CN110211323A - Forest fire recognition methods based on cascade sort - Google Patents
Forest fire recognition methods based on cascade sort Download PDFInfo
- Publication number
- CN110211323A CN110211323A CN201910456498.5A CN201910456498A CN110211323A CN 110211323 A CN110211323 A CN 110211323A CN 201910456498 A CN201910456498 A CN 201910456498A CN 110211323 A CN110211323 A CN 110211323A
- Authority
- CN
- China
- Prior art keywords
- fire
- classifier
- candidate region
- cascade
- recognition methods
- 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 42
- 238000012706 support-vector machine Methods 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 15
- 238000012795 verification Methods 0.000 claims abstract description 12
- 239000000284 extract Substances 0.000 claims description 9
- 230000009467 reduction Effects 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 238000007619 statistical method Methods 0.000 claims description 2
- 238000012549 training Methods 0.000 abstract description 16
- 238000002372 labelling Methods 0.000 abstract description 3
- 238000012937 correction Methods 0.000 abstract description 2
- 238000012216 screening Methods 0.000 abstract 1
- 230000011218 segmentation Effects 0.000 description 6
- 238000013461 design Methods 0.000 description 5
- 238000012544 monitoring process Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 238000013475 authorization Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 235000019504 cigarettes Nutrition 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 239000000779 smoke Substances 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/005—Fire alarms; Alarms responsive to explosion for forest fires, e.g. detecting fires spread over a large or outdoors area
-
- G—PHYSICS
- G08—SIGNALLING
- G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
- G08B17/00—Fire alarms; Alarms responsive to explosion
- G08B17/12—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions
- G08B17/125—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions by using a video camera to detect fire or smoke
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Business, Economics & Management (AREA)
- Emergency Management (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biodiversity & Conservation Biology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Fire-Detection Mechanisms (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of forest fire recognition methods based on cascade sort, and moving target is partitioned into from original video by dynamic background modeling algorithm；By carrying out connected component labeling to moving target, fire candidate region is obtained；Building two-stage cascade classifier classifies to candidate region, and the first order carries out fire hazard classification using the color model based on fire, and the second level carries out fire hazard classification using the support vector machines based on histogram of gradients feature；Further, by multiframe verify screening cascade classifier as a result, continuous multiple frames are detected as with the situation of fire, just output fire behavior alarm.System includes: candidate region generation module, classifier off-line training module, cascade sort module and multiframe correction verification module.The present invention can take into account check system false-alarm and detection speed.
Description
Technical field
The invention belongs to computer visions and pattern-recognition, image procossing and Intellective Fire Alarm System field, especially relate to
And a kind of forest fire recognition methods based on cascade sort.
Background technique
The forest data ecological resources important as the mankind have very important effect life, the production of the mankind.And
Forest fire is larger to the destruction of the forest reserves, or even threatens life and property safety of people.Currently, being based on video
The forest fire of monitoring identifies aspect, usually relies on manpower mode on duty and carries out fire behavior monitoring, is easy to fatigue, not can guarantee whole day
Wait monitoring.And the intelligent fire identification function based on monitor video, it is avoided that drawbacks described above.Therefore the intelligence fire based on monitor video
The research of calamity identification is directed to the demand in society and market.
In field of target recognition, the description of target is a key link, and forest rocket identifies that field is no exception.It is mentioned
The feature taken is most important to the availability of this feature to the characterization ability and its computation complexity of dynamic object, for example, multiple
Miscellaneous feature may have preferable characterization ability but often computation complexity is excessively high, it is difficult to meet requirement of real-time.Zhang Qingjie
Deng (design of Zhang Qingjie, Zheng Ergong, Xu Liang, et al. forest fire prevention unmanned aerial vehicle system and the research of forest fires recognizer [J]
Electronic measurement technique, 2017 (1)) unmanned plane image is utilized, using discrete cosine transform conspicuousness detection algorithm, obtain forest
Pyrotechnics candidate region, then extracts space-time piecemeal and covariance matrix feature combination supporting vector machine realizes forest fire detection,
This method does not utilize the time response of video, to obtain candidate region based on dynamic background modeling, and does not design different branches
Support vector machines, to cope with the otherness of the pyrotechnics calamity target scale of different scale.Similarly, patent one kind is based on intelligence view
Forest rocket detection method and its dedicated unit (the Chinese patent Authorization Notice No.: CN201210332403.7, authorization of frequency analysis
The day for announcing: on 01 09th, 2013) dynamic analysis are carried out to successive frame based on intensive optical flow method and extract dynamic motion spy
Property, realize that forest rocket identifies, does not also design the support vector machines of different branches using single branch's support vector machines, to cope with not
With the otherness of the pyrotechnics calamity target scale of scale.
Forest rocket identifies field, since system has requirement of real-time, so designed recognition methods needs have
Preferable operational efficiency, in view of this, (Gao Na, Li Liang the fire image Feature Fusion Algorithm neural network based such as high Na
[J] computer system application, 2010,19 (1): 86-89.) propose using area of flame variation, centroid position, wedge angle number,
The features combination neural fusion fire identification such as circularity, although this method computing cost is smaller, extracted feature
Fairly simple, system robustness is difficult to ensure.
(the forest fire identification side of Zhu Sisi, Ding Dehong, Chen Chaoying, the et al. based on image procossing such as Zhu Sisi
Method research [J] infrared technique, 2016,38 (5)) by whether having red area in detection image, it is then red according to gained
The area in color region whether be in sustainable growth state, final marriage pyrotechnics circularity realizes forest fire identification, this method by
In the sustainable growth feature for depending on conflagration area, therefore the system delay of fire identification is larger.It is calculated using slow moving object segmentation
Method obtains the candidate region of cigarette, then using least-mean-square error algorithm in combination with detection, the shadow Detection for rising cigarette district domain
With removal, the detection of smog color region to realize that forest smog identifies, this method only has studied the knowledge of the smoke target in forest
Other problem, the not yet fire target in research forest fire detection identify problem, in addition, this method is special due to extracted smog
Sign is simpler, and easy grey trees and Part Wild animal identification by background are smog, causes false-alarm.
In field of target recognition, cascade classifier has the real-time of acceleration system and the false-alarm of inhibition system important
Effect.Patent forest-fire remote video monitoring firework identification method (Chinese patent Authorization Notice No.:
CN201010040086.2, authorized announcement date: on 07 07th, 2010) according to color, shape and the texture in image object
The combination of feature realizes that forest fire identifies using single classifier, although this method has reused Fusion Features institute band
The precision improvement come with acceleration system and reduces false-alarm however, being not designed to the mode of cascade classifier.
Although in conclusion forest fire recognition methods based on video monitoring and achieve certain achievement,
In order to reach the requirement of practical application, there is an urgent need to pound verification and measurement ratio, real-time aspect makes further improvement.
Summary of the invention
The embodiment of the present invention is designed to provide a kind of forest fire recognition methods based on cascade sort, it is intended to solve
The recognition accuracy of existing forest fire recognition methods is not up to standard and/or not fully up to expectations, real-time needs to improve, robustness
Not strong problem.
A kind of forest fire recognition methods based on cascade sort, it is characterised in that designing peak flow color model, building one
The classifier of a two-stage cascade, and reduction system false-alarm is verified by multiframe, it specifically includes:
Step 1 obtains candidate region based on dynamic background modeling algorithm；
Step 2, extracts fire color model feature to candidate region, carries out first order classification using color model classifier；
Step 3 extracts gradient towards histogram feature to candidate region, and the support vector machine classifier of four branches carries out third
Grade classification；
Step 4 carries out multiframe verification to classification results, continuous multiple frames is detected as with the situation of fire, just exports fire behavior alarm.
Further, the forest fire recognition methods described in claim 1 based on cascade sort, which is characterized in that step 1
In dynamic background modeling algorithm obtain candidate region refer to, open image using input picture first and establish background model, subsequent figure
As obtaining the segmentation result of moving region by established background model.Segmentation result is extracted using 8- connection labelling method
All connected regions, to obtain fire candidate region.
Further, the forest fire recognition methods described in claim 1 based on cascade sort, which is characterized in that step 2
The color model feature refers to, is counted by the relationship of red channel and green channel to 5000 Zhang Hanhuo Target Photos
Analyze resulting color model；The color model classifier refers to that the red pixel number ratio of picture is greater than green pixel
Number ratio (be greater than 0.13), and when red channel number of pixels is greater than 15, this grade of separator just will be considered that the candidate region
For fiery classification.
Further, the forest fire recognition methods described in claim 1 based on cascade sort, which is characterized in that step 3
The support vector machines of four branch refers to, exports picture space according to the first order, is divided into four mutually not according to the height of picture
The subset of intersection is classified using four support vector machines.
Further, the forest fire recognition methods described in claim 1 based on cascade sort, which is characterized in that step 4
The multiframe verification refers in continuous multiple frames image, is continuously judged as that the classification results of fire are counted to classifier is cascaded
Number is greater than 10 testing result to count value, is just considered fire detection result.
Forest fire recognition methods provided by the invention based on cascade sort identifies skill with existing video forest rocket
Art is compared, the invention has the advantages that and effect: according to dynamic background modeling method carry out image segmentation, efficiently utilize
The dynamic of forest fire target is not only able to maintain preferable segmentation effect, but also full figure can effectively be avoided to carry out picture search
Computing cost；Using the classifier based on color model, color characteristic simple based on fire target, effective is quickly divided
Class, Heterosis exist: a large amount of non-duration and degree of heating favored areas are inhibited, and the post-processing of only candidate region has not been saved calculating and opened
Pin, moreover it is possible to reduce system false-alarm.Four branch's support vector machine classifiers based on histogram of gradients feature, can not only describe simultaneously
The profile and texture information of fire target, the sub characteristic present ability to fire target of enhancing description, and sample can be reduced
Difference in class promotes the generalization ability of classifier；The method that classification results are carried out multiframe verification can further decrease system void
It is alert.While can ensure that verification and measurement ratio, the Forest Fire currently based on video is effectively reduced by the design of cascade classifier in the present invention
The false alarm rate of calamity identifying system, and the acceleration system speed of service, can preferably meet practical application request.
Detailed description of the invention
Fig. 1 is the forest fire recognition methods flow chart provided in an embodiment of the present invention based on cascade sort；
Fig. 2 is the forest fire recognition methods system structure diagram provided in an embodiment of the present invention based on cascade sort；
In figure: A, candidate region generation module；B, classifier off-line training module；C, cascade sort module；D, multiframe calibration mode
Block；
Fig. 3 is the implementation example figure of four branched structures support vector machine classifier structure provided in an embodiment of the present invention；
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
With reference to the accompanying drawing and specific embodiment is further described application principle of the invention.
As shown in Figure 1, the forest fire recognition methods based on cascade sort of the embodiment of the present invention the following steps are included:
S101 obtains candidate region using based on dynamic background modeling algorithm；
S102 realizes that the first order is classified using the classifier based on color model；
S103 realizes that the second level is classified using four branch's support vector machine classifiers based on gradient towards histogram feature；
S104 carries out spininess verification to classification results, inhibits system false-alarm.
Dynamic background modeling algorithm in step S101 refers to, opens image using input picture first and establishes background model, after
Continuous image obtains the segmentation result of moving region by established background model.The candidate region, which refers to be connected in 8-, to be marked
All connected regions that method is extracted.
Color model classifier described in step S102 refers to that the red pixel number ratio of picture is greater than green pixel number ratio
Rate (be greater than 0.13), and when red channel number of pixels is greater than 15, this grade of separator just will be considered that the candidate region is fire class
Not.
The support vector machine classifier of four branched structures described in step S103, which refers to, divides sample sky according to candidate region height
Between, constitute four mutually disjoint sample sets；In each sample set, the histogram of gradients for extracting candidate region is special
Sign, and pass through the classifier of linear SVM learning algorithm four branched structures of acquisition.
The verification of multiframe described in step S104 refers in continuous multiple frames image, the classification to being cascaded classifier and being judged as fire
As a result confirmed, when confirming that number is greater than 15, just issue final FIRE WARNING.
As shown in Fig. 2, the forest fire identification based on cascade classifier of the embodiment of the present invention is mainly given birth to by candidate region
At modules A, classifier off-line training module B, cascade sort module C, multiframe correction verification module D composition.
Modules A is chosen in candidate region, for modeling partitioning algorithm and 8- connected component labeling technology using dynamic background, from
Candidate region is generated in original video.
Classifier off-line training module B, it is offline true for collecting sample, dividing sample, scaling and training classifier
The relevant parameter for determining color model classification is greater than the threshold value and red pixel number of green pixel ratio including red pixel ratio
Classification thresholds.
Cascade sort module C chooses modules A with candidate region and classifier off-line training module B is connect, for candidate
Region carries out cascade sort.
Multiframe verification and neighborhood matching module D, connect, for being judged as to being cascaded classifier with cascade sort module C
The classification results number of fire counts, and the result for being identified as fire to continuous multiple frames just exports alarm of fire.
Specific embodiments of the present invention:
The overall flow of the method for the present invention is as shown in Figure 1, the method for the present invention main body includes two parts: 1. two-stage cascade classifiers；
2. multiframe verifies reduction system false-alarm.
1. two-stage cascade classifier
Two-stage cascade classifier, that is, color model classifier and based on gradient towards four branch's support vector machines of histogram feature
Classifier cascade is constituted.Every first-level class device includes two parts of classifier off-line training and on-line checking.
1.1 classifier off-line trainings
1.1.1 training sample prepares
For scale Forest Scene monitor video, the artificial fire target minimum circumscribed rectangle that cuts obtains fire target picture sample.From
The biggish sample of hand picking class outer pattern differentials forms training set D in the sample set that sample obtains, and in total includes just
Sample 5000 is opened, and negative sample 20000 is opened；Further, training set D is divided by four mutually disjoint sons according to sample height
Collection, the sample for (containing 50 pixels) highly between 10 to 50 pixels is classified as distant view sample, highly at 50 to 100
The sample for (containing 100 pixels) between pixel is classified as middle scape sample, (contains 150 highly between 100 to 150 pixels
Pixel) sample be classified as two-shot sample, be highly classified as super close shot sample more than or equal to the sample of 150 pixels.
1.1.1.1 color model classifier training
Red channel and green channel to all positive samples in training sample set D it is for statistical analysis, specifically, statistics institute
There are positive sample (the totally 5000) quantity of Green pixel and the quantity of red pixel.Obtain red pixel lower limit value (for
15).Red pixel number ratioR 1Greater than the ratio of green pixel numberR 2Degree, i.e.,R 2-R 1>0.13。
1.1.1.2 four branch's support vector machine classifiers training
By arest neighbors interpolation algorithm to it is remote, in, close, super close shot sample carry out rescaling, sample size adjusted is distinguished
For 30 × 30 pixels2, 75 × 75 pixels2, 125 × 125 pixels2, 170 × 170 pixels2, on the training set of this kind of scale
Training, towards the line support vector machine classifier of histogram feature, obtains four branch's support vector machine classifiers, such as based on gradient
Shown in Fig. 3, for it is remote, in, close, super four branches of close shot, gradient is specifically configured to towards histogram feature parameter: four points
The block size of branch is 2 × 2, and block moving step length is 1 cell element, and chest number is 9, and projection angle range is [0, π]；Four
The complaint size of a branch is respectively (3,4), (6,8), (8,8) and (8,8).Therefore four obtained characteristic dimensions of branch
Respectively 756,756,1024 and 1024.
The present invention first designs simple effective color feature fire, then extracts to the fire of different distance different
Gradient, can quick, accurate description fire target feature towards histogram feature.
1.2 on-line checking
Fire detection belongs to two classification problem, and the first step generates video fire hazard candidate region, second step, fire online classification.
1.2.1 candidate region generates
Visible light video image is read in frame by frame, candidate region all in the image is generated using method of the present invention, is had
Body implementing procedure is as follows:
Using neighborhood territory pixel come background model, prospect is detected by comparing background model and current input pixel value, is had
Body can be subdivided into three steps: the first step, initialize the background model of each pixel in single-frame images.I.e. for a picture
Vegetarian refreshments possesses the spatial characteristics of close pixel value in conjunction with neighbor pixel, the pixel value of the random neighborhood point for selecting it
As its model sample value.Second step carries out foreground object segmentation operation, specifically, background mould to subsequent image sequence
Type is that each background dot stores a sample set, and then whether each new pixel value and sample set multilevel iudge belong to background,
If approximate sample point number is greater than threshold value, then it is assumed that new pixel is background.Detection process is mainly determined by three parameters:
Sample set numberN20 are set as, threshold value is set as 2, the threshold value of closely located judgementRIt is set as 5.Third step, background model
It updates, pixel is counted, if some pixel is continuousNIt is secondary to be detected as prospect, then it is updated to background dot.
1.2.2 cascade sort
Cascade sort of the present invention is cascaded by color model classification and four branch's support vector cassifications, in cascade sort rank
Section, the candidate region that 2 classifiers being only cascaded are classified as fire is just fire in classifier Stage Classification, otherwise,
Current candidate region is cascaded classifier and is judged as non-fire.The implementation process of the classification of cascade 2 classifiers:
1) color model is classified
Extract the red ratio of candidate regionR1, green ratio featureR2 and red pixel numberC(image infrared channel is greater than
200 are considered red), ifR1>R2AndC> 15, then it is judged as fire in color model sorting phase, otherwise in color mould
Type sorting phase judges it for non-fire.
2) four branch's support vector cassification
To certain candidate region, according to its height, a support vector machines branch is chosen by Fig. 3, extracts gradient towards histogram spy
Sign, classifies according to the decision function of linear SVM shown in formula (1)；
(1)
Whereinw It is the weight vector of linear SVM,x Be candidate region gradient towards histogram feature vector,b
It is constant offset, input vectorx Response bef (x), iff (xSentence then in four branch's support vector cassification stages) > 0
Breaking, it is otherwise fire in the cascade sort stage, judges it for non-fire.
2. multiframe verifies reduction system false-alarm
Classify only with above-mentioned 2 grades of cascade classifiers, in fact it could happen that be segmented correctly but be cascaded classifier mistake point
The fire target of class, the present embodiment can reduce system false-alarm in this respect, specifically implement in two steps: the first step, to cascade point
The output of class device carries out multiframe verification, obtains the higher testing result of confidence level, i.e., by establishing fire target chained list, to fire
The number that target is essentially continuouslypIt is counted, when all having fire detection result in adjacent two frame, then continuous detection time
Numberp It is automatic to increase by 1, when picture fire continuously detects numberp When greater than 15, it is believed that obtain the higher fire of confidence level
Testing result.On this basis, chained list is emptied every 200 frames, in case EMS memory occupation is excessive.
Upper only the embodiment of the present invention, the present invention are not fully so limited, all institutes within principle of the present invention
Any modifications, equivalent replacements, and improvements etc. of work, should all be included in the protection scope of the present invention.
Claims (4)
1. a kind of forest fire recognition methods based on cascade sort, it is characterised in that one two-stage cascade classifier of building, and
Reduction system false-alarm is verified by multiframe, is specifically included:
Step 1 obtains candidate region based on dynamic background modeling algorithm；
Step 2, extracts fire color model feature to candidate region, carries out first order classification using color model classifier；
Step 3 extracts gradient towards histogram feature to candidate region, and the support vector machine classifier of four branches carries out third
Grade classification；
Step 4 carries out multiframe verification to classification results, continuous multiple frames is detected as with the situation of fire, just exports fire behavior alarm.
2. the forest fire recognition methods described in claim 1 based on cascade sort, which is characterized in that color mould described in step 2
Type feature refers to that the relationship by red channel and green channel to 5000 Zhang Hanhuo Target Photos is for statistical analysis resulting
Color model；The color model classifier refers to that it is (big that the red pixel number ratio of picture is greater than green pixel ratio
0.13) when, and red channel number of pixels is greater than 15, this grade of separator just will be considered that the candidate region for fiery classification.
3. the forest fire recognition methods described in claim 1 based on cascade sort, which is characterized in that four branch described in step 3
Support vector machines refer to, according to the first order export picture space, four mutually disjoint subsets are divided into according to the height of picture,
Classified using four support vector machines.
4. the forest fire recognition methods described in claim 1 based on cascade sort, which is characterized in that multiframe school described in step 4
Finger is tested in continuous multiple frames image, is continuously judged as that the classification results of fire count to classifier is cascaded, to count value
Testing result greater than 10 is just considered fire detection result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910456498.5A CN110211323A (en) | 2019-05-29 | 2019-05-29 | Forest fire recognition methods based on cascade sort |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910456498.5A CN110211323A (en) | 2019-05-29 | 2019-05-29 | Forest fire recognition methods based on cascade sort |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110211323A true CN110211323A (en) | 2019-09-06 |
Family
ID=67789390
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910456498.5A Pending CN110211323A (en) | 2019-05-29 | 2019-05-29 | Forest fire recognition methods based on cascade sort |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110211323A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111814617A (en) * | 2020-06-28 | 2020-10-23 | 智慧眼科技股份有限公司 | Video-based fire determination method and device, computer equipment and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101315667A (en) * | 2008-07-04 | 2008-12-03 | 南京航空航天大学 | Multi-characteristic synthetic recognition method for outdoor early fire disaster |
| CN101441712A (en) * | 2008-12-25 | 2009-05-27 | 北京中星微电子有限公司 | Flame video recognition method and fire hazard monitoring method and system |
| CN103198332A (en) * | 2012-12-14 | 2013-07-10 | 华南理工大学 | Real-time robust far infrared vehicle-mounted pedestrian detection method |
| CN103632158A (en) * | 2013-11-20 | 2014-03-12 | 北京环境特性研究所 | Forest fire prevention monitor method and forest fire prevention monitor system |
| CN104091171A (en) * | 2014-07-04 | 2014-10-08 | 华南理工大学 | Vehicle-mounted far infrared pedestrian detection system and method based on local features |
| KR20160091709A (en) * | 2015-01-26 | 2016-08-03 | 창원대학교 산학협력단 | Fire detection System and Method using Features of Spatio-temporal Video Blocks |
| US9443149B2 (en) * | 2013-07-23 | 2016-09-13 | Korea Advanced Institute Of Science And Technology | Method and apparatus for detecting smoke from image |
| CN108447219A (en) * | 2018-05-21 | 2018-08-24 | 中国计量大学 | System and method for detecting fire hazard based on video image |
| CN108765833A (en) * | 2018-04-27 | 2018-11-06 | 天津大学 | Based on the incipient fire detection algorithm for improving mixed Gaussian and machine learning |
| CN109191761A (en) * | 2018-09-06 | 2019-01-11 | 中国石油大学(华东) | A kind of method for recognizing fire disaster based on flame multiple features fusion |
-
2019
- 2019-05-29 CN CN201910456498.5A patent/CN110211323A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101315667A (en) * | 2008-07-04 | 2008-12-03 | 南京航空航天大学 | Multi-characteristic synthetic recognition method for outdoor early fire disaster |
| CN101441712A (en) * | 2008-12-25 | 2009-05-27 | 北京中星微电子有限公司 | Flame video recognition method and fire hazard monitoring method and system |
| CN103198332A (en) * | 2012-12-14 | 2013-07-10 | 华南理工大学 | Real-time robust far infrared vehicle-mounted pedestrian detection method |
| US9443149B2 (en) * | 2013-07-23 | 2016-09-13 | Korea Advanced Institute Of Science And Technology | Method and apparatus for detecting smoke from image |
| CN103632158A (en) * | 2013-11-20 | 2014-03-12 | 北京环境特性研究所 | Forest fire prevention monitor method and forest fire prevention monitor system |
| CN104091171A (en) * | 2014-07-04 | 2014-10-08 | 华南理工大学 | Vehicle-mounted far infrared pedestrian detection system and method based on local features |
| KR20160091709A (en) * | 2015-01-26 | 2016-08-03 | 창원대학교 산학협력단 | Fire detection System and Method using Features of Spatio-temporal Video Blocks |
| CN108765833A (en) * | 2018-04-27 | 2018-11-06 | 天津大学 | Based on the incipient fire detection algorithm for improving mixed Gaussian and machine learning |
| CN108447219A (en) * | 2018-05-21 | 2018-08-24 | 中国计量大学 | System and method for detecting fire hazard based on video image |
| CN109191761A (en) * | 2018-09-06 | 2019-01-11 | 中国石油大学(华东) | A kind of method for recognizing fire disaster based on flame multiple features fusion |
Non-Patent Citations (6)
| Title |
|---|
| 孙水发等: "《视频前景检测及其在水电工程监测中的应用》", 31 December 2014, 国防工业出版社 * |
| 庄家俊: ""面向汽车辅助驾驶的远红外行人检测关键技术研究"", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 曼陀罗彼岸花: ""ViBe算法原理和代码解析"", 《CSDN博客》 * |
| 汤琳等: ""基于HOG特征优化的夜间行人快速识别方案"", 《计算机工程》 * |
| 王丰斌等: ""一种高可靠性的头部校验夜间行人快速识别方案"", 《四川大学学报（自然科学版）》 * |
| 王国华: ""车载单目远红外行人检测关键技术研究"", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111814617A (en) * | 2020-06-28 | 2020-10-23 | 智慧眼科技股份有限公司 | Video-based fire determination method and device, computer equipment and storage medium |
| CN111814617B (en) * | 2020-06-28 | 2023-01-31 | 智慧眼科技股份有限公司 | Fire determination method and device based on video, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108052859B (en) | Abnormal behavior detection method, system and device based on clustering optical flow characteristics | |
| CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
| CN105787472B (en) | A kind of anomaly detection method based on the study of space-time laplacian eigenmaps | |
| CN101389004B (en) | Moving target classification method based on on-line study | |
| CN110135296A (en) | Airfield runway FOD detection method based on convolutional neural networks | |
| CN101142584A (en) | Method for facial features detection | |
| CN111222478A (en) | Construction site safety protection detection method and system | |
| CN102521565A (en) | Garment identification method and system for low-resolution video | |
| CN104134079A (en) | Vehicle license plate recognition method based on extremal regions and extreme learning machine | |
| CN109918971A (en) | Number detection method and device in monitor video | |
| CN112183472A (en) | Method for detecting whether test field personnel wear work clothes or not based on improved RetinaNet | |
| CN104268528A (en) | Method and device for detecting crowd gathered region | |
| CN112287827A (en) | Complex environment pedestrian mask wearing detection method and system based on intelligent lamp pole | |
| CN112163572A (en) | Method and device for identifying object | |
| Indrabayu et al. | Blob modification in counting vehicles using gaussian mixture models under heavy traffic | |
| CN109993061A (en) | A kind of human face detection and tracing method, system and terminal device | |
| CN102867183A (en) | Method and device for detecting littered objects of vehicle and intelligent traffic monitoring system | |
| CN107832730A (en) | Improve the method and face identification system of face recognition accuracy rate | |
| CN110827434A (en) | Community security patrol recording system and method for grid target identification | |
| CN113095301A (en) | Road occupation operation monitoring method, system and server | |
| CN113191273A (en) | Oil field well site video target detection and identification method and system based on neural network | |
| CN111339927A (en) | Intelligent work state identification system for personnel in electric power business hall | |
| CN109919068B (en) | Real-time monitoring method for adapting to crowd flow in dense scene based on video analysis | |
| Wang et al. | Vision-based highway traffic accident detection | |
| Arafah et al. | Face recognition system using Viola Jones, histograms of oriented gradients and multi-class support vector machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190906 |