CN110070007A - Video smoke recognition methods, device, computer equipment and storage medium - Google Patents
Video smoke recognition methods, device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN110070007A CN110070007A CN201910266782.6A CN201910266782A CN110070007A CN 110070007 A CN110070007 A CN 110070007A CN 201910266782 A CN201910266782 A CN 201910266782A CN 110070007 A CN110070007 A CN 110070007A
- Authority
- CN
- China
- Prior art keywords
- video
- images
- recognized
- atmosphere
- transmittance
- 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
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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of video smoke recognition methods, comprising: extracts the images to be recognized in video to be identified, estimates atmosphere brightness and dark primary in the images to be recognized；According to the atmosphere brightness and the dark primary, the transmittance figure picture of the images to be recognized is obtained；Motion detection is carried out to the video that the transmittance figure picture is formed, obtains motion feature vector；The motion feature vector is inputted into default classifier, identification obtains the smoke region in the video to be identified.By transmitance image, location of smoke can be completely positioned independent of illumination variation, effectively improve the accuracy rate of smog positioning.
Description
Technical field
The present invention relates to field of image processings, set more particularly to a kind of video smoke recognition methods, device, computer
Standby and storage medium.
Background technique
Fire is one of most common major disaster of the mankind, but is compared with other natural calamities, danger brought by fire
Evil can reduce.In order to reduce injury caused by fire, most important one be exactly as early as possible to fire to
Give early warning.Fire early period because fuel generally can not full combustion, be often accompanied by the appearance of smog, and based on visible light
Video monitoring has become the important means that protection people live safely, therefore the Smoke Detection based on video is fire alarm
An important research and application direction.
For the Smoke Detection in video, many detection algorithms, the mainly view by extracting smog are currently existed
Feature, such as color, edge and texture are felt, but these features have relied on the raw information of visible light video, so seriously
Dependent on the variation of other movement things in external light source and video, cause algorithm accuracy not high.
Summary of the invention
The purpose of the present invention is to provide a kind of video smoke recognition methods, device, computer equipment and readable storage mediums
Matter can effectively improve the accuracy of Smoke Detection in video.
The purpose of the present invention is achieved through the following technical solutions:
A kind of video smoke recognition methods, which comprises
The images to be recognized in video to be identified is extracted, estimates atmosphere brightness and dark original in the images to be recognized
Color；
According to the atmosphere brightness and the dark primary, the transmittance figure picture of the images to be recognized is obtained；
Motion detection is carried out to the video that the transmittance figure picture is formed, obtains motion feature vector；
The motion feature vector is inputted into default classifier, identification obtains the smoke region in the video to be identified.
In one embodiment, the step of atmosphere brightness estimated in the images to be recognized, comprising:
Mini-value filtering is carried out to the Minimal color weight of the images to be recognized, exports minimum value pixel；
Edge extracting, which is carried out, with gray component of the default edge detection operator to the images to be recognized obtains edge graph
Picture, and block statistics are carried out to the edge image, obtain statistics ratio；
According to atmosphere brightness described in the minimum value pixel and the statistics ratio-dependent.
In one embodiment, described the step of block statistics are carried out to the edge image, obtain statistics ratio, packet
It includes:
Calculate neighborhood inward flange pixel number in the edge image centered on each pixel and neighborhood total pixel number it
Than obtaining the statistics ratio.
In one embodiment, the atmosphere according to the minimum value pixel and the statistics ratio-dependent is bright
The step of spending, comprising:
According to the candidate atmosphere light area in images to be recognized described in the minimum value pixel and the statistics ratio-dependent
Domain；
To the candidate atmosphere light zone marker connected component, the connected component of predeterminated position is chosen as atmosphere light area
Domain；
Using the max pixel value in the atmosphere light region as the atmosphere brightness.
In one embodiment, described according to the atmosphere brightness and the dark primary, obtain the images to be recognized
Transmittance figure as the step of, comprising:
The dark primary and atmosphere brightness are handled according to default atmospherical scattering model, obtain the first transmissivity
Function；
Using default bilateral filtering function, first transmittance function is handled, the second transmittance figure is obtained
Picture；
The transmittance figure picture of the images to be recognized is obtained according to the second transmittance figure picture.
In one embodiment, described that the motion feature vector is inputted default classifier, identification obtains described wait know
Before the step of smoke region in other video, further includes:
Support vector machines is trained using the positive sample image block and negative sample image block of preset ratio, obtains institute
State default classifier.
In one embodiment, the preset ratio of the positive sample image block and negative sample image is 9:1.
A kind of video smoke identification device, described device include:
Atmosphere light illumination estimate module estimates the figure to be identified for extracting the images to be recognized in video to be identified
Atmosphere brightness and dark primary as in；
Transmissivity image collection module, for obtaining described to be identified according to the atmosphere brightness and the dark primary
The transmittance figure picture of image；
Motion detection block, the video for being formed to the transmittance figure picture carry out motion detection, obtain motion feature
Vector；
Smog identification module, for the motion feature vector to be inputted default classifier, identification obtains described to be identified
Smoke region in video.
A kind of computer equipment, including memory and processor, the memory are stored with computer program, the place
Manage above-mentioned steps when device executes the computer program.
A kind of computer readable storage medium, is stored thereon with computer program, and the computer program is held by processor
Above-mentioned steps are realized when row.
Video smoke recognition methods provided by the invention, extracts the images to be recognized in video to be identified, described in estimation
Atmosphere brightness and dark primary in images to be recognized；According to the atmosphere brightness and the dark primary, obtain described wait know
The transmittance figure picture of other image；Motion detection is carried out to the video that the transmittance figure picture is formed, obtains motion feature vector；
The motion feature vector is inputted into default classifier, identification obtains the smoke region in the video to be identified.By saturating
Rate image is crossed, can completely position location of smoke independent of illumination variation, effectively improves the standard of smog positioning
True rate.
Detailed description of the invention
Fig. 1 is the applied environment figure of video smoke recognition methods in one embodiment；
Fig. 2 is the flow diagram of video smoke recognition methods in one embodiment；
Fig. 3 is the flow diagram of video smoke recognition methods in another embodiment；
Fig. 4 is the structural block diagram of video smoke identification device in another embodiment；
Fig. 5 is the internal structure chart of computer equipment in one embodiment.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, with reference to the accompanying drawings and embodiments, right
The present invention is described in further detail.It should be appreciated that the specific embodiments described herein are only used to explain this hair
It is bright, and the scope of protection of the present invention is not limited.
Video smoke recognition methods provided by the present application can be applied in application environment as shown in Figure 1.This applies ring
Border includes server 104 and photographic device 102, and server 104 obtains video to be identified from photographic device 102, and extract to
It identifies the images to be recognized in video, estimates atmosphere brightness and the dark primary in the images to be recognized；Server 104
According to the atmosphere brightness and the dark primary, the transmittance figure picture of the images to be recognized is obtained；Server 104 is to described
The video that transmittance figure picture is formed carries out motion detection, obtains motion feature vector；Server 104 by the motion feature to
Amount inputs default classifier, and identification obtains the smoke region in the video to be identified.Wherein, server can be with independent
The server cluster of server either multiple servers composition is realized；Photographic device can use camera, camera, hand
There is machine etc. the device of camera function to realize.
In one embodiment, as shown in Fig. 2, providing a kind of video smoke recognition methods, it is applied to figure in this way
It is illustrated for server in 1, comprising the following steps:
Step S202 extracts the images to be recognized in video to be identified, estimates that the atmosphere in the images to be recognized is bright
Degree and dark primary.
In the specific implementation process, transmissivity is only by atmospheric scattering coefficient and depth of field function without the shadow that is shone by atmosphere light
It rings, and the object that other in smog and scene move is the basic reason for influencing transmitance in original scene.So if can
Transmitance image is estimated, it can be independent of illumination variation completely in conjunction with the prior information of smog movement
Position location of smoke.
In this step, in the regional area of non-atmosphere light, certain some pixel always has at least one Color Channel tool
There is very low value, in other words, the minimum value of the region luminous intensity is the number of a very little, i.e. dark primary.
In one embodiment, the atmosphere brightness in the estimation images to be recognized of step S202, comprising:
A carries out mini-value filtering to the Minimal color weight of the images to be recognized, exports minimum value pixel.
In the specific implementation process, atmosphere light region has 3 characteristics: brightness is higher, gray scale is flat and position is on the upper side.
The pixel set for meeting above 3 conditions is determined as atmosphere light region.
Firstly, the Minimal color weight to color image carries out mini-value filtering, as following formula indicates:
In formula, c ∈ { R, G, B } respectively indicates R, G, B color channel；Ω (x) indicates the neighborhood centered on pixel x,
Adaptive, the scale factor 0.025 directly proportional to the minimum value in image height and width of size.
B carries out edge extracting with gray component of the default edge detection operator to the images to be recognized and obtains edge graph
Picture, and block statistics are carried out to the edge image, obtain statistics ratio；
Specifically, the step of carrying out block statistics to the edge image, obtaining statistics ratio, comprising: calculate the side
The ratio between neighborhood inward flange pixel number and neighborhood total pixel number in edge image centered on each pixel, obtain the statistics ratio.
In the specific implementation process, side is carried out using gray component of the Canny edge detection operator to color image f (x)
Edge extracts, and carries out block statistics to edge image E (x), for each pixel x, calculates in the neighborhood centered on pixel x
The statistics ratio of neighborhood total pixel number shared by edge pixel number, is denoted as Nedge(x), it may be expressed as:
In formula, # indicates number.In edge image E (x), the value of edge pixel is 1, and the value of non-edge pixels is 0.
C, according to atmosphere brightness described in the minimum value pixel and the statistics ratio-dependent.
Specifically, the step of atmosphere brightness according to the minimum value pixel and the statistics ratio-dependent, packet
It includes:
C1, according to the candidate atmosphere light in images to be recognized described in the minimum value pixel and the statistics ratio-dependent
Region；
C2 chooses the connected component of predeterminated position as atmosphere light to the candidate atmosphere light zone marker connected component
Region；
C3, using the max pixel value in the atmosphere light region as the atmosphere brightness.
In the present embodiment, while meeting Θ (x) > TvAnd Nedge(x) < TpPixel set be established as candidate atmosphere light
Region.Set luminance threshold TvIt is 95% of maximum value in Θ (x), flat threshold TpIt is 0.001.Finally to candidate atmosphere light area
Field mark connected component is generally present in the prior information above image according to atmosphere light region, chooses one above image
Max pixel value in atmosphere light region is determined as the estimated value of atmosphere brightness A as atmosphere light region by connected component.
Step S204 obtains the transmittance figure of the images to be recognized according to the atmosphere brightness and the dark primary
Picture.
In one embodiment, step S204 according to the atmosphere brightness and the dark primary, obtain described wait know
The transmittance figure picture of other image, comprising:
D handles the dark primary and atmosphere brightness according to default atmospherical scattering model, obtains the first transmission
Rate function.
Specifically, in the specific implementation process, generallys use default atmospherical scattering model and retouch and be illustrated in object through smog
Imaging mechanism, preset atmospherical scattering model it is as follows:
I (x)=A ρ (x) e-βd(x)+A(1-e-βd(x)) (3)
Wherein: I (x) is scene imaging, and A is atmosphere brightness, and β is atmospheric scattering coefficient, and ρ (x) is the field at coordinate x
Scape albedo；D (x) is the depth of field at coordinate x, wherein transmissivity t (x)=1-e-βd(x), unobstructed natural image J (x)=A ρ
(x)。
First item J (x) (1-t (x)) on the right of equation (3) is called direct attenuation term.Since the scattering of atmospheric particles is made
With a part in target surface reflected light is lost because of scattering, and unscattered part directly reaches imaging sensor, is arrived
The light intensity reached exponentially decays with the increase of propagation distance.Section 2 At (x) is then atmosphere light ingredient, this is because greatly
Gas particle causes atmosphere to show the characteristic of light source the scattering of natural light.
Dark primary priori is in the regional area for assume at least one Color Channel, and scene albedo goes to zero:
To image J, define:
JcSome Color Channel of J is represented, and Ω (x) is one piece of square region centered on x.Through statistical observation
It obtains, JdarkIntensity it is always very low and level off to 0.If J is outdoor image, JdarkThe referred to as dark primary of J.
Mini-value filtering is carried out simultaneously at left and right sides of peer-to-peer (3), available:
As can be seen from the above formula that needing estimation and the t of accurate atmosphere brightness A to accurately estimate t (x)
(x) refinement estimation.
In this step, the estimation of transmissivity t (x) is generally divided into two steps: rough estimate and refinement are estimated.In order to avoid block
Effect improves formula (5), obtains the first transmittance function, is shown below:
It is unrelated with albedo since atmospheric transmittance function is only the function of the depth of field, t (x) is refined and is needed when estimating
Carry out the edge details that depth of field mutation is kept while segment smoothing processing.Therefore t (x) refinement estimation can regard one as and put down
Sliding problem.The purpose of edge preserving smoothing is to keep output image similar as far as possible to input picture, but cross larger gradient
Region should be smooth as far as possible.Since bilateral filtering is theoretical simple, and there is fast algorithm, therefore use quick bilateral filtering
Method estimates t (x).
E handles first transmittance function, is obtained the second transmittance figure using default bilateral filtering function
Picture.
Specifically, bilateral filtering is a kind of non-iterative smooth filtering method that edge is kept, its weight is by airspace and value
The product of domain smooth function provides, at a distance from center pixel and the increase of gray scale difference value, the weight of field pixel by
It is decrescence small, Gaussian bilateral filtering is used in the present embodiment, i.e. airspace and codomain smooth function is Gaussian function, for saturating
Cross the rough estimate of rateIt is smoothed, be may be expressed as: using Gaussian bilateral filtering
Wherein, tbIt (x) is the refinement of bilateral filtering as a result, normalization coefficient WbFor
In formula,WithFor Gaussian function, σsFor the size of airspace Gaussian template, σrFor codomain Gaussian function
Scale.For the lesser pixel of closely located and gray scale difference value with center pixel, bilateral filtering assigns biggish weight；And
For closely located, but the biggish pixel of gray scale difference value, assign lesser weight.
F obtains the transmittance figure picture of the images to be recognized according to the second transmittance figure picture.
In the present embodiment, bilateral filtering can be very good keep image border, thus effectively inhibit in restoration result by
In the Halo effect that gray scale is mutated and is introduced in edge.
Step S206 carries out motion detection to the video that the transmittance figure picture is formed, obtains motion feature vector.
In this step, motion detection is carried out using ViBe algorithm；ViBe algorithm is a kind of imparametrization cluster background modeling
Method, has good adaptability and real-time to fixed camera under various circumstances, and detection effect is obvious.The algorithm it is only
Special place is the more new strategy of background model.Random chooses whether to update background model, and random selection updates corresponding back
The pixel of scape model, random selection update the pixel of neighborhood background model.It is main including the following steps:
1) background model indicates: the background model for belonging to each pixel in background is formed with N number of background sample, defines I
It (x) is the pixel in theorem in Euclid space at x, IkFor the sampled pixel of selection, then the corresponding background model M (x) of pixel I (x)
It is shown below:
M (x)={ I1,I2,…,IN} (9)
2) background model initializing: unlike other background difference algorithms, ViBe algorithm initializes plan using single frames
Slightly.In the first frame from the eight neighborhood N of I (x)G(x) N number of pixel value is repeatedly chosen at random in, is stored in corresponding background model
N number of sample in.It is shown below, defines M0(x) as the background model of first frame.
M0(x)={ I0(y)|y∈NG(x)} (10)
3) pixel classifications: in 2-D theorem in Euclid space, S is definedR(I (x)) is centered on pixel I (x), and distance R is half
The set of diameter, and given threshold #min.As set M (x) and SRWhen the intersection of (I (x)) is greater than #min, then determine I (x) for back
Scene element, on the contrary it is foreground pixel.
4) background model updates:
The update of background model be exactly so that background model can adapt to the continuous variation of background, such as the variation of illumination,
Change of background object etc..Conservative more new strategy is that foreground point is never used to fill background model, can be caused dead
Lock, if such as initialization when one piece of static region by mistake be detected as movement, under this policy it
It is taken as the object of movement forever to treat；Blind strategy be it is insensitive to deadlock, prospect background can update background
Model, the disadvantage is that the object slowly moved, which can incorporate in background, to be detected.ViBe algorithm use more new strategy be
Conservative more new strategy adds foreground point method of counting.Pixel is counted, if the continuous n times of some pixel are tested
Surveying is prospect, then is updated to background dot.
Specific update method are as follows:
If I (x) is classified as background pixel, ViBe algorithm will be to set the background model of probability updating I (x), i.e., from I
(x) a sample I is randomly selected in background model M (x)k, then substituted with I (x).This randomization update mode is protected
The life cycle for having demonstrate,proved each sample is successively decreased in Smoothness Index, avoids the defect of first in first out more new strategy.In order to guarantee
By the background of foreground occlusion, ViBe algorithm will be with same probability updating I (x) neighborhood for neighborhood of pixels Space Consistency and recovery
Background model, i.e., from the background model M of I (x) neighborhoodG(x) a sample I is randomly selected ink, then substituted with I (x).
In the specific implementation process, it according to the transmittance function and atmospherical scattering model having estimated that, can derive
Corresponding albedo function is as follows:
Wherein ρ (x) is exactly the albedo of corresponding position, and I (x) is original pixel value, and A is the strong of the natural light of estimation
Degree, t (x) are the transmittance function of estimation.Because albedo reflection is that scene without atmospheric attenuation directly reaches detection
The energy of device, therefore can really reflect Energy distribution of the object in scene under no smog circumstance of occlusion.
The motion feature vector is inputted default classifier by step S208, and identification obtains in the video to be identified
Smoke region.
In this step, by motion detection, the object moved in available video, then pass through support vector machines
Classification distinguishes the target of smog and other movements, because colouring information is most direct effective information in visible light, makes
The feature vector of smog color is described with colour moment and color histogram.
In addition it is reduced because smog can generally make background thicken so as to cause the high fdrequency component of image, it is small
The variation of the high fdrequency component of wave conversion can be used as feature and be used to describe smog.
W (x, y)=| HL (x, y) |2+|LH(x,y)|2+|HH(x,y)|2 (13)
For the candidate region for the smog divided according to motion information, four component images are obtained by wavelet transformation,
In include horizontal direction (HL) high fdrequency component, vertical direction (LH) high fdrequency component, diagonal (HH) high fdrequency component and one
Low-resolution image is opened, and calculates high fdrequency component using formula (12) and formula (13), wherein.W (x, y) is the high frequency division of present frame
Spirogram picture, RiEach pixel coordinate in representative image region.Because high fdrequency component is a static nature, but background is by mould
Paste is a dynamic process, so needing to count N frame image statistics high-frequency information in video, the high frequency component values group of this N frame
It is used to reflect the variation of high-frequency energy at a feature vector.
As shown in figure 3, in one embodiment, the motion feature vector is inputted default classifier by step S208, know
Before the step of not obtaining the smoke region in the video to be identified, further includes:
Step S207 instructs support vector machines using the positive sample image block and negative sample image block of preset ratio
Practice, obtains the default classifier.
In the specific implementation process, for the feature vector that extracts need a strong classifier to different features to
Amount is classified, and uses support vector machines as the classifier, which use 15000 image blocks as positive sample training
The preset ratio of the classifier, positive sample image block and negative sample image is 9:1.
Above-mentioned video smoke recognition methods, by extracting the images to be recognized in video to be identified, estimation is described wait know
Atmosphere brightness and dark primary in other image；According to the atmosphere brightness and the dark primary, the figure to be identified is obtained
The transmittance figure picture of picture；Motion detection is carried out to the video that the transmittance figure picture is formed, obtains motion feature vector；By institute
It states motion feature vector and inputs default classifier, identification obtains the smoke region in the video to be identified.Pass through transmitance
Image can completely position location of smoke independent of illumination variation, effectively improve the accurate of smog positioning
Rate.
As shown in figure 4, Fig. 4 is the structural schematic diagram of video smoke identification device in one embodiment, mentioned in the present embodiment
For a kind of video smoke identification device, including atmosphere light illumination estimate module 401, transmissivity image collection module 402, movement
Detection module 403 and smog identification module 404, in which:
Atmosphere light illumination estimate module 401, for extracting the images to be recognized in video to be identified, estimation is described wait know
Atmosphere brightness and dark primary in other image；
Transmissivity image collection module 402, for according to the atmosphere brightness and the dark primary, obtain it is described to
Identify the transmittance figure picture of image；
Motion detection block 403, the video for being formed to the transmittance figure picture carry out motion detection, are moved
Feature vector；
Smog identification module 404, for the motion feature vector to be inputted default classifier, identification obtain it is described to
Identify the smoke region in video.
Specific about video smoke identification device limits the limit that may refer to above for video smoke recognition methods
Fixed, details are not described herein.Modules in above-mentioned video smoke identification device can fully or partially through software, hardware and
A combination thereof is realized.Above-mentioned each module can be embedded in the form of hardware or independently of in the processor in computer equipment, can also
Be stored in the memory in computer equipment in a software form, the above modules pair are executed in order to which processor calls
The operation answered.
As shown in figure 5, Fig. 5 is the schematic diagram of internal structure of computer equipment in one embodiment.The computer equipment packet
Include processor, non-volatile memory medium, memory and the network interface connected by device bus.Wherein, which sets
Standby non-volatile memory medium is stored with operating device, database and computer-readable instruction, can be stored with control in database
Part information sequence when the computer-readable instruction is executed by processor, may make processor to realize a kind of video smoke identification
Method.The processor of the computer equipment supports the operation of entire computer equipment for providing calculating and control ability.It should
Computer-readable instruction can be stored in the memory of computer equipment, it, can when which is executed by processor
So that processor executes a kind of video smoke recognition methods.The network interface of the computer equipment is used for and terminal connection communication.
It will be understood by those skilled in the art that structure shown in Fig. 5, the only frame of part-structure relevant to application scheme
Figure, does not constitute the restriction for the computer equipment being applied thereon to application scheme, specific computer equipment can wrap
It includes than more or fewer components as shown in the figure, perhaps combines certain components or with different component layouts.
In one embodiment it is proposed that a kind of computer equipment, computer equipment include memory, processor and deposit
The computer program that can be run on a memory and on a processor is stored up, processor realizes following step when executing computer program
It is rapid: to extract the images to be recognized in video to be identified, estimate atmosphere brightness and dark primary in the images to be recognized；Root
According to the atmosphere brightness and the dark primary, the transmittance figure picture of the images to be recognized is obtained；To the transmittance figure
As the video progress motion detection formed, motion feature vector is obtained；The motion feature vector is inputted into default classifier,
Identification obtains the smoke region in the video to be identified.
Processor executes when computer program in the estimation images to be recognized in one of the embodiments,
The step of atmosphere brightness, comprising: mini-value filtering is carried out to the Minimal color weight of the images to be recognized, output is minimum
It is worth pixel；Edge extracting, which is carried out, with gray component of the default edge detection operator to the images to be recognized obtains edge image,
And block statistics are carried out to the edge image, obtain statistics ratio；According to the minimum value pixel and the statistics ratio
Determine the atmosphere brightness.
Processor executes described to the edge image progress piecemeal when computer program in one of the embodiments,
The step of counting, obtaining statistics ratio, comprising: calculate the neighborhood inward flange picture in the edge image centered on each pixel
The ratio between prime number and neighborhood total pixel number obtain the statistics ratio.
Processor executes described according to the minimum value pixel and institute when computer program in one of the embodiments,
The step of stating atmosphere brightness described in statistics ratio-dependent, comprising: according to the minimum value pixel and the statistics ratio-dependent
Candidate atmosphere light region in the images to be recognized；To the candidate atmosphere light zone marker connected component, default position is chosen
The connected component set is as atmosphere light region；It is bright using the max pixel value in the atmosphere light region as the atmosphere
Degree.
According to the atmosphere brightness and described dark when processor executes computer program in one of the embodiments,
Primary colors, obtain the transmittance figure of the images to be recognized as the step of, comprising: according to default atmospherical scattering model to described dark
Primary colors and atmosphere brightness are handled, and the first transmittance function is obtained；Using default bilateral filtering function, to described first
Transmittance function is handled, and the second transmittance figure picture is obtained；It is obtained according to the second transmittance figure picture described to be identified
The transmittance figure picture of image.
Processor executes described by the motion feature vector input when computer program in one of the embodiments,
Before the step of default classifier, identification obtains the smoke region in the video to be identified, further includes: use preset ratio
Positive sample image block and negative sample image block support vector machines is trained, obtain the default classifier.
Processor executes the positive sample image block and negative sample figure when computer program in one of the embodiments,
The preset ratio of picture is 9:1.
In one embodiment it is proposed that a kind of storage medium for being stored with computer-readable instruction, this is computer-readable
When instruction is executed by one or more processors, so that one or more processors execute following steps: extracting video to be identified
In images to be recognized, estimate atmosphere brightness and the dark primary in the images to be recognized；According to the atmosphere brightness and
The dark primary obtains the transmittance figure picture of the images to be recognized；The video formed to the transmittance figure picture is transported
Dynamic detection, obtains motion feature vector；The motion feature vector is inputted into default classifier, identification obtains described to be identified
Smoke region in video.
The estimation figure to be identified when computer-readable instruction is executed by processor in one of the embodiments,
The step of atmosphere brightness as in, comprising: mini-value filtering is carried out to the Minimal color weight of the images to be recognized, it is defeated
Minimum value pixel out；Edge extracting, which is carried out, with gray component of the default edge detection operator to the images to be recognized obtains side
Edge image, and block statistics are carried out to the edge image, obtain statistics ratio；According to the minimum value pixel and the system
Count atmosphere brightness described in ratio-dependent.
When computer-readable instruction is executed by processor in one of the embodiments, it is described to the edge image into
Row block statistics, the step of obtaining statistics ratio, comprising: calculate in the neighborhood in the edge image centered on each pixel
The ratio between edge pixel number and neighborhood total pixel number obtain the statistics ratio.
It is described according to the minimum value picture when computer-readable instruction is executed by processor in one of the embodiments,
The step of atmosphere brightness described in the plain and described statistics ratio-dependent, comprising: according to the minimum value pixel and statistics ratio
Example determines the candidate atmosphere light region in the images to be recognized；To the candidate atmosphere light zone marker connected component, choose
The connected component of predeterminated position is as atmosphere light region；Using the max pixel value in the atmosphere light region as the atmosphere
Brightness.
When computer-readable instruction is executed by processor in one of the embodiments, according to the atmosphere brightness and
The dark primary, obtain the transmittance figure of the images to be recognized as the step of, comprising: according to default atmospherical scattering model pair
The dark primary and atmosphere brightness are handled, and the first transmittance function is obtained；Using default bilateral filtering function, to institute
It states the first transmittance function to be handled, obtains the second transmittance figure picture；According to the second transmittance figure picture acquisition
The transmittance figure picture of images to be recognized.
When computer-readable instruction is executed by processor in one of the embodiments, it is described by the motion feature to
Before the step of amount inputs default classifier, and identification obtains the smoke region in the video to be identified, further includes: using pre-
If the positive sample image block and negative sample image block of ratio are trained support vector machines, the default classifier is obtained.
The positive sample image block and negative when computer-readable instruction is executed by processor in one of the embodiments,
The preset ratio of sample image is 9:1.
It should be understood that although each step in the flow chart of attached drawing is successively shown according to the instruction of arrow,
These steps are not that the inevitable sequence according to arrow instruction successively executes.Unless expressly stating otherwise herein, these steps
Execution there is no stringent sequences to limit, can execute in the other order.Moreover, in the flow chart of attached drawing at least
A part of step may include that perhaps these sub-steps of multiple stages or stage are not necessarily same to multiple sub-steps
Moment executes completion, but can execute at different times, and execution sequence is also not necessarily and successively carries out, but can be with
It is executed in turn or alternately at least part of the sub-step or stage of other steps or other steps.
The above is only some embodiments of the invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (10)
1. a kind of video smoke recognition methods, which is characterized in that the described method includes:
The images to be recognized in video to be identified is extracted, estimates atmosphere brightness and dark primary in the images to be recognized；
According to the atmosphere brightness and the dark primary, the transmittance figure picture of the images to be recognized is obtained；
Motion detection is carried out to the video that the transmittance figure picture is formed, obtains motion feature vector；
The motion feature vector is inputted into default classifier, identification obtains the smoke region in the video to be identified.
2. the method according to claim 1, wherein the atmosphere brightness in the estimation images to be recognized
The step of, comprising:
Mini-value filtering is carried out to the Minimal color weight of the images to be recognized, exports minimum value pixel；
Edge extracting is carried out with gray component of the default edge detection operator to the images to be recognized and obtains edge image, and right
The edge image carries out block statistics, obtains statistics ratio；
According to atmosphere brightness described in the minimum value pixel and the statistics ratio-dependent.
3. according to the method described in claim 2, it is characterized in that, it is described to the edge image carry out block statistics, obtain
The step of statistics ratio, comprising:
The ratio between neighborhood inward flange pixel number and the neighborhood total pixel number in the edge image centered on each pixel are calculated, is obtained
The statistics ratio.
4. according to the method described in claim 2, it is characterized in that, described according to the minimum value pixel and the statistics ratio
The step of determining the atmosphere brightness, comprising:
According to the candidate atmosphere light region in images to be recognized described in the minimum value pixel and the statistics ratio-dependent；
To the candidate atmosphere light zone marker connected component, the connected component of predeterminated position is chosen as atmosphere light region；
Using the max pixel value in the atmosphere light region as the atmosphere brightness.
5. the method according to claim 1, wherein described according to the atmosphere brightness and the dark primary,
Obtain the transmittance figure of the images to be recognized as the step of, comprising:
The dark primary and atmosphere brightness are handled according to default atmospherical scattering model, obtain the first transmittance function；
Using default bilateral filtering function, first transmittance function is handled, obtains the second transmittance figure picture；
The transmittance figure picture of the images to be recognized is obtained according to the second transmittance figure picture.
6. the method according to claim 1, wherein described input default classification for the motion feature vector
Before the step of device, identification obtains the smoke region in the video to be identified, further includes:
Support vector machines is trained using the positive sample image block and negative sample image block of preset ratio, is obtained described default
Classifier.
7. according to the method described in claim 6, it is characterized in that, the default ratio of the positive sample image block and negative sample image
Example is 9:1.
8. a kind of video smoke identification device, which is characterized in that described device includes:
Atmosphere light illumination estimate module is estimated in the images to be recognized for extracting the images to be recognized in video to be identified
Atmosphere brightness and dark primary；
Transmissivity image collection module, for obtaining the images to be recognized according to the atmosphere brightness and the dark primary
Transmittance figure picture；
Motion detection block, the video for being formed to the transmittance figure picture carry out motion detection, obtain motion feature vector；
Smog identification module, for the motion feature vector to be inputted default classifier, identification obtains the video to be identified
In smoke region.
9. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In the step of processor realizes any one of claims 1 to 7 the method when executing the computer program.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of method described in any one of claims 1 to 7 is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910266782.6A CN110070007A (en) | 2019-04-03 | 2019-04-03 | Video smoke recognition methods, device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910266782.6A CN110070007A (en) | 2019-04-03 | 2019-04-03 | Video smoke recognition methods, device, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110070007A true CN110070007A (en) | 2019-07-30 |
Family
ID=67367081
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910266782.6A Pending CN110070007A (en) | 2019-04-03 | 2019-04-03 | Video smoke recognition methods, device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110070007A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652184A (en) * | 2020-06-19 | 2020-09-11 | 成都通甲优博科技有限责任公司 | Smoke identification method and device, storage medium and data processing equipment |
| CN114648852A (en) * | 2022-05-24 | 2022-06-21 | 四川九通智路科技有限公司 | Tunnel fire monitoring method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103985091A (en) * | 2014-04-30 | 2014-08-13 | 西安理工大学 | Single image defogging method based on luminance dark priori method and bilateral filtering |
| CN104050478A (en) * | 2014-07-09 | 2014-09-17 | 湖南大学 | Smog detection method and system |
| CN105046666A (en) * | 2015-07-24 | 2015-11-11 | 中国科学技术大学 | Dark channel prior based traffic video real-time defogging method |
| CN105654494A (en) * | 2016-01-04 | 2016-06-08 | 杭州中威电子股份有限公司 | Smog detection method for detecting smog in video image |
| CN107045723A (en) * | 2017-03-10 | 2017-08-15 | 北京环境特性研究所 | Smog recognition methods based on transmissivity dynamic detection |
-
2019
- 2019-04-03 CN CN201910266782.6A patent/CN110070007A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103985091A (en) * | 2014-04-30 | 2014-08-13 | 西安理工大学 | Single image defogging method based on luminance dark priori method and bilateral filtering |
| CN104050478A (en) * | 2014-07-09 | 2014-09-17 | 湖南大学 | Smog detection method and system |
| CN105046666A (en) * | 2015-07-24 | 2015-11-11 | 中国科学技术大学 | Dark channel prior based traffic video real-time defogging method |
| CN105654494A (en) * | 2016-01-04 | 2016-06-08 | 杭州中威电子股份有限公司 | Smog detection method for detecting smog in video image |
| CN107045723A (en) * | 2017-03-10 | 2017-08-15 | 北京环境特性研究所 | Smog recognition methods based on transmissivity dynamic detection |
Non-Patent Citations (1)
| Title |
|---|
| 张志勇 等: ""基于小波变换和ICA的运动目标分割"", 《计算机辅助设计与图形学学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652184A (en) * | 2020-06-19 | 2020-09-11 | 成都通甲优博科技有限责任公司 | Smoke identification method and device, storage medium and data processing equipment |
| CN114648852A (en) * | 2022-05-24 | 2022-06-21 | 四川九通智路科技有限公司 | Tunnel fire monitoring method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tripathi et al. | Single image fog removal using bilateral filter | |
| CN104299192B (en) | A kind of single image to the fog method based on atmospheric light scattering physical model | |
| CN105913441B (en) | It is a kind of for improving the shadow removal method of target detection performance in video | |
| CN111062293B (en) | Unmanned aerial vehicle forest flame identification method based on deep learning | |
| CN109389569B (en) | Monitoring video real-time defogging method based on improved DehazeNet | |
| CN111598791B (en) | Image defogging method based on improved dynamic atmospheric scattering coefficient function | |
| CN110163818A (en) | A kind of low illumination level video image enhancement for maritime affairs unmanned plane | |
| CN105959510B (en) | A kind of video rapid defogging method | |
| CN105447825B (en) | Image defogging method and its system | |
| CN107451966A (en) | A kind of real-time video defogging method realized using gray-scale map guiding filtering | |
| CN106056557B (en) | A kind of single image rapid defogging method based on improvement atmospherical scattering model | |
| CN107045723A (en) | Smog recognition methods based on transmissivity dynamic detection | |
| CN110211067A (en) | One kind being used for UUV Layer Near The Sea Surface visible images defogging method | |
| CN105023246B (en) | A kind of image enchancing method based on contrast and structural similarity | |
| CN110070007A (en) | Video smoke recognition methods, device, computer equipment and storage medium | |
| CN112164010A (en) | Multi-scale fusion convolution neural network image defogging method | |
| Sahu et al. | Image dehazing based on luminance stretching | |
| CN110223240A (en) | Image defogging method, system and storage medium based on color decaying priori | |
| CN107818547A (en) | The minimizing technology of the spiced salt and Gaussian mixed noise in a kind of sequence towards twilight image | |
| CN109345479B (en) | Real-time preprocessing method and storage medium for video monitoring data | |
| Chen et al. | Improve transmission by designing filters for image dehazing | |
| Khan et al. | Recent advancement in haze removal approaches | |
| CN112950589A (en) | Dark channel prior defogging algorithm of multi-scale convolution neural network | |
| CN112907461A (en) | Defogging and enhancing method for infrared degraded image in foggy day | |
| CN107451962A (en) | A kind of image defogging method and device |
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: 20190730 |