CN106778488B - Low-light (level) smog video detecting method based on image correlation - Google Patents
Low-light (level) smog video detecting method based on image correlation Download PDFInfo
- Publication number
- CN106778488B CN106778488B CN201611032108.4A CN201611032108A CN106778488B CN 106778488 B CN106778488 B CN 106778488B CN 201611032108 A CN201611032108 A CN 201611032108A CN 106778488 B CN106778488 B CN 106778488B
- Authority
- CN
- China
- Prior art keywords
- image
- region
- frame
- smog
- correlation
- 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
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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides the low-light (level) smog video detecting method based on image correlation, first the video image of typing is compressed and is converted into gray level image, and the moving target in four frame difference operations extraction video image is carried out to resulting gray level image, suitable threshold value, which is chosen, by the obtained result of inter-frame difference is converted into bianry image；Then median filtering is carried out to resulting bianry image, tentatively reduction picture noise；By successively calculating the frame-to-frame correlation of each connected region in bianry image, the small non-smog movement region of image correlation is rejected, accurate smoke region image is obtained.Effect of the present invention is can to can be visually seen smog development, to exclude the influence of the factors such as temperature, humidity, air pressure, smell, improves the accuracy of Smoke Detection, conservative estimation false alarm rate can maintain within 10%.Meanwhile cigarette source position can be accurately positioned by the smoke region development process that playback is extracted.
Description
Technical field
The present invention relates to Smoke Detection fields, more particularly to the camera of included infrared auxiliary light source is utilized under low-light (level)
The method for carrying out video smoke detection.
Background technique
Fire hazard aerosol fog detection technique based on video has intuitive visual, fast and reliable advantage, can meet the morning of fire
Phase detects demand, becomes the research hotspot in fire detection field.Researchers at home and abroad propose many effective cigarettes for many years
Mist detection method, but these methods are primarily directed to the condition of illumination abundance, for the ring of the low-light (level)s such as confined space or night
Smog detection method research under border is not enough, but practical application is again highly desirable.Such as the common cargo hold smog of current aircraft
Detection system uses light scattering type smoke detector, and such detector is vulnerable to factors such as humidity, dust, overpowering odor, elaioleucites
It influences, false alarm rate is higher.When receiving cargo hold smoke alarm, unit lacks the true and false of effective means identification verifying alarm, often leads
It causes aircraft to make a return voyage, make preparation for dropping, increases operation cost, or even threaten flight safety.How the cargo hold smoke detection system of video is utilized
The supplement or alarm verifying means united as conventional airplane cargo hold smoke detection system, utilize effective smog video detecting method
It is urgently researched and developed to improve flight safety, reduce false alarm rate.
Summary of the invention
The present invention provides a kind of low-light (level) smog video detecting method based on image correlation, to be regarded for low-light (level)
Smoke Detection under frequency monitoring, realizes fire alarm.
To achieve the goals above, the present invention provides the image processing method in a kind of smog video detection, smog view
Detection method includes the following steps for frequency:
Target video image is compressed and is converted into grayscale image by step 1；
Step 2 carries out four frame difference operations to gray level image, extracts more complete moving region in video image；
Step 3 removes the noise spot in bianry image with median filtering；
Step 4, the frame-to-frame correlation setting dynamic threshold based on each connected region reject non-smog connected region.
It is comprised the steps of: in step 4 described in it
Step 4.1 marks each connected region in each frame image with the bwlabel function in MATLAB:
The label that connected region is carried out to the image after the denoising of step 3 median filtering judges of connected region in image
Several and size.
Step 4.2, the correlation size for calculating each connected region:
Since the size for calculating each connected domain with the correlation of corresponding region in previous frame image the second frame image.
The small connected region of step 4.3, removal correlation:
Given threshold Q, if the frame-to-frame correlation CR of connected regionK(i) it is less than threshold value Q, the then connected region that will be marked
Domain is rejected, on the contrary then retain.
Effect of the invention is can quickly to be detected the smoke region in low-light (level) video using the detection method, be passed through
The small connected region of correlation is rejected, the noise that low-light (level) generates is reduced, is patrolled suitable for low-light (level)s such as aircraft holds, without personnel
Depending on, the environment shorter to the response time requirement of smog.Conventional detection devices are vulnerable to the factors shadow such as humidity, dust, overpowering odor
It rings and causes false alarm rate high, this method can quickly, effectively extract the smoke region in video on the basis of video monitoring,
It is tested through experiment, the time that this method handles 1 frame video image is about 0.4s, can meet application requirement.Relative to traditional cigarette
Fog detector can be visually seen smog development, to exclude the influence of the factors such as temperature, humidity, air pressure, smell, improve cigarette
The accuracy of mist detection, conservative estimation false alarm rate can maintain within 10%.Meanwhile it being sent out by the smoke region that playback is extracted
Cigarette source position can be accurately positioned in exhibition process.Corner for such as large cargo hold, basement, garage, resident's corridor
Deng accuracy and real-time with higher provide technical support for the fire safety under low light environment.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention；
Fig. 2 a1 to Fig. 2 j1 and Fig. 2 a2 to Fig. 2 j2 is the smog movement region obtained based on the method for the present invention and unused figure
As correlation rejects the comparison diagram of the image in non-smog movement region.
Specific embodiment
The low-light (level) smog video detecting method of the invention based on image correlation is subject in detail below in conjunction with attached drawing
Explanation.
Low-light (level) smog video detecting method based on image correlation of the invention, be particularly suitable for carry it is infrared auxiliary
Help the Smoke Detection under the camera head monitor of light source.The detailed process of its Smoke Detection is roughly divided into following five steps:
The target video image of typing is compressed and is converted into grayscale image by step 1:
The smog video image size of this camera typing is 576*720 pixel, in order to reduce the time used in entire algorithm
To increase the real-time of Smoke Detection, image is compressed in MATLAB software first, MATLAB software is the U.S.
The business mathematics software that Mathworks company produces, based on the exploitation of algorithm, data visualization, data analysis and numerical value
Calculate advanced techniques computational language and interactive environment.Image is carried out 0.5 times of compression by this experiment, and the size of image is made to become 288*
360 pixels, the compression method used is closest interpolation method.Closest interpolation method is that one of image procossing is most basic, most
Simple image zooming method, is realized using imresize function in matlab.In order to subsequent image processing convenience then
Image is subjected to gray processing.
Step 2 carries out four frame difference operations to gray level image, extracts more complete moving region in video image:
The step 2 the following steps are included:
Step 2.1, the continuous four frame images read in video sequence, carry out calculus of differences:
Inter-frame difference is more typical a kind of algorithm in moving object detection algorithm, and principle is to adjacent two frame or more
Frame carries out difference, obtains the absolute value of two field pictures pixel value, it is compared with detection threshold value, so that it is determined that video sequence
In whether have moving object appearance.Video frame selected by inter-frame difference is four frame images of continuous adjacent in this experiment, with k-th frame
For, successively k-th frame, K+1 frame, K+2 frame, K+3 frame is selected to do calculus of differences.Image before note calculus of differences is followed successively by IK、
IK+1、IK+2、IK+3, the image after calculus of differences is DK、DK+1, difference rule it is as follows:
I.e. by continuous four frame first frame and third frame do difference；Second frame and the 4th frame do difference, avoid
Because two continuous frames interval time is too short to extract moving region the problem of.
The grayscale image binaryzation that step 2.2, the suitable threshold value of selection obtain calculus of differences；
Need to meet can be (preceding by moving region for selected threshold value when image binaryzation that above-mentioned inter-frame difference is obtained
Scape) and background more clearly separate, the region greater than threshold value T is considered as prospect, i.e. smog diffusion zone；Less than threshold value T's
Region is considered as background.The selection of threshold value need to be depending on practical smog movement speed situation, according to difference in real process
The mean value of the gray value of image is adjusted after operation.The threshold value T=Im+0.4 of this experimental selection, wherein Im is difference image picture
The average value of element value.That is:
Wherein threshold value T is calculated according to the specific image of each frame, is the amount of a dynamic change.
Step 2.3 is done two width bianry images and operation:
Two width difference image D can be obtained after step 2.1 calculus of differencesKAnd DK+1, then through step 2.2 to DKAnd DK+1Two
Moving region and the separated binary map of background are obtained after value, bianry image is done and operation, and D is as a result denoted asIk:
DIk=DK∩DK+1
Step 3 removes noise spot with median filtering:
The image DI that step 2 is obtainedKMedian filtering denoising is carried out, GI is as a result denoted asK, this experimental selection median filtering
Window size is 3*3.
Step 4 rejects non-smog connected region based on frame-to-frame correlation setting dynamic threshold, include the following three steps:
Step 4.1: each connected region in each frame image is marked with the bwlabel function in MATLAB:
The label that connected region is carried out to the image after the denoising of step 3 median filtering judges of connected region in image
Several and position judges connected region using four connection standards.The sum for remembering connected region is N, then the whole in each connected region
Pixel value is successively labeled as 1 to N.
Step 4.2, the correlation size for calculating each connected region:
The size of each connected domain correlation is calculated since the second frame image.I connected domain C of k-th frameKi, k-th frame connected region
Sum is N.If the minimum rectangular area comprising this connected domain isK-1 frame withThe same target area of sitting of interior pixel isTwo rectangular areasWithThe sum of products of interior respective coordinates pixel value is
The then correlation of i-th of connected domainWherein i≤N.
The small connected region of step 4.3, removal correlation:
As shown in Figure 1, setting dynamic threshold Q, judges since first connected domain, if the interframe of connected region is related
Property CR (i) be less than threshold value Q, then the connected region marked is rejected, it is on the contrary then retain, then judge next connected domain, directly
Terminate to judge after all having handled to all connected domains.Selected threshold value is 0.6 times of frame-to-frame correlation maximum value in connected region,
That is Q=0.6CR (i)max。
As shown in Fig. 2, 10 frame smog video images in this experiment is selected to detect after the method for the present invention is handled
Smog development process successively such as Fig. 2 a1-j1, compares the smog image such as Fig. 2 a2-j2 obtained only with step 1-3, can obviously see
The effect of step 4 out.From in the image handled without step 4 it can be seen that other than smoke region also including in bianry image
There is the noise region of non-smog, and the position of noise region and size are random, and passes through step 4 treated bianry image
Contain only accurate smoke region.
The present invention is had been described in detail by examples detailed above, and the description above is not considered as to limit of the invention
System, all equivalent structures done using description of the invention and accompanying drawing content or process transformation, or be used in directly, indirectly
Other related technical areas should all be included in the protection scope of the present invention.
Claims (1)
1. a kind of low-light (level) smog video detecting method based on image correlation, this method is based on Honeywell 1080P net
The infrared gun shaped video camera of network is connected with D.C. regulated power supply, and the device that the network port RJ45 of video camera is connected with PC machine is examined
It surveys, under night low-illumination environment, the typing of smog video is carried out using the high-definition camera of included infrared auxiliary light source, then
The video of typing is sent to PC end by the network port, phase finally is carried out to video image on the MATLAB platform on the end PC
Pass is handled, and first then rejects image using frame-to-frame correlation with lesser noise in median filtering removal image in treatment process
In non-smog connected domain, finally extract accurate smog movement region, the connected domain in bianry image is as interconnected
Region composed by the identical point of pixel value, this connected region include non-smog caused by random noise other than smoke region
Region, to obtain accurate smoke region must remove this non-smog connected region, it is characterized in that: this method includes following step
It is rapid:
Step 1: target video image is compressed and is converted into gray level image
High-definition camera by carrying infrared auxiliary light source carries out typing to smog video, takes continuous 200 frame figure in video
As being tested, the pixel of the image of typing is 576*720, in order to reduce time used in entire algorithm, the compression of use
Method is arest neighbors interpolation method, image is carried out 0.5 times of compression, i.e., compressed image pixel is 288*360, in order to facilitate after
Compressed image is converted gray level image by the processing of phase；
Step 2: obtaining gray level image to step 1 carries out four frame difference operations, complete smog movement in video image is extracted
Region
Smog movement region is extracted using four frame difference methods, four frame difference operations, the company of taking are done to the gray level image that step 1 obtains
Continue adjacent four frames image and do calculus of differences:
Wherein IK、IK+1、IK+2、IK+3For continuous four frames gray level image, DK、DK+1For difference image, wherein 1≤K≤197, select threshold
Value T carries out binary conversion treatment to difference image, and moving region and background are separated, and the size of threshold value T is in (0,1) range, choosing
The threshold value T=Im+0.4 selected, wherein Im is the average value of difference image pixel value, and the region less than threshold value T is denoted as background, is greater than
The region of threshold value T is denoted as smog movement region, it may be assumed that
Two obtained width binary images are done and are denoted as D with operation resultIk:
DIk=DK∩DK+1
Step 3: with the noise spot in median filtering removal bianry image
In order to which the random noise point reduced in the obtained binary image of step 2 is adopted with reducing the operand of next step
The bianry image is denoised with common median filter method, using the nonlinear smoothing performance of median filter method,
The value of pixel each in bianry image is set to the intermediate value of all pixels point in window selected by median filtering, to eliminate orphan
Vertical noise spot, the window size of selected median filter method are 3*3, i.e., the value of each pixel after median filtering
All take be using the pixel as center size all pixels point value in the region 3*3 median, if bianry image before median filtering
In certain point be isolated noise point, then after median filtering this point will become background dot, treated, and image is still denoted as DIk；
Step 4: marking above-mentioned bianry image DIkIn connected region, frame-to-frame correlation based on each connected region sets dynamic
Threshold value rejects non-smog connected region, operates as follows:
Step 4.1 is marked in each frame image respectively with the connected component labeling function bwlabel in the MATLAB software on the aforementioned end PC
A connected region:
The label that connected region is carried out to the image after the denoising of step 3 median filtering, i.e., if connected region in current frame image
Domain one shares N number of, then whole pixels in each connected region is successively labeled as 1 to N, judges connected region using four connection standards
That is, say that they are connections if some pixel value is identical as it in the upper and lower, left and right four direction of a pixel in domain
, connected domain is region composed by the pixel of all connections；
Step 4.2, the correlation size for calculating above-mentioned each connected region:
The size of each connected domain correlation is calculated since the second frame image, correlation is defined as completely including this connected domain most
Total pixel in the sum of products of pixel value in pixel value and previous frame in small rectangular area in this region and the rectangular area
The ratio between number, DIkIn i-th of connected domain beMinimum rectangular area comprising this connected domain isIn K-1 frame image
WithThe same target area of sitting of interior pixel isTwo rectangular areasWithInterior respective coordinates pixel value multiplies
It is the sum of long-pending to beWherein M, N are the line number of the minimum rectangular area comprising connected domain
And columns, total pixel of rectangular area are denoted as S, i.e. S=M*N, then the correlation of i-th of connected domain
The small connected region of step 4.3, removal correlation:
After step 4.2 finds out the correlation CR (i) of each connected domain, given threshold Q rejects the small connected region of correlation,
The threshold value Q is set as 0.6 times of the maximum value in all connected domain correlations of present frame, i.e. Q=0.6CR (i)maxIt is dynamic for one
The size of the value of state, threshold value Q is different because of the pixel value of each frame difference, if the frame-to-frame correlation CR (i) of connected region is small
In threshold value Q, then the connected region marked is rejected, it is on the contrary then retain.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611032108.4A CN106778488B (en) | 2016-11-22 | 2016-11-22 | Low-light (level) smog video detecting method based on image correlation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611032108.4A CN106778488B (en) | 2016-11-22 | 2016-11-22 | Low-light (level) smog video detecting method based on image correlation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106778488A CN106778488A (en) | 2017-05-31 |
| CN106778488B true CN106778488B (en) | 2019-07-16 |
Family
ID=58971017
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611032108.4A Expired - Fee Related CN106778488B (en) | 2016-11-22 | 2016-11-22 | Low-light (level) smog video detecting method based on image correlation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106778488B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107895454B (en) * | 2017-11-16 | 2020-02-14 | 江苏理工学院 | Early fire smoke video detection method based on wavelet energy |
| CN107967448B (en) * | 2017-11-16 | 2021-05-18 | 江苏理工学院 | Early fire smoke real-time detection method and system |
| CN109034161B (en) * | 2018-07-12 | 2022-06-17 | 中国船舶重工集团公司第七二四研究所 | Sea ice identification method based on radar video image interframe correlation |
| CN109142176B (en) * | 2018-09-29 | 2024-01-12 | 佛山市云米电器科技有限公司 | Smoke subarea space rechecking method based on space association |
| CN111145234A (en) * | 2019-12-25 | 2020-05-12 | 沈阳天眼智云信息科技有限公司 | Fire smoke detection method based on binocular vision |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101515326A (en) * | 2009-03-19 | 2009-08-26 | 浙江大学 | Method for identifying and detecting fire flame in big space |
| CN102509307A (en) * | 2011-10-12 | 2012-06-20 | 西安理工大学 | Method for searching moving target based on longitude and latitude location and image registration |
| CN102509414A (en) * | 2011-11-17 | 2012-06-20 | 华中科技大学 | Smog detection method based on computer vision |
| CN104978733A (en) * | 2014-04-11 | 2015-10-14 | 富士通株式会社 | Smoke detection method and smoke detection device |
| CN105741480A (en) * | 2016-03-17 | 2016-07-06 | 福州大学 | Fire and smoke detection method based on image identification |
-
2016
- 2016-11-22 CN CN201611032108.4A patent/CN106778488B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101515326A (en) * | 2009-03-19 | 2009-08-26 | 浙江大学 | Method for identifying and detecting fire flame in big space |
| CN102509307A (en) * | 2011-10-12 | 2012-06-20 | 西安理工大学 | Method for searching moving target based on longitude and latitude location and image registration |
| CN102509414A (en) * | 2011-11-17 | 2012-06-20 | 华中科技大学 | Smog detection method based on computer vision |
| CN104978733A (en) * | 2014-04-11 | 2015-10-14 | 富士通株式会社 | Smoke detection method and smoke detection device |
| CN105741480A (en) * | 2016-03-17 | 2016-07-06 | 福州大学 | Fire and smoke detection method based on image identification |
Non-Patent Citations (1)
| Title |
|---|
| 《基于 Surendra 背景减除法和四帧差分法的目标检测算法》;纪青华等;《计算机应用与软件》;20141215;第31卷(第12期);第242-244页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106778488A (en) | 2017-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106778488B (en) | Low-light (level) smog video detecting method based on image correlation | |
| CN108846365B (en) | Detection method and device for fighting behavior in video, storage medium and processor | |
| CN101271514B (en) | Image detection method and device for fast object detection and objective output | |
| CN102708651B (en) | A kind of image type smoke fire disaster detection and system | |
| CN113591968A (en) | Infrared weak and small target detection method based on asymmetric attention feature fusion | |
| CN110334660A (en) | A kind of forest fire monitoring method based on machine vision under the conditions of greasy weather | |
| Zhao et al. | Early smoke detection of forest fire video using CS Adaboost algorithm | |
| CN114202646A (en) | Infrared image smoking detection method and system based on deep learning | |
| CN104301585A (en) | Method for detecting specific kind objective in movement scene in real time | |
| Dwivedi et al. | An approach for unattended object detection through contour formation using background subtraction | |
| Malbog et al. | Edge detection comparison of hybrid feature extraction for combustible fire segmentation: a Canny vs Sobel performance analysis | |
| CN114241379A (en) | Passenger abnormal behavior identification method, device and equipment and passenger monitoring system | |
| Saponara et al. | Exploiting R-CNN for video smoke/fire sensing in antifire surveillance indoor and outdoor systems for smart cities | |
| CN102999994A (en) | Flame detection device based on video image analysis | |
| CN103020588B (en) | Based on the flame detecting method of video image analysis | |
| CN112184751A (en) | Object identification method and system and electronic equipment | |
| CN105869184A (en) | Forest fire smoke image detection method based on path analysis | |
| CN103020587A (en) | View analyzing method based on video image analysis | |
| Gao et al. | Design and implementation of fire detection system using new model mixing | |
| CN111191575B (en) | Naked flame detection method and system based on flame jumping modeling | |
| KR20220107494A (en) | Drone detection system based on deep learning using SWIR camera | |
| Gao | An intelligent video surveillance system | |
| Mudjirahardjo et al. | Temporal analysis for fast motion detection in a crowd | |
| CN104794738A (en) | Method and device for detecting smoke and fire | |
| Hu et al. | Motion objects detection based on higher order statistics and hsv color space |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190716 Termination date: 20211122 |