CN106709521A - Fire pre-warning method and fire pre-warning system based on convolution neural network and dynamic tracking - Google Patents
Fire pre-warning method and fire pre-warning system based on convolution neural network and dynamic tracking Download PDFInfo
- Publication number
- CN106709521A CN106709521A CN201611219996.0A CN201611219996A CN106709521A CN 106709521 A CN106709521 A CN 106709521A CN 201611219996 A CN201611219996 A CN 201611219996A CN 106709521 A CN106709521 A CN 106709521A
- Authority
- CN
- China
- Prior art keywords
- flame
- region
- early warning
- convolutional neural
- neural networks
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
- G06F18/2111—Selection of the most significant subset of features by using evolutionary computational techniques, e.g. genetic algorithms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Physiology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a fire pre-warning method based on a convolution neural network and dynamic tracking, comprising the following steps: S1, carrying out Gaussian blurring on a video image to get an image after Gaussian blurring; S2, carrying out pixel-value-based candidate area extraction on the image after Gaussian blurring to extract candidate flame areas; S3, segmenting the candidate fire areas to get the coordinate positions of the bounding rectangles of final candidate flame areas; S4, judging whether the areas are a flame clump tracked, making a tracking judgment if the areas are a flame clump tracked, or, going to S5; S5, inputting the image at the coordinate position of the bounding rectangle of each final candidate flame area to a convolution neural network classifier, sending the images to a judgment layer after the features of the images are extracted by the convolution neural network classifier, and outputting the probability that each rectangle is a flame and the probability that each rectangle is not a flame; and S6, if the flame probability is greater than a set value, sending a pre-warning signal. The method and the system can prevent missing detection and false detection, and are low in cost and easy to use.
Description
Technical field
It is the present invention relates to field of computer technology more particularly to a kind of based on convolutional neural networks and dynamic tracking
Flame method for early warning and system.
Background technology
The main flow of prior art is:
The profile and behavioral characteristics of the textural characteristics such as method one, the symbiosis gray matrix based on image or flame, with reference to point
Class recognizes scheduling algorithm, carries out the fire defector in image.
Method two, special camera is used, such as add polarized lenses, high pass low pass filter, the letter such as infrared auxiliary
Breath carries out the judgement of flame.
The shortcoming that above method is present:
Method one:Because the unstability of manual feature, the degree of accuracy in complex environment is relatively low, when occur it is red or
Easily occur erroneous judgement during the object of crocus, and easily missing inspection occur when flame color is thin.
Method two:The camera for needing installation new could obtain this function, and original common monitoring camera can not be used,
Need to spend extra cost and abandon original equipment that the function of fire defector could be obtained.
Therefore, the prior art is defective, it is necessary to improve.
The content of the invention
The technical problems to be solved by the invention are:There is provided one kind can prevent missing inspection and flase drop, and cost is relatively low, user
Just the flame method for early warning tracked based on convolutional neural networks and dynamic and system.
Technical scheme is as follows:A kind of flame method for early warning based on convolutional neural networks and dynamic tracking, bag
Include following steps:S1:Gaussian Blur is carried out to video image, the image after Gaussian Blur is obtained;S2:To the figure after Gaussian Blur
Extracted as carrying out the candidate region based on pixel value, extract candidate's flame region;S3:Candidate's flame region is split,
Obtain the boundary rectangle coordinate position of final flame candidate region;S4:Determine whether the flame agglomerate in tracking, be to carry out
Tracking judgement, otherwise continues step S5;S5:By the image where the boundary rectangle coordinate position of each final flame candidate region,
Input convolutional neural networks grader, by sending into diagnostic horizon after the feature extraction of convolutional neural networks grader, finally exports
Be flame probability and be not flame probability;S6:When the probability of flame is more than setting value, early warning signal is sent.
Above-mentioned technical proposal is applied to, in described flame method for early warning, in step S6, the same of early warning signal is being sent
When, also transmit the video of front and rear ten seconds that flame finds, and system time during discovery flame.
Each above-mentioned technical proposal is applied to, in described flame method for early warning, after receiving this early warning signal, is also passed through
Websocket or http connections long notify webpage client or cell-phone customer terminal, by webpage client or cell-phone customer terminal, bullet
Go out pop-up and generate a warning data.
Each above-mentioned technical proposal is applied to, in described flame method for early warning, after pop-up is ejected, also by warning data
It is sent to by way of short message on the mobile phone of advance association.
Each above-mentioned technical proposal is applied to, in described flame method for early warning, also including step S7:Also to wrong report early warning
Fed back, and this region is designated as nonflame, be stored in feedback database.
Each above-mentioned technical proposal is applied to, in described flame method for early warning, after step S7, S71 is also performed:When anti-
When data volume in feedback database is more than N, then all feedback pictures in database are extracted, and add original flame to train number
According to storehouse, and over-sampling is carried out to feedback picture, convolutional neural networks grader is trained using new data set.
Each above-mentioned technical proposal is applied to, in described flame method for early warning, in step S5, is also regularly updated and practical
New convolutional neural networks grader.
Each above-mentioned technical proposal is applied to, in described flame method for early warning, in step S4, its step is specifically included:
S41:When in one region of input comprising flame, tracker contrasts this region with recorded flame region;S42:Judge
This region is higher than threshold value T1 with the Duplication of certain sole zone of record, then it is assumed that this region has been considered to flame, no longer
The judgement of convolutional neural networks is carried out, and the posting field for overlapping is replaced with the position of this new region;S43:Judge this area
Domain is both less than threshold value T2 with the Duplication in any region of record, then it is assumed that this region is newfound flame agglomerate, by this area
Domain is recorded into tracker;S44:The Duplication in the multiple regions in judging this region and recording is higher than threshold value T1, then it is assumed that former fire
The multiple in flame region is fused into current region, then take the minimum enclosed rectangle meter of all posting fields that current region is overlapped
Enter posting field, and the overlapping region of all records is deleted from tracker;S45:Judge certain in preceding multiple regions and record
One Duplication in region is higher than threshold value T1, then it is assumed that former flame region has split into current multiple regions, records all current
Region enters tracker and the posting field Chong Die with current multiple regions is deleted from tracker;S46:Judge outside this minimum
Rectangle is connect with the value that intersect of certain region in the flame agglomerate of tracking more than 80%, then it is assumed that this rectangle has been judged as fire
Flame, no longer carries out the judgement of convolutional neural networks, is directly used in renewal flame tracking device to save computing resource;It is otherwise next
Stepping enters the judgement of convolutional neural networks.
Each above-mentioned technical proposal is applied to, the corresponding flame early warning system for using of more than one any flame method for early warning
System, including fire defector subsystem and early warning subsystem, wherein, fire defector subsystem includes:Gauss is carried out to video image
It is fuzzy, obtain the Gaussian Blur module of the image after Gaussian Blur;The time based on pixel value is carried out to the image after Gaussian Blur
Favored area extracts the extraction module of candidate's flame region;Candidate's flame region is split, final flame candidate region is obtained
Boundary rectangle coordinate position segmentation module;To selecting flame region to be tracked the tracking determining device of judgement;And will
Image where the boundary rectangle coordinate position of each final flame candidate region, is input into convolutional neural networks grader, through pulleying
Product neural network classifier feature extraction after send into diagnostic horizon, finally output be flame probability and be not flame probability
Convolutional neural networks grader;Early warning subsystem includes flame early warning sending module and receives client.
Each above-mentioned technical proposal is applied to, in described flame early warning system, also including a pair of wrong report early warning is carried out
Feedback, and this region is designated as the user feedback subsystem and connected feedback database of nonflame.
Using such scheme, the present invention passes through dynamically to track the method combined with convolutional neural networks, to by pixel value
The flame candidate region of pre-selection is judged that can extract more rich and careful feature, the original craft feature of reduction is descriptive
The missing inspection for not enough bringing by force and flase drop.
Also, the function of fire defector early warning is obtained with using only monitoring camera, it is not necessary to replace hardware, and cost
It is relatively low, it is easy to use.
Brief description of the drawings
Fig. 1 is the flowage structure schematic diagram of Flame early warning system of the present invention;
Fig. 2 is the flowage structure schematic diagram of early warning subsystem in the present invention;
Fig. 3 is the flowage structure schematic diagram of user feedback subsystem in the present invention.
Specific embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.
Present embodiments provide a kind of flame method for early warning based on convolutional neural networks and dynamic tracking and system, this reality
Apply example and pass through dynamically to track the method combined with convolutional neural networks, the flame candidate region to being preselected by pixel value is sentenced
It is disconnected, more rich and careful feature can be extracted, reduce the original craft descriptive missing inspection for not enough bringing by force of feature and flase drop.
And embodiment is obtained with the function of fire defector early warning using only monitoring camera, it is not necessary to replace hardware, and
Cost is relatively low, easy to use.
The present embodiment combines traditional feature extraction and deep neural network, i.e. convolutional neural networks grader is to regarding
Frequency carries out implementation analysis, and the flame to detecting carries out early warning, it is ensured that monitor area has precognition energy in advance to fire
Power.When wrong report occurring or failing to report, can be fed back by the user of system, the data of feedback are used for Optimized model, improve
The stability and accuracy of system.
The system is mainly made up of three parts:Flame detector system, user feedback system and early warning system.
The work of flame detector system does primary treatment pre- mainly by the feature of image in video using manual feature
Candidate region is selected, candidate region is judged using convolutional neural networks then, whether drawn in video image containing flame
Judgement.
Early warning system is coupled with flame detector system, and when flame detector system produces flame, early warning system can handle
Detect the video segment of ten seconds before and after flame, and timestamp when detecting flame be collected by websocket or
The channel feedback of http connections long is to client.Wherein client includes mobile phone app ends and web terminal.And when client leaves
During phone number information, alarm can also be sent to Client handset in the form of short message.Timely analysis feedback can reduce the condition of a fire
The risk of expansion.
User feedback system is then for optimized algorithm on line, after client receives alarm, if user feels this
It is false alarm to alarm, then this warning message can be fed back by reporting feedback button by mistake.System then can be causing this
The image information of wrong report is labeled as nonflame and is stored in feedback database.
Flame detector system
Flame detector system mainly include Gaussian Blur, candidate region based on pixel value extract, candidate region segmentation with
And convolutional neural networks judge four parts.
Gaussian Blur:Noise for reducing image
σ=0.15*ksize-0.35
X, y are image coordinate, and ksize is the size of Gaussian Blur core.Calculated after Gaussian Blur core to every using above formula
Individual pixel does the image after convolution operation obtains Gaussian Blur.
Extract flame candidate region based on pixel value
Gray=b+g+r
Gray value gray and brightness value s is first calculated from rgb pixel values according to two above-mentioned formula.Then test following
Inequality condition.Wherein red_threshold is the lowest threshold of red component, and saturation_threshold is bright
Spend the lowest threshold of component.Inequality as follows is detected, it is possible to be considered candidate's meet inequality demand
Flame region.
1 > red_threshold
R > g
G > b
s≥(255-r)*saturation_threshold/red_threshold
Split candidate region
Extracted by candidate region above, we can obtain the mask of binaryzation, it is all to be tested by pixel value
Pixel is set to 255, and remaining is set to 0.
We are corroded and expansive working removal noise to mask images afterwards;
Again each piece of position of the mask of connection is obtained with connected region filtering;
Region of the removal connection area less than area_threshold, thinks that it is unlikely to be flame for too small region
Region, can remove if red bright spot etc in this step;
Calculate the barycenter of profile, the coordinate value of barycenter be exactly in profile all values be 255 point coordinate average value;
Start unrestrained water filling by sub-pixel point of profile barycenter in original RBG images, waited with flame of taking off as far as possible
The part of favored area.Principle is, since barycenter, detects the color of each point, if its neighbor pixel color and its color
Gap be considered as then flame pixels in threshold value, otherwise it is assumed that being non-flame pixels.
The flame candidate region extended after unrestrained water filling is extracted, and records its minimum enclosed rectangle.
If the value that this minimum enclosed rectangle intersects with certain region in the flame agglomerate of tracking is more than 80%, then it is assumed that
This rectangle has been judged as flame, no longer carries out the judgement of convolutional neural networks, be directly used in renewal flame tracking device with
Save computing resource;Otherwise next step enters the judgement of convolutional neural networks.The flame candidate region rectangle of this frame acquisition is set
Coordinate be (x, y, w, h), the computational methods that flame rectangular area in tracking is intersected for (x1, y1, w1, h1) wherein region are such as
Under, the wide and height in two rectangle intersection regions is first calculated, then ask the area of intersecting area and the ratio of itself area.
Cross_x=min (x+w, x1+w1)-max (x, x1)
Cross_y=min (y+h, y1+h1)-max (y, y1)
Ratio=cross_x*cross_y/ (w*h) if (cross_x > 0 and cross_y > 0)
Convolutional neural networks judge
Image where the boundary rectangle position of previous step is input into convolutional neural networks, by the spy of convolutional neural networks
Levy extraction after send into diagnostic horizon, finally output be flame probability and be not flame probability.
Tracking comprehensive descision
In view of flame is smaller in the scope that interframe is shifted under monitor video.Can use mass tracking pair it is determined that
Flame is tracked, in order to avoid convolutional neural networks are used for multiple times same flame is carried out to judge to waste computing resource.
And tracking can make up the mistake of convolutional neural networks, when convolutional neural networks previous frame extremely determines that this region is sent out
Showed flame, but present frame but judge it is not flame, such case probability of happening is smaller, but tracking can be caused at this
In the case of kind, flame detector system has the related output of sequential, the probability that reduction is made mistakes.
When in one region of input comprising flame, tracker contrasts this region with recorded flame region;
Survival:If this region is higher than threshold value T1 with the Duplication of certain sole zone of record, then it is assumed that this region is
It is considered as flame, no longer carries out the judgement of convolutional neural networks, and the record for overlapping is replaced with the position of this new region
Region;
It is newly-built:If this region is both less than threshold value T2 with the Duplication in any region of record, then it is assumed that this region is new
It was found that flame agglomerate, this regional record is entered into tracker;
Fusion:If the Duplication in the multiple regions in this region and record is higher than threshold value T1, then it is assumed that former flame region
Multiple be fused into current region, then take all posting field minimum enclosed rectangles that current region overlaps and count recording areas
Domain, and the overlapping region of all records is deleted from tracker;
Division:If current multiple region is higher than threshold value T1 with the Duplication in some region in record, then it is assumed that former fire
Flame regional split into current multiple regions, therefore record all current regions into tracker and from tracker delete with
The posting field that current multiple region overlaps.
Early warning system
When flame detector system detects newfound flame, then early warning system is notified, and transmit before flame finds
The system time during flame of the video of ten seconds, and discovery afterwards.
After receiving this early warning, backstage is connected by websocket and notifies webpage client, ejects pop-up and generation one
Warning data.Webpage client can see video and the time of flame generation in the early warning page.
Cell-phone customer terminal can also receive pre-alert notification by identical connection method.
If the user bound of camera provides phone number, can also be produced by SMS notification flame early warning signal,
Please user check in order to avoid causing unnecessary loss in time.
3rd, user feedback system
After user receives flame early warning, if it is considered to this time flame early warning is wrong report, wrong report feedback can be carried out.System
The flame region for causing this time alarm can be found out, and this region is designated as nonflame, be stored in feedback database.
When the data volume in feedback database is more than N, all feedback pictures in database can be extracted, and add original
Flame tranining database in, and to feedback picture carry out over-sampling, with improve feedback picture weights, reduce new model herein
The probability made a mistake in class data.Then convolutional neural networks grader is trained using new data set.
Training step is as follows:
Data are entered into row stochastic rotation, affine transformation is carried out;
Size of data is normalized into 70*70;
Data are cut out at random, rear size is cut out for 60*60;
Convolutional neural networks structure is entered data into, and is learnt using back-propagation algorithm, use neural network forecast value
Gap with label value carries out the weights of each level of backpropagation adjustment network as error;
Last network output be classification and its belong to such other probable value accordingly.
These are only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all it is of the invention spirit and
Any modification, equivalent and improvement for being made within principle etc., should be included within the scope of the present invention.
Claims (10)
1. it is a kind of based on convolutional neural networks and dynamic tracking flame method for early warning, it is characterised in that comprise the following steps:
S1:Gaussian Blur is carried out to video image, the image after Gaussian Blur is obtained;
S2:The candidate region based on pixel value is carried out to the image after Gaussian Blur to extract, and extracts candidate's flame region;
S3:Candidate's flame region is split, the boundary rectangle coordinate position of final flame candidate region is obtained;
S4:Determine whether the flame agglomerate in tracking, be to be tracked judgement, otherwise continue step S5;
S5:By the image where the boundary rectangle coordinate position of each final flame candidate region, input convolutional neural networks classification
Device, by sending into diagnostic horizon after the feature extraction of convolutional neural networks grader, finally output is the probability of flame and is not fire
The probability of flame;
S6:When the probability of flame is more than setting value, early warning signal is sent.
2. flame method for early warning according to claim 1, it is characterised in that:In step S6, the same of early warning signal is being sent
When, also transmit the video of front and rear ten seconds that flame finds, and system time during discovery flame.
3. flame method for early warning according to claim 2, it is characterised in that:After receiving this early warning signal, also pass through
Websocket or http connections long notify webpage client or cell-phone customer terminal, by webpage client or cell-phone customer terminal, bullet
Go out pop-up and generate a warning data.
4. flame method for early warning according to claim 3, it is characterised in that:After pop-up is ejected, also warning data is led to
The mode for crossing short message is sent on the mobile phone of advance association.
5. flame method for early warning according to claim 2, it is characterised in that:Also include step S7:Also wrong report early warning is entered
Row feedback, and this region is designated as nonflame, it is stored in feedback database.
6. flame method for early warning according to claim 5, it is characterised in that:After step S7, S71 is also performed:Work as feedback
When data volume in database is more than N, then all feedback pictures in database are extracted, and add original flame training data
In storehouse, and over-sampling is carried out to feedback picture, convolutional neural networks grader is trained using new data set.
7. flame method for early warning according to claim 6, it is characterised in that:In step S5, also regularly update and practical new
Convolutional neural networks grader.
8. flame method for early warning according to claim 1, it is characterised in that:In step S4, its step is specifically included:
S41:When in one region of input comprising flame, tracker contrasts this region with recorded flame region;
S42:Judge that this region is higher than threshold value T1 with the Duplication of certain sole zone of record, then it is assumed that this region has been considered to
It is flame, no longer carries out the judgement of convolutional neural networks, and the posting field for overlapping is replaced with the position of this new region;
S43:Judge the Duplication both less than threshold value T2 in this region and any region of record, then it is assumed that this region is newfound
Flame agglomerate, tracker is entered by this regional record;
S44:The Duplication in the multiple regions in judging this region and recording is higher than threshold value T1, then it is assumed that the multiple of former flame region
Current region is fused into, then the minimum enclosed rectangle for taking all posting fields that current region is overlapped counts posting field,
And the overlapping region of all records is deleted from tracker;
S45:Multiple region and the Duplication in some region in record are higher than threshold value T1 before judging, then it is assumed that former flame region point
Current multiple regions are cleaved into, all current regions have been recorded into tracker and is deleted from tracker and current multiple regions
The posting field of overlap;
S46:Judge that the value that this minimum enclosed rectangle intersects with certain region in the flame agglomerate of tracking is more than 80%, then it is assumed that
This rectangle has been judged as flame, no longer carries out the judgement of convolutional neural networks, be directly used in renewal flame tracking device with
Save computing resource;Otherwise next step enters the judgement of convolutional neural networks.
9. the flame early warning systems for being used corresponding to more than one power any flame method for early warning of 1-8, it is characterised in that:Including fire
Flame detects subsystem and early warning subsystem, wherein, fire defector subsystem includes:
Gaussian Blur is carried out to video image, the Gaussian Blur module of the image after Gaussian Blur is obtained;
The extraction module that candidate's flame region is extracted in the candidate region based on pixel value is carried out to the image after Gaussian Blur;
Candidate's flame region is split, the segmentation module of the boundary rectangle coordinate position of final flame candidate region is obtained;
To selecting flame region to be tracked the tracking determining device of judgement;
And by the image where the boundary rectangle coordinate position of each final flame candidate region, input convolutional neural networks are classified
Device, by sending into diagnostic horizon after the feature extraction of convolutional neural networks grader, finally output is the probability of flame and is not fire
The convolutional neural networks grader of the probability of flame;
Early warning subsystem includes flame early warning sending module and receives client.
10. flame early warning system according to claim 9, it is characterised in that:Also including a pair of wrong report early warning is carried out instead
Feedback, and this region is designated as the user feedback subsystem and connected feedback database of nonflame.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611219996.0A CN106709521A (en) | 2016-12-26 | 2016-12-26 | Fire pre-warning method and fire pre-warning system based on convolution neural network and dynamic tracking |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611219996.0A CN106709521A (en) | 2016-12-26 | 2016-12-26 | Fire pre-warning method and fire pre-warning system based on convolution neural network and dynamic tracking |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106709521A true CN106709521A (en) | 2017-05-24 |
Family
ID=58903356
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611219996.0A Pending CN106709521A (en) | 2016-12-26 | 2016-12-26 | Fire pre-warning method and fire pre-warning system based on convolution neural network and dynamic tracking |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106709521A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107349538A (en) * | 2017-08-03 | 2017-11-17 | 电子科技大学 | The intelligence anticipation system and method that a kind of fire isolation strip is configured |
CN108171318A (en) * | 2017-11-30 | 2018-06-15 | 河南大学 | One kind is based on the convolutional neural networks integrated approach of simulated annealing-Gaussian function |
CN108257347A (en) * | 2018-01-10 | 2018-07-06 | 安徽大学 | A kind of flame image sequence sorting technique and device using convolutional neural networks |
CN108985374A (en) * | 2018-07-12 | 2018-12-11 | 天津艾思科尔科技有限公司 | A kind of flame detecting method based on dynamic information model |
CN109214353A (en) * | 2018-09-27 | 2019-01-15 | 云南大学 | A kind of facial image based on beta pruning model quickly detects training method and device |
CN111639620A (en) * | 2020-06-08 | 2020-09-08 | 深圳航天智慧城市***技术研究院有限公司 | Fire disaster analysis method and system based on visible light image recognition |
US20200368568A1 (en) * | 2018-06-04 | 2020-11-26 | Gaoli Ge | Electric heater fire extinguishing strategy customization mechanism |
CN112069974A (en) * | 2020-09-02 | 2020-12-11 | 安徽铜峰电子股份有限公司 | Image recognition method and system for recognizing defects of components |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101887587A (en) * | 2010-07-07 | 2010-11-17 | 南京邮电大学 | Multi-target track method based on moving target detection in video monitoring |
CN103150856A (en) * | 2013-02-28 | 2013-06-12 | 江苏润仪仪表有限公司 | Fire flame video monitoring and early warning system and fire flame detection method |
CN103473788A (en) * | 2013-07-31 | 2013-12-25 | 中国电子科技集团公司第三十八研究所 | Indoor fire and flame detection method based on high-definition video images |
CN104008368A (en) * | 2014-05-13 | 2014-08-27 | 重庆大学 | Fire recognition method based on maximum entropy threshold segmentation |
CN106097391A (en) * | 2016-06-13 | 2016-11-09 | 浙江工商大学 | A kind of multi-object tracking method identifying auxiliary based on deep neural network |
-
2016
- 2016-12-26 CN CN201611219996.0A patent/CN106709521A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101887587A (en) * | 2010-07-07 | 2010-11-17 | 南京邮电大学 | Multi-target track method based on moving target detection in video monitoring |
CN103150856A (en) * | 2013-02-28 | 2013-06-12 | 江苏润仪仪表有限公司 | Fire flame video monitoring and early warning system and fire flame detection method |
CN103473788A (en) * | 2013-07-31 | 2013-12-25 | 中国电子科技集团公司第三十八研究所 | Indoor fire and flame detection method based on high-definition video images |
CN104008368A (en) * | 2014-05-13 | 2014-08-27 | 重庆大学 | Fire recognition method based on maximum entropy threshold segmentation |
CN106097391A (en) * | 2016-06-13 | 2016-11-09 | 浙江工商大学 | A kind of multi-object tracking method identifying auxiliary based on deep neural network |
Non-Patent Citations (2)
Title |
---|
毛星云 等: "《OpenCV3编程入门》", 28 February 2015 * |
高荣: "隧道火灾红外双目视觉定位***研究", 《中国优秀硕士学位论文全文数据库信息科技辑(月刊)》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107349538B (en) * | 2017-08-03 | 2019-12-27 | 电子科技大学 | Intelligent pre-judging system and method for fire isolation zone arrangement place |
CN107349538A (en) * | 2017-08-03 | 2017-11-17 | 电子科技大学 | The intelligence anticipation system and method that a kind of fire isolation strip is configured |
CN108171318A (en) * | 2017-11-30 | 2018-06-15 | 河南大学 | One kind is based on the convolutional neural networks integrated approach of simulated annealing-Gaussian function |
CN108257347A (en) * | 2018-01-10 | 2018-07-06 | 安徽大学 | A kind of flame image sequence sorting technique and device using convolutional neural networks |
CN108257347B (en) * | 2018-01-10 | 2020-09-29 | 安徽大学 | Flame image sequence classification method and device by using convolutional neural network |
US20200368568A1 (en) * | 2018-06-04 | 2020-11-26 | Gaoli Ge | Electric heater fire extinguishing strategy customization mechanism |
CN108985374A (en) * | 2018-07-12 | 2018-12-11 | 天津艾思科尔科技有限公司 | A kind of flame detecting method based on dynamic information model |
CN109214353B (en) * | 2018-09-27 | 2021-11-23 | 云南大学 | Training method and device for rapid detection of face image based on pruning model |
CN109214353A (en) * | 2018-09-27 | 2019-01-15 | 云南大学 | A kind of facial image based on beta pruning model quickly detects training method and device |
CN111639620A (en) * | 2020-06-08 | 2020-09-08 | 深圳航天智慧城市***技术研究院有限公司 | Fire disaster analysis method and system based on visible light image recognition |
CN111639620B (en) * | 2020-06-08 | 2023-11-10 | 深圳航天智慧城市***技术研究院有限公司 | Fire analysis method and system based on visible light image recognition |
CN112069974A (en) * | 2020-09-02 | 2020-12-11 | 安徽铜峰电子股份有限公司 | Image recognition method and system for recognizing defects of components |
CN112069974B (en) * | 2020-09-02 | 2023-04-18 | 安徽铜峰电子股份有限公司 | Image recognition method and system for recognizing defects of components |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106709521A (en) | Fire pre-warning method and fire pre-warning system based on convolution neural network and dynamic tracking | |
JP6474107B2 (en) | Video monitoring system, video processing apparatus, video processing method, and video processing program | |
CN110084165B (en) | Intelligent identification and early warning method for abnormal events in open scene of power field based on edge calculation | |
CN104737202A (en) | Fire detection method and apparatus | |
CN106559695A (en) | Barrage message processing method and device and electronic equipment | |
CN102348128A (en) | Surveillance camera system having camera malfunction detection function | |
CN110516529A (en) | It is a kind of that detection method and system are fed based on deep learning image procossing | |
CN106339657B (en) | Crop straw burning monitoring method based on monitor video, device | |
CN103246896A (en) | Robust real-time vehicle detection and tracking method | |
CN110718067A (en) | Violation behavior warning method and related device | |
CN103686092B (en) | Student stands up dual system detection method and system | |
CN107480653A (en) | passenger flow volume detection method based on computer vision | |
CN112422909B (en) | Video behavior analysis management system based on artificial intelligence | |
CN111553305B (en) | System and method for identifying illegal videos | |
Madhavi et al. | Traffic Congestion Detection from Surveillance Videos using Deep Learning | |
CN104715614B (en) | Method for capturing suspected fake-licensed cars | |
CN112069043A (en) | Terminal equipment state detection method, model generation method and device | |
CN113920585A (en) | Behavior recognition method and device, equipment and storage medium | |
CN111753587A (en) | Method and device for detecting falling to ground | |
CN113963316A (en) | Target event determination method and device, storage medium and electronic device | |
US20230156161A1 (en) | Failure identification and handling method, and system | |
CN111144260A (en) | Detection method, device and system of crossing gate | |
CN110647831A (en) | Court trial patrol method and system | |
TWI706381B (en) | Method and system for detecting image object | |
CN109409224A (en) | A kind of method of natural scene fire defector |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170524 |
|
RJ01 | Rejection of invention patent application after publication |