (en) * | 2007-04-27 | 2011-11-23 | 北京信心晟通科技发展有限公司 | Multimedia data processing system and method based on material exchange format |
CN102395984A (en) * | 2009-04-14 | 2012-03-28 | 皇家飞利浦电子股份有限公司 | Key frames extraction for video content analysis |
CN102693299B (en) * | 2012-05-17 | 2015-01-07 | 西安交通大学 | System and method for parallel video copy detection |
CN104463864B (en) * | 2014-12-05 | 2018-08-14 | 华南师范大学 | Multistage parallel key frame cloud extracting method and system |
-
2014
- 2014-12-05 CN CN201410731007.0A patent/CN104463864B/en active Active
-
2015
- 2015-10-21 WO PCT/CN2015/092349 patent/WO2016086731A1/en active Application Filing
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1719909A (en) * | 2005-07-15 | 2006-01-11 | 复旦大学 | Method for measuring audio-video frequency content change |
CN103064935A (en) * | 2012-12-24 | 2013-04-24 | 深圳先进技术研究院 | System and method for multimedia data parallel processing |
Non-Patent Citations (1)
Title |
---|
海量网络视频快速检索关键技术研究;刘守群;《中国博士学位论文全文数据库 信息科技辑》;20101015(第10期);摘要,正文第4-5页第1.3节,第12页第2.2节,第13页第2.2.1节,第14页第2.2.1节,第21页第2.4.1节,第31-32页第2.5.2节第1-2段,第62-63页第5.2节,第64页第5.2.1节,第66页第5.2.2节第4段,第67页第5.2.2节第2段,第74页第六章第1段,第77页第6.1.4节第2-3段,第80-86页第6.2节 * |
Also Published As
Publication number | Publication date |
---|---|
WO2016086731A1 (en) | 2016-06-09 |
CN104463864A (en) | 2015-03-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | Learning two-view correspondences and geometry using order-aware network | |
US11055555B2 (en) | Zero-shot object detection | |
CN106650789B (en) | Image description generation method based on depth LSTM network | |
US11727053B2 (en) | Entity recognition from an image | |
Zaech et al. | Learnable online graph representations for 3d multi-object tracking | |
Ding et al. | Violence detection in video by using 3D convolutional neural networks | |
Wen et al. | CF-SIS: Semantic-instance segmentation of 3D point clouds by context fusion with self-attention | |
CN110728294A (en) | Cross-domain image classification model construction method and device based on transfer learning | |
CN105335368B (en) | A kind of product clustering method and device | |
JP7242994B2 (en) | Video event identification method, apparatus, electronic device and storage medium | |
Petkos et al. | Graph-based multimodal clustering for social event detection in large collections of images | |
CN104463864B (en) | Multistage parallel key frame cloud extracting method and system | |
Li et al. | Co-saliency detection based on hierarchical consistency | |
Liu et al. | Place-centric visual urban perception with deep multi-instance regression | |
CN114328988A (en) | Multimedia data feature extraction method, multimedia data retrieval method and device | |
CN113963303A (en) | Image processing method, video recognition method, device, equipment and storage medium | |
Zheng et al. | Clustering matters: Sphere feature for fully unsupervised person re-identification | |
Cao et al. | GMN: generative multi-modal network for practical document information extraction | |
Yan et al. | Geometrically based linear iterative clustering for quantitative feature correspondence | |
CN106844338B (en) | method for detecting entity column of network table based on dependency relationship between attributes | |
CN114723652A (en) | Cell density determination method, cell density determination device, electronic apparatus, and storage medium | |
Liao et al. | Depthwise grouped convolution for object detection | |
CN111241326B (en) | Image visual relationship indication positioning method based on attention pyramid graph network | |
CN112257689A (en) | Training and recognition method of face recognition model, storage medium and related equipment | |
JP2019086979A (en) | Information processing device, information processing method, and program |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |