CN108389359B - Deep learning-based urban fire alarm method - Google Patents
Deep learning-based urban fire alarm method Download PDFInfo
- Publication number
- CN108389359B CN108389359B CN201810315652.2A CN201810315652A CN108389359B CN 108389359 B CN108389359 B CN 108389359B CN 201810315652 A CN201810315652 A CN 201810315652A CN 108389359 B CN108389359 B CN 108389359B
- Authority
- CN
- China
- Prior art keywords
- fire
- deep learning
- image
- cloud server
- training set
- 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.)
- Expired - Fee Related
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 53
- 238000000034 method Methods 0.000 title claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 38
- 238000012544 monitoring process Methods 0.000 claims abstract description 15
- 238000011161 development Methods 0.000 claims description 12
- 230000018109 developmental process Effects 0.000 claims description 12
- 238000012790 confirmation Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000011835 investigation Methods 0.000 abstract description 3
- 238000001514 detection method Methods 0.000 description 9
- 239000000779 smoke Substances 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 230000026676 system process Effects 0.000 description 1
Images
Classifications
-
- 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
-
- 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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The invention discloses a city fire alarm method based on deep learning, which is characterized in that a monitoring camera, a cloud server, an automatic alarm module and a monitoring center are combined, a deep learning network is firstly used for training fire images and non-fire images, then the images shot in real time are transmitted to the cloud server through the monitoring camera, the cloud server determines whether the shot images are suspected fire images through the deep learning network, if the shot images are suspected fire images, the cloud server controls the automatic alarm module to perform early warning prompt, meanwhile, the cloud server transmits the suspected fire images to the monitoring center to be displayed, at the moment, a fire fighter only needs to judge the suspected fire images, the fire dangerous cases can be subjected to online investigation, and the alarm can be given out in time after the fire occurs.
Description
Technical Field
The invention relates to an urban fire alarm method, in particular to an urban fire alarm method based on deep learning.
Background
Target detection based on deep learning refers to the application of deep learning in computer vision. Unlike traditional visual target detection, which only marks positions, target detection based on deep learning not only predicts the positions of objects, but also classifies the objects. The traditional visual detection algorithm only has a good detection effect on face and pedestrian recognition at present, but has poor robustness, so that the target detection effect is poor in a dark or complex environment. The target detection based on deep learning has strong robustness, and meanwhile, the positioning and classification problems can be combined, and multiple targets can be detected and classified simultaneously. Object detection using deep learning may be more efficient and reliable.
The traditional urban fire alarm system generally comprises a fire detector, an area alarm and a centralized alarm; and the fire extinguishing system can also be linked with various fire extinguishing facilities and communication devices according to the requirements of engineering so as to form a central control system. Namely, a complete fire control system is formed by automatic alarm, automatic fire extinguishing, safe evacuation induction, system process display, fire control archive management and the like. A fire detector is an apparatus for detecting a fire since smoke, high temperature and fire light are generated along with the occurrence of the fire at a stage of the fire. The smoke, heat and light can be converted into electric signals by a detector to alarm or an automatic fire extinguishing system is started to extinguish the fire in time. The regional alarm can convert the signal sent by the detector of the floor into sound and light alarm, and can monitor the output signal of the centralized alarm of a plurality of floors or control an automatic fire extinguishing system.
At present, most of urban fire alarm systems are arranged in rooms or large indoor public places, most of detection equipment of the urban fire alarm systems are smoke alarms, the alarms are frequently triggered by mistake, and if people smoke indoors, the fire alarm systems are triggered by mistake. After the fire alarm, people can troubleshoot dangerous situations, and the inaccurate alarm can lead the fire fighters to do nothing. In addition, if a fire disaster occurs outdoors, an indoor fire detector is not suitable, and smoke generated by the fire disaster and influenced by external airflow can be dispersed in a very short time when the fire disaster occurs outdoors, so that the smoke detector cannot acquire smoke information, and further, the fire disaster occurring outdoors cannot be warned; in addition, although the conditions of each region can be observed by each monitoring camera arranged in the city, the condition that the regions are different from each other cannot be monitored in real time by manpower, so that the fire disaster which occurs outdoors cannot be early warned in time.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a deep learning-based urban fire alarm method, which adopts a deep learning network, images are shot through a monitoring camera, and are only displayed through a monitoring center after being analyzed and processed by a cloud server, so that fire fighters can perform online investigation on the dangerous case of fire only by judging the images of suspected fire, and can timely give an alarm after the occurrence of the fire is determined.
In order to achieve the purpose, the invention adopts the technical scheme that: a city fire alarm method based on deep learning adopts an alarm system which comprises a monitoring camera, a cloud server, an automatic alarm module and a monitoring center, wherein the monitoring camera is connected with the cloud server, the cloud server is respectively connected with the automatic alarm module and the monitoring center, and the method specifically comprises the following steps:
A. building a deep learning network in a cloud server;
B. shooting or collecting fire scene pictures of various different conditions, and storing the obtained fire scene pictures into a cloud server;
C. intercepting flame images in each fire scene graph in a cloud server as a positive sample training set of a deep learning network, and storing the flame images; intercepting other bright spots or highlight objects in each fire scene graph to serve as a negative sample training set of the deep learning network, and storing the negative sample training set;
D. respectively reading a positive sample training set and a negative sample training set by the deep learning network, assigning a value of 1 to each image in the positive sample training set and assigning a value of 0 to each image in the negative sample training set, and determining a classification network model of the deep learning network after training;
E. an image shot by a monitoring camera in real time is transmitted to a cloud server, the cloud server converts an RGB color image into a gray image for each frame of shot image, then binaryzation is carried out after threshold segmentation of the gray image, and a target contour is extracted from a black and white image after binaryzation;
F. according to the shape of the extracted target contour, on the premise that the central point is not changed, the width and the height of the target contour are doubled, the coordinates of the expanded area are recorded, the area is extracted from the corresponding RGB color image and is sent to a deep learning network for processing after training;
G. the deep learning network compares each frame of image with the positive sample training set and the negative sample training set and assigns values, all the extracted regions of each frame of image respectively obtain a value α, and the value is more than or equal to 0 and less than or equal to α and less than or equal to 1, the closer the value of α is to 1, the higher the possibility that the region is a fire scene is, the closer the value of α is to 0, the lower the possibility that the region is a fire scene is;
H. and setting a threshold b, comparing the α value of each frame of image with the threshold b by the cloud server, determining that the image is a suspected fire area image when α is more than or equal to b, controlling an automatic alarm module by the cloud server to perform early warning, transmitting the suspected fire area image to a monitoring center through the cloud server to be displayed, and determining subsequent treatment measures after a firefighter observes the image for further confirmation.
Further, the shot or collected fire scene picture is a flame picture when a fire just occurs.
Further, the specific process of building the deep learning network is as follows: the method comprises the steps of installing python or a third-party library of python in a cloud server as a deep learning development language, installing a toolkit used by OpenCV in python, installing a deep learning development framework of either cache, tenserflow or pyrrch, and installing pycharm as an IDE development tool of python.
Further, in the step F, OpenCV is used as a computer vision aided development tool to extract a highlight area of each frame of image.
Further, the value range of the threshold b is 0.6-0.9.
Compared with the prior art, the method adopts a mode of combining the monitoring camera, the cloud server, the automatic alarm module and the monitoring center, firstly trains a fire image and a non-fire image on the deep learning network, then transmits the image shot in real time to the cloud server through the monitoring camera, the cloud server determines whether the shot image is a suspected fire image or not through the deep learning network, if the shot image is determined to be the suspected fire image, the shot image is transmitted to the monitoring center through the cloud server to be displayed, at the moment, a fire fighter can perform online investigation on the fire dangerous situation only by judging the suspected fire image, and can timely warn after the fire occurs; due to the high-precision classification of the deep learning network, a large number of non-fire scene graphs can be eliminated, and the workload of fire fighters is reduced conveniently.
Drawings
FIG. 1 is an overall frame diagram of the present invention;
fig. 2 is a flow chart of the operation of the present invention.
Detailed Description
The present invention will be further explained below.
As shown in fig. 1 to 2, the alarm system adopted in the present invention includes a monitoring camera, a cloud server, an automatic alarm module, and a monitoring center, the monitoring camera is connected to the cloud server, the cloud server is connected to the automatic alarm module and the monitoring center, respectively, and the method specifically includes the steps of:
A. building a deep learning network in a cloud server;
B. shooting or collecting fire scene pictures of various different conditions, and storing the obtained fire scene pictures into a cloud server;
C. intercepting flame images in each fire scene graph in a cloud server as a positive sample training set of a deep learning network, and storing the flame images; intercepting other bright spots or highlight objects in each fire scene graph to serve as a negative sample training set of the deep learning network, and storing the negative sample training set;
D. respectively reading a positive sample training set and a negative sample training set by the deep learning network, assigning a value of 1 to each image in the positive sample training set and assigning a value of 0 to each image in the negative sample training set, and determining a classification network model of the deep learning network after training;
E. the method comprises the steps that an image shot by a monitoring camera in real time is transmitted to a cloud server, because the brightness of flame is very high, the cloud server converts an RGB color image into a gray image for each frame of shot image through a computer vision technology, then binarizes the gray image after threshold segmentation, and extracts a target contour for a binarized black and white image;
F. according to the shape of the extracted target contour, on the premise that the central point is not changed, the width and the height of the target contour are doubled, the coordinates of the expanded area are recorded, the area is extracted from the corresponding RGB color image and is sent to a deep learning network for processing after training;
G. according to the classification result of the classification network, the convolutional neural network used in deep learning is considered to be a complex calculation formula, the region extracted from the picture in the step F is used as input, an output value α is obtained through calculation, thousands of parameters exist in the formula, the parameters are randomly assigned before network training, in the process of iterating n times of training, the values of the parameters are continuously adjusted, the adjusted parameters enable the result α of training a positive sample through network calculation in the next training to tend to 1, the result α of calculating a negative sample tends to 0, the deep learning network compares each frame of image with the training set of the positive sample and the training set of the negative sample and assigns values, all the extracted regions of each frame of image respectively obtain a value α, wherein the value is more than or equal to 0 and less than or equal to α and less than or equal to 1, the closer to 1 of α, the probability that the region is the fire scene is greater, and the closer to 0 of α, the probability that the region is the fire scene is smaller;
H. and setting a threshold b, comparing the α value of each frame of image with the threshold b by the cloud server, determining that the image is a suspected fire area image when α is more than or equal to b, controlling an automatic alarm module by the cloud server to perform early warning, transmitting the suspected fire area image to a monitoring center through the cloud server to be displayed, and determining subsequent treatment measures after a firefighter observes the image for further confirmation.
Further, the shot or collected fire scene picture is a flame picture when a fire just occurs.
Further, the specific process of building the deep learning network is as follows: the method comprises the steps of installing python or a third-party library of python in a cloud server as a deep learning development language, installing a toolkit used by OpenCV in python, installing a deep learning development framework of either cache, tenserflow or pyrrch, and installing pycharm as an IDE development tool of python.
Further, in the step F, OpenCV is used as a computer vision aided development tool to extract a highlight area of each frame of image.
Further, the value range of the threshold b is 0.6-0.9.
Claims (5)
1. A city fire alarm method based on deep learning adopts an alarm system which comprises a monitoring camera, a cloud server, an automatic alarm module and a monitoring center, wherein the monitoring camera is connected with the cloud server, and the cloud server is respectively connected with the automatic alarm module and the monitoring center, and is characterized in that the method specifically comprises the following steps:
A. building a deep learning network in a cloud server;
B. shooting or collecting fire scene pictures of various different conditions, and storing the obtained fire scene pictures into a cloud server;
C. intercepting flame images in each fire scene graph in a cloud server as a positive sample training set of a deep learning network, and storing the flame images; intercepting other bright spots or highlight objects in each fire scene graph to serve as a negative sample training set of the deep learning network, and storing the negative sample training set;
D. respectively reading a positive sample training set and a negative sample training set by the deep learning network, assigning a value of 1 to each image in the positive sample training set and assigning a value of 0 to each image in the negative sample training set, and determining a classification network model of the deep learning network after training;
E. an image shot by a monitoring camera in real time is transmitted to a cloud server, the cloud server converts an RGB color image into a gray image for each frame of shot image, then binaryzation is carried out after threshold segmentation of the gray image, and a target contour is extracted from a black and white image after binaryzation;
F. according to the shape of the extracted target contour, on the premise that the central point is not changed, the width and the height of the target contour are doubled, the coordinates of the expanded area are recorded, the area is extracted from the corresponding RGB color image and is sent to a deep learning network for processing after training;
G. the deep learning network compares each frame of image with the positive sample training set and the negative sample training set and assigns values, all the extracted regions of each frame of image respectively obtain a value α, and the value is more than or equal to 0 and less than or equal to α and less than or equal to 1, the closer the value of α is to 1, the higher the possibility that the region is a fire scene is, the closer the value of α is to 0, the lower the possibility that the region is a fire scene is;
H. and setting a threshold b, comparing the α value of each frame of image with the threshold b by the cloud server, determining that the image is a suspected fire area image when α is more than or equal to b, controlling an automatic alarm module by the cloud server to perform early warning, transmitting the suspected fire area image to a monitoring center through the cloud server to be displayed, and determining subsequent treatment measures after a firefighter observes the image for further confirmation.
2. The deep learning-based urban fire alarm method according to claim 1, wherein the shot or collected fire scene picture is a picture of the fire just before the fire.
3. The deep learning-based urban fire alarm method according to claim 1, wherein the specific process of building the deep learning network is as follows: the method comprises the steps of installing python or a third-party library of python in a cloud server as a deep learning development language, installing a toolkit used by OpenCV in python, installing a deep learning development framework of either cache, tenserflow or pyrrch, and installing pycharm as an IDE development tool of python.
4. The deep learning-based urban fire alarm method according to claim 1, wherein OpenCV is used as a computer vision aided development tool to extract a highlight region of each frame of image in step F.
5. The deep learning-based urban fire alarm method according to claim 1, wherein the threshold value b ranges from 0.6 to 0.9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810315652.2A CN108389359B (en) | 2018-04-10 | 2018-04-10 | Deep learning-based urban fire alarm method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810315652.2A CN108389359B (en) | 2018-04-10 | 2018-04-10 | Deep learning-based urban fire alarm method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108389359A CN108389359A (en) | 2018-08-10 |
| CN108389359B true CN108389359B (en) | 2020-03-24 |
Family
ID=63073772
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810315652.2A Expired - Fee Related CN108389359B (en) | 2018-04-10 | 2018-04-10 | Deep learning-based urban fire alarm method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108389359B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109167963A (en) * | 2018-09-28 | 2019-01-08 | 广东马上到网络科技有限公司 | A kind of monitoring method and system of public civilization |
| CN109410497B (en) * | 2018-11-20 | 2021-01-19 | 江苏理工学院 | Bridge opening space safety monitoring and alarming system based on deep learning |
| CN109543631A (en) * | 2018-11-28 | 2019-03-29 | 公安部沈阳消防研究所 | A kind of fire image detection alarm method based on machine learning |
| CN110007666A (en) * | 2019-04-16 | 2019-07-12 | 山东大学 | A kind of intelligent patrol detection fire early-warning system and its working method based on fish-eye camera trolley |
| CN110314314A (en) * | 2019-05-27 | 2019-10-11 | 安徽中科中涣防务装备技术有限公司 | A kind of gas station's Intelligent preventive control apparatus and system |
| CN110517441A (en) * | 2019-09-26 | 2019-11-29 | 华南师范大学 | Based on the frame-embedded smog of deep learning and flame video alarming system and method |
| CN112509270A (en) * | 2020-11-19 | 2021-03-16 | 北京城市轨道交通咨询有限公司 | Fire monitoring linkage method, device and system for train compartment |
| CN113570817A (en) * | 2021-08-05 | 2021-10-29 | 广东电网有限责任公司 | Fire safety alarm method and device, computer equipment and storage medium |
| CN117095506B (en) * | 2023-10-18 | 2023-12-15 | 潍坊市平安消防工程有限公司 | Fire safety monitoring system and method based on alarm area model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103630235A (en) * | 2013-11-22 | 2014-03-12 | 中国地质大学（武汉） | ICP (Inductively Coupled Plasm) light source flame state detection device and detection method |
| CN104834920A (en) * | 2015-05-25 | 2015-08-12 | 成都通甲优博科技有限责任公司 | Intelligent forest fire recognition method and device based on multispectral image of unmanned plane |
| CN107230329A (en) * | 2017-06-15 | 2017-10-03 | 深圳市泰和安科技有限公司 | A kind of scaling method of fire detector, device and equipment |
| CN107862333A (en) * | 2017-11-06 | 2018-03-30 | 哈尔滨工程大学 | A kind of method of the judgment object combustion zone under complex environment |
-
2018
- 2018-04-10 CN CN201810315652.2A patent/CN108389359B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103630235A (en) * | 2013-11-22 | 2014-03-12 | 中国地质大学（武汉） | ICP (Inductively Coupled Plasm) light source flame state detection device and detection method |
| CN104834920A (en) * | 2015-05-25 | 2015-08-12 | 成都通甲优博科技有限责任公司 | Intelligent forest fire recognition method and device based on multispectral image of unmanned plane |
| CN107230329A (en) * | 2017-06-15 | 2017-10-03 | 深圳市泰和安科技有限公司 | A kind of scaling method of fire detector, device and equipment |
| CN107862333A (en) * | 2017-11-06 | 2018-03-30 | 哈尔滨工程大学 | A kind of method of the judgment object combustion zone under complex environment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108389359A (en) | 2018-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108389359B (en) | Deep learning-based urban fire alarm method | |
| US20210049885A1 (en) | Method and system for monitoring fire | |
| CN106600888B (en) | Automatic forest fire detection method and system | |
| US7859419B2 (en) | Smoke detecting method and device | |
| CN112069975A (en) | Comprehensive flame detection method based on ultraviolet, infrared and vision | |
| CN107437318B (en) | Visible light intelligent recognition algorithm | |
| CN109637068A (en) | Intelligent pyrotechnics identifying system | |
| CN111488799B (en) | Falling object identification method and system based on image identification | |
| CN111523397B (en) | Intelligent lamp post visual identification device, method and system and electronic equipment thereof | |
| KR102407327B1 (en) | Apparatus for Monitoring Fire And System having the same | |
| CN104463253B (en) | Passageway for fire apparatus safety detection method based on adaptive background study | |
| CN104599427A (en) | Intelligent image type fire alarming system for highway tunnel | |
| CN112785809B (en) | Fire re-ignition prediction method and system based on AI image recognition | |
| Wong et al. | Experimental study of video fire detection and its applications | |
| CN111899452A (en) | Forest fire prevention early warning system based on edge calculation | |
| CN107688793A (en) | A kind of outside transformer substation fire automatic monitoring method for early warning | |
| EP2000952A2 (en) | Smoke detecting method and device | |
| CN109034038B (en) | Fire identification device based on multi-feature fusion | |
| CN105574468A (en) | Video flame detection method, device and system | |
| CN110853287A (en) | Flame real-time monitoring system and method based on Internet of things distributed architecture | |
| CN114120171A (en) | Fire smoke detection method, device and equipment based on video frame and storage medium | |
| CN107729811B (en) | Night flame detection method based on scene modeling | |
| US9997038B2 (en) | Smoke detection apparatus, method for detecting smoke and computer program | |
| CN113553922A (en) | Mask wearing state detection method based on improved convolutional neural network | |
| CN108765833A (en) | Based on the incipient fire detection algorithm for improving mixed Gaussian and machine learning |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200324 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |