CN107633238A - A kind of video analysis method and intellectual analysis server - Google Patents
A kind of video analysis method and intellectual analysis server Download PDFInfo
- Publication number
- CN107633238A CN107633238A CN201710949174.6A CN201710949174A CN107633238A CN 107633238 A CN107633238 A CN 107633238A CN 201710949174 A CN201710949174 A CN 201710949174A CN 107633238 A CN107633238 A CN 107633238A
- Authority
- CN
- China
- Prior art keywords
- data
- frame
- subject object
- real time
- module
- 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
- 238000004458 analytical method Methods 0.000 title claims abstract description 92
- 238000012806 monitoring device Methods 0.000 claims abstract description 28
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000013528 artificial neural network Methods 0.000 claims abstract description 6
- 238000001514 detection method Methods 0.000 claims description 76
- 238000007405 data analysis Methods 0.000 claims description 41
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 238000001914 filtration Methods 0.000 claims description 14
- 238000000034 method Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 238000013507 mapping Methods 0.000 claims description 12
- 238000012544 monitoring process Methods 0.000 claims description 8
- 238000009432 framing Methods 0.000 claims description 7
- 230000009286 beneficial effect Effects 0.000 abstract 1
- 238000012360 testing method Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000007812 deficiency Effects 0.000 description 2
- 230000009191 jumping Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000003032 molecular docking Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of video analysis method, it comprises the following steps：S10, the historical data of access monitoring device storage, it is frame queue that historical data is carried out into sub-frame processing by openCV and ffmpeg；Wherein, historical data is the video data of the past generation of monitoring device record；S20, real time data is gathered by the monitoring device at scene, real time data decoded by ffmpeg, and sub-frame processing is frame queue；Wherein, real time data is the monitor video data that monitoring device gathers in real time；S30, multilayer volume integral neural network is established according to the frame queue of the frame queue of historical data and real time data, every frame picture in frame queue is tested and analyzed, identifies the subject object in picture.The beneficial effects of the present invention are the accuracy that can be improved to the subject object identification in video, the rate of false alarm of reduction monitor supervision platform or system.
Description
Technical field
The present invention relates to the technical field of video analysis, more particularly to a kind of video analysis method and intellectual analysis service
Device.
Background technology
Safety monitoring is preventing and fought crime, and safeguards civil order, disaster prevention accident, reduces the personal and collective person
Injury and property loss etc. serve very positive effect.Such as utilize safety monitoring and intelligent video analysis technology counterweight
Want region to carry out effective monitoring, on the one hand can deter the action of offender to a certain extent, play preventative；The opposing party
Face, once the criminal activities such as invasion occur, can find and alarm in time, record scene of a crime automatically, assist staff and
When solve a case, greatly improve efficiency, while save substantial amounts of human and material resources and financial resources.
Wherein, intelligent video analysis is an important technology in safety monitoring, its refer to by by background in scene and
Target separates, and then analyzes and follow the trail of the concern target occurred in camera scene.With intelligent video analysis in some weights
Want region to carry out target detection, duty personnel can be replaced to work for 24 hours, realized unattended.But in practical application, intelligence
Energy video analytics server is all to use traditional algorithm, and illumination variation, complexity, the target of target motion in actual environment
It is blocked, situations such as target is similar to background color, background is mixed and disorderly can all increase the difficulty of object detecting and tracking algorithm design
Degree, so as to cause a large amount of false alarms, it appears in general traditional algorithm can not meet to use needs, while in general traditional algorithm
Target is detected both for property, pays close attention to object variations, while needs to be modified algorithm, can not be popularized.
The content of the invention
To solve the above problems, the main object of the present invention is to provide a kind of video analysis method, all improving to video
In subject object identification accuracy, reduce the rate of false alarm of monitor supervision platform or system.
To achieve the above object, video analysis method proposed by the present invention, it comprises the following steps：
S10, the historical data of access monitoring device storage, historical data is carried out at framing by openCV and ffmpeg
Manage as frame queue.Wherein, historical data is the video data of the past generation of monitoring device record.
S20, real time data is gathered by the monitoring device at scene, real time data decoded by ffmpeg, and point
Frame processing is frame queue.Wherein, real time data is the monitor video data that monitoring device gathers in real time.
S30, multilayer volume integral neural network is established according to the frame queue of the frame queue of historical data and real time data, to frame
Every frame picture in queue is tested and analyzed, and identifies the subject object in picture.
Preferably, in step s 30, the process of the analysis detection of multilayer convolutional neural networks is as follows：
S31, every frame image data in frame queue is added as into convolutional layer with the filter stack of multiple different weights.
S32, scan per frame image data, and carry out convolution biasing when locally connecting and put computing, extract local feature, obtain
Obtain the Feature Mapping array of image.
S33, redundancy is removed by nonlinear function to obtain new output array, and feature will be generated in pond layer
Input source of the data as next convolutional layer.
S34, repeat step S32~step S33, until extracting combination abstract characteristics or the characteristic of global characteristics mapping
According to.
S35, using full connected mode, the characteristic array for the deep layer finally extracted is delivered into grader, identifies picture
In subject object.
Preferably, in step S30, when every frame picture in frame queue tests and analyzes, according to monitoring device institute
The scene at place and application scenario, target signature model corresponding with its scene is selected to carry out analysis detection to frame queue.
Preferably, the video analysis method also includes：S40, passage time sequence and continuous multiple frames relation pair multilayer convolution
Neural network recognization to subject object filtered.
Preferably, in step s 40, the subject object recognized to multilayer convolutional neural networks carries out filter process：
S41, judge whether the subject object recognized between adjacent two frame is concern target, if so, then carrying out down
One step, if it is not, then excluding the subject object recognized.
S42, judge whether the Duplication between adjacent two frame is reasonable, if rationally, carrying out in next step, if unreasonable,
Exclude the subject object recognized.
S43, judge whether the maxima and minima of subject object that recognizes is reasonable, if maxima and minima closes
Reason, then retain the subject object recognized.Otherwise, the subject object recognized is excluded.
The present invention also proposes a kind of intellectual analysis server, and it includes：Historical data AM access module, real-time data imputing system mould
Block, data analysis detection module and output module.
Historical data AM access module is connected with the first input end of data analysis detection module, and it is used to receive monitoring device
The historical data of storage, and historical data is subjected to sub-frame processing, exported in the form of frame queue to data analysis detection module.
Real-time data imputing system module is connected with the second input of data analysis detection module, and it is used to receive monitoring device
The real time data gathered in real time, and real time data decoded, sub-frame processing, exported in the form of frame queue to data analysis
Detection module.
Multilayer convolutional neural networks algorithm routine is provided with data analysis detection module, for every frame figure in frame queue
Piece is tested and analyzed and identifies the subject object in picture.
Identification target output module is connected with the output end of data analysis detection module, and it is used for from historical data and reality
When data in the subject object that recognizes be sent to monitor supervision platform or system.
Preferably, the process of data analysis detection module analysis detection is as follows：
S51, every frame image data in frame queue is added as into convolutional layer with the filter stack of multiple different weights.
S52, scan per frame image data, and carry out convolution biasing when locally connecting and put computing, extract local feature, obtain
Obtain the Feature Mapping array of image.
S53, redundancy is removed by nonlinear function to obtain new output array, and feature will be generated in pond layer
Input source of the data as next convolutional layer.
S54, repeat step S52~step S53, until extracting combination abstract characteristics or the characteristic of global characteristics mapping
According to.
S55, using full connected mode, the characteristic array for the deep layer finally extracted is delivered into grader, identifies picture
In subject object.
Preferably, the intellectual analysis server also includes target signature model fitting module, and it detects mould with data analysis
The 3rd input connection of block, for the scene according to residing for monitoring device and application scenario, is selected corresponding with its scene
Target signature model carries out analysis detection to frame queue.
Preferably, the intellectual analysis server also includes data filtering module, its input and data analysis detection module
Output end connection, its output end with identification target output module input be connected, for data analysis detection module knowledge
The subject object being clipped to is filtered.
Preferably, the filter process of data filtering module is as follows:
S61, judge whether the subject object recognized between adjacent two frame is concern target, if so, then carrying out down
One step, if it is not, then excluding the subject object recognized.
S62, judge whether the Duplication between adjacent two frame is reasonable, if rationally, carrying out in next step, if unreasonable,
Exclude the subject object recognized.
S63, judge whether the maxima and minima of subject object that recognizes is reasonable, if maxima and minima closes
Reason, then retain the subject object recognized.Otherwise, the subject object recognized is excluded.
Video analysis method provided by the invention and intellectual analysis server, can be to having occurred and that and saving as phase obstruction and rejection
The audio frequency and video historical data of formula file carries out analysis detection, can also analyze the current occurent data flow of detection in real time；Using
Convolutional neural networks algorithms, relative to traditional algorithm, analysis detectability is higher, and rate of false alarm is lower；Target signature model can
Voluntarily select to match according to different scenes and application environment, without modifying or upgrading to system, make analysis detection more
Have definition and accuracy, and ease of use；By carrying out multi-filtering to the subject object recognized, wrong report is reduced
Rate.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the required accompanying drawing used in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Structure according to these accompanying drawings obtains other accompanying drawings.
Fig. 1 is the schematic flow sheet of the embodiment of video analysis method one of the present invention；
Fig. 2 is the structural representation of the embodiment of intellectual analysis server one of the present invention；
The object of the invention is realized, functional characteristics and advantage will be described further referring to the drawings in conjunction with the embodiments.
Embodiment
The present invention proposes a kind of video analysis method.
Reference picture 1, Fig. 1 are the schematic flow sheet of the embodiment of video analysis method one of the present invention.
As shown in figure 1, in embodiments of the present invention, the video analysis method comprises the following steps：
S10, the historical data of access monitoring device storage, historical data is carried out at framing by openCV and ffmpeg
Manage as frame queue.Wherein, historical data is monitoring device record or depositing in interior other storage devices storage of LAN
The video data of video data, such as AVI, WMV, MPEG or MP4 form that past occurs.
In the present embodiment, video format identification first is carried out to historical data, then passes through openCV and ffmpeg framings
Picture relatively independent one by one is compressed into, and is stored in frame queue.
S20, real time data is gathered by the monitoring device at scene, real time data decoded by ffmpeg, and point
Frame processing is frame queue.Wherein, real time data is monitoring device, such as video monitoring platform, video camera, real by monitoring device
When the monitor video data that gather.
In the present embodiment, by network interface receiving real-time data, and real time data is carried out in real time by ffmpeg
Property video decoding, framing is compressed into picture relatively independent one by one, then is stored in frame queue.
S30, multilayer volume integral neural network is established according to the frame queue of the frame queue of historical data and real time data, to frame
Every frame picture in queue is tested and analyzed, and identifies the subject object in picture.
In the present embodiment, frame queue is to analyze state demarcation as analysis frame, analysis frame and frame to be analyzed.Wherein,
Analysis frame is the object for having been subjected to analysis detection, and analysis frame analyzes detection object to be current, and frame to be analyzed is next time
Analyze the object of detection.
Analysis detection is carried out to analysis frame by multilayer convolutional neural networks, its detailed process is as follows：
S31, every frame image data in frame queue is added as into convolutional layer with the filter stack of multiple different weights.
S32, scan per frame image data, and carry out convolution biasing when locally connecting and put computing, extract local feature, obtain
Obtain the Feature Mapping array of image.
S33, redundancy is removed by nonlinear function to obtain new output array, and feature will be generated in pond layer
Input source of the data as next convolutional layer.
S34, repeat step S32~step S33, until extracting combination abstract characteristics or the characteristic of global characteristics mapping
According to.
S35, using full connected mode, the characteristic array for the deep layer finally extracted is delivered into grader, identifies picture
In subject object.
Further, in step S30, when every frame picture in frame queue tests and analyzes, according to monitoring device
Residing scene and application scenario, target signature model corresponding with its scene is selected to carry out analysis detection to frame queue.Example
Such as, this analysis detection needs to pay close attention to " people ", then the target signature model for people may be selected, as this analysis detection needs to pay close attention to
" vehicle ", then it may be selected to be the target signature model of vehicle.It should be noted that various types of target signature models can be advance
Obtained by deep learning training data.
Further, the video analysis method also includes：S40, passage time sequence and continuous multiple frames relation pair multilayer volume
Product neural network recognization to subject object filtered.
Specifically, the subject object recognized to multilayer convolutional neural networks carries out filter process：
S41, judge whether the subject object recognized between adjacent two frame is concern target, if so, then carrying out down
One step, if it is not, then excluding the subject object recognized.
S42, judge whether the Duplication between adjacent two frame is reasonable, if rationally, carrying out in next step, if unreasonable,
Exclude the subject object recognized.
S43, judge whether the maxima and minima of subject object that recognizes is reasonable, if maxima and minima closes
Reason, then retain the subject object recognized.Otherwise, the subject object recognized is excluded.
In the present embodiment, by whether judging subject object that adjacent analysis frame recognizes all to pay close attention to target, with row
Except contingency result；Afterwards, continue to judge the Duplication between adjacent two frame, to exclude the very big result of jumping characteristic；Finally,
Whether the maximum and minimum value of judged result exceed actual application environment.So pass through multi-filtering, can effectively filter with
Machine, the testing result of contingency, the deficiency of target identification ability is also compensate for a certain degree, reduces rate of false alarm.
Compared with prior art, video analysis method provided by the invention, can be to having occurred and that and saving as related pattern
The audio frequency and video historical data of file carries out analysis detection, can also analyze the current occurent data flow of detection in real time；Employ
Convolutional neural networks algorithm, relative to traditional algorithm, analysis detectability is higher, and rate of false alarm is lower；Target signature model can root
Voluntarily select to match according to different scenes and application environment, without modifying or upgrading to system, have more analysis detection
Definition and accuracy, and ease of use；By carrying out multi-filtering to the subject object recognized, rate of false alarm is reduced.
The present invention also proposes a kind of intellectual analysis server.
Reference picture 2, Fig. 2 are the structural representation of the embodiment of intellectual analysis server one of the present invention.
As shown in figure 1, in embodiments of the present invention, the intellectual analysis server includes：Historical data AM access module 100,
Real-time data imputing system module 200, data analysis detection module 300 and output module.
Wherein, historical data AM access module 100 is connected with the first input end of data analysis detection module 300, and it is used for
The historical data of monitoring device storage is received, and historical data is subjected to sub-frame processing, is exported in the form of frame queue to data
Analyze detection module 300.
In the present embodiment, historical data is to have occurred and that and save as the data of corresponding format file, as AVI, WMV,
The video datas such as MPEG, MP4, the voice data such as including MP3, WAV, the historical data can be stored in intellectual analysis server,
Also other storage devices in LAN can be deposited in.Historical data AM access module 100 first passes through carries out video lattice to historical data
Formula identifies, picture relatively independent one by one is then compressed into by openCV and ffmpeg framings, then export with frame queue
To data analysis detection module 300.And for the time domain signal such as MP3, WAV, then the signal that can intercept set time length is defeated
Go out to data analysis detection module 300.
Real-time data imputing system module 200 is connected with the second input of data analysis detection module 300, and it is used to receive prison
The real time data that equipment gathers in real time is controlled, and real time data is decoded, sub-frame processing, is exported in the form of frame queue to number
According to analysis detection module 300.
In the present embodiment, real-time data imputing system module 200 is led to by network interface receiving real-time data to real time data
Cross ffmpeg and carry out real-time video decoding, framing, be compressed into picture relatively independent one by one, and in the form of frame queue
Export to data analysis detection module 300.Real time data derives from the data for current occurent related system or equipment
Stream, the video data that such as video monitoring platform, video camera monitoring device gather in real time, the real-time data imputing system module 200
International ONVIF, GB/T28181 consensus standards, and the proprietary protocol of most of businessman are held, can be carried out with a variety of monitoring devices pair
Connect.
Multilayer convolutional neural networks algorithm routine is provided with data analysis detection module 300, for every in frame queue
Frame picture is tested and analyzed and identifies the subject object in picture.
In the present embodiment, data analysis detection module 300 is mainly by deep learning framework caffe and convolutional neural networks
Algorithm constitution, the data combining target feature come in historical data AM access module 100 and the conveying of real-time data imputing system module 200
Model carries out analysis detection.
The data that historical data AM access module 100 and the conveying of real-time data imputing system module 200 are come in are in the form of frame queue
Performance.Specifically, frame queue is to analyze state demarcation as analysis frame, analysis frame and frame to be analyzed.Analysis frame is
The object of the analysis detection of data analysis detection module 300 is crossed, analysis frame is that data analysis detection module 300 is currently analyzing inspection
Object is surveyed, frame to be analyzed is the object that data analysis detection module 300 analyzes detection next time.
Data analysis detection module 300 carries out analysis detection to analysis frame, and its detailed process is as follows：
S51, every frame image data in frame queue is added as into convolutional layer with the filter stack of multiple different weights.
S52, scan per frame image data, and carry out convolution biasing when locally connecting and put computing, extract local feature, obtain
Obtain the Feature Mapping array of image.
S53, redundancy is removed by nonlinear function to obtain new output array, and feature will be generated in pond layer
Input source of the data as next convolutional layer.
S54, repeat step S52~step S53, until extracting combination abstract characteristics or the characteristic of global characteristics mapping
According to.
S55, using full connected mode, the characteristic array for the deep layer finally extracted is delivered into grader, identifies picture
In subject object.
Identification target output module 400 is connected with the output end of data analysis detection module 300, and it is used for from history number
Monitor supervision platform or system are sent to according to the subject object with being recognized in real time data.
In the present embodiment, for historical data, the data containing subject object can be directly separated out and be resident locally or lead to
Cross network interface protocols and be sent to related system, and for real time data, result can be led to according to its support docking monitor supervision platform
Cross network interface protocols and be transmitted directly to corresponding monitor supervision platform.
Further, in the present embodiment, the intellectual analysis server also includes target signature model fitting module 500,
It is connected with the 3rd input of data analysis detection module 300, for the scene and applied field according to residing for monitoring device
Close, select target signature model corresponding with its scene to carry out analysis detection to frame queue.
In the present embodiment, the internal memory of target signature model fitting module 500 contains the target signature of some different types
Model, and the interface provided with the perpetual object corresponding with each target signature model.It may be selected for different perpetual objects
Different target signature models, such as this analysis detection perpetual object is " people ", then the target signature model for people may be selected,
If this analysis detection perpetual object is " vehicle ", the target signature model of vehicle may be selected to be.It is it should be noted that various
The target signature model of type can obtain beforehand through deep learning training data.
Further, the intellectual analysis server also includes data filtering module 600, and its input detects with data analysis
The output end connection of module 300, its output end is with identifying that the input of target output module 400 is connected, for data analysis
The subject object that detection module 300 recognizes is filtered.
Specifically, the filter process of data filtering module 600 is as follows:
S61, judge whether the subject object recognized between adjacent two frame is concern target, if so, then carrying out down
One step, if it is not, then excluding the subject object recognized.
S62, judge whether the Duplication between adjacent two frame is reasonable, if rationally, carrying out in next step, if unreasonable,
Exclude the subject object recognized.It should be noted that the rational numerical value model of Duplication is preset with data filtering module 600
Enclose, once detect that Duplication between adjacent two frame not in the number range, that is, judges that Duplication is unreasonable.
S63, judge whether the maxima and minima of subject object that recognizes is reasonable, if maxima and minima closes
Reason, then retain the subject object recognized.Otherwise, the subject object recognized is excluded.It should be noted that data filtering module
Maximum threshold values and minimum threshold values are preset with 600, when the maximum for detecting the subject object recognized is more than maximum threshold values, or
When person detects that the minimum value of the subject object recognized is less than minimum threshold values, that is, judge that the subject object that recognizes is unreasonable.
In the present embodiment, by whether judging subject object that adjacent analysis frame recognizes all to pay close attention to target, with row
Except contingency result；Afterwards, continue to judge the Duplication between adjacent two frame, to exclude the very big result of jumping characteristic；Finally,
Whether the maximum and minimum value of judged result exceed actual application environment.So pass through multi-filtering, can effectively filter with
Machine, the testing result of contingency, the deficiency of target identification ability is also compensate for a certain degree, reduces rate of false alarm.
Compared with prior art, intellectual analysis server provided by the invention, can be to having occurred and that and saving as phase obstruction and rejection
The audio frequency and video historical data of formula file carries out analysis detection, supports current generic video stream protocol, supports network transmission, can be real-time
The current occurent data flow of analysis detection, or access teledata carry out analysis detection；Data analysis detection module 300
Inside build and increased income deep learning framework caffe and used convolutional neural networks algorithm, relative to traditional algorithm, analysis detection
Ability is higher, and rate of false alarm is lower；Target signature model in target signature model fitting module 500 can be according to different scenes and should
Voluntarily selected to match with environment, without modifying or upgrading to system, analysis detection is had more definition and accuracy,
Ease of use；Multi-filtering is carried out to the subject object result recognized by data filtering module 600, reduces rate of false alarm；
Can be the object that the output analysis of traditional monitor supervision platform recognizes meanwhile the intelligent analyzer is also supported to dock traditional monitor supervision platform
Target.
Module in the intellectual analysis server can realize it is highly integrated, installation and it is easy to use, input data can obtain
To analysis testing result.Compared to existing intelligent video analysis product, cumbersome exploitation debugging process is greatly reduced.
In addition, in the present embodiment, data analysis detection module 300 using the GPU servers of high performance parallel computing as
Analysis foundation, operational capability is strong, and arithmetic speed is fast.And the GPU servers carry CUDA universal parallel Computational frame, parallel fortune
Calculation ability and the far super general processor CPU of floating-point operation ability, make the analysis detectability of data analysis detection module 300 more
By force, more efficiency, the speed that data analysis detection module 300 carries out mass data analysis detection can be effectively improved.
The preferred embodiments of the present invention are the foregoing is only, are not intended to limit the scope of the invention, it is every at this
Under the inventive concept of invention, the equivalent structure transformation made using description of the invention and accompanying drawing content, or directly/use indirectly
It is included in other related technical areas in the scope of patent protection of the present invention.
Claims (10)
- A kind of 1. video analysis method, it is characterised in that comprise the following steps：S10, the historical data of access monitoring device storage, it is by historical data progress sub-frame processing by openCV and ffmpeg Frame queue；Wherein, the historical data is the video data of the past generation of monitoring device record；S20, real time data is gathered by the monitoring device at scene, real time data decoded by ffmpeg, and at framing Manage as frame queue；Wherein, the real time data is the monitor video data that monitoring device gathers in real time；S30, multilayer volume integral neural network is established according to the frame queue of the frame queue of historical data and real time data, to frame queue In every frame picture tested and analyzed, identify picture in subject object.
- 2. video analysis method as claimed in claim 1, it is characterised in that in step s 30, multilayer convolutional neural networks The process for analyzing detection is as follows：S31, every frame image data in frame queue is added as into convolutional layer with the filter stack of multiple different weights；S32, scan per frame image data, and carry out convolution biasing when locally connecting and put computing, extract local feature, schemed The Feature Mapping array of picture；S33, redundancy is removed by nonlinear function to obtain new output array, and characteristic will be generated in pond layer Input source as next convolutional layer；S34, repeat step S32~step S33, until extracting combination abstract characteristics or the characteristic of global characteristics mapping；S35, using full connected mode, the characteristic array for the deep layer finally extracted is delivered into grader, identified in picture Subject object.
- 3. video analysis method as claimed in claim 1, it is characterised in that every in frame queue in the step S30 When frame picture is tested and analyzed, scene and application scenario according to residing for monitoring device, mesh corresponding with its scene is selected Mark characteristic model carries out analysis detection to frame queue.
- 4. the video analysis method as described in claims 1 to 3 any one, in addition to：S40, passage time sequence and continuous The subject object that multiframe relation pair multilayer convolutional neural networks recognize is filtered.
- 5. video analysis method as claimed in claim 4, in the step S40, multilayer convolutional neural networks are recognized Subject object carry out filter process be：S41, judge whether the subject object recognized between adjacent two frame is concern target, if so, then carry out in next step, If it is not, then exclude the subject object recognized；S42, judge whether the Duplication between adjacent two frame is reasonable, if rationally, in next step, if unreasonable, exclude The subject object recognized；S43, judge whether the maxima and minima of subject object that recognizes is reasonable, if maxima and minima is reasonable, Then retain the subject object recognized；Otherwise, the subject object recognized is excluded.
- A kind of 6. intellectual analysis server, it is characterised in that including：Historical data AM access module, real-time data imputing system module, number According to analysis detection module and output module；The historical data AM access module is connected with the first input end of the data analysis detection module, and it is used to receive monitoring The historical data of equipment storage, and historical data is subjected to sub-frame processing, exported to data analysis and detected in the form of frame queue Module；The real-time data imputing system module is connected with the second input of the data analysis detection module, and it is used to receive monitoring The real time data that equipment gathers in real time, and real time data decoded, sub-frame processing, exported in the form of frame queue to data Analyze detection module；Multilayer convolutional neural networks algorithm routine is provided with the data analysis detection module, for every frame figure in frame queue Piece is tested and analyzed and identifies the subject object in picture；The identification target output module is connected with the output end of the data analysis detection module, and it is used for from historical data Subject object with being recognized in real time data is sent to monitor supervision platform or system.
- 7. intellectual analysis server as claimed in claim 6, the process of the data analysis detection module analysis detection is as follows：S51, every frame image data in frame queue is added as into convolutional layer with the filter stack of multiple different weights；S52, scan per frame image data, and carry out convolution biasing when locally connecting and put computing, extract local feature, schemed The Feature Mapping array of picture；S53, redundancy is removed by nonlinear function to obtain new output array, and characteristic will be generated in pond layer Input source as next convolutional layer；S54, repeat step S52~step S53, until extracting combination abstract characteristics or the characteristic of global characteristics mapping；S55, using full connected mode, the characteristic array for the deep layer finally extracted is delivered into grader, identified in picture Subject object.
- 8. intellectual analysis server as claimed in claim 6, it is characterised in that also including target signature model fitting module, It is connected with the 3rd input of the data analysis detection module, for the scene and applied field according to residing for monitoring device Close, select target signature model corresponding with its scene to carry out analysis detection to frame queue.
- 9. the intellectual analysis server as described in claim 6~8 any one, it is characterised in that also including data filtering mould Block, its input are connected with the output end of the data analysis detection module, its output end and the identification target output module Input connection, the subject object for being recognized to the data analysis detection module filters.
- 10. intellectual analysis server as claimed in claim 9, it is characterised in that the filter process of the data filtering module It is as follows:S61, judge whether the subject object recognized between adjacent two frame is concern target, if so, then carry out in next step, If it is not, then exclude the subject object recognized；S62, judge whether the Duplication between adjacent two frame is reasonable, if rationally, in next step, if unreasonable, exclude The subject object recognized；S63, judge whether the maxima and minima of subject object that recognizes is reasonable, if maxima and minima is reasonable, Then retain the subject object recognized；Otherwise, the subject object recognized is excluded.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710949174.6A CN107633238A (en) | 2017-10-12 | 2017-10-12 | A kind of video analysis method and intellectual analysis server |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710949174.6A CN107633238A (en) | 2017-10-12 | 2017-10-12 | A kind of video analysis method and intellectual analysis server |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107633238A true CN107633238A (en) | 2018-01-26 |
Family
ID=61104461
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710949174.6A Pending CN107633238A (en) | 2017-10-12 | 2017-10-12 | A kind of video analysis method and intellectual analysis server |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107633238A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109165558A (en) * | 2018-07-26 | 2019-01-08 | 浙江大学 | Unusual checking system in a kind of elevator based on deep learning |
| CN110309362A (en) * | 2019-07-05 | 2019-10-08 | 深圳中科云海科技有限公司 | A kind of video retrieval method and system |
| WO2019214019A1 (en) * | 2018-05-11 | 2019-11-14 | 深圳市鹰硕技术有限公司 | Online teaching method and apparatus based on convolutional neural network |
| CN110533896A (en) * | 2019-09-03 | 2019-12-03 | 北京旷视科技有限公司 | Processing method, device and the electronic equipment of warning information |
| CN111010534A (en) * | 2019-11-11 | 2020-04-14 | 中国建设银行股份有限公司 | Real-time asynchronous video analysis method and system |
| CN111178241A (en) * | 2019-12-27 | 2020-05-19 | 航天信息股份有限公司 | An intelligent monitoring system and method based on video analysis |
| CN111223011A (en) * | 2019-12-29 | 2020-06-02 | 航天信息股份有限公司 | A method and system for food safety supervision of catering enterprises based on video analysis |
| CN113096201A (en) * | 2021-03-30 | 2021-07-09 | 上海西井信息科技有限公司 | Embedded video image deep learning system, method, equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102375993A (en) * | 2010-08-23 | 2012-03-14 | 索尼公司 | Target detection method and device |
| CN105184258A (en) * | 2015-09-09 | 2015-12-23 | 苏州科达科技股份有限公司 | Target tracking method and system and staff behavior analyzing method and system |
| US20160275376A1 (en) * | 2015-03-20 | 2016-09-22 | Netra, Inc. | Object detection and classification |
| CN106295489A (en) * | 2015-06-29 | 2017-01-04 | 株式会社日立制作所 | Information processing method, information processor and video monitoring system |
| CN106375721A (en) * | 2016-09-14 | 2017-02-01 | 重庆邮电大学 | An Intelligent Video Surveillance System Based on Cloud Platform |
| CN106780546A (en) * | 2016-12-06 | 2017-05-31 | 南京航空航天大学 | The personal identification method of the motion blur encoded point based on convolutional neural networks |
| CN106874883A (en) * | 2017-02-27 | 2017-06-20 | 中国石油大学(华东) | A kind of real-time face detection method and system based on deep learning |
| CN107231547A (en) * | 2017-07-07 | 2017-10-03 | 广东中星电子有限公司 | Video monitoring system and method |
-
2017
- 2017-10-12 CN CN201710949174.6A patent/CN107633238A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102375993A (en) * | 2010-08-23 | 2012-03-14 | 索尼公司 | Target detection method and device |
| US20160275376A1 (en) * | 2015-03-20 | 2016-09-22 | Netra, Inc. | Object detection and classification |
| CN106295489A (en) * | 2015-06-29 | 2017-01-04 | 株式会社日立制作所 | Information processing method, information processor and video monitoring system |
| CN105184258A (en) * | 2015-09-09 | 2015-12-23 | 苏州科达科技股份有限公司 | Target tracking method and system and staff behavior analyzing method and system |
| CN106375721A (en) * | 2016-09-14 | 2017-02-01 | 重庆邮电大学 | An Intelligent Video Surveillance System Based on Cloud Platform |
| CN106780546A (en) * | 2016-12-06 | 2017-05-31 | 南京航空航天大学 | The personal identification method of the motion blur encoded point based on convolutional neural networks |
| CN106874883A (en) * | 2017-02-27 | 2017-06-20 | 中国石油大学(华东) | A kind of real-time face detection method and system based on deep learning |
| CN107231547A (en) * | 2017-07-07 | 2017-10-03 | 广东中星电子有限公司 | Video monitoring system and method |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019214019A1 (en) * | 2018-05-11 | 2019-11-14 | 深圳市鹰硕技术有限公司 | Online teaching method and apparatus based on convolutional neural network |
| CN109165558A (en) * | 2018-07-26 | 2019-01-08 | 浙江大学 | Unusual checking system in a kind of elevator based on deep learning |
| CN109165558B (en) * | 2018-07-26 | 2022-03-25 | 浙江大学 | An abnormal behavior detection system in elevators based on deep learning |
| CN110309362A (en) * | 2019-07-05 | 2019-10-08 | 深圳中科云海科技有限公司 | A kind of video retrieval method and system |
| CN110533896A (en) * | 2019-09-03 | 2019-12-03 | 北京旷视科技有限公司 | Processing method, device and the electronic equipment of warning information |
| CN110533896B (en) * | 2019-09-03 | 2022-05-03 | 北京旷视科技有限公司 | Processing method and device of early warning information and electronic equipment |
| CN111010534A (en) * | 2019-11-11 | 2020-04-14 | 中国建设银行股份有限公司 | Real-time asynchronous video analysis method and system |
| CN111178241A (en) * | 2019-12-27 | 2020-05-19 | 航天信息股份有限公司 | An intelligent monitoring system and method based on video analysis |
| CN111223011A (en) * | 2019-12-29 | 2020-06-02 | 航天信息股份有限公司 | A method and system for food safety supervision of catering enterprises based on video analysis |
| CN113096201A (en) * | 2021-03-30 | 2021-07-09 | 上海西井信息科技有限公司 | Embedded video image deep learning system, method, equipment and storage medium |
| CN113096201B (en) * | 2021-03-30 | 2023-04-18 | 上海西井信息科技有限公司 | Embedded video image deep learning method, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107633238A (en) | A kind of video analysis method and intellectual analysis server | |
| Singh et al. | Real-time anomaly recognition through CCTV using neural networks | |
| CN112287816B (en) | Dangerous work area accident automatic detection and alarm method based on deep learning | |
| CN211293956U (en) | AI-based identification and alarm system for abnormal agent on construction site | |
| CN112084963B (en) | A monitoring and early warning method, system and storage medium | |
| CN114926781B (en) | Multi-person time-space domain abnormal behavior positioning method and system supporting real-time monitoring scene | |
| CN109712127B (en) | A transmission line fault detection method for machine patrol video stream | |
| CN107911653A (en) | The module of intelligent video monitoring in institute, system, method and storage medium | |
| CN108090458A (en) | Tumble detection method for human body and device | |
| KR102149832B1 (en) | Automated Violence Detecting System based on Deep Learning | |
| CN110351244A (en) | A kind of network inbreak detection method and system based on multireel product neural network fusion | |
| CN107527617A (en) | Monitoring method, apparatus and system based on voice recognition | |
| CN117576632B (en) | Multi-mode AI large model-based power grid monitoring fire early warning system and method | |
| Jayaswal et al. | A Framework for Anomaly Classification Using Deep Transfer Learning Approach. | |
| CN110674790A (en) | A kind of abnormal scene processing method and system in video surveillance | |
| CN114724584A (en) | Abnormal sound identification model construction method, abnormal sound detection method and system | |
| CN113534276B (en) | Railway intrusion behavior detection method based on fast R-CNN | |
| CN112541393A (en) | Transformer substation personnel detection method and device based on deep learning | |
| CN119418272B (en) | An intelligent security system, method, device and storage medium based on AI analysis | |
| CN112487961A (en) | Traffic accident detection method, storage medium and equipment | |
| CN115294519A (en) | Abnormal event detection and early warning method based on lightweight network | |
| CN115953831B (en) | Abnormal behavior supervision method and system in interrogation scenarios based on ST-Transformer network | |
| CN117523668A (en) | Abnormal behavior detection method in spatiotemporal action network | |
| CN106341658A (en) | Intelligent city security state monitoring system | |
| CN119919887A (en) | A production line safety monitoring method and system based on machine vision |
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: 20180126 |