CN107992799B - Preprocess method towards Smoke Detection application - Google Patents
Preprocess method towards Smoke Detection application Download PDFInfo
- Publication number
- CN107992799B CN107992799B CN201711102833.9A CN201711102833A CN107992799B CN 107992799 B CN107992799 B CN 107992799B CN 201711102833 A CN201711102833 A CN 201711102833A CN 107992799 B CN107992799 B CN 107992799B
- Authority
- CN
- China
- Prior art keywords
- image
- carries out
- rate
- input picture
- value
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to technical field of computer vision, provide one kind towards Smoke Detection and answer used preprocess method.This method is divided into three steps, and the first step carries out operation of presorting to input picture, by input picture point fogless, mist and fogless three classes.Second step carries out defogging enhancing operation according to luminance information for mist image.Third step carries out sky cutting operation for the image and mist image after defogging, removes sky areas.Picture by pretreatment operation carries out Smoke Detection again, improves accuracy and detection efficiency.
Description
Technical field
The invention belongs to technical field of computer vision, are related to a kind of preprocess method towards Smoke Detection application.
Background technique
As the frequent report of haze, the improvement of atmosphere pollution are extremely urgent in recent years for cities in China.Air contaminant treatment
It is related to every aspect from prevention to purification.Start with from atmosphere pollution prevention direction, by blazoning camera in rural area field, in real time
Field status information is acquired, and is detected and positioned smoke target automatically to acquired image by smoke detection system,
Phenomenon is burned to straw from village and is monitored alarm, can achieve the purpose for reducing pollution source.The present invention relies on stalk and burns
Smoke Detection project is burnt, pretreatment operation, including three classification of mist figure, defogging, sky segmentation streams are carried out to monitor video
Journey.Improve detection efficiency and accuracy.
Pretreatment operation is related to three big technologies, is mist figure sorting technique first.
Judge whether image has mist, is the premise of intelligent defogging.According to system requirements, the figure for getting video camera is needed
As being simply divided into three classes: thick fog, mist, fogless.
Existing mist figure classification method extracts feature in Duo Shicong image, and SVM is recycled to classify.For example, Hu Zhongyi
Deng[15]The mist figure automatic testing method of proposition is collected big using the spectrum signature of image and gray scale symbiosis square as characteristic of division
The outdoor colored fog free images of amount carry out simulation plus mist, and the SVM model of mist figure classification is obtained by training.Furthermore it is also possible to utilize
These features are equally put by the predicted characteristics of image, such as half inverse figure prediction fog-zone ratio, the mist concentration feature of transmission plot prediction
Training obtains disaggregated model in SVM classifier.
Followed by defogging technology.
Et al. He. propose the defogging technology based on dark channel prior principle, image is estimated by the dark channel diagram of image
Transmission plot, and Image restoration is constructed according to imaging model, recovers former fog free images.But the method for He. et al. is not
The brightness of image can be exported according to the automatic brightness adjustment of input picture.
Three types of technology is sky cutting techniques.
Sky areas detection and segmentation correlative study status are as follows:
Hongping Lietal is based on maximum variance between clusters (OSTU) and proposes a kind of cloud detection model.One is collected first
The apparent sky image of serial sky and cloud region, then calculate the color characteristics of these image corresponding regions, spectrum and
Textural characteristics choose optimal threshold using OSTU, construct cloud identification model finally according to feature.
Lei Qin etc.[12]A kind of sky partitioning algorithm is proposed based on sea image defogging, using average drifting method to sea level chart
As being split, the zone boundary information of image carries out postsearch screening with the method for insertion confidence level edge detection after segmentation, most
The morphological operations such as the image expansion and corrosion that obtain afterwards extract sky areas.
According to McCartney imaging model, scene depth is bigger, and the value of transmission plot is smaller.When the value of transmission plot levels off to 0
When, the pixel value of image is equivalent to the value A of atmosphere light, and day vacancy distance be scene depth be it is infinitely great, just meet saturating
It penetrates figure and levels off to 0.According to this principle, sky areas is identified the problem of being converted into the value for seeking atmosphere light A by Bo Jiangetal.
For land Algorithms of Robots Navigation System, YehuShenet al proposes a kind of single image sky detection method, based on figure
The gradient magnitude information of picture, the sky segmentation threshold optimized according to energy function, then by contextual information etc., exclude nothing
Sky image refines partitioning boundary.
Smoke detection system requires real-time detection smog, therefore the requirement of sky partitioning portion is quickly succinct, the present invention is based on
Algorithm above proposes a kind of succinct, quickly and effectively sky partitioning algorithm for the project demands.
Summary of the invention
Present invention aim to address the shortcomings of smoke detection system.In smoke detection system, processing target is
Image, processing intent are to detect smoke target in the picture.To obtain accurately detailed information from image, and carry out
Judgement, tested altimetric image need target clear, therefore first carry out pretreatment to the video flowing that video camera captures and improve image matter
Amount carries out detection again and is necessary.Pretreatment operation of the invention mainly solves the problems, such as three classes: 1. according to analysis, in detection process
In, influencing picture quality principal element is that image caused by haze weather obscures, contrast declines.Therefore, existing go is being studied
On the basis of mist technology, the defogging technology based on dark channel prior principle is selected, based on the practical application of Smoke Detection in fact
It is analyzed with property, and with regard to encountering in Smoke Detection the problem of is improved.The DCP defogging algorithm for proposing optimization, to be detected
Picture carries out defogging and enhancing operation.2. during Smoke Detection, since puff profile and cloud form are very much like, meeting
Cause cloud erroneous detection.To avoid cloud erroneous detection, removal sky areas process is added in pretreatment operation.3. when mist excessive concentration, nothing
Method distinguishes smoke and fog, and system is caused frequently to be reported by mistake.Therefore sort operation is added in pretreatment operation to be divided into image to be detected
Fogless, mist and thick fog three classes, fogless direct detection are detected again after mist defogging, and thick fog does not detect, and improve detection efficiency and just
True rate, and prevent from frequently reporting staff's bring trouble by mistake.
Classify first to image to be detected: if thick fog class, then system does not detect, and prompting needs in artificial detection video flowing
With the presence or absence of crop straw burning phenomenon;If mist, then Smoke Detection is carried out again after carrying out defogging processing to image;It is straight if fogless
It connects and image is detected.After detecting smog, secondary detection is carried out after image is carried out sky dividing processing, to exclude cloud
Piece erroneous detection.
Beneficial effects of the present invention: the picture by pretreatment operation carries out Smoke Detection again, improve accuracy and
Detection efficiency.
Detailed description of the invention
Fig. 1 is a kind of flow chart of preprocess method towards Smoke Detection application.
Fig. 2 is the image divided after sky.
Specific embodiment
The present invention is based on opencv2.4.9, by frame image in real-time interception monitor video, are located in advance to interception image
Reason operation, then send Smoke Detection program back to and carry out Smoke Detection.Three steps of pretreatment operation point, steps are as follows:
The first step carries out sort operation to input picture.Purpose is divided into fogless, mist and thick fog three classes.
Image used in the step is divided into two classes: one kind is input picture to be sorted, and one kind is for training pattern
Training image.
Step1: four features: S are extracted to training imagemean、Srate、Tmean、Trate。
Wherein Smean、SrateIt is to be extracted according to half inverse figure of training image.
Obtain half inverse figure of training image
Wherein c represents channel value, c ∈ { r, g, b }.X is location of pixels.
Then half inverse figure is found out respectivelyWith the tone illustration of original training imageTone illustration subtracts each other,
Obtain Isi-src:
Thus S is obtainedmean:
Wherein N represents total number-of-pixels.
Second feature SrateIt is defined as follows:
Wherein r (Isi-src(x)) it is defined as follows:
Wherein τ is the threshold value determined according to actual conditions.
Third feature is obtained from the transmission plot of training image.Third feature TmeanIt is defined as follows:
WhereinFor the transmission plot of training image.
4th feature TrateIt is defined as follows:
WhereinAre as follows:
tlIt is a predefined threshold value.Less than tlPixel value mistiness degree with higher.TrateExpression value is less than tl's
Pixel proportion in whole picture figure, while also providing the information of the proportion in the overall situation of the region with suitable concentration.
tlDetermination be the crucial step of comparison.Pass through tlAdjustment, can be TrateSubstantially it is divided in three active domains
It is interior.We determine t by calculating the data of training imagel.We enable tlThe step for being 0.05 with step-length in [0.2,0.6] range
Amplitude variation, it is therefore an objective to find a t 'lIt can make TrateValue more clearly falls in three and intersects in lesser domain.
It is determined by experiment, as t 'lWhen=0.35, Trate[0.01,0.4] is substantially fallen in, [0.4,0.8], [0.8,1] three
In a range domain.Therefore we select t 'l=0.35.
Step2: four features of training image (including thick fog figure, mist figure, fogless figure) are extracted.SVM is trained.
Step3: aforementioned four feature is extracted to input image to be detected, is put into trained SVM classifier and is divided
Class.
Second step carries out defogging to the image that classification is " mist " by classification.The present invention is based on He. et al. to mention
The defogging algorithm based on dark channel prior principle out, proposes a kind of optimization method according to the automatic calculating parameter of input picture.
Step1: the air light value A of input picture is estimated.
Step2: parameter ω is estimated according to input picture:
ω=0.0017 × V2-0.1217×V+4.6028
Wherein V is V channel average value of the input picture in HSV space.
Step3: transmission plot is calculated
Wherein c represents channel value, c ∈ { r, g, b }.IcFor input picture.
Step4: restore image:
Wherein t0For threshold value, preventing denominator is 0.J is the image after restoring, i.e. fog free images.
Third step carries out Smoke Detection for the image after the completion of fog free images and defogging first, if smokelessly, exporting,
If there is cigarette, secondary detection is carried out again after carrying out sky segmentation.
The used sky partitioning algorithm of the present invention is based on dark channel prior principle.Specific step is as follows:
Step1: processing is exposed to image, increases overall exposing degree, allows sky areas to brighten, while reducing a day dead zone
Domain contrast.
Step2: to image I after exposureInputSeek dark channel diagram Idark;
Step3: to dark channel diagram IdarkIt carries out denoising and obtains IG;
Step4: with Sobel boundary operator to IGGradient magnitude figure is sought, and expansive working, output are carried out to gradient magnitude
Igrad;
Step5: setting heaven line function b (x), is from left to right scanned from top to bottom to gradient magnitude figure, scans each column
When, it encounters boundary point and y value at this time is stored in corresponding b (x);
Step6: according to heaven line function b (x) to input picture IInputIt is handled, boundary line above section triple channel in day is equal
It is assigned a value of zero, it is constant below day boundary line;
Step7: image I of the output through over-segmentationSegmented.
Image after over-segmentation sky is fed again into smoke detection system and carries out secondary detection.
Claims (1)
1. a kind of image pre-processing method towards Smoke Detection application, which is characterized in that steps are as follows:
The first step carries out sort operation to input picture, is divided into fogless, mist and thick fog three classes;
Image used in the step is divided into two classes: one kind is input picture to be sorted, another kind of for for training pattern
Training image;
Step1: four features: S are extracted to training imagemean、Srate、TmeanAnd Trate;
Wherein, SmeanAnd SrateIt is to be extracted according to half inverse figure of training image;
Obtain half inverse figure of training image
Wherein, c represents channel value, c ∈ { r, g, b };X is location of pixels;
Then half inverse figure is found out respectivelyWith the tone illustration of original training imageAnd Ihue, the two subtracts each other, obtains
Isi-src:
Thus S is obtainedmean:
Wherein N represents total number-of-pixels;
Second feature SrateIt is defined as follows:
Wherein, r (Isi-src(x)) it is defined as follows:
Wherein, τ is the threshold value determined according to actual conditions;
Third feature is obtained from the transmission plot of training image;Third feature TmeanIt is defined as follows:
Wherein,For the transmission plot of training image;
4th feature TrateIt is defined as follows:
WhereinAre as follows:
tlIt is a predefined threshold value, is less than tlPixel value mistiness degree with higher;TrateExpression value is less than tlPixel
The proportion in whole picture figure, while the information with region proportion in the overall situation of suitable concentration being also provided;
Pass through tlAdjustment, TrateIt is divided in three active domains, the data by calculating training image determine tl;Enable tlIn
The step change for being 0.05 with step-length in [0.2,0.6] range, it is therefore an objective to find a t 'lMake TrateValue is more clearly fallen in
Three intersect in lesser domain;
It is determined by experiment, as t 'lWhen=0.35, Trate[0.01,0.4] is fallen in, [0.4,0.8], [0.8,1] three range domains
It is interior, select t 'l=0.35;
Step2: extracting four features of training image, and training image includes thick fog figure, mist figure and fogless figure, carries out to SVM
Training;
Step3: aforementioned four feature is extracted to input image to be detected, is put into trained SVM classifier and classifies;
Second step carries out defogging to the image that classification is " mist " by classification;Using based on dark channel prior principle
Defogging algorithm proposes a kind of optimization method according to the automatic calculating parameter of input picture, the specific steps are as follows:
Step1: the air light value A of input picture is estimated;
Step2: parameter ω is estimated according to input picture:
ω=0.0017 × V2-0.1217×V+4.6028
Wherein, V is V channel average value of the input picture in HSV space;
Step3: transmission plot is calculated
Wherein, c represents channel value, c ∈ { r, g, b };IcFor input picture;
Step4: restore image:
Wherein, t0For threshold value, preventing denominator is 0;J is the image after restoring, i.e. fog free images;
Third step carries out Smoke Detection for the image after the completion of fog free images and defogging, if smokelessly, exporting, if having first
Cigarette carries out secondary detection after then carrying out sky segmentation again;Used sky partitioning algorithm is based on dark channel prior principle, specific to walk
It is rapid as follows:
Step1: processing is exposed to image, increases overall exposing degree, allows sky areas to brighten, while reducing sky areas pair
Degree of ratio;
Step2: to image I after exposureInputSeek dark channel diagram Idark;
Step3: to dark channel diagram IdarkIt carries out denoising and obtains IG;
Step4: with Sobel boundary operator to IGGradient magnitude figure is sought, and expansive working is carried out to gradient magnitude, exports Igrad;
Step5: setting heaven line function b (x), is scanned from top to bottom, from left to right to gradient magnitude figure, when scanning each column,
It encounters boundary point y value at this time is stored in corresponding b (x);
Step6: according to heaven line function b (x) to input picture IInputIt is handled, the equal assignment of day boundary line above section triple channel
It is zero, it is constant below day boundary line;
Step7: image I of the output through over-segmentationSegmented。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711102833.9A CN107992799B (en) | 2017-11-10 | 2017-11-10 | Preprocess method towards Smoke Detection application |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711102833.9A CN107992799B (en) | 2017-11-10 | 2017-11-10 | Preprocess method towards Smoke Detection application |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107992799A CN107992799A (en) | 2018-05-04 |
CN107992799B true CN107992799B (en) | 2019-11-08 |
Family
ID=62030471
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711102833.9A Active CN107992799B (en) | 2017-11-10 | 2017-11-10 | Preprocess method towards Smoke Detection application |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107992799B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111192332B (en) * | 2018-11-14 | 2023-04-07 | 九阳股份有限公司 | Smoke detection-based smoke machine control method and smoke machine |
CN111928304B (en) * | 2019-05-13 | 2022-03-29 | 青岛海尔智能技术研发有限公司 | Oil smoke concentration identification method and device and range hood |
CN114821151A (en) * | 2022-03-23 | 2022-07-29 | 苏州英特模汽车科技有限公司 | Training method of smoke detection model, smoke detection device and system |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105574830A (en) * | 2016-02-04 | 2016-05-11 | 沈阳工业大学 | Low-quality image enhancement method under extreme weather conditions |
CN105976338A (en) * | 2016-05-12 | 2016-09-28 | 山东大学 | Dark channel prior defogging method based on sky recognition and segmentation |
CN106548463A (en) * | 2016-10-28 | 2017-03-29 | 大连理工大学 | Based on dark and the sea fog image automatic defogging method and system of Retinex |
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 |
CN106991400A (en) * | 2017-04-05 | 2017-07-28 | 北京中燕信息技术有限公司 | A kind of fire hazard smoke detecting method and device |
CN107301624A (en) * | 2017-06-05 | 2017-10-27 | 天津大学 | The convolutional neural networks defogging algorithm pre-processed based on region division and thick fog |
-
2017
- 2017-11-10 CN CN201711102833.9A patent/CN107992799B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105574830A (en) * | 2016-02-04 | 2016-05-11 | 沈阳工业大学 | Low-quality image enhancement method under extreme weather conditions |
CN105976338A (en) * | 2016-05-12 | 2016-09-28 | 山东大学 | Dark channel prior defogging method based on sky recognition and segmentation |
CN106548463A (en) * | 2016-10-28 | 2017-03-29 | 大连理工大学 | Based on dark and the sea fog image automatic defogging method and system of Retinex |
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 |
CN106991400A (en) * | 2017-04-05 | 2017-07-28 | 北京中燕信息技术有限公司 | A kind of fire hazard smoke detecting method and device |
CN107301624A (en) * | 2017-06-05 | 2017-10-27 | 天津大学 | The convolutional neural networks defogging algorithm pre-processed based on region division and thick fog |
Non-Patent Citations (3)
Title |
---|
Guided Image Filtering;He Kaiming,Sun Jian,Tang Xiaoou;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20131231;第35卷(第6期);第1397-1409页 * |
基于天空区域分割的单幅海面图像去雾方法;雷琴,施朝健,陈婷婷;《计算机工程》;20150515;第41卷(第5期);第237-242页 * |
基于物理模型的单幅雾天图像去雾方法;旷文龙,叶青,宋赞;《计算机与现代化》;20140326(第3期);第123-130页 * |
Also Published As
Publication number | Publication date |
---|---|
CN107992799A (en) | 2018-05-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10565479B1 (en) | Identifying and excluding blurred areas of images of stained tissue to improve cancer scoring | |
CN106651872B (en) | Pavement crack identification method and system based on Prewitt operator | |
Cheng et al. | Automatic pavement distress detection system | |
CN109753890B (en) | Intelligent recognition and sensing method for road surface garbage and implementation device thereof | |
CN104951784B (en) | A kind of vehicle is unlicensed and license plate shading real-time detection method | |
Nienaber et al. | Detecting potholes using simple image processing techniques and real-world footage | |
US10489903B2 (en) | Stepwise refinement detection method for pavement cracks | |
CN112434695B (en) | Upper pull rod fault detection method based on deep learning | |
CN109635758B (en) | Intelligent building site video-based safety belt wearing detection method for aerial work personnel | |
CN108564030A (en) | Classifier training method and apparatus towards vehicle-mounted thermal imaging pedestrian detection | |
CN115082683A (en) | Injection molding defect detection method based on image processing | |
CN109086687A (en) | The traffic sign recognition method of HOG-MBLBP fusion feature based on PCA dimensionality reduction | |
Li et al. | Research on vehicle license plate location based on neural networks | |
CN107992799B (en) | Preprocess method towards Smoke Detection application | |
CN104021574A (en) | Method for automatically identifying pavement diseases | |
CN103903018A (en) | Method and system for positioning license plate in complex scene | |
CN107909012A (en) | A kind of real-time vehicle tracking detection method and device based on disparity map | |
CN105893970A (en) | Nighttime road vehicle detection method based on luminance variance characteristics | |
CN112560807A (en) | Crowd gathering detection method based on human head detection | |
Ibrahim et al. | Characterization of cracking in pavement distress using image processing techniques and k-nearest neighbour | |
CN115294377A (en) | System and method for identifying road cracks | |
CN112508911A (en) | Rail joint touch net suspension support component crack detection system based on inspection robot and detection method thereof | |
CN113673614B (en) | Metro tunnel foreign matter intrusion detection device and method based on machine vision | |
CN110991245A (en) | Real-time smoke detection method based on deep learning and optical flow method | |
CN107038690A (en) | A kind of motion shadow removal method based on multi-feature fusion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |