CN105844295A - Video smog fine classification method based on color model and motion characteristics - Google Patents
Video smog fine classification method based on color model and motion characteristics Download PDFInfo
- Publication number
- CN105844295A CN105844295A CN201610162064.0A CN201610162064A CN105844295A CN 105844295 A CN105844295 A CN 105844295A CN 201610162064 A CN201610162064 A CN 201610162064A CN 105844295 A CN105844295 A CN 105844295A
- Authority
- CN
- China
- Prior art keywords
- smoke
- color model
- suspicious
- video
- smog
- 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.)
- Granted
Links
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- 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
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a video smog fine classification method based on a color model and motion characteristics, comprising steps of differentiating and constructing a color model on the basis of a gauss mixing model and a background, detecting a motion pixel point and performing initial screening, obtaining a motion object connected region, extracting a motion changing characteristic, setting a threshold and determining whether a suspicious smoke area exists, performing pre-processing on the suspicious smoke area, extracting a SIFT characteristic on the basis of an image block, combining with an SVM optimization random forest algorithm, performing training based on the suspicious smoke area image block and thus realizing the fine classification of the fire hazard smoke, the cigarette smoke, the water vapor, etc. The video smog fine classification method based on color model and motion characteristics performs video smoke detection based on the color model and the motion characteristic and by targeting the fine classification, realizes real-time fast smoke detection, effectively eliminates the interference having similar effect interference sources with the fire hazard smoke, and improves the detection efficiency and accuracy.
Description
Technical field
The present invention relates to Computer Image Processing, Intelligent Recognition field, particularly relate to a kind of based on color model and motion
The video smoke sophisticated category method of feature.
Background technology
In recent years, along with video monitoring system is popularized in city and fire-fighting with laying special stress on protecting unit, artificial intelligence and pattern
Identification technology constantly develops, and fire detection method based on video more and more comes into one's own.Video fire hazard detection method has
Efficiently, in real time, the advantage such as intelligence, low cost, convenience, realize fire detection in real time, for earlier fire being detected thus
Bigger loss is avoided to have the most positive meaning.
Video fire hazard detection method, mainly includes flame detecting method and smog detection method.Smoke detection method is to visit
The smog that fire detecting calamity produces detects fire, it is judged that whether also have smoke region in video image.General fire early period of origination
Experience is glowed the stage, and smoldering in this stage is the main sign of fire.Obviously, this method is it appeared that the rank of glowing of fire
Section, can detect fire earlier, can reduce the loss of people's property, therefore video fire hazard based on smog detection more
Technology more can meet the demand in market, also becomes study hotspot instantly.
During realizing the present invention, although inventor finds prior art at least to there is problems in that at present and has
The correlational study of many video smoke detections, and achieve certain achievement, but affected by technology maturity and cost, based on
The video fire hazard detection technique of smog can't be used widely.Existing algorithm is to there being similar effect to do with fire hazard aerosol fog
Disturbing source (such as smoke from cigarette, mosquito incense smog, water vapour etc.), there is no algorithm at present can realize preferably getting rid of effect, and this
Interference source cannot be got rid of, and that just cannot promote accuracy rate the most further and reduce false alert rate；The ring that background is complex
Smoke region is extracted and detection by border, and impact is relatively big, and existing algorithm effect when processing this background is poor；More existing possess
The preferably algorithm required time complexity of detection performance can not fully meet the real time handling requirement of monitoring system, thus
Lose the most important meaning of video smoke detection technique.
Summary of the invention
In view of the above problems, it is proposed that the present invention in case provide one overcome the problems referred to above or at least in part solve or
Person slows down the video smoke sophisticated category method based on color model Yu motion feature of the problems referred to above.
The technical solution used in the present invention is: a kind of video smoke sophisticated category side based on color model Yu motion feature
Method, the method step is as follows:
Step (1), obtain monitoring video flow in real time, the motor image vegetarian refreshments in detection current video frame, and build color mould
Type, carries out Preliminary screening to motor image vegetarian refreshments；
Step (2), acquisition connection moving region, and calculate its kinematic parameter, and obtain suspicious smog according to threshold value differentiation
Region；
Step (3), suspicious smoke region carries out generating after Image semantic classification image block, extract SIFT feature, then foundation
The vision code book given birth to is mapped as rectangular histogram, obtains the characteristic vector of image block；
Step (4), carry out features training based on image block, in conjunction with svm classifier algorithm, the node of random forests algorithm is divided
The process of splitting is optimized, the ballot parallelization to decision tree, thus realizes the sophisticated category of all kinds of smog.
Wherein, described step (1) farther includes:
Step (a), for frame of video, based on gauss hybrid models and background difference algorithm, carry out moving target respectively
Extract, result is merged, to tackle different environment；
Step (b), structure color model, carry out principium identification and retain suspicious smog pixel motor image vegetarian refreshments.
Described step (b) farther includes:
The translucence of smog makes smog its color when the thinnest very nearly the same with background object color, therefore first
First the moving region detected is carried out the contrast differentiation of foreground and background colour: (i j), calculates foreground picture for pixel x
With the difference of Background respective channel, obtain diffR、diffG、diffBIf meeting:
diffR± α=diffG± α=diffB±α
Then think that this pixel in foreground picture is suspicious smog pixel, continue the most according to the following steps to differentiate: be not subject to
If the interference of background object color, smog color is generally canescence, and under rgb space, triple channel value is closely, and the biggest
In a certain threshold value, the most satisfied:
R ± β=G ± β=B ± β, R (G, B) ＞ γ
Then think that this pixel is suspicious smog pixel, otherwise exclude.
Wherein, described step (2) farther includes:
Step (a), take a series of region contour processing method, obtain connection moving region；
Step (b), the connection moving region obtained according to process, calculate its kinematic parameter, and differentiate according to threshold value,
Obtain suspicious smoke region.
Described step (b) farther includes:
Select motion target area area change rate as a kinematic parameter: to calculate one-shot change rate every N frame sampling, that
At t0Time, the smog area change rate sampled between adjacent two two field pictures is represented by:
Select motion target area geometry barycenter rate of change as a kinematic parameter: at t0Time, vertical direction barycenter becomes
The meansigma methods of rate is represented by:
Wherein, described step (3) farther includes:
Step (a), vision code book generation phase, by the pretreatment of training image collection, extract SIFT feature, Jin Erju
Class generates vision code book；
Step (b), the present frame suspicious smoke region character representation stage, by its pretreatment, generation image block, extraction
SIFT feature, obtains the map histograms of image block according to vision code book, then obtain suspicious smoke region characteristic of correspondence to
Amount.
Described step (b) farther includes:
Remember that suspicious smoke region generates P image block after pretreatment, carry out SIFT feature extraction, obtain image block
SIFT feature vector, calculate each SIFT feature in image block to the Euclidean between vision word under its vision code book away from
From, and be mapped as closest vision word, will the corresponding word frequency+1 of this vision word, after completing this step, often
One image block just maps for a rectangular histogram corresponding with vision word sequence, the suspicious smoke region of final testing image
Each image block be represented as the characteristic vector of K dimension, then this image has just had P K dimensional feature vector.
Wherein, described step (4) farther includes:
Step (a), in the fission process of node, select strong two-value grader SVM, strengthen the classification of decision tree with this
Reliability；
Step (b), in order to improve the prediction classification speed of random forests algorithm, the present invention is pre-each decision tree of random forest
During surveying classification results, use the implementation method of multi-threaded parallel.
Described step (a) farther includes:
Traditional random forests algorithm is during generating decision tree, and the division for decision tree nodes uses weak
Grader, directly sets threshold value, and the size of eigenvalue Yu threshold value by comparing current selected feature, by the sample in present node
Originally left and right child node it is divided into.The present invention intends selecting strong two-value grader SVM in the fission process of node, strengthens decision-making with this
The reliability of the classification of tree.Trained each intermediate node in decision tree by SVM classifier, obtain by this way determines
Plan tree is strongr.
Present invention major advantage compared with the conventional method is:
A kind of based on color model Yu motion feature the video smoke sophisticated category method that the present invention provides, by optimizing
Moving object detection algorithm, builds suitable smog color model, in conjunction with the most careful regional processing method, obtains motion even
Motion change feature is also extracted in logical region, and merges random forest and carry out sophisticated category with SVM, it is achieved that smog inspection real-time
Survey, effectively eliminate and have the impact of similar effect interference source with fire hazard aerosol fog, improve efficiency and the accuracy rate of detection.
Described above is only the general introduction of technical solution of the present invention, in order to better understand the technological means of the present invention,
And can be practiced according to the content of description, and in order to allow above and other objects of the present invention, the feature and advantage can
Become apparent, below especially exemplified by the detailed description of the invention of the present invention.
Accompanying drawing explanation
By reading the detailed description of hereafter preferred implementation, various other advantage and benefit common for this area
Technical staff will be clear from understanding.Accompanying drawing is only used for illustrating the purpose of preferred implementation, and is not considered as the present invention
Restriction.In the accompanying drawings:
Fig. 1 shows the basic flow sheet of the present invention；
Fig. 2 shows the flow chart of the smoke region detection algorithm of the present invention；
Fig. 3 shows the flow chart of the image feature representation of the present invention.
Detailed description of the invention
Below with reference to the accompanying drawings, embodiments of the invention are described in detail.
First the method overall process of the present invention is illustrated.
Fig. 1 shows the basic flow sheet of example of the present invention, is therefrom expressly understood that four committed steps of the present invention:
(1) monitoring video flow is obtained in real time, the motor image vegetarian refreshments in detection current video frame, and build color model pixel is carried out
Preliminary screening；(2) obtain connection moving region, and calculate its kinematic parameter, and obtain suspicious smoke region according to threshold value differentiation；
(3) generate image block after suspicious smoke region is carried out Image semantic classification, extract SIFT feature, then according to having given birth to the vision arrived
Code book is mapped as rectangular histogram, obtains the characteristic vector of image block；(4) carry out features training based on image block, divide in conjunction with SVM
The node split process of random forests algorithm is optimized by class algorithm, the ballot parallelization to decision tree, thus realizes all kinds of
The sophisticated category of smog.
Fig. 2 shows the motor image vegetarian refreshments detection algorithm of the embodiment of the present invention and the flow process of suspicious smoke region detection algorithm
Figure.
As can be seen from the figure the first step of motor image vegetarian refreshments detection algorithm is exactly: for frame of video, mix based on Gauss
Matched moulds type and background difference algorithm, carry out the extraction of moving target respectively, merge result, to tackle different environment.
Using existing gauss hybrid models that current image frame is carried out background modeling, test parameters chooses applicable parameter
Extract motor image vegetarian refreshments；In view of gauss hybrid models moving direction fixed and the recognition effect of the object the most slowly of moving
Poor, the present invention is also adopted by background subtraction to extract motor image vegetarian refreshments.Above two moving object detection algorithm should in reality
The environmental condition being each suitable for is had in, in conjunction with the moving parameter information of subsequent acquisition, and monitoring shooting time point, it is two
The result that kind detection method obtains arranges different weights.
In this step, it is a more crucial step that background frames updates, and selectes static background frame in conjunction with frame differential method,
If the pixel number in the difference image of N continuous consecutive frame is less than specifying threshold value, that selected present frame is updated to background frames.Prospect
Background pixel point judges: use otsu maximum variance between clusters to generate the optimal threshold of present frame, more than the pixel of this threshold value
It is confirmed as foreground pixel point.
Fig. 2 show motor image vegetarian refreshments detection algorithm followed by build color model, to moving object detection algorithm
The motor image vegetarian refreshments detected differentiates, retains suspicious smog pixel.
Finding in investigation, indoor smog color is generally Lycoperdon polymorphum Vitt white, needs when detecting smog pixel to build rationally
Smog color model.The translucence of smog makes smog its color when the thinnest differ nothing with background object color
Several, the contrast that first moving region detected carries out foreground and background colour differentiates:
For pixel x, (i j), calculates the difference of foreground picture and Background respective channel, obtains diffR、diffG、
diffBIf meeting:
diffR± α=diffG± α=diffB±α
Then think that this pixel in foreground picture is suspicious smog pixel, otherwise continue to differentiate according to (b) step.It is not subject to
If the interference of background object color, smog color is generally canescence, and under rgb space, triple channel value is closely, and the biggest
In a certain threshold value, the most satisfied:
R ± β=G ± β=B ± β
R (G, B) ＞ γ
Then think that this pixel is suspicious smog pixel, otherwise exclude.
Shown by Fig. 2, whether it is being at the beginning of suspicious smog pixel is carried out by color model to motor image vegetarian refreshments
After step judges, enter suspicious smoke region detection algorithm flow process, obtain connection moving region.
Suspicious smog pixel point set carrying out a series of process to obtain motion communication region, these comprise intermediate value in processing
Filtering, Morphological scale-space, profile process etc..Utilize medium filtering to remove noise, utilize morphologic opening and closing operation by phase
Neighbouring region merges, and on this basis, utilizes profile to process each region entirety, not only removes those profiles less than regulation
The connected region of threshold value, and those satisfactory profile convex closures or polynomial fitting curve are replaced, to extract
Complete motion communication region.
In Fig. 2 example, the second step of suspicious smoke region detection algorithm is: the connection motor region obtained according to process
Territory, calculates its kinematic parameter, and differentiates according to threshold value, obtain suspicious smoke region, provide for ensuing sophisticated category
More efficiently data.
Select motion target area area change rate as a kinematic parameter: target area area can pass through statistic mixed-state
The bianry image pixel number of rear acquisition is calculated.In view of the problem of amount of calculation in actual treatment, every N frame sampling meter
Calculate one-shot change rate, additionally remember that obtaining the area of target area from present frame is St(t=1,2 ..., i), then at t0Time, sampling
Smog area change rate between adjacent two two field pictures is represented by:
IfSet threshold value more than it, then can add up into statistic, as the factor being determined as smog.When tired
After adding and exceed its setting threshold value, it is believed that this frame sequence meets this motion feature of region area rate of change.
Select motion target area geometry barycenter rate of change as a kinematic parameter: t frame motion target area is corresponding
Centroid position (xo, yo) calculate such as formula:
In formula, DR (t) represents the motion target area of t frame.Then at t0Time, the meansigma methods of vertical direction barycenter rate of change
It is represented by:
In formula, H is picture frame vertical height, ifSet threshold value more than it, then can add up into statistic, as sentencing
Not Wei a factor of smog, when this statistic is cumulative set threshold value more than it after, it is believed that this sequence frame meets moving target district
Territory geometry barycenter vertical direction rate of change.
By the investigation to sophisticated category method, the present invention uses precise image based on code book (Codebook-based)
Classification mode, the most first becomes many local patch (can be split by rigidity each width image division in training data
Can also be based on SIFT critical point detection), so, each image is just represented, then to image by an a lot of patch
Each patch carries out feature extraction, and each feature is as a vocabulary, and image is then by the side of visual word bag Bag-of-Words
Method is indicated.Feature further according to all of training image determines the dictionary/code of one group of visual vocabulary comprising characteristics of image
This (Visual Dictionary or Codebook) distribution situation to its visual vocabulary of image study of each classification, so
Afterwards by statistical model or the middle semanteme of generative probabilistic model study category image, the last classification according to image is semantic
Constitute the classification determining image.
The present invention carries out feature extraction training based on image block, and in the feature of image block, the present invention uses image
The rotation of scaling, target, affine transformation, illumination variation etc. have the SIFT feature of invariance, set up based on image block SIFT
The BOW model (Bag of words model) of feature.Such as Fig. 3, the suspicious smoke region obtained for differentiation, the present invention is entered
The step of row character representation comprises two Main Stage: vision code book generation phase and input picture character representation stage.
At vision code book generation phase, to carrying out SIFT feature extraction after training image collection pretreatment, then carry out K-
Means clusters, and is gathered into some bunches that quantity is bigger, has higher similarity in bunch, and bunch between similarity relatively low, the most often
Individual bunch is counted as a vision word, and this vision word can be used for representing certain that the deposited key point within this race has jointly
Local pattern.Cluster is divided into K bunch, and cluster centre is also K, and the length of code book is the most just for K.The step for can be trained
The vision code book of image set.
In the input picture character representation stage, suspicious smoke region is carried out Image semantic classification, obtain the suspicious cigarette of alignment
Use the mode of even partition to generate image block behind territory, fog-zone again, use consistent fixed mesh, such each image to obtain
Image block all has same size.Remember that suspicious smoke region generates P image block after pretreatment, carry out SIFT feature extraction,
Obtain the SIFT feature vector of image block.Calculate each SIFT feature in image block to vision word under its vision code book it
Between Euclidean distance, and be mapped as closest vision word, will the corresponding word frequency+1 of this vision word, complete this
After one step, each image block just maps for a rectangular histogram corresponding with vision word sequence, and final testing image can
The each image block doubting smoke region is represented as the characteristic vector of a K dimension, then this image has just had P K dimensional feature
Vector.
By Fig. 3 example shown step, the present invention has carried out feature extraction expression based on code book to suspicious smoke region, with
This provides effective smoke characteristics model for sophisticated category.Next the present invention just uses random forests algorithm algorithm, based on figure
As block carries out features training.
Traditional random forests algorithm is during generating decision tree, and the division for decision tree nodes uses weak
Grader, directly sets threshold value, and the size of eigenvalue Yu threshold value by comparing current selected feature, by the sample in present node
Originally left and right child node it is divided into.The present invention selects strong two-value grader SVM in the fission process of node, strengthens decision tree with this
The reliability of classification.The each intermediate node in decision tree, the decision-making obtained by this way is trained by SVM classifier
Set strongr.
It addition, for a test sample, traditional random forests algorithm is the Voting Model using serial, the most often
Individual decision tree classifier votes to one by one this test sample, such as first decision tree classifier to test sample poll closing it
After, second decision tree classifier starts to vote this test sample again, the like.When serial ballot is to compare consumption
Between, particularly when Numerous and the test sample huge amount of decision tree classifier when.The most gloomy in order to improve
The prediction classification speed of woods algorithm, intends, during random forest each decision tree prediction classification results, using multithreading also herein
The implementation method of rowization.
In sum, a kind of based on color model Yu motion feature the video smoke sophisticated category side that the present invention provides
Method, by optimizing moving object detection algorithm, builds suitable smog color model, in conjunction with the most careful regional processing side
Method, obtains motion communication region and also extracts motion change feature, and merges random forest and carry out sophisticated category with SVM, it is achieved that
Smoke Detection real-time, effectively eliminates and has the impact of similar effect interference source with fire hazard aerosol fog, improve the effect of detection
Rate and accuracy rate.
Through the above description of the embodiments, those of ordinary skill in the art can be apparent from other advantages
And amendment.Embodiment of above is merely to illustrate the present invention, and not limitation of the present invention, about the common skill of technical field
Art personnel, without departing from the spirit and scope of the present invention, it is also possible to make a variety of changes and modification, the most all etc.
Same technical scheme falls within scope of the invention, and the scope of patent protection of the present invention should be defined by the claims.
Claims (9)
1. a video smoke sophisticated category method based on color model Yu motion feature, it is characterised in that step is as follows:
Step (1), obtain monitoring video flow in real time, the motor image vegetarian refreshments in detection current video frame, and build color model, right
Motor image vegetarian refreshments carries out Preliminary screening；
Step (2), acquisition connection moving region, and calculate its kinematic parameter, and obtain suspicious smoke region according to threshold value differentiation；
Step (3), suspicious smoke region carries out generating after Image semantic classification image block, extract SIFT feature, then according to
The vision code book given birth to is mapped as rectangular histogram, obtains the characteristic vector of image block；
Step (4), carry out features training based on image block, in conjunction with the svm classifier algorithm node split mistake to random forests algorithm
Journey is optimized, the ballot parallelization to decision tree, thus realizes the sophisticated category of all kinds of smog.
A kind of video smoke sophisticated category method based on color model Yu motion feature the most according to claim 1, its
Being characterised by, described step (1) farther includes:
Step (a), for frame of video, based on gauss hybrid models and background difference algorithm, carry out the extraction of moving target respectively,
Result is merged, to tackle different environment；
Step (b), structure color model, carry out principium identification and retain suspicious smog pixel motor image vegetarian refreshments.
A kind of video smoke sophisticated category method based on color model Yu motion feature the most according to claim 1, its
Being characterised by, described step (2) farther includes:
Step (a), take a series of region contour processing method, obtain connection moving region；
Step (b), the connection moving region obtained according to process, calculate its kinematic parameter, and differentiate according to threshold value, obtain
Suspicious smoke region.
A kind of video smoke sophisticated category method based on color model Yu motion feature the most according to claim 1, its
Being characterised by, described step (3) farther includes:
Step (a), vision code book generation phase, by the pretreatment of training image collection, extract SIFT feature, and then cluster is raw
Become vision code book；
Step (b), the present frame suspicious smoke region character representation stage, by its pretreatment, generation image block, extraction SIFT
Feature, obtains the map histograms of image block according to vision code book, then obtains suspicious smoke region characteristic of correspondence vector.
A kind of video smoke sophisticated category method based on color model Yu motion feature the most according to claim 1, its
Being characterised by, described step (4) farther includes:
Step (a), in the fission process of node, select strong two-value grader SVM, strengthen classification reliable of decision tree with this
Property；
Step (b), in order to improve the prediction classification speed of random forests algorithm, each decision tree of random forest predict classification results
During, use the implementation method of multi-threaded parallel.
A kind of video smoke sophisticated category method based on color model Yu motion feature the most according to claim 2, its
Being characterised by, described step (b) farther includes:
The translucence of smog makes smog its color when the thinnest very nearly the same with background object color, the most right
The moving region detected carries out the contrast differentiation of foreground and background colour: for pixel x, (i j), calculates foreground picture and the back of the body
The difference of scape figure respective channel, obtains diffR、diffG、diffBIf meeting:
diffR± α=diffG± α=diffB±α
Then think that this pixel in foreground picture is suspicious smog pixel, continue the most according to the following steps to differentiate: not by background
If object color interference, smog color is generally canescence, and under rgb space, triple channel value closely, and is all higher than certain
One threshold value, the most satisfied:
R ± β=G ± β=B ± β, R (G, B) ＞ γ
Then think that this pixel is suspicious smog pixel, otherwise exclude.
A kind of video smoke sophisticated category method based on color model Yu motion feature the most according to claim 3, its
Being characterised by, described step (b) farther includes:
Select motion target area area change rate as a kinematic parameter: to calculate one-shot change rate every N frame sampling, then
t0Time, the smog area change rate sampled between adjacent two two field pictures is represented by:
Select motion target area geometry barycenter rate of change as a kinematic parameter: at t0Time, vertical direction barycenter rate of change
Meansigma methods is represented by:
A kind of video smoke sophisticated category method based on color model Yu motion feature the most according to claim 4, its
Being characterised by, described step (b) farther includes:
Remembering that suspicious smoke region generates P image block after pretreatment, carry out SIFT feature extraction, the SIFT obtaining image block is special
Levy vector, the Euclidean distance between each SIFT feature in calculating image block to vision word under its vision code book, and incite somebody to action
It is mapped as closest vision word, will the corresponding word frequency+1 of this vision word, after completing this step, each image
Block just maps for a rectangular histogram corresponding with vision word sequence, each figure of the suspicious smoke region of final testing image
As block is represented as the characteristic vector of a K dimension, then this image has just had P K dimensional feature vector.
A kind of video smoke sophisticated category method based on color model Yu motion feature the most according to claim 5, its
Being characterised by, described step (a) farther includes:
Strong two-value grader SVM is selected in the fission process of node, the reliability of classification strengthening decision tree with this, pass through
SVM classifier trains each intermediate node in decision tree, and the decision tree obtained by this way is strongr.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610162064.0A CN105844295B (en) | 2016-03-21 | 2016-03-21 | A kind of video smoke sophisticated category method based on color model and motion feature |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610162064.0A CN105844295B (en) | 2016-03-21 | 2016-03-21 | A kind of video smoke sophisticated category method based on color model and motion feature |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105844295A true CN105844295A (en) | 2016-08-10 |
| CN105844295B CN105844295B (en) | 2019-09-27 |
Family
ID=56588151
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610162064.0A Active CN105844295B (en) | 2016-03-21 | 2016-03-21 | A kind of video smoke sophisticated category method based on color model and motion feature |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105844295B (en) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106355566A (en) * | 2016-10-10 | 2017-01-25 | 浙江树人大学 | Smoke and flame detection method applied to fixed camera dynamic video sequence |
| CN106446957A (en) * | 2016-10-08 | 2017-02-22 | 常熟理工学院 | Haze image classification method based on random forest |
| CN106599874A (en) * | 2016-12-26 | 2017-04-26 | 清华大学苏州汽车研究院（吴江） | Agglomerate fog detection method based on video analysis |
| CN106682635A (en) * | 2016-12-31 | 2017-05-17 | 中国科学技术大学 | Smoke detecting method based on random forest characteristic selection |
| CN107609470A (en) * | 2017-07-31 | 2018-01-19 | 成都信息工程大学 | The method of outdoor fire disaster early-stage smog video detection |
| CN108319964A (en) * | 2018-02-07 | 2018-07-24 | 嘉兴学院 | A kind of fire image recognition methods based on composite character and manifold learning |
| CN108428324A (en) * | 2018-04-28 | 2018-08-21 | 温州大学激光与光电智能制造研究院 | The detection device of smog in a kind of fire scenario based on convolutional network |
| 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 |
| CN109086647A (en) * | 2018-05-24 | 2018-12-25 | 北京飞搜科技有限公司 | Smog detection method and equipment |
| CN109740639A (en) * | 2018-12-15 | 2019-05-10 | 中国科学院深圳先进技术研究院 | A kind of wind and cloud satellite remote-sensing image cloud detection method of optic, system and electronic equipment |
| CN110120142A (en) * | 2018-02-07 | 2019-08-13 | 中国石油化工股份有限公司 | A kind of fire hazard aerosol fog video brainpower watch and control early warning system and method for early warning |
| CN110135374A (en) * | 2019-05-21 | 2019-08-16 | 吉林大学 | It is identified using image block characteristics and returns the fire hazard smoke detecting method classified |
| CN110503006A (en) * | 2019-07-29 | 2019-11-26 | 恒大智慧科技有限公司 | A kind of community smoking management-control method, system and its storage medium |
| CN110658118A (en) * | 2018-06-29 | 2020-01-07 | 九阳股份有限公司 | Cooking smoke detection method and range hood |
| CN111553405A (en) * | 2020-04-24 | 2020-08-18 | 青岛杰瑞工控技术有限公司 | Clustering fog recognition algorithm based on pixel density K-means |
| CN112163121A (en) * | 2020-11-03 | 2021-01-01 | 南京邦峰智能科技有限公司 | Video content information intelligent analysis processing method based on big data |
| CN112580396A (en) * | 2019-09-29 | 2021-03-30 | 东北林业大学 | Forest fire recognition method |
| CN112784642A (en) * | 2019-11-08 | 2021-05-11 | 杭州海康威视数字技术股份有限公司 | Vehicle detection method and device |
| CN113593170A (en) * | 2021-08-11 | 2021-11-02 | 张佳 | Intelligent early warning system based on remote smoke detection |
| CN116416200A (en) * | 2022-01-12 | 2023-07-11 | 中国消防救援学院 | Forest fire smoke detection method based on region matching |
| CN112784642B (en) * | 2019-11-08 | 2024-06-21 | 杭州海康威视数字技术股份有限公司 | Vehicle detection method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101908141A (en) * | 2010-08-04 | 2010-12-08 | 丁天 | Video smoke detection method based on mixed Gaussian model and morphological characteristics |
| CN102708651A (en) * | 2012-05-23 | 2012-10-03 | 无锡蓝天电子有限公司 | Image type smoke fire disaster detection method and system |
| US20130094699A1 (en) * | 2011-10-12 | 2013-04-18 | Industry Academic Cooperation Foundation Keimyung University | Forest fire smoke detection method using random forest classification |
| CN104408745A (en) * | 2014-11-18 | 2015-03-11 | 北京航空航天大学 | Real-time smog scene detection method based on video image |
-
2016
- 2016-03-21 CN CN201610162064.0A patent/CN105844295B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101908141A (en) * | 2010-08-04 | 2010-12-08 | 丁天 | Video smoke detection method based on mixed Gaussian model and morphological characteristics |
| US20130094699A1 (en) * | 2011-10-12 | 2013-04-18 | Industry Academic Cooperation Foundation Keimyung University | Forest fire smoke detection method using random forest classification |
| CN102708651A (en) * | 2012-05-23 | 2012-10-03 | 无锡蓝天电子有限公司 | Image type smoke fire disaster detection method and system |
| CN104408745A (en) * | 2014-11-18 | 2015-03-11 | 北京航空航天大学 | Real-time smog scene detection method based on video image |
Non-Patent Citations (2)
| Title |
|---|
| C YU 等: "Erratum to: Video Fire Smoke Detection Using Motion and Color Features", 《FIRE TECHNOLOGY》 * |
| 郑璐 等: "基于运动和颜色的视频烟雾检测算法", 《计算机工程与设计》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106446957A (en) * | 2016-10-08 | 2017-02-22 | 常熟理工学院 | Haze image classification method based on random forest |
| CN106446957B (en) * | 2016-10-08 | 2019-02-22 | 常熟理工学院 | A kind of haze image classification method based on random forest |
| CN106355566A (en) * | 2016-10-10 | 2017-01-25 | 浙江树人大学 | Smoke and flame detection method applied to fixed camera dynamic video sequence |
| CN106599874A (en) * | 2016-12-26 | 2017-04-26 | 清华大学苏州汽车研究院（吴江） | Agglomerate fog detection method based on video analysis |
| CN106599874B (en) * | 2016-12-26 | 2019-07-30 | 清华大学苏州汽车研究院（吴江） | A kind of group's mist detection method based on video analysis |
| CN106682635B (en) * | 2016-12-31 | 2019-01-04 | 中国科学技术大学 | A kind of smog detection method based on random forest feature selecting |
| CN106682635A (en) * | 2016-12-31 | 2017-05-17 | 中国科学技术大学 | Smoke detecting method based on random forest characteristic selection |
| CN107609470A (en) * | 2017-07-31 | 2018-01-19 | 成都信息工程大学 | The method of outdoor fire disaster early-stage smog video detection |
| CN107609470B (en) * | 2017-07-31 | 2020-09-01 | 成都信息工程大学 | Method for detecting early smoke of field fire by video |
| CN108319964A (en) * | 2018-02-07 | 2018-07-24 | 嘉兴学院 | A kind of fire image recognition methods based on composite character and manifold learning |
| CN110120142A (en) * | 2018-02-07 | 2019-08-13 | 中国石油化工股份有限公司 | A kind of fire hazard aerosol fog video brainpower watch and control early warning system and method for early warning |
| CN108319964B (en) * | 2018-02-07 | 2021-10-22 | 嘉兴学院 | Fire image recognition method based on mixed features and manifold learning |
| 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 |
| CN108428324A (en) * | 2018-04-28 | 2018-08-21 | 温州大学激光与光电智能制造研究院 | The detection device of smog in a kind of fire scenario based on convolutional network |
| CN109086647A (en) * | 2018-05-24 | 2018-12-25 | 北京飞搜科技有限公司 | Smog detection method and equipment |
| CN110658118B (en) * | 2018-06-29 | 2023-05-02 | 九阳股份有限公司 | Cooking smoke detection method and smoke machine |
| CN110658118A (en) * | 2018-06-29 | 2020-01-07 | 九阳股份有限公司 | Cooking smoke detection method and range hood |
| CN109740639A (en) * | 2018-12-15 | 2019-05-10 | 中国科学院深圳先进技术研究院 | A kind of wind and cloud satellite remote-sensing image cloud detection method of optic, system and electronic equipment |
| CN110135374A (en) * | 2019-05-21 | 2019-08-16 | 吉林大学 | It is identified using image block characteristics and returns the fire hazard smoke detecting method classified |
| CN110503006A (en) * | 2019-07-29 | 2019-11-26 | 恒大智慧科技有限公司 | A kind of community smoking management-control method, system and its storage medium |
| CN112580396A (en) * | 2019-09-29 | 2021-03-30 | 东北林业大学 | Forest fire recognition method |
| CN112784642A (en) * | 2019-11-08 | 2021-05-11 | 杭州海康威视数字技术股份有限公司 | Vehicle detection method and device |
| CN112784642B (en) * | 2019-11-08 | 2024-06-21 | 杭州海康威视数字技术股份有限公司 | Vehicle detection method and device |
| CN111553405A (en) * | 2020-04-24 | 2020-08-18 | 青岛杰瑞工控技术有限公司 | Clustering fog recognition algorithm based on pixel density K-means |
| CN111553405B (en) * | 2020-04-24 | 2023-08-18 | 青岛杰瑞工控技术有限公司 | Group fog recognition algorithm based on pixel density K-means clustering |
| CN112163121A (en) * | 2020-11-03 | 2021-01-01 | 南京邦峰智能科技有限公司 | Video content information intelligent analysis processing method based on big data |
| CN113593170A (en) * | 2021-08-11 | 2021-11-02 | 张佳 | Intelligent early warning system based on remote smoke detection |
| CN113593170B (en) * | 2021-08-11 | 2022-08-30 | 张佳 | Intelligent early warning system based on remote smoke detection |
| CN116416200A (en) * | 2022-01-12 | 2023-07-11 | 中国消防救援学院 | Forest fire smoke detection method based on region matching |
| CN116416200B (en) * | 2022-01-12 | 2023-10-27 | 中国消防救援学院 | Forest fire smoke detection method based on region matching |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105844295B (en) | 2019-09-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105844295A (en) | Video smog fine classification method based on color model and motion characteristics | |
| CN104008370B (en) | A kind of video face identification method | |
| CN104866829B (en) | A kind of across age face verification method based on feature learning | |
| CN109919122A (en) | A kind of timing behavioral value method based on 3D human body key point | |
| CN103886344B (en) | A kind of Image Fire Flame recognition methods | |
| CN102982344B (en) | Based on the support vector machine classification method merging various visual angles feature and many label informations simultaneously | |
| CN109359684A (en) | Fine granularity model recognizing method based on Weakly supervised positioning and subclass similarity measurement | |
| CN108268859A (en) | A kind of facial expression recognizing method based on deep learning | |
| CN105335966B (en) | Multiscale morphology image division method based on local homogeney index | |
| CN102629328B (en) | Probabilistic latent semantic model object image recognition method with fusion of significant characteristic of color | |
| CN106504064A (en) | Clothes classification based on depth convolutional neural networks recommends method and system with collocation | |
| CN106096602A (en) | A kind of Chinese licence plate recognition method based on convolutional neural networks | |
| CN104504362A (en) | Face detection method based on convolutional neural network | |
| CN108961675A (en) | Fall detection method based on convolutional neural networks | |
| CN112734775A (en) | Image annotation, image semantic segmentation and model training method and device | |
| CN106909902A (en) | A kind of remote sensing target detection method based on the notable model of improved stratification | |
| CN102682302A (en) | Human body posture identification method based on multi-characteristic fusion of key frame | |
| CN104616006B (en) | A kind of beard method for detecting human face towards monitor video | |
| CN109448015A (en) | Image based on notable figure fusion cooperates with dividing method | |
| CN109886161A (en) | A kind of road traffic index identification method based on possibility cluster and convolutional neural networks | |
| CN103886589A (en) | Goal-oriented automatic high-precision edge extraction method | |
| CN105046195A (en) | Human behavior identification method based on asymmetric generalized Gaussian distribution model (AGGD) | |
| CN113239801B (en) | Cross-domain action recognition method based on multi-scale feature learning and multi-level domain alignment | |
| CN102073841A (en) | Poor video detection method and device | |
| CN106934386A (en) | A kind of natural scene character detecting method and system based on from heuristic strategies |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 100080 c-95, 4th floor, 48 Haidian West Street, Haidian District, Beijing Patentee after: Beijing Zhongke ningkun Technology Co.,Ltd. Address before: 100080 c-95, 4th floor, 48 Haidian West Street, Haidian District, Beijing Patentee before: Beijing Zhongke Jinshu Technology Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210415 Address after: 100080 c-95, 4th floor, 48 Haidian West Street, Haidian District, Beijing Patentee after: Beijing Zhongke Jinshu Technology Co.,Ltd. Address before: 100191 No. 37, Haidian District, Beijing, Xueyuan Road Patentee before: BEIHANG University |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20220906 Address after: A14-5, 13th Floor, Building A, Building J1, Phase II, Innovation Industrial Park, No. 2800, Chuangxin Avenue, High-tech Zone, Hefei, China (Anhui) Pilot Free Trade Zone, 230088 Patentee after: Anhui aiguan Vision Technology Co.,Ltd. Address before: 100080 c-95, 4th floor, 48 Haidian West Street, Haidian District, Beijing Patentee before: Beijing Zhongke ningkun Technology Co.,Ltd. |