CN107707975A - Video intelligent clipping method based on monitor supervision platform - Google Patents

Video intelligent clipping method based on monitor supervision platform Download PDF

Info

Publication number
CN107707975A
CN107707975A CN201710855228.2A CN201710855228A CN107707975A CN 107707975 A CN107707975 A CN 107707975A CN 201710855228 A CN201710855228 A CN 201710855228A CN 107707975 A CN107707975 A CN 107707975A
Authority
CN
China
Prior art keywords
frame
video
human body
sport foreground
classifier
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
Application number
CN201710855228.2A
Other languages
Chinese (zh)
Inventor
王霞
李岳楠
张为
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University
Original Assignee
Tianjin University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin University filed Critical Tianjin University
Priority to CN201710855228.2A priority Critical patent/CN107707975A/en
Publication of CN107707975A publication Critical patent/CN107707975A/en
Pending legal-status Critical Current

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/43Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/44Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs
    • H04N21/44008Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs involving operations for analysing video streams, e.g. detecting features or characteristics in the video stream
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/40Scenes; Scene-specific elements in video content
    • G06V20/41Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • G06V20/42Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/23Processing of content or additional data; Elementary server operations; Server middleware
    • H04N21/234Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs
    • H04N21/23418Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs involving operations for analysing video streams, e.g. detecting features or characteristics
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/47End-user applications
    • H04N21/472End-user interface for requesting content, additional data or services; End-user interface for interacting with content, e.g. for content reservation or setting reminders, for requesting event notification, for manipulating displayed content
    • H04N21/47205End-user interface for requesting content, additional data or services; End-user interface for interacting with content, e.g. for content reservation or setting reminders, for requesting event notification, for manipulating displayed content for manipulating displayed content, e.g. interacting with MPEG-4 objects, editing locally
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00Television systems
    • H04N7/18Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Human Computer Interaction (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The present invention relates to a kind of video intelligent clipping method based on monitor supervision platform, including:Sport foreground, and the sport foreground point number in statistics frame are extracted, carries out threshold process;Extract sport foreground profile and carry out human testing;To the doubtful frame of video containing personnel, its quantity is counted, if the frame number after statistics is very few, video clipping processing below is no longer carried out, saves as former video;Conversely, then continuing each video frame images of multiple dimensioned scanning input, it is detected and judged with the human testing assembled classifier of pre-training, will be deemed as the frame of video comprising human body and be defined as key frame of video;The key frame of video that preceding step is obtained carries out front and rear extension multi-frame processing, and then the corresponding frame number of these key frame of video after expanding frame processing is stored entirely in same crucial frame number queue;Editing.

Description

Video intelligent clipping method based on monitor supervision platform
Technical field
Field of intelligent video surveillance in terms of the invention belongs to security protection, specifically belong to a kind of and be based on existing monitor supervision platform Video early stage Intelligent treatment technology.
Background technology
In traditional safety-security area, relevant staff need to carry out video more than times check repeatedly could be at some Some favourable clues are found in details, it is such so as to which further video is analyzed and handled according to these clues Requirement of the process to relevant staff is very high, it is necessary to the staring concern screen of staff, spiritual need high concentration, So that the working strength of related personnel is excessive;On the other hand, the notice focus of people has cycle regular hour, reaches After regular hour length, the situation of distraction just occurs in people, be thus likely to result in video contain compared with The key frame of high price value information it is ignored fall, be unable to reach initial purpose and demand.
Based on such present situation, if can be carried out just to some more tediously long monitor videos by some effective algorithms The editing processing of step, shortens video length, then gives that relevant staff is further to be handled, and so can both mitigate significantly The burden of staff, it can increase work efficiency again.
With the continuous development of the technologies such as image steganalysis, computer vision, various intelligent algorithms are constantly suggested, closely In the past few years, machine learning, deep learning are also concerned by people further, and attract increasing people to bound oneself to it, correlative study Deepen continuously.Personage in video is analyzed, on condition that carrying out Preliminary detection to the personage in video, and presently, there are Some personnel's detection algorithms provide strong theories integration to solve actual engineering problem.The conventional side of human testing at present Method is extraction human body key feature and the corresponding grader of training, and the classical conventional method of early stage is the gradient of Dalal et al. designs The combination of histogram feature and SVM classifier, this method promotes the breakthrough in pedestrian detection field, similar, right That answers also has the features such as LBP, Harr, and the grader designed the methods of Adaboost graders, neutral net.In addition, work as It is pedestrian detection that this field of the preceding detection to people is most of, and the substantially pedestrian in parallel views, human body contour outline Than more complete, but because the position of monocular cam in practice is fixed, enter the monitoring field of camera from pedestrian and take the photograph to walking out As the monitoring field of head, the profile of pedestrian is not necessarily always complete, and sometimes only part human body is appeared in monitor video, Moreover, with the change of distance, the human body image in video is not of uniform size, at this time only fears using only certain single grader Fearness can not meet the needs of actual.For such case, if the people for being adapted to a variety of environment in monitor video can be designed Grader is surveyed in physical examination, retains more key frames by reaching higher personnel targets verification and measurement ratio, it is possible to is preferably ensured Video quality after editing processing.
Meanwhile concern of the China for intelligent security guard field is also increasingly deep, video monitoring system spreads over production and living In, and the monitor video to obtaining be further analyzed processing just become a vital task.By using one kind compared with Early stage editing processing is carried out to video for effective personnel's detection algorithm, can greatly mitigate the work load of staff, And the analytical cycle of video is greatly shortened, the work efficiency of relevant departments is improved, this has one for intelligent security guard field Determine meaning.
The content of the invention
It is an object of the invention to overcome the shortcomings of the high human cost of analyzing and processing monitor video in existing safety-security area, One kind is provided and is based on existing monitor supervision platform, can be as follows to the method for the intelligent editing processing of video progress, technical scheme:
A kind of video intelligent clipping method based on monitor supervision platform, including following step:
1) input video, video is read frame by frame, all frame of video are normalized into unified size, then frame of video carried out Pretreatment, so that subsequent step is further handled;
2) to pretreated frame of video, sport foreground, and the sport foreground point number in statistics frame are extracted;
3) threshold process is carried out to foreground point number, retains the frame of video that foreground point number is more than threshold value, exclude foreground point The very few frame of video of number;
4) to the video frame extraction for meeting threshold condition its sport foreground profile remained, sport foreground profile is entered The appropriate Morphological scale-space of row, then draws its bounding rectangles frame, according to detection demand in practice, takes and suitably amplify its side The strategy of boundary's rectangle frame, the width of rectangle frame amplify 1.1 times, highly enlarged 1.2 times of rectangle frame;
5) obtained sport foreground bounding rectangles frame is further processed, calculates the face of sport foreground bounding rectangles frame Product and the ratio of width to height, if sport foreground bounding rectangles frame area is too small or aspect ratio does not meet the general features of human body, The sport foreground is excluded, and then excludes the frame of video;
6) discriminatory analysis in summary, a part of doubtful frame of video containing personnel is tentatively remained, counts its quantity, If the frame number after statistics is very few, video clipping processing below is no longer carried out, saves as former video;Conversely, then continue more Each video frame images of dimensional scan input, are detected and are sentenced to it with the human testing assembled classifier of pre-training It is disconnected, it will be deemed as the frame of video comprising human body and be defined as key frame of video;
Herein, the human testing assembled classifier of pre-training is as follows:
Multiple dimensioned scanning is first carried out to each video frame images of input, then, by using being suitable in monitor video The human testing assembled classifier of a variety of environment, human testing assembled classifier include upper half of human body detection cascade classifier Cascade classifier is detected with the human body lower part of the body, and upper half of human body detection cascade classifier contains the head-and-shoulder area of human body and right The body frame structural information answered, it is ensured that higher human testing rate, when only existing part human body in video image, especially It is when human body head information is not present, can be then compensated its defect using lower part of the body detection cascade classifier;
7) information correlativity between frame of video is considered, the key frame of video that preceding step is obtained carries out front and rear extend Multi-frame processing, these corresponding frame numbers of key frame of video after expanding frame processing are then stored entirely in same key frame In number queue;
8) read former video frame by frame again, judge that the frame number of each frame of video whether there is in the crucial frame number being previously obtained In queue, if in the presence of, correspond to retain the key frame of video, most at last these key frame of video remained reconfigure for One new video preserves.
Color information of the present invention independent of video frame images, the monitor video at night can also be handled, by more Completely key frame of video of the overwhelming majority comprising personnel is preserved, it is rational to remove the video without larger break-up value Frame, video length is shortened, effectively realizes the early stage editing processing of video, significantly reduce the work load of staff, Human cost is reduced, improves efficiency.
Brief description of the drawings
The video monitoring system block diagram that Fig. 1 is carried by the inventive method
Fig. 2 is the detecting system of human body block diagram designed by the inventive method
Fig. 3 is the partial test result figure of the detecting system of human body designed by the inventive method
Fig. 4 is the flow chart of the inventive method
Embodiment
The general processing framework that video monitoring system is treated in existing safety-security area is:The picture shot by analog video camera Face is directly transmitted to monitor by a cable part and shown, another part is transmitted to DVR.Into the mould of DVR Intend signal and be changed into digital code stream, on the one hand encoded, be stored in document form in DVR;On the other hand, may be used DVR is connected by network at any time, extraction code stream is shown, analyzed.It is specific as shown in Figure 1.Itd is proposed based on the present invention Method formed video early stage processing software, be by extraction be stored in DVR the video after transcoding text Part, then the video is pre-processed, input detecting system of human body, by multiple dimensioned scan video two field picture frame by frame and make Detected with the human testing assembled classifier of pre-training, will be determined as that the frame of video comprising personnel is remained as key Frame, and to its appropriate expansion frame, the key frame after expanding frame processing is reassembled into new video, by removing without larger analysis The frame of video of value, video length is shortened, effectively realize the early stage intelligence editing processing of video.
The inventive method is specifically as shown in Figure 4.
Various pieces are described in detail below:
1. frame of video pre-processes
The problem of in view of computational efficiency, the frame of video of reading is normalized, it is big that CIF forms are uniformly processed into It is small, because monitor video is much night video, so frame of video is converted into gray scale bitmap-format by rgb signal, so With independent of color information, and carry out appropriate histogram equalization processing and avoid disturbing caused by frame of video jump in brightness.
2. sport foreground is extracted
What is taken in this programme is the foreground extracting method of mixed Gauss model, by carrying out background modeling and the in good time back of the body Scape updates, and carries out appropriate morphologic filtering processing to the sport foreground of extraction, includes the processing such as dilation erosion, can obtain To more complete sport foreground.
3. foreground point counts and processing
The foreground point number in the sport foreground extracted is counted, and appropriate threshold value constraint is carried out to it, gives up and includes The very few frame of video of foreground point number.
4. sport foreground contours extract
To qualified video frame extraction its sport foreground profile remained, it is carried out at appropriate morphology Reason, then draws its bounding rectangles frame, according to detection demand in practice, takes the strategy for suitably amplifying its bounding rectangles frame, The width of rectangle frame amplifies about 1.1 times, highly enlarged about 1.2 times of rectangle frame, so can walk follow-up detection of classifier Rapid verification and measurement ratio is higher, improves precision;
5. frame of video is chosen in advance
Obtained sport foreground bounding rectangles frame is further processed, calculates its area and size, if fortune Dynamic prospect bounding rectangles frame area is too small or aspect ratio does not meet the general features of human body, then excludes the sport foreground, enter And exclude the frame of video;
6. detection of classifier
In summary discriminatory analysis, a part of doubtful frame of video containing personnel is tentatively remained, counts its quantity, such as Frame number after fruit statistics is very few, then no longer carries out video clipping processing below, save as former video;Conversely, then continue it It is input in detecting system of human body below, each video frame images of multiple dimensioned scanning input, with the people trained Physical examination surveys grader and it is detected and judged, will be deemed as the frame of video comprising human body and is defined as key frame of video.
Herein, the good human testing classifier design process of pre-training is as shown in Figure 2.
Various pieces are described in detail below:
In view of the profile information of human body in actual monitored video, the not necessarily complete and profile of human body is not of uniform size, herein Take following scheme:Multiple dimensioned scanning is first carried out to each video frame images of input, then, by using fitting for design The human testing assembled classifier of a variety of environment in monitor video is answered, reaches higher human testing rate, by most video Key frame more completely preserves.
For the consideration to the Video processing time, what is taken herein is the strategy of cascade classifier, specifically, people's physical examination Survey assembled classifier and include upper half of human body detection cascade classifier and human body lower part of the body detection cascade classifier.Human body upper half Body detection cascade classifier contains the head-and-shoulder area of human body and corresponding body frame structural information, it is ensured that higher Human testing rate, but, when only existing part human body in video image, especially in the absence of human body head information when, under Half body detection cascade classifier can then compensate its defect.
Specifically, there is employed herein histogram of gradients feature extraction and cascade Adaboost algorithm design assembled classifier Scheme.It is to cascade multiple strong classifiers to cascade Adaboost algorithm, and each strong classifier therein is by multiple weak What classifiers combination formed, the two kinds of cascade classifiers designed herein are set as 18 grades.By training substantial amounts of positive negative sample Preliminary the upper half of human body detection cascade classifier for having obtained better performances and lower part of the body detection cascade classifier, then by this two Acted on after kind classifiers combination on test set, analyze test result.According to test result, increase corresponding to this paper and perfect Positive negative sample, including the upper half of human body positive sample at a variety of visual angles and the negative sample for easily causing interference, such as trees, building, electricity Line bar, roadblock etc., the diversity of sample is improved, effectively improves classifier performance, improves verification and measurement ratio, reduce wrong report Rate.
Generally speaking, by multiple dimensioned scan video two field picture and using human testing assembled classifier, original can be regarded Key frame of the overwhelming majority comprising personnel more completely preserves in frequency, the video matter being effectively guaranteed after editing processing Amount.
7. key frame stores
In view of the information correlativity between frame of video, the key frame of video that preceding step is obtained carries out front and rear extension 15 Frame processing, these corresponding frame numbers of key frame of video after expanding frame processing are then stored entirely in same crucial frame number In queue;
8. it is reassembled as new video
Again former video is read frame by frame, judges that the frame number of each frame of video whether there is in the crucial frame number team being previously obtained In row, if in the presence of corresponding to retain the key frame of video, most these key frame of video remained are reconfigured as one at last Individual new video preserves.By more completely remaining the key frame of video containing key message, eliminate without larger point The frame of video of value is analysed, shortens video length, effectively realizes the early stage intelligence editing processing of video.

Claims (1)

1. a kind of video intelligent clipping method based on monitor supervision platform, including following step:
1) input video, video is read frame by frame, all frame of video are normalized into unified size, then frame of video located in advance Reason, so that subsequent step is further handled;
2) to pretreated frame of video, sport foreground, and the sport foreground point number in statistics frame are extracted;
3) threshold process is carried out to foreground point number, retains the frame of video that foreground point number is more than threshold value, exclude foreground point number Very few frame of video;
4) to the video frame extraction for meeting threshold condition its sport foreground profile remained, sport foreground profile is fitted When Morphological scale-space, then draw its bounding rectangles frame, according to detection demand in practice, take and suitably amplify its shape based moment The strategy of shape frame, the width of rectangle frame amplify 1.1 times, highly enlarged 1.2 times of rectangle frame;
5) obtained sport foreground bounding rectangles frame is further processed, calculate sport foreground bounding rectangles frame area and The ratio of width to height, if sport foreground bounding rectangles frame area is too small or aspect ratio does not meet the general features of human body, exclude The sport foreground, and then exclude the frame of video;
6) discriminatory analysis in summary, a part of doubtful frame of video containing personnel is tentatively remained, counts its quantity, if Frame number after statistics is very few, then no longer carries out video clipping processing below, save as former video;Conversely, then continue multiple dimensioned Each video frame images of input are scanned, it is detected and judged with the human testing assembled classifier of pre-training, will It is judged as that the frame of video comprising human body is defined as key frame of video;
Herein, the human testing assembled classifier of pre-training is as follows:
Multiple dimensioned scanning is first carried out to each video frame images of input, it is then, a variety of in monitor video by using being suitable for The human testing assembled classifier of environment, human testing assembled classifier include upper half of human body detection cascade classifier and people The body lower part of the body detects cascade classifier, and upper half of human body detection cascade classifier contains the head-and-shoulder area of human body and corresponding Body frame structural information, it is ensured that higher human testing rate, when only existing part human body in video image, especially not When human body head information being present, it can then be compensated its defect using lower part of the body detection cascade classifier;
7) information correlativity between frame of video is considered, the key frame of video that preceding step is obtained carries out front and rear extension multiframe Processing, these corresponding frame numbers of key frame of video after expanding frame processing are then stored entirely in same crucial frame number team In row;
8) read former video frame by frame again, judge that the frame number of each frame of video whether there is in the crucial frame number queue being previously obtained In, if in the presence of corresponding to retain the key frame of video, most these key frame of video remained are reconfigured as one at last New video preserves.
CN201710855228.2A 2017-09-20 2017-09-20 Video intelligent clipping method based on monitor supervision platform Pending CN107707975A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710855228.2A CN107707975A (en) 2017-09-20 2017-09-20 Video intelligent clipping method based on monitor supervision platform

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710855228.2A CN107707975A (en) 2017-09-20 2017-09-20 Video intelligent clipping method based on monitor supervision platform

Publications (1)

Publication Number Publication Date
CN107707975A true CN107707975A (en) 2018-02-16

Family

ID=61173163

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710855228.2A Pending CN107707975A (en) 2017-09-20 2017-09-20 Video intelligent clipping method based on monitor supervision platform

Country Status (1)

Country Link
CN (1) CN107707975A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109583441A (en) * 2018-10-19 2019-04-05 天津大学 Subtitle recognition method based on monitor supervision platform
CN111611944A (en) * 2020-05-22 2020-09-01 创新奇智(北京)科技有限公司 Identity recognition method and device, electronic equipment and storage medium
CN111738196A (en) * 2020-06-30 2020-10-02 上海松鼠课堂人工智能科技有限公司 Learning state tracking and analysis device and system

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060256210A1 (en) * 2005-04-28 2006-11-16 Kathleen Ryall Spatio-temporal graphical user interface for querying videos
US20100092037A1 (en) * 2007-02-01 2010-04-15 Yissum Research Develpoment Company of the Hebrew University of Jerusalem Method and system for video indexing and video synopsis
CN101908150A (en) * 2010-06-25 2010-12-08 北京交通大学 Human body detection method
CN102521582A (en) * 2011-12-28 2012-06-27 浙江大学 Human upper body detection and splitting method applied to low-contrast video
CN103413330A (en) * 2013-08-30 2013-11-27 中国科学院自动化研究所 Method for reliably generating video abstraction in complex scene
CN103646094A (en) * 2013-12-18 2014-03-19 上海紫竹数字创意港有限公司 System and method for automatic extraction and generation of audiovisual product content abstract

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060256210A1 (en) * 2005-04-28 2006-11-16 Kathleen Ryall Spatio-temporal graphical user interface for querying videos
US20100092037A1 (en) * 2007-02-01 2010-04-15 Yissum Research Develpoment Company of the Hebrew University of Jerusalem Method and system for video indexing and video synopsis
CN101908150A (en) * 2010-06-25 2010-12-08 北京交通大学 Human body detection method
CN102521582A (en) * 2011-12-28 2012-06-27 浙江大学 Human upper body detection and splitting method applied to low-contrast video
CN103413330A (en) * 2013-08-30 2013-11-27 中国科学院自动化研究所 Method for reliably generating video abstraction in complex scene
CN103646094A (en) * 2013-12-18 2014-03-19 上海紫竹数字创意港有限公司 System and method for automatic extraction and generation of audiovisual product content abstract

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109583441A (en) * 2018-10-19 2019-04-05 天津大学 Subtitle recognition method based on monitor supervision platform
CN111611944A (en) * 2020-05-22 2020-09-01 创新奇智(北京)科技有限公司 Identity recognition method and device, electronic equipment and storage medium
CN111738196A (en) * 2020-06-30 2020-10-02 上海松鼠课堂人工智能科技有限公司 Learning state tracking and analysis device and system

Similar Documents

Publication Publication Date Title
CN110070033B (en) Method for detecting wearing state of safety helmet in dangerous working area in power field
CN109829443A (en) Video behavior recognition methods based on image enhancement Yu 3D convolutional neural networks
CN109902558B (en) CNN-LSTM-based human health deep learning prediction method
Fu et al. Combined thresholding and neural network approach for vein pattern extraction from leaf images
Raut et al. Plant disease detection in image processing using MATLAB
CN109670430A (en) A kind of face vivo identification method of the multiple Classifiers Combination based on deep learning
CN107527351A (en) A kind of fusion FCN and Threshold segmentation milking sow image partition method
CN113553979B (en) Safety clothing detection method and system based on improved YOLO V5
CN114359727B (en) Tea disease identification method and system based on lightweight optimization Yolo v4
CN105513053B (en) One kind is used for background modeling method in video analysis
CN106845415A (en) A kind of pedestrian based on deep learning becomes more meticulous recognition methods and device
CN104021523B (en) A kind of method of the image super-resolution amplification based on marginal classification
CN112380982A (en) Integrated monitoring method for progress and quality of infrastructure project in power industry
CN107798304B (en) Method for rapidly auditing video
CN111401310B (en) Kitchen sanitation safety supervision and management method based on artificial intelligence
CN108491807B (en) Real-time monitoring method and system for oestrus of dairy cows
CN107707975A (en) Video intelligent clipping method based on monitor supervision platform
Monwar et al. Pain recognition using artificial neural network
CN105023272A (en) Crop leaf insect pest detection method and system
CN107330478A (en) A kind of cherry tomato sorting technique, system and cherry tomato on-line sorting system
CN101650782A (en) Method for extracting front human face outline based on complexion model and shape constraining
CN106570885A (en) Background modeling method based on brightness and texture fusion threshold value
CN107247934A (en) A kind of round-the-clock yawn detection method and system based on swift nature point location
CN108898098A (en) Early stage video smoke detection method based on monitor supervision platform
Zhang et al. Hand gesture detection and segmentation based on difference background image with complex background

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20180216