CN111161505A - Method for carrying out fire early warning through video monitoring - Google Patents
Method for carrying out fire early warning through video monitoring Download PDFInfo
- Publication number
- CN111161505A CN111161505A CN201811326974.3A CN201811326974A CN111161505A CN 111161505 A CN111161505 A CN 111161505A CN 201811326974 A CN201811326974 A CN 201811326974A CN 111161505 A CN111161505 A CN 111161505A
- Authority
- CN
- China
- Prior art keywords
- smoke
- area
- region
- color
- extracted
- 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
- 238000012544 monitoring process Methods 0.000 title claims abstract description 14
- 238000000034 method Methods 0.000 title claims abstract description 13
- 239000000779 smoke Substances 0.000 claims abstract description 83
- 238000004458 analytical method Methods 0.000 claims abstract description 17
- 230000003068 static effect Effects 0.000 claims abstract description 13
- 239000000203 mixture Substances 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 description 14
- 230000033001 locomotion Effects 0.000 description 10
- 238000009792 diffusion process Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000002265 prevention Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 239000004576 sand Substances 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 239000006227 byproduct Substances 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 230000000391 smoking effect Effects 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
- 238000009423 ventilation Methods 0.000 description 1
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
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B29/00—Checking or monitoring of signalling or alarm systems; Prevention or correction of operating errors, e.g. preventing unauthorised operation
- G08B29/18—Prevention or correction of operating errors
- G08B29/185—Signal analysis techniques for reducing or preventing false alarms or for enhancing the reliability of the system
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
Abstract
The invention provides a method for fire early warning through video monitoring, which detects smoke by utilizing two characteristics of static characteristics and dynamic characteristics of the smoke, firstly performs static characteristic analysis, extracts a region similar to the smoke color from a video image, then performs dynamic characteristic analysis, detects the extracted region, judges whether the region has irregularity, diffusivity and translucency characteristics, and uses the region not conforming to the characteristics as a false smoke region, thereby excluding a non-smoke region similar to the smoke color.
Description
Technical Field
The invention relates to the field of video monitoring, in particular to a method for smoke detection through video monitoring.
Background
Fires can cause significant economic losses and loss of life and personal injury. In the public safety field, many fire detection technologies are used to avoid fires, and many of them are based on ion detection, temperature detection, relative humidity detection, air ventilation detection, and conventional ultraviolet and infrared fire detectors. Most objects can generate smoke before burning, and early warning and alarming of fire can be realized by smoke detection. Conventional smoke detectors require both placement near the fire site and fail to provide site specific information, such as: the location of the fire, the size of the fire, the rate of spread of the fire, etc. Also, if combustion by-products generated by other means (e.g. smoking) are detected by the smoke detector, a fire alarm is also generated, but this fire alarm is false. Thus, conventional smoke detectors suffer from unreliability.
Disclosure of Invention
The invention aims to provide a more reliable fire detection technology based on a video monitoring technology, which can fully utilize video monitoring data of key fire prevention areas, realize the functions of remote monitoring, intelligent monitoring, early warning and real-time alarming, ensure the large-scale fire prevention monitoring requirement in the public safety field and improve the monitoring intelligence level.
The technical scheme of the invention is as follows:
a method for carrying out fire early warning through video monitoring is characterized in that: the method comprises the steps of detecting smoke by utilizing two characteristics of static characteristics and dynamic characteristics of the smoke, firstly carrying out static characteristic analysis, extracting an area similar to the smoke color from a video image, then carrying out dynamic characteristic analysis, and detecting the extracted area, thereby excluding a non-smoke area similar to the smoke color.
The static characteristic analysis comprises the following specific steps: in RGB color space, creating a mixed Gaussian model, and initializing the Gaussian model
Where i represents the second Gaussian model, the first pixel value of the observed object is taken as the mean value mu, and a larger variance sigma is initialized2And a low weight ωiThen, matching the next pixel with the initialized Gaussian model one by one, and if the pixel is not matched with the existing Gaussian model, adding a new Gaussian model; matching the color of the current pixel point with the ith Gaussian distribution in turn from high to low according to the priority until a matched distribution is found; and judging whether the pixels in the image to be detected belong to the established Gaussian mixture model or not, thereby extracting the image area with the smoke color.
The dynamic characteristic analysis comprises the following specific steps: and detecting the extracted smoke color area, judging whether the extracted smoke color area has irregularity, diffusivity and translucency characteristics, and filtering the area which does not accord with the characteristics as a false smoke area.
Many video-based fire detection technologies detect and alert fires but in many practical situations smoke detection will alert fires sooner than flame detection, reducing economic losses and casualties. The invention provides a method for detecting smoke by using two characteristics, wherein the static characteristics are used for extracting an area similar to the color of the smoke from a video image, and the dynamic characteristics are used for detecting the extracted area, so that a non-smoke area similar to the color of the smoke can be eliminated, the false alarm rate is reduced, the reliability of early alarm can be improved, the fire prevention and control level in a public safety scene is improved, and a fire scene can be timely and effectively found and controlled. The invention utilizes the existing monitoring platform, does not need to increase hardware equipment, fully utilizes the existing resources and can realize rapid and large-scale deployment.
Detailed Description
The invention detects smoke by utilizing two characteristics of static characteristic and dynamic characteristic of smoke, firstly performs static characteristic analysis, extracts an area with similar color to the smoke from a video image, and then performs dynamic characteristic analysis to detect the extracted area, thereby excluding a non-smoke area with similar color to the smoke.
1. Static feature analysis
Color is one of the features in smoke image information, and a region having a smoke color in an image can be extracted by color detection. Smoke is reflective to light so the color of the surrounding environment has an effect on the smoke color. For example: at night, the flames become the primary light source and the smoke appears red. A mixture gaussian model is created in the RGB color space to detect smoke color pixels. The system can adapt to the change of environment and detect the smog more effectively like this.
1.1 establishment of Gaussian mixture model
Initializing Gaussian model
(i denotes the second Gaussian model), with the first pixel value of the observed object as the mean value μ, and initializing a larger variance σ2And a low weight ωi. And then matching the next pixel with the initialized Gaussian model one by one, and if the pixel is not matched with the existing Gaussian model, adding a new Gaussian model.
1.2 Gaussian distribution matching
And matching the color of the current pixel point with the ith Gaussian distribution until a matched distribution is found. The conditions that match the ith distribution are:
if the matched model exists, updating the mean value and the variance of the model:
a, β is learning rate, a is constant, β is calculated as:
the weight value updating formula is as follows:
ωi＝(1-a)ωi+aMi
Mifor the matching flags of the ith model, the distribution M of the matchingiOther models M1i0. The above equation shows that only the weight of the gaussian model matching X is increased, and the weights of other models are all decreased.
Updating the priority of Gaussian modelsBy weight ωiSum variance σiUpdating is carried out, and the formula is as follows:
pi＝ωi/σi
when the number of the established Gaussian models reaches the upper limit, if no Gaussian model matched with X is found, removing the Gaussian model with the minimum priority, introducing a new Gaussian model according to X, giving a smaller weight and a larger variance, and updating the weights of all the Gaussian models. Creating a Gaussian mixture model requires using a set of smoke sample images FC ═ c1,c2,…cnAnd (c) training. After the image area is established, whether the pixel in the image to be detected belongs to the established Gaussian mixture model can be judged, and therefore the image area with the smoke color is extracted.
2. Dynamic feature analysis
The smoke dynamics are formed by irregular diffusion, and in addition, the smoke presents a translucent characteristic while dynamically diffusing. It includes irregularities in shape and diffusivity, and the translucent nature blurs the edge information of the background. We combine the irregularity and translucency characteristics to detect smoke. And detecting the extracted fog color area, judging whether the area has irregularity, diffusivity and translucency characteristics, and filtering the area which does not accord with the characteristics as a false fog area. The detection of dynamic characteristics can greatly improve the accurate alarm rate.
2.1 irregularity of the Smoke
The shape of the smoke is constantly changing due to the influence of the airflow, so that measuring the shape of the smoke is difficult to realize. Therefore, a method of extracting the region from above with a perimeter ratio to an area above may be employed. When their ratio is greater than a threshold (STD), it is a smoke region, otherwise it is a non-smoke region. The formula is as follows:
SEP/STD≥STD
where SEP is the smoke region perimeter and STD is the smoke region area.
2.2 diffusion of Smoke
Due to the diffusion of the smoke, the smokeWill increase continuously. Therefore, we calculate the growth rate (Δ a) of the extraction region over timedi) The diffusibility of smoke was judged. In the digital image, the area (P) of the smoke can be represented by the number of pixels in the region, and the time interval can be represented by the number of frames in the interval. The formula is as follows:
in the formula piThe total number of pixels representing the possible smoke region in the ith image of the image sequence, (i + k) -i i.e. dt represents k frames of images,
i.e. represents deltaAdiThe rate of change of the number of pixels of the extracted possible smoke region between the i-th to i + k-th frames. Since the smoke region size is affected by the airflow, we use the calculation of the average growth rate over a period of time (10 frames). This can improve the detection accuracy. The formula is as follows:
in the formula
I.e. for n deltas AdiAnd (6) averaging. If the average growth rate
Above a threshold (STD), the region is a smoke region, otherwise it is a non-smoke region.
2.3 directionality of Smoke diffusion
The smoke has certain directivity in the diffusion process, and the smoke density is generally lower than the air density, and the movement direction of the smoke is generally from bottom to top, so the smoke floating direction is analyzed, and the smoke movement can be effectively distinguished from other movements. The motion direction of a rectangular block (obtained by expanding the boundary coordinates of a single motion region) is calculated by using a discrete motion direction search, and each direction angle corresponds to a direction code. The motion direction of the rectangular block is determined by searching two continuous frames of images in discrete motion directions with absolute error and minimum criteria.
In the formula Ik-1And IkFor gray-scale images of two successive frames, WsAnd WeStarting and ending values, H, respectively, of the width of the rectangular blocksAnd HeThe starting value and the ending value of the height of the rectangular block are respectively, i and j are respectively the horizontal component and the vertical component of the search template, the values of i and j are-1, 0 and 1, and the calculation formula of the directional coding DC is as follows:
DC＝3×j+i+5
the absolute error and the i, j value of the minimum direction in the discrete motion direction search are obtained through the two formulas, and the absolute error and the i, j value are converted into corresponding direction codes, namely the motion direction of the rectangular block is obtained. The fluttering characteristic of smoke from bottom to top can be easily obtained, and the value of the directional code DC is concentrated on 1,2 and 3.
2.4 translucency of the Smoke
The smoke has translucency. In a scene area blocked by smoke, the edge of the scene area becomes fuzzy and high-frequency information is slowly reduced; when the rigid object shelters the scene, the high-frequency information of the scene has a relatively obvious steep change. According to the characteristic, the interference of moving people and vehicles similar to the color of smoke can be effectively removed by utilizing wavelet analysis. The wavelet transform of an image is a time-to-scale (time-to-frequency) analysis method of a signal, which is a time-frequency localization analysis method that a window size is fixed and constant, but a shape thereof is changeable, and a time window and a frequency window are changeable, that is, a low frequency portion has a higher frequency resolution and a lower time resolution, and a high frequency portion has a higher time resolution and a lower frequency resolution. Each level of decomposition produces wavelet coefficients representing a relatively coarse (low frequency image) and a relatively fine (high frequency image) image, and an image may be decomposed into a low resolution image and a number of sub-images representing the details of the image with resolution going from low to high. The sub-images of different resolutions correspond to different frequencies, the useful signal usually appearing as a low frequency signal and the noise signal usually appearing as a high frequency signal. Therefore, the texture background features of the original target image can be extracted by decomposing the high-frequency signals of the image, and the target can be extracted in a low-frequency area.
The static characteristics and the dynamic characteristics are extracted to be used as feature vectors, judgment is made through an SVM (support vector machine) classifier, the smoke features appearing in the video are detected, and early warning and alarming functions are provided.
Claims (4)
1. A method for carrying out fire early warning through video monitoring is characterized in that: the method comprises the steps of detecting smoke by utilizing two characteristics of static characteristics and dynamic characteristics of the smoke, firstly carrying out static characteristic analysis, extracting an area similar to the smoke color from a video image, then carrying out dynamic characteristic analysis, and detecting the extracted area, thereby excluding a non-smoke area similar to the smoke color.
2. The method for fire early warning through video surveillance according to claim 1, wherein: the static characteristic analysis comprises the following specific steps: in RGB color space, creating a mixed Gaussian model, and initializing the Gaussian model
Where i represents the second Gaussian model, the first pixel value of the observed object is taken as the mean value mu, and a larger variance sigma is initialized2And a low weight ωiThen, matching the next pixel with the initialized Gaussian model one by one, and if the pixel is not matched with the existing Gaussian model, adding a new Gaussian model; the color of the current pixel point and its existing ith Gaussian distribution are calculatedMatching the priority levels from high to low in sequence until a matched distribution is found; and judging whether the pixels in the image to be detected belong to the established Gaussian mixture model or not, thereby extracting the image area with the smoke color.
3. The method for fire early warning through video surveillance according to claim 1, wherein: the dynamic characteristic analysis comprises the following specific steps: and detecting the extracted smoke color area, judging whether the extracted smoke color area has irregularity, diffusivity and translucency characteristics, and filtering the area which does not accord with the characteristics as a false smoke area.
4. A method of fire early warning through video surveillance as claimed in claim 3, wherein: the irregularity characteristics are determined by comparing the perimeter of the extracted region with the upper area, and when the ratio is greater than a set threshold value, the extracted region is a smoke region, otherwise, the extracted region is a non-smoke region, and the formula is as follows:
SEP/STD≥STD；
in the above formula, SEP is the perimeter of the smoke region, and STD is the area of the smoke region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811326974.3A CN111161505A (en) | 2018-11-08 | 2018-11-08 | Method for carrying out fire early warning through video monitoring |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811326974.3A CN111161505A (en) | 2018-11-08 | 2018-11-08 | Method for carrying out fire early warning through video monitoring |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111161505A true CN111161505A (en) | 2020-05-15 |
Family
ID=70555126
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811326974.3A Pending CN111161505A (en) | 2018-11-08 | 2018-11-08 | Method for carrying out fire early warning through video monitoring |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111161505A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111882807A (en) * | 2020-06-22 | 2020-11-03 | 杭州后博科技有限公司 | Method and system for identifying regional fire occurrence area |
| CN113378629A (en) * | 2021-04-27 | 2021-09-10 | 阿里云计算有限公司 | Method and device for detecting abnormal vehicle in smoke discharge |
| CN114792459A (en) * | 2021-01-25 | 2022-07-26 | 杭州申弘智能科技有限公司 | Remote fire monitoring management system and smoke detection method |
-
2018
- 2018-11-08 CN CN201811326974.3A patent/CN111161505A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111882807A (en) * | 2020-06-22 | 2020-11-03 | 杭州后博科技有限公司 | Method and system for identifying regional fire occurrence area |
| CN111882807B (en) * | 2020-06-22 | 2022-03-15 | 杭州后博科技有限公司 | Method and system for identifying regional fire occurrence area |
| CN114792459A (en) * | 2021-01-25 | 2022-07-26 | 杭州申弘智能科技有限公司 | Remote fire monitoring management system and smoke detection method |
| CN113378629A (en) * | 2021-04-27 | 2021-09-10 | 阿里云计算有限公司 | Method and device for detecting abnormal vehicle in smoke discharge |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11468660B2 (en) | Pixel-level based micro-feature extraction | |
| Habiboglu et al. | Real-time wildfire detection using correlation descriptors | |
| KR101168760B1 (en) | Flame detecting method and device | |
| Ko et al. | Fire detection based on vision sensor and support vector machines | |
| JP5325899B2 (en) | Intrusion alarm video processor | |
| Zhou et al. | Wildfire smoke detection based on local extremal region segmentation and surveillance | |
| US7868772B2 (en) | Flame detecting method and device | |
| US9047515B2 (en) | Method and system for wildfire detection using a visible range camera | |
| CN103473788A (en) | Indoor fire and flame detection method based on high-definition video images | |
| CN111161505A (en) | Method for carrying out fire early warning through video monitoring | |
| CN109214392A (en) | A method of Smoke Detection is carried out by video monitoring | |
| CN107437318B (en) | Visible light intelligent recognition algorithm | |
| CN107301375B (en) | Video image smoke detection method based on dense optical flow | |
| CN106127148A (en) | A kind of escalator passenger's unusual checking algorithm based on machine vision | |
| CN108629935B (en) | Method and system for detecting burglary of climbing stairs and turning windows based on video monitoring | |
| US8665329B2 (en) | Apparatus for automatically ignoring cast self shadows to increase the effectiveness of video analytics based surveillance systems | |
| CN110874592A (en) | Forest fire smoke image detection method based on total bounded variation | |
| EP2000998B1 (en) | Flame detecting method and device | |
| CN109086682A (en) | A kind of intelligent video black smoke vehicle detection method based on multi-feature fusion | |
| CN115311604A (en) | Fire fighting method based on Internet of things | |
| Cheng et al. | A multiscale parametric background model for stationary foreground object detection | |
| CN109754411A (en) | Building pivot frame larceny detection method and system are climbed based on optical flow method target following | |
| Baldini et al. | A simple and robust method for moving target tracking | |
| CN108520200B (en) | A kind of effective coal-mine fire detection system | |
| KR101300130B1 (en) | System and method for detecting smoke using surveillance camera |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200515 |