CN110517441A - Based on the frame-embedded smog of deep learning and flame video alarming system and method - Google Patents
Based on the frame-embedded smog of deep learning and flame video alarming system and method Download PDFInfo
- Publication number
- CN110517441A CN110517441A CN201910916328.0A CN201910916328A CN110517441A CN 110517441 A CN110517441 A CN 110517441A CN 201910916328 A CN201910916328 A CN 201910916328A CN 110517441 A CN110517441 A CN 110517441A
- Authority
- CN
- China
- Prior art keywords
- smog
- fire behavior
- flame
- deep learning
- behavior information
- 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
- 238000013135 deep learning Methods 0.000 title claims abstract description 63
- 238000000034 method Methods 0.000 title claims abstract description 23
- 230000006399 behavior Effects 0.000 claims abstract description 104
- 238000012544 monitoring process Methods 0.000 claims abstract description 23
- 238000012545 processing Methods 0.000 claims description 23
- 235000019504 cigarettes Nutrition 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 7
- 239000003595 mist Substances 0.000 claims description 2
- 238000012549 training Methods 0.000 description 17
- 238000001514 detection method Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 230000002265 prevention Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000004044 response Effects 0.000 description 4
- 238000010792 warming Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000009434 installation Methods 0.000 description 3
- 230000035945 sensitivity Effects 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 230000005012 migration Effects 0.000 description 2
- 238000013508 migration Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 230000007717 exclusion Effects 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 208000030523 mesoaxial synostotic syndactyly with phalangeal reduction Diseases 0.000 description 1
- 230000004297 night vision Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000005622 photoelectricity Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000036632 reaction speed Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 239000000779 smoke Substances 0.000 description 1
- 238000011895 specific detection Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000032258 transport Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/12—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions
- G08B17/125—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions by using a video camera to detect fire or smoke
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- Theoretical Computer Science (AREA)
- Alarm Systems (AREA)
Abstract
The present invention discloses a kind of based on the frame-embedded smog of deep learning and flame video alarming system comprising sequentially connected Image Acquisition end, embeded processor, control centre, Cloud Server and mobile terminal；The image information of Image Acquisition end acquisition ambient enviroment is simultaneously transferred to embeded processor；Embeded processor is judged according to deep learning intelligent algorithm containing smog or/and flame characteristic in image information, and forms fire behavior information, and then embeded processor transmits fire behavior information to control centre；Fire behavior information carries out fire behavior alarm based on the received for control centre, and transmits fire behavior information to Cloud Server；Cloud Server transmits fire behavior information to mobile terminal；Present invention further teaches one kind to be based on the frame-embedded smog of deep learning and flame video alarm method.The fire alarm system that the application is low in cost, identification is accurate, is quick on the draw, and can directly be transformed in original video monitoring system, is suitable for popularization and application.
Description
Technical field
The present invention relates to fire alarm technical fields, and in particular, to one kind is based on the frame-embedded smog of deep learning
And flame video alarming system and method.
Background technique
Fire threatens people's life and property safety huge, and discovery fire early simultaneously carries out control and can protect the people
Safety of life and property, it is significant.In order to find fire early, although having had already appeared several fire in the prior art
Detection system, but still remain different defects.
Cigarette sense one is the fire detecting system formed by traditional cigarette sense and warming fire detector, in such system
It is generally less than 12 meters with the radius of investigation of the small wherein smoke alarm of the investigative range of warming fire detector, single temperature detection
The investigative range of device is generally smaller than 50 square metres.In addition, the response time of cigarette sense and warming fire detector is longer, generally exist
Fire just responds after occurring 30 seconds.Moreover, cigarette sense and warming fire detector are also vulnerable to the influence of external environment, such as wind speed
And building structure, the ability of anti-interference are poor.Another kind is the fire detecting system of GPU server, and such system is inherently
It is expensive, and many erection recondition expenses are also needed, it is with high costs.
There are also one is the high-definition image type flame detector as disclosed in Publication No. CN206893057U, which is visited
The structure for surveying device is: acquisition component and embedded intelligence image flame analysis platform are fixed by the bracket in main body, wherein regarding
Frequency acquisition component includes a black-white CCD mould group, a high clear colorful mould group, and the bimodulus group in video acquisition component is located at main body
Front end the same horizontal position, embedded intelligence fire disaster analyzing platform include core processor, video acquisition unit, network transmission list
Member, input and output control and alarm output unit；Core processor is DSP+ARM dual core processor；The video of acquisition component
Output is connected in the interface of video acquisition component of analysis platform.The working principle of the fire detector is: video acquisition group
Collected vision signal is input in embedded fire analysis platform by part, is realized coding and decoding video, video image analysis, is led to
With functions such as input and output control, data storage and network communications, wherein DSP specializes in data processing, and ARM, which is specialized in, is
It is under the overall leadership reason, control and data communication, processing result can be transmitted by alarm output unit, video is transferred to rear end by Ethernet
In monitor supervision platform.Such high-definition image type flame detector has the disadvantages that the 1, system using combination picture flame detecting
Method, install the black-white CCD mould group and high clear colorful mould group of infrared fileter additional using front end, therefore structure is complicated, cost
Costly, and then the cost of sensor is increased.Although 2, black-white CCD mould group front end installation infrared fileter can be by major part
Interference source filters out, but due in sunlight there is also infrared light radiation, therefore, it is difficult to by the interference elimination of sunlight, if will be compared with
This will increase equipment cost again with the infrared fileter that good exclusion sunlight interference then needs installation more high-end, and also be easy
The flammule at initial stage, which is filtered out, leads to missing inspection.3, the system mainly carries out pyrotechnics by increasing hardware module on video camera
Identification only determines whether fire exception on flame identification algorithm simply by CRT technology judgement, this
Recognition methods is easy the interference by environment, thus recognition accuracy is lower, and it is also few to be applicable in scene.4, the system is due to using
Infrared filter, therefore can only carry out the detection of flame, time of fire alarming has limitation, and cigarette is the omen of fire, if can be
The stage that smog generates can detect, can generate alarm in advance.5, the video acquisition component and embedded intelligence of the system
Fire disaster analyzing platform is fixed by the bracket in a shell, so that an entirety is constituted, to the place for being already installed with CATV
For, it is then Chong Die with existing video monitoring function to install the set system, can not upgrade in existing video monitoring system and change
It makes.
Therefore, be badly in need of now it is a kind of it is low in cost, identification is accurate, is quick on the draw, and can be in original video monitoring system
On the fire alarm system be directly transformed.
Summary of the invention
The present invention provides a kind of based on the frame-embedded smog of deep learning and flame video in view of the deficiencies of the prior art
Alarm system and method.
One kind disclosed by the invention includes image based on the frame-embedded smog of deep learning and flame video alarming system
Collection terminal, embeded processor, control centre, Cloud Server and mobile terminal；Image Acquisition end is connect with embeded processor,
Embeded processor is connect with control centre, and control centre connect with Cloud Server, and Cloud Server is connect with mobile terminal；Image
The image information of collection terminal acquisition ambient enviroment is simultaneously transferred to embeded processor；Embeded processor is artificial according to deep learning
Intelligent algorithm is judged containing smog or/and flame characteristic in image information, and forms fire behavior information, then embeded processor
Fire behavior information is transmitted to control centre；Fire behavior information carries out fire behavior alarm based on the received for control centre, and transmits fire behavior information
To Cloud Server；Cloud Server transmits fire behavior information to mobile terminal.
It according to an embodiment of the present invention, further include video memory；Video memory is connect with embeded processor；
Video memory receives fire behavior information and is stored.
According to an embodiment of the present invention, Image Acquisition end is the web camera of video monitoring system.
According to an embodiment of the present invention, embeded processor includes processing unit and storage unit；Deep learning people
The algorithm model of work intelligent algorithm is stored in storage unit, and processing unit receives the image information of Image Acquisition end transmitting, and adjusts
The algorithm model of the deep learning intelligent algorithm stored with storage unit carries out input calculating, and deep learning artificial intelligence is calculated
The algorithm model output image information of method contains the fire behavior information of smog or/and flame characteristic.
According to an embodiment of the present invention, fire behavior information include the image containing smog or/and flame characteristic callout box with
And the information of fire behavior Alert Level.
According to an embodiment of the present invention, control centre includes control platform and warning end, and control platform receives fire behavior letter
Breath, and warning end is controlled according to fire behavior information and carries out corresponding picture and text or/and audible alarm.
One kind disclosed by the invention includes following based on the frame-embedded smog of deep learning and flame video alarm method
Step:
Acquire the image information of ambient enviroment；
Judged according to deep learning intelligent algorithm containing smog or/and flame characteristic in image information, and is formed
Fire behavior information；
Fire behavior alarm is carried out according to fire behavior information；
Upload fire behavior information；
Issue fire behavior information.
According to an embodiment of the present invention, judge to contain smog in image information according to deep learning intelligent algorithm
And flame characteristic, comprising the following steps:
The algorithm model of image information input deep learning intelligent algorithm；
The characteristics of image of doubtful smog or/and flame in image information is extracted, and forms object；
Detect the threshold values of object；
Judge whether the confidence level of object is greater than threshold values；If it has not, then re-entering image information to deep learning people
The algorithm model of work intelligent algorithm；If it has, then judging that the object in image information for smog or/and flame, and forms fire
Feelings information.
According to an embodiment of the present invention, fire behavior information is formed, comprising:
According between the smog of the area and object of the smog of object or/and flame or/and the area of flame
Pace of change forms the fire behavior information containing fire behavior Alert Level.
According to an embodiment of the present invention, when detecting the threshold values of object, object is labeled, and is believed in image
Callout box is formed in breath；Fire behavior information further includes the image containing smog or/and flame characteristic callout box.
The application is low in cost, identification is accurate, is quick on the draw, and can be directly transformed in original video monitoring system
Fire alarm system is suitable for popularization and application.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present application, constitutes part of this application, this Shen
Illustrative embodiments and their description please are not constituted an undue limitation on the present application for explaining the application.In the accompanying drawings:
Fig. 1 is that the frame-embedded smog of deep learning and flame video alarming system control block diagram are based in embodiment one；
Fig. 2 is the flow chart based on the frame-embedded smog of deep learning and flame video alarm method in embodiment two；
Fig. 3 is to be judged to contain smog and flame in image information according to deep learning intelligent algorithm in embodiment two
The flow chart of feature.
Description of symbols:
1, Image Acquisition end；2, embeded processor；21, processing unit；22, storage unit；3, control centre；4, cloud takes
Business device；5, mobile terminal；6, video memory.
Specific embodiment
Multiple embodiments of the invention will be disclosed with schema below, as clearly stated, the details in many practices
It will be explained in the following description.It should be appreciated, however, that the details in these practices is not applied to limit the present invention.Also
It is to say, in some embodiments of the invention, the details in these practices is non-essential.In addition, for the sake of simplifying schema,
Some known usual structures and component will be painted it in the drawings in simply illustrative mode.
It is to be appreciated that the directional instruction such as up, down, left, right, before and after ... of institute is only used in the embodiment of the present invention
In explaining relative positional relationship, motion conditions etc. in the case where a certain particular pose is as shown in the picture between each component, if the spy
When determining posture and changing, then directionality instruction also correspondingly changes correspondingly.
In addition, the description for being such as related to " first ", " second " in the present invention is used for description purposes only, not especially censure
The meaning of order or cis-position, also non-to limit the present invention, the component described just for the sake of difference with same technique term
Or operation, it is not understood to indicate or imply its relative importance or implicitly indicates the number of indicated technical characteristic
Amount." first " is defined as a result, the feature of " second " can explicitly or implicitly include at least one of the features.In addition, each
Technical solution between a embodiment can be combined with each other, but must can be implemented as base with those of ordinary skill in the art
Plinth will be understood that the combination of this technical solution is not present when conflicting or cannot achieve when occurs in the combination of technical solution,
Also not the present invention claims protection scope within.
In order to further understand the content, features and effects of the present invention, the following examples are hereby given, and cooperate attached drawing
Detailed description are as follows:
Embodiment one
Referring to Fig.1, Fig. 1 is that the frame-embedded smog of deep learning and flame video alarming system control are based in embodiment one
Block diagram processed.In the present embodiment includes Image Acquisition end based on the frame-embedded smog of deep learning and flame video alarming system
1, embeded processor 2, control centre 3, Cloud Server 4 and mobile terminal 5.Image Acquisition end 1 and embeded processor 2 connect
It connects, embeded processor 2 is connect with control centre 3, and control centre 3 connect with Cloud Server 4, Cloud Server 4 and mobile terminal 5
Connection.Image Acquisition end 1 acquires the image information of ambient enviroment and is transferred to embeded processor 2.2 basis of embeded processor
Deep learning intelligent algorithm is judged containing smog or/and flame characteristic in image information, and forms fire behavior information, then
Embeded processor 2 transmits fire behavior information to control centre 3.Fire behavior information carries out fire behavior alarm based on the received for control centre 3,
And fire behavior information is transmitted to Cloud Server 4.Cloud Server 4 transmits fire behavior information to mobile terminal 5.
It is set by the cooperation of Image Acquisition end 1, embeded processor 2, control centre 3, Cloud Server 4 and mobile terminal 5
Set, provide it is a kind of it is low in cost, identification is accurate, be quick on the draw, and can be directly transformed in original video monitoring system
Fire alarm system is suitable for popularization and application.
Again referring to Fig.1, further, Image Acquisition end 1 is the web camera of video monitoring system.It is understood that
It is that video monitoring system has widely been promoted as a kind of technology of maturation, such as CATV；And web camera conduct
The video acquisition end of video monitoring system has covered among existing video surveillance network and has needed each of fire preventing
In kind scene and environment.Image Acquisition end 1 in the present embodiment can use the web camera of video monitoring network system, and two
The public video acquisition end of person, does not need the Image Acquisition end 1 for re-laying this system, in original video surveillance network system
Be transformed on system so that the corresponding video acquisition end with a video monitoring network system of each embeded processor 2 into
Row connection, formed one-to-one relationship, multiple existing video acquisition ends can be realized in the present embodiment based on deep learning
The all standing of frame-embedded smog and flame video alarming system.Above-mentioned transformation process is simple and convenient, and expense is low, without weight
Original video monitoring system is installed or replaced again, nor will affect original video monitoring function, compatibility is high.Such as
This, can save the hardware cost at Image Acquisition end 1, also not need to be equipped with additional photoelectricity, infrared energy monitoring modular, at low cost
It is honest and clean.Especially relative to the fire detecting system of GPU server, the equipment cost of tens of thousands of members can be saved.In addition, being taken the photograph with network
Camera can reach 1 meter to 100 meters as Image Acquisition end 1, detection monitoring distance, and specific detection monitoring distance can be according to net
The parameter of network video camera determines, so may make the coverage area of this system farther, can monitor fire condition remotely,
Increase the monitoring range of this system.Moreover, the web camera as Image Acquisition end 1 also has night vision function, in low photograph
The night-environment of degree can also carry out the acquisition of image information, handle by this system, and the automatic knowledge of fire equally may be implemented
Not, the application range of this system has been expanded.Web camera also has the environment such as wind speed and building structure stronger simultaneously
Adaptability.Image information in the present embodiment, for the video image of web camera shooting, single network video camera each second can
Transmit the video image of 30 frames.
Again referring to Fig.1, further, embeded processor 2 includes processing unit 21 and storage unit 22.Deep learning
The algorithm model of intelligent algorithm is stored in storage unit 22, and processing unit 21 receives the image letter transmitted at Image Acquisition end 1
Breath, and the algorithm model for the deep learning intelligent algorithm for calling storage unit 22 to store carries out input calculating, deep learning
The algorithm model output image information of intelligent algorithm contains the fire behavior information of smog or/and flame characteristic.
Specifically, processing unit 21 is equipped with Ubuntu system, the algorithm mould of deep learning intelligent algorithm in advance
Type is copied into storage unit 22 after the completion of external world's training.Multicore processing can be used in processing unit 21 in the present embodiment
Memory can be used in chip, storage unit 22.The image information that Image Acquisition end 1 acquires is transmitted to processing unit 21, for example, net
The video image of network video camera 30 frames of transmission per second.The Ubuntu system of processing unit 21 can call depth in storage unit 22
The algorithm model of intelligent algorithm is practised, and a frame video image is transmitted as input, then by algorithm mould using Image Acquisition end 1
Type carries out intelligence computation, and last algorithm model can export the fire behavior information containing smog or/and flame characteristic.The output of fire behavior information
Later, processing unit 21 continues to transmit fire behavior information to control centre 3.
In specific application, the training of the algorithm model of deep learning intelligent algorithm and migration process are as follows: step
One, train smog fire defector algorithm model.On GPU server using Caffe deep learning frame by tens of thousands of or
The picture of hundreds of thousands of smog, flame and the pyrotechnics marked optimizes the parameters of algorithm model, the artificial mind of training
Through network, the detection model of accuracy rate high smog and flame object is trained.Step 2, in the Ubuntu system of processing unit 21
AI inference engine Tengine is installed in system.The source code that Tengine is downloaded from github, into Tengine's after the completion of downloading
Makefile.config file is modified under catalogue, is then executed make instruction and is compiled the installation for completing Tengine.Step
Three, the network structure file mobilenet_ of mobilenet.caffemodel algorithm model and model that training is obtained
Deploy.prototxt is copied in the Tengine catalogue of storage unit 22, modifies the mssd.cpp text under Tengine catalogue
Part, the program in mssd.cpp are write with C Plus Plus, it is main realize read camera, detection judges flame smog and fire report
The cloud upload function of alert information executes make compiling of instruction program after the completion of modification and obtains MSSD.exe executable file, transports
Row this document can start fire detection system, so far complete migration process.
The detection of video image is carried out by deep learning algorithm of target detection, algorithm model has used tens of thousands of or number
100000 smog or flame image under different scenes optimizes training, and this system is enabled to accurately identify smog fire
Flame, strong antijamming capability reduce the false alarm rate of fire identification alarm.In addition, algorithm model takes into account the instruction of smog and flame
Practice study, when calculating judges, fire real time video image can be jointly processed by conjunction with smog and flame image, can greatly improved
The accuracy of fire judgement.By optimizing training to artificial neural network, image detection speed is promoted, and then promote this and be
The reaction sensitivity of system.Meanwhile the transplanting mode of the algorithm model using deep learning intelligent algorithm, it reduces and height is matched
Set the dependence of server, it is possible to reduce network transfer speeds, blocking bottleneck limitation.
Preferably, fire behavior information includes the image containing smog or/and flame characteristic callout box and fire behavior Alert Level
Information.It is understood that there is the overlay area of oneself as the web camera at Image Acquisition end 1, by fire behavior
The image containing smog or/and flame characteristic callout box is formed among information, convenient for people when obtaining fire behavior information, is convenient for people
There are more intuitive impression, convenient for directly making further judgement, fire in the generation position that fire in callout box occurs for work
Processing is very urgent thing, only shortens and judges the time, can handle in time, protects personal safety as well as the property safety.Meanwhile it is deep
The algorithm model of degree study intelligent algorithm can also increase characteristic according to area of flame, for example, the range that area of flame increases
And the speed increased, different fire behavior Alert Levels is marked off, the rank of fire behavior Alert Level is higher, the fire behavior of subsequent sending
Alarm is stronger, to allow supervisor to draw attention.
Again referring to Fig.1, further, control centre 3 includes control platform and warning end, and control platform receives fire behavior letter
Breath, and warning end is controlled according to fire behavior information and carries out corresponding picture and text or/and audible alarm.Specifically, embeded processor 2
The connection relationship that data information can interact, the processing unit of embeded processor 2 are formed with control centre 3 by the network switch
The fire behavior information of 21 algorithm models output passes to the control platform of control centre 3, for example, control platform can be for display
The end PC of screen, control platform show received fire behavior information on a display screen, and according to fire behavior Alert Level control
Warning end processed is alarmed.Warning end can be used loudspeaker and with the LED display for rolling picture and text showing, fire behavior alarm grade
Not higher, the audio warning of the sending of loudspeaker just has that louder and frequency is higher, while the picture and text of LED display flash frequency
Rate also can be higher.Wherein, the position and fire behavior Alert Level that fire occurs can be shown in LED picture and text showing, preferred
Scheme in, can be prompted with the position of the fire service inventory to monitoring area range, for example, show near fire send out
The position of the urgent fire service inventory in raw region, such as fire extinguisher, the position of outdoor water pipe head.At this point, the work people of control centre 3
Member can further be judged according to the pictorial information shown in control platform, if there is the case where fire erroneous judgement, be closed
Fire behavior alarm, if not fire is judged by accident, then processing of being evacuated, put out a fire or reported a fire in time, because, in LED picture and text showing
It has had shown that the position of the position of fire, size and extinguishment fire suppression articles, has so just been handled for quick extinguishing and condition is provided.
The response time of this system is small, for fire recognition speed less than 1 second, it is fast by the speed of network transmission, generally worse
Network transmission environment under, response speed in early detection fire and can alarm again smaller than 10 seconds, and substantially increase
Administration of the prevention and control timeliness and safety.Moreover, this system other than carrying out identification judgement to flame, is combined to smog figure
The identification of picture judges that cigarette is the omen of fire, takes into account identification for smog, this is extremely important for the judgement of incipient fire, energy
The response time for enough promoting this system, further improve administration of the prevention and control timeliness and safety.
Again referring to Fig.1, further, control centre 3 can also upload fire behavior information to cloud while carrying out fire behavior alarm
Server 4.In specific application, control centre 3 can establish wireless network communication by gateway and Cloud Server 4, and then realize
The interaction of data information.Cloud Server 4 can receive fire behavior information, and be issued to mobile terminal 5.Mobile terminal 5 is the palm
It is held in user hand, for example, the safety director of video monitoring regional, security personnel or region fire alarm responsible person etc., user is according to shifting
The fire behavior information of dynamic terminal 5 may know that there is a situation where positions and fire to occur for fire at a distance, can do source in time
Reason.The smart phone with APP can be used in mobile terminal 5 in the present embodiment, can carry out to the image in fire behavior information
It has been shown that, so that user can intuitively see the field condition of fire behavior alarm region, judges processing in time.Further, it is also possible to
The form of voice SMS is pushed to the smart phone of user.
Cloud Server 4 can provide data-pushing interface for this system, enable the fire behavior information and control of mobile terminal 5
The fire behavior information at center 3 processed synchronizes update, so where no matter user can know warning message and fire in time
Calamity field condition realizes multi-party coordinated management, and then the reliability of this system can be improved.By the fire behavior information at scene, such as have
There is the picture of callout box to be transferred to for the mobile terminal 5 in hand, allow user that can see on-site actual situations at the first time, reduces
The generation report by mistake, failed to report.In addition, Cloud Server 4 also there is big data to manage platform, fire behavior information can be stored, be made
Algorithm model can be advanced optimized, increase and calculate as training material by obtaining the picture containing smog or/flame in fire behavior information
The judging nicety rate of method model.In addition, being used as transfer by Cloud Server 4, extinguishing pipe is reduced by cloud telecommunication
Manage cost.
It is multiple referring to Fig.1, in the present embodiment based on the frame-embedded smog of deep learning and flame video alarming system also
Including video memory 6.Video memory 6 is connect with embeded processor 2, and video memory 6 receives fire behavior information and carries out
Storage.Video memory 6 is specifically to store to containing smog or/and flame characteristic image information, in embeded processor 2
What the deep learning intelligent algorithm of storage can be stored with video memory 6 believes containing smog or/and flame characteristic image
Breath carries out lasting deep learning, to increase the accuracy of deep learning intelligent algorithm, so that in the present embodiment
Also it is capable of the knowledge of Continuous optimization itself in use process based on the frame-embedded smog of deep learning and flame video alarming system
Other sensitivity and accuracy.
It is understood that in order to avoid failing to report, in the preferred scheme, the image information that Image Acquisition end 1 acquires
What can be continued is transferred to control centre and mobile terminal, manually to be checked, avoids the occurrence of and fails to report situation.
Embodiment two
It is in embodiment two based on the frame-embedded smog of deep learning and flame video alarm method referring to Fig. 2, Fig. 2
Flow chart.The frame-embedded smog of deep learning and flame video alarm method are based in the present embodiment, comprising the following steps:
Acquire the image information of ambient enviroment.It is carried out particular by image information of the Image Acquisition end 1 to ambient enviroment
Acquisition.
Judged according to deep learning intelligent algorithm containing smog or/and flame characteristic in image information, and is formed
Fire behavior information.The algorithm model of the deep learning intelligent algorithm in storage unit 21 is called particular by processing unit 21
Whether judge containing smog or/and flame characteristic in the image information of input, if being judged as YES, then by algorithm model
Output fire behavior information continues to judge the image information newly inputted if it has not, being then not processed.
Fire behavior alarm is carried out according to fire behavior information.Fire behavior information carries out fire behavior alarm, monitoring based on the received for control centre 3
Personnel carry out subsequent reply processing according to fire behavior alarm.
Upload fire behavior information.Fire behavior information is uploaded to Cloud Server 4.
Issue fire behavior information.Cloud Server 4 issues fire behavior information to mobile terminal 5 again, multiple users of mobile terminal 5 into
Row coordinated management controls fire, improves administration of the prevention and control timeliness and safety.
With continued reference to Fig. 3, Fig. 3 is to judge to contain in image information according to deep learning intelligent algorithm in embodiment two
There is the flow chart of smog and flame characteristic.In the present embodiment, image information is judged according to deep learning intelligent algorithm
In contain smog and flame characteristic, comprising the following steps:
The algorithm model of image information input deep learning intelligent algorithm.
The characteristics of image of doubtful smog or/and flame in image information is extracted, and forms object.
Detect the threshold values of object.
Judge whether the confidence level of object is greater than threshold values.If it has not, then re-entering image information to deep learning people
The algorithm model of work intelligent algorithm.If it has, then judging that the object in image information for smog or/and flame, and forms fire
Feelings information.
It, can be to tens of thousands of or hundreds of thousands of Zhang Hanyou cigarettes during forming the algorithm model of deep learning intelligent algorithm
Mist, flame and and the fire picture of pyrotechnics be trained, be primarily directed to color in picture, contrast, variation and
Characteristic point carries out recognition training.For example, the color contrast of front and back, the brightness pair of conflagration area and non-conflagration area occur for fire
Than the change in shape comparison of front and back, the Feature point recognition of smog and flame, such as the spike of flame or zigzag fashion occur for fire
Edge feature etc..In specific training, the color of smog and flame, contrast, change in the image after each fire generation
Change and characteristic point is as positive sample training set, and each image that smog and flame do not occur is as negative sample training set.According to
Above-mentioned training content carries out assignment to each image in training set, for example, each image in positive sample training set is assigned a value of
1, each image in negative sample training set is assigned a value of 0.Then by deep learning intelligent algorithm to every frame image with just
Sample training collection and the comparison of negative sample training set and assignment, all regions being extracted of every frame image respectively obtain a valve
Value, the range of threshold values is between 0-1, and the threshold values is closer to illustrating that a possibility that region is fire is bigger in 1, conversely, valve
Value closer in 0, illustrate the region be fire a possibility that it is smaller.Threshold values can be preset among algorithm model, for example, 0.8,
It can certainly be 0.9 or 0.7, the numerical value is higher, and the judgement of algorithm model will be sensitiveer, but more high also more is likely to occur
The case where failing to judge, it is preferred that threshold values is set as 0.8.
The characteristics of image of doubtful smog or/and flame in extracting image information, and in the step of forming object, mesh
Each region will appear according to each characteristics of image inside mark object.Later, algorithm model can first detect the threshold values of object,
Object threshold values herein is the threshold values for detecting object each region, and the threshold values of object each region can be identical
It is different.Then, judge whether the confidence level of object is greater than threshold values, specifically to object the inside, the confidence level in each region
Judgement is compared with the threshold values of corresponding region.Specifically, each object being detected has confidence level, this
Confidence level is the object that detects practical a possibility that being smog or/and flame, by the confidence level compared with threshold values, such as
Threshold value setting is 0.8, when confidence level be greater than 0.8 when, then judge the object in image information for smog or/and flame, and
Form fire behavior information.
Preferably, it forms fire behavior information and contains fire behavior Alert Level.Specifically according to the smog of object or/and flame
Pace of change between the area of the smog or/and flame of area and object forms the fire behavior containing fire behavior Alert Level
Information.For example, setting detects that the area of the object for smog or/and flame is level-one fire alarm for the first time；After the unit time,
It is expanded twice relative to the area for detecting the object for smog or/and flame for the first time, then becomes multiple alarm；If
After the unit time of half, four times are expanded relative to the area for detecting the object for smog or/and flame for the first time,
It is then three-level fire alarm.It is of course also possible to set rank for smog and flame respectively, or fire is carried out with other setting means
Alert grade setting so that the higher expression of rank that there is a situation where fire is more serious.Preferably, in specific application, can distinguish
3 ranks are set for smog, are set as 4 ranks for flame.
Preferably, when detecting the threshold values of object, object is labeled, and forms mark in image information
Frame；Fire behavior information further includes the image containing smog or/and flame characteristic callout box.When convenient for finally issuing fire behavior information, use
Family can obtain more more vivid judgement information.
To sum up, present invention, can monitor the fire abnormal conditions in environment in real time, and by fire abnormal conditions with
Intuitive image forms push monitoring center and into the mobile phone of user, and user can be allowed intuitively to see the practical feelings at scene
Condition realizes multi-party collaboration administration of the prevention and control, improves reliability to failing to report, reporting by mistake and fire condition is accurately handled.The depth of use
The reaction speed for improving main body system of the algorithm model of degree study intelligent algorithm judges sensitivity and accuracy of judgement
Degree.And can be transformed in existing video monitoring system, reduce hardware cost.
Upper is only embodiments of the present invention, is not intended to restrict the invention.To those skilled in the art,
The invention may be variously modified and varied.All any modifications made in spirit and principles of the present invention, equivalent replacement,
Improve etc., it should all be included within scope of the presently claimed invention.
Claims (10)
1. one kind is based on the frame-embedded smog of deep learning and flame video alarming system, which is characterized in that adopted including image
Collect end, embeded processor, control centre, Cloud Server and mobile terminal；Described image collection terminal and the embedded processing
Device connection, the embeded processor are connect with the control centre, and the control centre connect with the Cloud Server, described
Cloud Server is connect with the mobile terminal；The image information of described image collection terminal acquisition ambient enviroment is simultaneously transferred to described embedding
Enter formula processor；The embeded processor is judged to contain cigarette in described image information according to deep learning intelligent algorithm
Mist or/and flame characteristic, and fire behavior information is formed, then the embeded processor transmits the fire behavior information to the control
Center；The fire behavior information carries out fire behavior alarm based on the received for the control centre, and transmits the fire behavior information to described
Cloud Server；The Cloud Server transmits the fire behavior information to the mobile terminal.
2. according to claim 1 be based on the frame-embedded smog of deep learning and flame video alarming system, feature
It is, further includes video memory；Described image memory is connect with the embeded processor；Described image memory connects
It receives the fire behavior information and is stored.
3. according to claim 1 be based on the frame-embedded smog of deep learning and flame video alarming system, feature
It is, described image collection terminal is the web camera of video monitoring system.
4. according to claim 1 be based on the frame-embedded smog of deep learning and flame video alarming system, feature
It is, the embeded processor includes processing unit and storage unit；The algorithm of the deep learning intelligent algorithm
Model is stored in the storage unit, and the processing unit receives the described image information of described image collection terminal transmitting, and adjusts
The algorithm model of the deep learning intelligent algorithm stored with the storage unit carries out input calculating, the depth
Practise the fire behavior information that the algorithm model output described image information of intelligent algorithm contains smog or/and flame characteristic.
5. according to claim 4 be based on the frame-embedded smog of deep learning and flame video alarming system, feature
It is, the fire behavior information includes the information of the image containing smog or/and flame characteristic callout box and fire behavior Alert Level.
6. according to claim 1 be based on the frame-embedded smog of deep learning and flame video alarming system, feature
It is, the control centre includes control platform and warning end, and the control platform receives the fire behavior information, and according to described
Fire behavior information controls the warning end and carries out corresponding picture and text or/and audible alarm.
7. one kind is based on the frame-embedded smog of deep learning and flame video alarm method, which is characterized in that including following step
It is rapid:
Acquire the image information of ambient enviroment；
Judged according to deep learning intelligent algorithm containing smog or/and flame characteristic in described image information, and is formed
Fire behavior information；
Fire behavior alarm is carried out according to the fire behavior information；
Upload the fire behavior information；
Issue the fire behavior information.
8. according to claim 7 be based on the frame-embedded smog of deep learning and flame video alarm method, feature
It is, is judged to contain smog and flame characteristic in described image information according to deep learning intelligent algorithm, including following
Step:
The algorithm model of deep learning intelligent algorithm described in described image information input；
The characteristics of image of doubtful smog or/and flame in described image information is extracted, and forms object；
Detect the threshold values of object；
Judge whether the confidence level of the object is greater than threshold values；If it has not, then re-entering described image information to the depth
The algorithm model of degree study intelligent algorithm；If it has, then judging the object in described image information for smog or/and fire
Flame, and form fire behavior information.
9. according to claim 8 be based on the frame-embedded smog of deep learning and flame video alarm method, feature
It is, forms fire behavior information, comprising:
According to the area of the smog of the area and the object of the smog of the object or/and flame or/and flame it
Between pace of change, formed the fire behavior information containing fire behavior Alert Level.
10. according to claim 9 be based on the frame-embedded smog of deep learning and flame video alarm method, feature
It is, when detecting the threshold values of object, the object is labeled, and form callout box in described image information；
The fire behavior information further includes the image containing smog or/and flame characteristic callout box.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910916328.0A CN110517441A (en) | 2019-09-26 | 2019-09-26 | Based on the frame-embedded smog of deep learning and flame video alarming system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910916328.0A CN110517441A (en) | 2019-09-26 | 2019-09-26 | Based on the frame-embedded smog of deep learning and flame video alarming system and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110517441A true CN110517441A (en) | 2019-11-29 |
Family
ID=68633707
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910916328.0A Pending CN110517441A (en) | 2019-09-26 | 2019-09-26 | Based on the frame-embedded smog of deep learning and flame video alarming system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110517441A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112071016A (en) * | 2020-09-14 | 2020-12-11 | 广州市几米物联科技有限公司 | Fire monitoring method, device, equipment and storage medium |
| CN112466083A (en) * | 2020-10-15 | 2021-03-09 | 中船重工远舟（北京）科技有限公司 | Marine fire monitoring and alarming method and system |
| CN112735083A (en) * | 2021-01-19 | 2021-04-30 | 齐鲁工业大学 | Embedded gateway for flame detection by using YOLOv5 and OpenVINO and deployment method thereof |
| CN112767644A (en) * | 2020-12-31 | 2021-05-07 | 千方捷通科技股份有限公司 | Method and device for early warning of fire in highway tunnel based on video identification |
| CN112801148A (en) * | 2021-01-14 | 2021-05-14 | 西安电子科技大学 | Fire recognition and positioning system and method based on deep learning |
| CN113011833A (en) * | 2021-01-15 | 2021-06-22 | 广州穗能通能源科技有限责任公司 | Safety management method and device for construction site, computer equipment and storage medium |
| CN113450530A (en) * | 2021-06-03 | 2021-09-28 | 河北华电石家庄热电有限公司 | Safety early warning system based on intelligent video analysis algorithm |
| CN113877124A (en) * | 2021-11-15 | 2022-01-04 | 应急管理部天津消防研究所 | Intelligent control system for jet flow falling point of fire monitor |
| CN114519918A (en) * | 2022-03-01 | 2022-05-20 | 华南师范大学 | Very early thermal imaging fire alarm system in chemical, petrochemical and pharmaceutical industries |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN204143592U (en) * | 2014-11-04 | 2015-02-04 | 无锡北斗星通信息科技有限公司 | Based on the forest fire detection platform of satellite remote sensing images |
| CN108389359A (en) * | 2018-04-10 | 2018-08-10 | 中国矿业大学 | A kind of Urban Fires alarm method based on deep learning |
| CN109359858A (en) * | 2018-10-15 | 2019-02-19 | 温州洪启信息科技有限公司 | Intelligent campus security-protection management system Internet-based |
| KR20190103855A (en) * | 2018-02-28 | 2019-09-05 | 주식회사 비젼하이텍 | Apparatus and method of searching fire image based on imaging area of the ptz camera |
-
2019
- 2019-09-26 CN CN201910916328.0A patent/CN110517441A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN204143592U (en) * | 2014-11-04 | 2015-02-04 | 无锡北斗星通信息科技有限公司 | Based on the forest fire detection platform of satellite remote sensing images |
| KR20190103855A (en) * | 2018-02-28 | 2019-09-05 | 주식회사 비젼하이텍 | Apparatus and method of searching fire image based on imaging area of the ptz camera |
| CN108389359A (en) * | 2018-04-10 | 2018-08-10 | 中国矿业大学 | A kind of Urban Fires alarm method based on deep learning |
| CN109359858A (en) * | 2018-10-15 | 2019-02-19 | 温州洪启信息科技有限公司 | Intelligent campus security-protection management system Internet-based |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112071016A (en) * | 2020-09-14 | 2020-12-11 | 广州市几米物联科技有限公司 | Fire monitoring method, device, equipment and storage medium |
| CN112466083A (en) * | 2020-10-15 | 2021-03-09 | 中船重工远舟（北京）科技有限公司 | Marine fire monitoring and alarming method and system |
| CN112767644A (en) * | 2020-12-31 | 2021-05-07 | 千方捷通科技股份有限公司 | Method and device for early warning of fire in highway tunnel based on video identification |
| CN112767644B (en) * | 2020-12-31 | 2022-05-24 | 千方捷通科技股份有限公司 | Method and device for early warning fire in highway tunnel based on video identification |
| CN112801148A (en) * | 2021-01-14 | 2021-05-14 | 西安电子科技大学 | Fire recognition and positioning system and method based on deep learning |
| CN113011833A (en) * | 2021-01-15 | 2021-06-22 | 广州穗能通能源科技有限责任公司 | Safety management method and device for construction site, computer equipment and storage medium |
| CN112735083A (en) * | 2021-01-19 | 2021-04-30 | 齐鲁工业大学 | Embedded gateway for flame detection by using YOLOv5 and OpenVINO and deployment method thereof |
| CN113450530A (en) * | 2021-06-03 | 2021-09-28 | 河北华电石家庄热电有限公司 | Safety early warning system based on intelligent video analysis algorithm |
| CN113877124A (en) * | 2021-11-15 | 2022-01-04 | 应急管理部天津消防研究所 | Intelligent control system for jet flow falling point of fire monitor |
| CN114519918A (en) * | 2022-03-01 | 2022-05-20 | 华南师范大学 | Very early thermal imaging fire alarm system in chemical, petrochemical and pharmaceutical industries |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110517441A (en) | Based on the frame-embedded smog of deep learning and flame video alarming system and method | |
| CN101334924B (en) | Fire hazard probe system and its fire hazard detection method | |
| CN103150856B (en) | Fire flame video monitoring and early warning system | |
| CN204288415U (en) | Intelligent early-warning firefighting robot | |
| CN107911653A (en) | The module of intelligent video monitoring in institute, system, method and storage medium | |
| CN104867265A (en) | Camera apparatus, and fire detection alarm system and method | |
| KR20140127574A (en) | Fire detecting system using unmanned aerial vehicle for reducing of fire misinformation | |
| KR20070028813A (en) | Method and system for monitoring forest fire | |
| CN204706130U (en) | Camera head and fire detection warning system | |
| CN114724330A (en) | Implementation method of self-adaptive mode switching multi-channel video fire real-time alarm system | |
| CN112257554B (en) | Forest fire recognition method, system, program and storage medium based on multiple spectra | |
| KR20210097500A (en) | Apparatus for Monitoring Fire And System having the same | |
| CN212782246U (en) | Smoke and fire recognition system based on artificial intelligence | |
| CN204856852U (en) | Intelligent security system based on thing networking | |
| CN201091014Y (en) | Fire detecting device | |
| CN106817560A (en) | It is a kind of to distributed terminal display device real-time monitoring system and monitoring method | |
| JP5042177B2 (en) | Image sensor | |
| CN107688793A (en) | A kind of outside transformer substation fire automatic monitoring method for early warning | |
| CN209015366U (en) | A kind of remote supervision system for fire protection warning | |
| CN110853287A (en) | Flame real-time monitoring system and method based on Internet of things distributed architecture | |
| CN110930632B (en) | Early warning system based on artificial intelligence | |
| CN116434533A (en) | AI wisdom highway tunnel synthesizes monitoring platform based on 5G | |
| CN108898781A (en) | A kind of intelligent fire Internet of things system based on public cloud | |
| KR102333955B1 (en) | Method, system and application for indicating safe area and navigating safe path for socially vulnerable people | |
| CN207440956U (en) | Fire Actuality Monitoring system |
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: 20191129 |