CN108460785A - Flame detecting method - Google Patents
Flame detecting method Download PDFInfo
- Publication number
- CN108460785A CN108460785A CN201611144635.4A CN201611144635A CN108460785A CN 108460785 A CN108460785 A CN 108460785A CN 201611144635 A CN201611144635 A CN 201611144635A CN 108460785 A CN108460785 A CN 108460785A
- Authority
- CN
- China
- Prior art keywords
- pictures
- flame
- model
- seq
- sequence
- 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
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000001514 detection method Methods 0.000 claims abstract description 8
- 230000009466 transformation Effects 0.000 claims abstract description 7
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 230000001186 cumulative effect Effects 0.000 claims description 10
- 239000000284 extract Substances 0.000 claims description 8
- 239000000654 additive Substances 0.000 claims description 3
- 230000000996 additive effect Effects 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 abstract 1
- 238000001914 filtration Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 4
- 101100243951 Caenorhabditis elegans pie-1 gene Proteins 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 2
- 238000002485 combustion reaction Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000009191 jumping Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of flame detecting methods, 25 frame pictures generate a continuous sequence of pictures before storing video first, it takes exercises detection and the extraction Area generation doubtful flame region similar with flame color to the 25th pictures in sequence, wavelet transformation is done to each pixel in sequence of pictures again, part of the small wave frequency rate more than the region of threshold value and the intersection of doubtful flame region is extracted, determines flame region.The present invention is used for the fire defector of video monitoring environment, realizes fire alarm, reduces property loss, the scope of application is wider, stability higher, more convenient to use.
Description
Technical field
A kind of real-time flame detecting method based on video image of the present invention, be directed to a kind of color based on flame,
Three kinds of features such as movement, small wave frequency rate and the flame detecting method for establishing Cumulative probability model.
Background technology
The method that existing flame detecting method combines similar flame color region detection by motion detection mostly.It is this kind of
Method be limited by extraction the single generalization of feature it is not strong, so can not effectively be distinguished under complex scene doubtful flame object and
Flame object, and robustness does not occur easily reporting by mistake and failing to report by force.
Invention content
In order to effectively solve the above problems, the present invention provides a kind of flame detecting method, the flame detecting method include with
Lower step：
Step 1, the preceding 25 frame picture of storage generate sequence of pictures Seq_pic；
Step 2, the flame majority generated that burnt due to object are rendered as cerise, so R points therein in RGB channel
Amount is far longer than G and B component, and is greater than a threshold value (our R component threshold value according to the numerical value of many experiments R component
It is 160).And in the channels HSV, brightness value is greater than the average brightness of surrounding, and also greater than one fixed threshold of saturation degree
Value (our saturation degree threshold value is 0.6).So according to above-mentioned analysis, we are to the 25th frame picture Seq_ in sequence of pictures
It is Mask that pic25, which extracts its flame color characteristic seal,color；
It is as follows to extract flame color Mask rules：
Wherein R G B respectively represent the red channel of image, green channel, blue channel.S represents saturation degree, and V represents bright
Degree, mean (V) represent luminance mean value.
Step 3, due to flame during exercise, edge jitter is more violent, but the flame core position of flame is more stable.One
As completely can not effectively obtain the edge of flame as method for testing motion traditional Vibe and GMM etc.., so using quick
The higher frame difference method of sensitivity is used as the algorithm of motion detection (we are detected using three frame difference methods herein).Rule is as follows：
A=Seq_pic25 (x, y)-Seq_pic24 (x, y)
B=Seq_pic24 (x, y)-Seq_pic23 (x, y)
Wherein A represents the difference of the 25th figure and each respective pixel value of the 24th pictures, and wherein B represents the 24th figure and the
The difference of each respective pixel value of 23 pictures.TH is a threshold value.
Step 4, flame are in burning, it has been found that its jumping frequency rate is that have mark that can follow, so we utilize low pass filtered
Wave Lo_D and high-pass filtering Hi_D does wavelet transformation to the sequence of pictures Seq_pic of continuous 25 frame on time shaft, extracts the Seq_
The small echo frequecy characteristic of pic.
Detailed process is as follows：
1. setting low-pass filtering Lo_D=[0.25,0.5,0.25]；High-pass filtering Hi_D=[- 0.25 0.5, -0.25]；Point
Continuous pixel value vector Seq_pic (x, y) (sets the pixel value of (x, y) in t pictures in the other Seq_pic to sequence of pictures
For Seq_pic t (x, y), then its continuous pixel value vector in sequence of pictures Seq_pic is Seq_pic (x, y)={ Seq_
Pic1 (x, y) ... .Seq_pic t (x, y) ... Seq_pic 25 (x, y) }) it does convolution algorithm and obtains HL.
2. obtaining the high fdrequency component H and low frequency component L of small echo signal respectively to HL interval samplings again.
It then proceedes to do wavelet transformation again to low frequency component L and obtains high fdrequency component H_1.
3. the value of the spike point of two high-frequency signals of H and H_1 is found out, if it is greater than time of threshold value TH (our TH is 30)
For number in [2 8], we then think that Seq_pic (x, y) is to meet the region of the small wave frequency rate of flame, and extract the region and be denoted as
Maskwave。
Step 5：Typically active when due to flame combustion, so when there is fire, flame all can be in fixed area
Burning a period of time spreads again.According to this feature, if continuously there is three kinds of regions Mask intersection of flame in same region
Position then its be flame maximum probability, so it is proposed that generate color motion frequency three Mask probability add up mould
Type can greatly exclude doubtful flame object by this model, solve traditional a large amount of mistakes based on video image fire defector
The problem of report.
Detailed process is as follows：
1. generated in the above process three Mask are overlapped three kinds of Cumulative probability models of generation respectively
Specific update rule is as follows：
modelcolor(x, y)=modelcolor(x,y)+Maskcolor(x,y)
modelmotion(x, y)=modelmotion(x,y)+Maskmotion(x,y)
modelwave(x, y)=modelwave(x,y)+Maskwave(x,y)
Above-mentioned formula modelcolor(x,y)、modelmotion(x,y)、modelwave(x, y) be respectively color, movement, small
Its initial value of the Cumulative probability model of wave frequency rate feature is all 0, THcolor、THmotion、THwaveRepresent color, movement, small wave frequency
Accumulation threshold of the rate in probability statistics model.
2. after carrying out above-mentioned steps, the first pictures Seq_pic1 of sequence of pictures Seq_pic is abandoned, then read and regard
The next frame picture of frequency, is put into sequence of pictures Seq_pic, is recorded as Seq_pic25, and before in sequence of pictures Seq_pic
Picture sequence numbers respectively subtract 1 (for example Seq_pic2 becomes Seq_pic1) again to newly-generated sequence of pictures Seq_pic extractions
Three Mask of the features such as color, movement, small wave frequency rate, and be superimposed with respective model before, generating probability adds up model
model。
3. the value of model is infinitely superimposed in order to prevent, we are applied with a decaying weight for model, to be prevented with this
It is infinitely superimposed
Rule is as follows：
if modelt(x, y)=0＆＆modelt-1(x, y)=0
modelt(x, y)=modelt(x,y)*0.5
Model in above-mentioned formulat-1(x, y) indicates the Cumulative probability model at t-1 moment, modelt(x, y) indicates t moment
Cumulative probability model.
4. in order to make this algorithm more efficient stable, we apply one to the add up additive process of model model of probability
Time window, when the time add up frame number more than we then by the numerical value reset all of model be 0. specific rules it is as follows：
If t ＞ 100
T=0
Model (x, y)=0
Step 6:According to test of many times we determined that being usually flame region when the value aggregate-value of model (x, y) is more than 10.
Description of the drawings
Fig. 1, system flow schematic diagram of the invention
Specific implementation mode
The flame detecting method comprises the steps of：
Step 1, the preceding 25 frame picture of storage generate sequence of pictures Seq_pic；
Step 2, the flame majority generated that burnt due to object are rendered as cerise, so R points therein in RGB channel
Amount is far longer than G and B component, and is greater than a threshold value (our R component threshold value according to the numerical value of many experiments R component
It is 160).And in the channels HSV, brightness value is greater than the average brightness of surrounding, and also greater than one fixed threshold of saturation degree
Value (our saturation degree threshold value is 0.6).So according to above-mentioned analysis, we are to the 25th frame picture Seq_ in sequence of pictures
It is Mask that pic25, which extracts its flame color characteristic seal,co1or；
It is as follows to extract flame color Mask rules：
Wherein R G B respectively represent the red channel of image, green channel, blue channel.S represents saturation degree, and V represents bright
Degree, mean (V) represent luminance mean value.
Step 3, due to flame during exercise, edge jitter is more violent, but the flame core position of flame is more stable.One
As completely can not effectively obtain the edge of flame as method for testing motion traditional Vibe and GMM etc.., so using quick
The higher frame difference method of sensitivity is used as the algorithm of motion detection (we are detected using three frame difference methods herein).Rule is as follows：
A=Seq_pic25 (x, y)-Seq_pic24 (x, y)
B=Seq_pic24 (x, y)-Seq_pic23 (x, y)
Wherein A represents the difference of the 25th figure and each respective pixel value of the 24th pictures, and wherein B represents the 24th figure and the
The difference of each respective pixel value of 23 pictures.TH is a threshold value.
Step 4, flame are in burning, it has been found that its jumping frequency rate is that have mark that can follow, so we utilize low pass filtered
Wave Lo_D and high-pass filtering Hi_D does wavelet transformation to the sequence of pictures Seq_pic of continuous 25 frame on time shaft, extracts the Seq_
The small echo frequecy characteristic of pic.
Detailed process is as follows：
4. setting low-pass filtering Lo_D=[0.25,0.5,0.25]；High-pass filtering Hi_D=[- 0.25 0.5, -0.25]；Point
Continuous pixel value vector Seq_pic (x, y) (sets the pixel value of (x, y) in t pictures in the other Seq_pic to sequence of pictures
For Seq_pic t (x, y), then its continuous pixel value vector in sequence of pictures Seq_pic is Seq_pic (x, y)={ Seq_
Pic1 (x, y) ... .Seq_pic t (x, y) ... Seq_pic 25 (x, y) }) it does convolution algorithm and obtains HL.
5. obtaining the high fdrequency component H and low frequency component L of small echo signal respectively to HL interval samplings again.
It then proceedes to do wavelet transformation again to low frequency component L and obtains high fdrequency component H_1.
6. the value of the spike point of two high-frequency signals of H and H_1 is found out, if it is greater than time of threshold value TH (our TH is 30)
For number in [2 8], we then think that Seq_pic (x, y) is to meet the region of the small wave frequency rate of flame, and extract the region and be denoted as
Maskwave。
Step 5：Typically active when due to flame combustion, so when there is fire, flame all can be in fixed area
Burning a period of time spreads again.According to this feature, if continuously there is three kinds of regions Mask intersection of flame in same region
Position then its be flame maximum probability, so it is proposed that generate color motion frequency three Mask probability add up mould
Type can greatly exclude doubtful flame object by this model, solve traditional a large amount of mistakes based on video image fire defector
The problem of report.
Detailed process is as follows：
5. generated in the above process three Mask are overlapped three kinds of Cumulative probability models of generation respectively
Specific update rule is as follows：
modelcolor(x, y)=modelcolor(x,y)+Maskcolor(x,y)
modelmotion(x, y)=modelmotion(x,y)+Maskmotion(x,y)
modelwave(x, y)=modelwave(x,y)+Maskwave(x,y)
Above-mentioned formula modelcolor(x,y)、modelmotion(x,y)、modelwave(x, y) be respectively color, movement, small
Its initial value of the Cumulative probability model of wave frequency rate feature is all 0, THcolor、THmotion、THwaveRepresent color, movement, small wave frequency
Accumulation threshold of the rate in probability statistics model.
6. after carrying out above-mentioned steps, the first pictures Seq_pic1 of sequence of pictures Seq_pic is abandoned, then read and regard
The next frame picture of frequency, is put into sequence of pictures Seq_pic, is recorded as Seq_pic25, and before in sequence of pictures Seq_pic
Picture sequence numbers respectively subtract 1 (for example Seq_pic2 becomes Seq_pic1) again to newly-generated sequence of pictures Seq_pic extractions
Three Mask of the features such as color, movement, small wave frequency rate, and be superimposed with respective model before, generating probability adds up model
model。
7. the value of model is infinitely superimposed in order to prevent, we are applied with a decaying weight for model, to be prevented with this
It is infinitely superimposed
Rule is as follows：
if modelt(x, y)=0＆＆modelt-1(x, y)=0
modelt(x, y)=modelt(x,y)*0.5
Model in above-mentioned formulat-1(x, y) indicates the Cumulative probability model at t-1 moment, modelt(x, y) indicates t moment
Cumulative probability model.
8. in order to make this algorithm more efficient stable, we apply one to the add up additive process of model model of probability
Time window, when the time add up frame number more than we then by the numerical value reset all of model be 0. specific rules it is as follows：
If t ＞ 100
T=0
Model (x, y)=0
Step 6:According to test of many times we determined that being usually flame region when the value aggregate-value of model (x, y) is more than 10.
Claims (6)
1. flame detecting method, which is characterized in that detecting step is as follows：
Step 1,25 frame pictures before video are generated into a continuous sequence of pictures；
Step 2, it takes exercises detection to the 25th pictures in sequence of pictures, while it is doubtful to extract the Area generation similar with flame color
Like flame region；
Step 3 carries out wavelet transformation to each pixel in sequence of pictures, extracts the region that small wave frequency rate is more than threshold value；
The cumulative figure of step 4, three mask probability for generating color and motion frequency, small wave frequency rate be more than threshold value and with doubtful flame
The part of region intersection, is determined as flame region.
2. flame detecting method as described in claim 1, it is characterised in that：In step 2, carried using RGB channel and the channels HSV
Take doubtful flame region.
3. flame detecting method as claimed in claim 2, it is characterised in that：R component threshold value is set in RGB channel as 160, if
It is 0.6 to determine saturation degree threshold value in the channels HSV.
4. flame detecting method as claimed in claim 2, it is characterised in that：In step 2, motion detection uses frame difference method, right
23 to 25 pictures of drop do three frames and look into motion detection generation movement mask.
5. flame detecting method as described in claim 1, which is characterized in that in step 3, Wavelet transformation is done to image sequence,
Extract the small wave frequency rate mask of Area generation that ratio is more than threshold value.
6. flame detecting method as described in claim 1, which is characterized in that step 4 further includes following steps：
Step 41, the first pictures Seq_pic1 of sequence of pictures Seq_pic is abandoned, then reads the next frame picture of video,
It is put into sequence of pictures Seq_pic, is recorded as Seq_pic25, and the picture sequence numbers in sequence of pictures Seq_pic respectively subtract before
1 (for example Seq_pic2 becomes Seq_pic1) is again to newly-generated sequence of pictures Seq_pic extractions color, movement, small wave frequency
Three Mask of the features such as rate, and be superimposed with respective model before, generating probability adds up model model；
The value of step 42, in order to prevent model are infinitely superimposed, we are applied with a decaying weight for model, to be prevented with this
It is infinitely superimposed；
Additive process one time window of application of step 43, the model model that adds up to probability, frame number is added up more than me when the time
Then by the numerical value reset all of model be 0.
Step 44, when model value aggregate-value be more than 10 be usually flame region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611144635.4A CN108460785A (en) | 2016-12-13 | 2016-12-13 | Flame detecting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611144635.4A CN108460785A (en) | 2016-12-13 | 2016-12-13 | Flame detecting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108460785A true CN108460785A (en) | 2018-08-28 |
Family
ID=63221820
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611144635.4A Pending CN108460785A (en) | 2016-12-13 | 2016-12-13 | Flame detecting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108460785A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127433A (en) * | 2019-12-24 | 2020-05-08 | 深圳集智数字科技有限公司 | Method and device for detecting flame |
-
2016
- 2016-12-13 CN CN201611144635.4A patent/CN108460785A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111127433A (en) * | 2019-12-24 | 2020-05-08 | 深圳集智数字科技有限公司 | Method and device for detecting flame |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108093221B (en) | Suture line-based real-time video splicing method | |
| CN106600888B (en) | Automatic forest fire detection method and system | |
| CN108447219A (en) | System and method for detecting fire hazard based on video image | |
| CN112785534B (en) | A method for removing ghosting and multi-exposure image fusion in dynamic scenes | |
| CN105374051B (en) | The anti-camera lens shake video moving object detection method of intelligent mobile terminal | |
| CN103702111B (en) | A kind of method detecting camera video color cast | |
| CN103942812B (en) | Moving object detection method based on Gaussian mixture and edge detection | |
| CN101441771A (en) | Video fire hazard smoke detecting method based on color saturation degree and movement mode | |
| CN102034110B (en) | Detection method of flame | |
| CN103020924B (en) | Low-illumination monitored image enhancement method based on similar scenes | |
| CN104700405B (en) | A kind of foreground detection method and system | |
| CN108256496A (en) | A kind of stockyard smog detection method based on video | |
| CN108961227A (en) | A kind of image quality evaluating method based on airspace and transform domain multiple features fusion | |
| CN105447890A (en) | Motion vehicle detection method resisting light effect | |
| CN107392879B (en) | A low-illumination surveillance image enhancement method based on reference frames | |
| CN113066077B (en) | Flame detection method and device | |
| CN104268888B (en) | A kind of image blurring detection method | |
| CN105405153B (en) | Intelligent mobile terminal anti-noise jamming Extracting of Moving Object | |
| CN108460785A (en) | Flame detecting method | |
| CN115830513A (en) | Method, device and system for determining image scene change and storage medium | |
| JP7608857B2 (en) | Anomaly detection method and anomaly detection device | |
| CN103079029B (en) | A Digital TV Mosaic Recognition Method Based on Macroblock Edge Information | |
| CN102685370B (en) | De-noising method and device of video sequence | |
| KR101733125B1 (en) | Method of chroma key image synthesis without background screen | |
| CN102143376A (en) | Method and device for detecting consistency of twin-channel video signals |
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: 20180828 |