CN105139429B - A kind of fire detection method based on flame notable figure and spatial pyramid histogram - Google Patents

A kind of fire detection method based on flame notable figure and spatial pyramid histogram Download PDF

Info

Publication number
CN105139429B
CN105139429B CN201510503877.7A CN201510503877A CN105139429B CN 105139429 B CN105139429 B CN 105139429B CN 201510503877 A CN201510503877 A CN 201510503877A CN 105139429 B CN105139429 B CN 105139429B
Authority
CN
China
Prior art keywords
mrow
flame
msub
block
sub
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
Application number
CN201510503877.7A
Other languages
Chinese (zh)
Other versions
CN105139429A (en
Inventor
陈喆
殷福亮
李政霖
耿晓馥
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dalian University of Technology
Original Assignee
Dalian University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Dalian University of Technology filed Critical Dalian University of Technology
Priority to CN201510503877.7A priority Critical patent/CN105139429B/en
Publication of CN105139429A publication Critical patent/CN105139429A/en
Application granted granted Critical
Publication of CN105139429B publication Critical patent/CN105139429B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B17/00—Fire alarms; Alarms responsive to explosion
    • G08B17/12—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions
    • G08B17/125—Actuation by presence of radiation or particles, e.g. of infrared radiation or of ions by using a video camera to detect fire or smoke
    • 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
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30232—Surveillance

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Emergency Management (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

A kind of fire detection method based on flame notable figure and spatial pyramid histogram, comprises the following steps：S1：The Strength Changes value and prospect degree of each pixel in image are calculated, make the difference method using successive frame obtains flame notable figure with Gaussian Mixture flame color model；S2：According to flame notable figure, the pixel of candidate's flame is filtered out using thresholding method, constructs the mask image containing candidate's flame pixels point；S3：Piecemeal processing is carried out to original image according to mask image and judges whether each sub-block contains flame；S4：Corresponding sub-block in previous frame image is found by sub-block, and the Distance Judgment image corresponded to using successive frame between the spatial pyramid histogram of sub-block whether there is flame.

Description

A kind of fire detection method based on flame notable figure and spatial pyramid histogram
Technical field
The present invention relates to technical field of image processing, more particularly to one kind to be based on flame notable figure and spatial pyramid Nogata The fire detection method of figure.
Background technology
Today's society, fire cause heavy losses to the safety of life and property of the people every year.It is annual that fire occurs in China Calamity about 40,000, cause people more than 2,000 dead, more than 3,000 people's injures and deaths, direct property loss is up to more than 10 hundred million yuan.Send out accurately and in time Existing fire (particularly initial fire disaster), and alarm is sent, the time that can not only allow trapped personnel and property to have abundance shifts, and And fire fighter can be allowed to implement to put out a fire to save life and property in the case where the condition of a fire is controllable.Therefore, the monitoring of fire be fire prevention and control field very Important problem.
The automatic fire detection technique of early stage applies smog or body heat sensor mostly, by particle, gas, temperature Or the sampling of radiant heat etc., to determine whether to have fire.Although this kind of fire detection technology realizes automatic detection substantially, Still there is significant limitation.First, the validity of such technology is confined to confined space, once it is applied to large area space When (particularly under outdoor environment), verification and measurement ratio will decline to a great extent；Secondly, such technology is affected by environment larger, wind speed, wind To environmental changes such as, sleet, the effect of fire detection can be all largely affected by；Again, detection time is longer, because gas Body, particle etc., which reach sensor, needs the regular hour；Meanwhile the cost of the sensor such as temperature, smog is higher, and easily damage It is bad.The popularization of monitoring camera, new possibility is provided for fire detection, it is initially more to carry out fire hazard monitoring, the party using artificial Formula not only wastes substantial amounts of human resources, and can not ensure monitoring effect.Usual monitoring personnel number is limited, can not accomplish each Locate exhaustive monitoring.Simultaneously as the notice of the mankind can not be concentrated for a long time, monitoring effect with the growth of working time and Decline.The limitation of above-mentioned prior art all cause they can not widespread adoption in reality fire detection.
Fire detection technology based on image procossing mainly has following four advantages：(1), should compared with traditional method Class technology has more preferable testing result, i.e., higher verification and measurement ratio and lower false drop rate；(2) when applied to larger space, Accurately and the presence of fire can be quickly detected；(3) almost free from the influence of the external environment, Detection results are stable；(4) Fire detection system can be added in existing monitoring system, cost is relatively low, can mitigate the finance of fire preventing and treating system Burden.Therefore, vast potential for future development is had based on image processing techniques fire detection technology.
Celi etc. is in document " Automatic fire detection in video sequences " in the prior art Judged using flame profile and area with time fast-changing feature.First with background updating and flame Color model screens to pixel, then carries out morphologic corrosion and expansion process to candidate region, and according to processing The spatial mean value and area of enclosed region afterwards change over time situation, to judge to whether there is flame in video.Although the skill Art can have higher verification and measurement ratio with the behavioral characteristics of accurate description flame.But presence of this method to noise is more Sensitivity, when video quality is poor or mobile chaff interference be present, this method has higher false drop rate.
2009, Habiboglu etc. was in document " Covariance matrix-based fire and flame It is proposed to change with time situation as feature by the use of covariance matrix to judge in detection method in video " With the presence or absence of flame.The technology is screened first with flame color model to pixel, then by its some two field picture The value of covariance matrix and three Color Channels trains grader, to candidate as feature using the training data demarcated manually Pixel is further classified.The situation that the technology application pixel covariance matrix changes over time is carried out as feature , although improving detection performance to a certain extent, the mobile chaff interference of similar flame color be present in classification It is lower easily to occur flase drop, and computation complexity it is higher.
The content of the invention
The problem of being existed according to prior art, the invention discloses one kind to be based on flame notable figure and spatial pyramid Nogata The fire detection method of figure, comprises the following steps：
S1：The Strength Changes value and prospect degree of each pixel in image are calculated, method and Gaussian Mixture are made the difference using successive frame Flame color model obtains flame notable figure；
S2：According to flame notable figure, the pixel of candidate's flame is filtered out using thresholding method, it is fiery that construction contains candidate The mask image of flame pixel；
S3：Original image is handled as follows according to mask image：If the value of pixel is 1 in mask image, corresponding Pixel in the original image of position retains initial value, and otherwise the value of pixel is set to zero in original image；Will be above-mentioned treated Original image be divided into multiple sub-blocks, calculate the candidate in row block number, row block number and each sub-block that image is divided out The number of flame pixels is so as to judging whether the sub-block contains flame；
S4：Corresponding sub-block in previous frame image is found by sub-block, and the space gold of sub-block is corresponded to using successive frame Distance Judgment image between word tower histogram whether there is flame.
It is specific in the following way that flame notable figure is obtained in S1：
S11：Calculate the Strength Changes value of each pixel in image：
Wherein：Pdiff(x, y, t) is the Strength Changes value for the pixel that t is located at (x, y) position, and I (x, y, t) is t Moment is located at the intensity level of the pixel of (x, y) position；
S12：The prospect degree of each pixel in image is calculated, before the prospect degree is pixel belongs in the foreground detection stage The probable value of scape：
Wherein, PF(x, y, t) represents the prospect degree of t position (x, y) place pixel, it be to before t N frames it is strong The weighting logarithm value summation of degree change, describes the characteristics of flame region intensity continuous changes；
S13：According to the Gaussian Mixture flame color model trained, each pixel color is calculated in RGB color For the probability of flame color：
Wherein, q (x, y, t) is color vector of the pixel at t position (x, y) place in RGB color, is represented Form is
Q (x, y, t)={ R (x, y, t), G (x, y, t), B (x, y, t) } (4)
K is the number of unimodal Gaussian density function component in the Gaussian Mixture flame color model trained, μkAnd Σk It is the mean vector and covariance matrix of k-th of component in the Gaussian Mixture flame color model trained respectively, αkIt is each The weight of Gaussian component；The expression formula of function η () is
Wherein, D is the dimension of vector, in the present invention, D=3；
S14：According to prospect degree and flame color probability structure flame notable figure：
The pixel prospect degree calculated according to formula (2) and the flame color probability calculation each pixel calculated by formula (3) Significance
fs(x, y, t)=PF(x,y,t)+logPc(q(x,y,t)) (6)
Wherein, fs(x, y, t) is the significance of t position (x, y) place pixel, and in flame notable figure (x, Y) pixel value of place's pixel.
In S2 in the following way：
According to the flame Saliency maps obtained by above-mentioned formula (6), noise spot is filtered out by Threshold segmentation, obtains including candidate The mask image of flame pixels point：
Wherein, fsFor the flame notable figure of inputted video image, τfFor threshold value obtained by experiment, if Ms(x, y, t)=1, table It may be flame pixels to show the pixel, need to further be judged；If Ms(x, y, t)=0, then it represents that the point is not necessarily in fire In flame region, without subsequent treatment.
S3 by sub-block to being handled in the following way：
S31：Width and high respectively W according to sub-blockbAnd HbPiecemeal is carried out to image, if the width of image and high respectively Wi And Hi, then image will be separated out come row block number NrWith row block number NcTo be
Wherein,Operator represents downward rounding operation；
S32：The number of candidate's flame pixels in each sub-block is counted in the following way, by flame pixels number The few sub-block of mesh is determined as nonflame sub-block, without subsequent treatment：
Wherein, Mb,i(t) possibility that i-th of sub-block of t includes flame is represented, the value is 1 i-th of son of expression t Block may contain flame, will carry out spatial pyramid statistics with histogram to it, and the value represents that the sub-block does not contain flame for 0, no With carrying out subsequent treatment；Bi(t) region that i-th of sub-block represents, T are representedbWhole sub-block pixel count is accounted for for candidate's flame pixels point Purpose proportion threshold value.
S4 falls into a trap the spatial pyramid histogram of each sub-block of nomogram picture, and according to corresponding blocks in continuous multiple image Whether the Distance Judgment sub-block between pyramid histogram contains flame：
S41：R channel images are selected to calculate spatial pyramid histogram, according to the mask figure containing candidate's flame pixels point Picture, necessarily initial value will be kept constant not at the pixel R channel value zero setting of flame region, candidate pixel point, i.e.,
S42：To it is each may be M containing the sub-block of flameb,i(t) it is empty that sub-block=1 calculates it on F (x, y, t) image Between pyramid histogram H (Bi(t) piecemeal operation again, l class resolution ratios), i.e., are carried out respectively according to 0~L of resolution ratio to sub-block It is lower by sub-block width and high two dimensions are all divided into 2lSection, wherein l=1,2 ... L, so as to which sub-block is separated into some fritters, unite Color histogram of each fritter on R passages is counted, the histogram vectors of each fritter are finally multiplied by weight betalHead and the tail phase afterwards Company forms the spatial pyramid histogram of the sub-block：
S43：Sub-block to may each include flame in present frame, finds corresponding sub-block in its previous frame image, I.e. in previous frame image by this frame be presently in reason sub-block centered on original position, in the range of-R~R search for calculate The spatial pyramid histogram of all sub-blocks, calculates the distance between the histogram of its sub-block handled by with present frame, and distance is minimum Sub-block be with present frame handled by the corresponding sub-block of sub-block；
S44：According to the corresponding sub-block determined in S43, the space that currently processed sub-block corresponds to sub-block with former frame is calculated The distance between pyramid histogram；As final judgement foundation after the result of calculation of multiframe is carried out averagely, such as formula (15) shown in：
Wherein Bc,i(t-1) corresponding sub-block of i-th of the sub-block of t in former frame, N are representedmExpression carries out average frame Number；
S45：According to the distance calculated in S44, judge to whether there is flame in currently processed sub-block, method such as formula (16) It is shown
Wherein, TsFor threshold value, if at least one sub-block is judged as flame being present in image, then it is assumed that the two field picture is deposited In flame.
By adopting the above-described technical solution, one kind provided by the invention is based on flame notable figure and spatial pyramid Nogata The fire detection method of figure, this method Flame notable figure establish process, have merged the prospect inspection for making poor method based on successive frame Survey stage and the soft-decision result of flame color model based on gauss hybrid models, avoid two stages difference hard decision bands The reduction of the verification and measurement ratio come.And the space pyramid histogram proposed can effectively describe the dynamic characteristic of flame, contribute to Flase drop caused by reducing flame color chaff interference.
Brief description of the drawings
, below will be to embodiment or existing in order to illustrate more clearly of the embodiment of the present application or technical scheme of the prior art There is the required accompanying drawing used in technology description to be briefly described, it should be apparent that, drawings in the following description are only this Some embodiments described in application, for those of ordinary skill in the art, on the premise of not paying creative work, Other accompanying drawings can also be obtained according to these accompanying drawings.
Fig. 1 is the flow chart of method disclosed by the invention；
Fig. 2 is to be based on spatial pyramid histogram part of test results figure in the present invention.
Embodiment
To make technical scheme and advantage clearer, with reference to the accompanying drawing in the embodiment of the present invention, to this Technical scheme in inventive embodiments is clearly completely described：
A kind of fire detection method based on flame notable figure and spatial pyramid histogram as shown in Figure 1, specific bag Include following steps：
S1：The Strength Changes value and prospect degree of each pixel in image are calculated, method and Gaussian Mixture are made the difference using successive frame Flame color model obtains flame notable figure；Obtaining pixel using flame model has the probability of flame color, two stages Soft-decision is carried out respectively, then result is integrated, and obtains a flame Saliency maps；Flame Saliency maps and image size phase Together, each point represents that relevant position pixel may be the probability of flame region pixel in image.S1 is specifically using such as lower section Formula：
S11：Calculate the Strength Changes value of each pixel in image：
Wherein：Pdiff(x, y, t) is the Strength Changes value for the pixel that t is located at (x, y) position, and I (x, y, t) is t Moment is located at the intensity level of the pixel of (x, y) position；
S12：The prospect degree of each pixel in image is calculated, before the prospect degree is pixel belongs in the foreground detection stage The probable value of scape：
Wherein, PF(x, y, t) represents the prospect degree of t position (x, y) place pixel, it be to before t N frames it is strong The weighting logarithm value summation of degree change, describes the characteristics of flame region intensity continuous changes；
S13：According to the Gaussian Mixture flame color model trained, each pixel color is calculated in RGB color For the probability of flame color：
Wherein, q (x, y, t) is color vector of the pixel at t position (x, y) place in RGB color, is represented Form is
Q (x, y, t)={ R (x, y, t), G (x, y, t), B (x, y, t) } (4)
K is the number of unimodal Gaussian density function component in the Gaussian Mixture flame color model trained, μkAnd Σk It is the mean vector and covariance matrix of k-th of component in the Gaussian Mixture flame color model trained respectively, αkIt is each The weight of Gaussian component；The expression formula of function η () is
Wherein, D is the dimension of vector, herein as 3.
S14：According to prospect degree and flame color probability structure flame notable figure：
The pixel prospect degree calculated according to formula (2) and the flame color probability calculation each pixel calculated by formula (3) Significance
fs(x, y, t)=PF(x,y,t)+logPc(q(x,y,t)) (6)
Wherein fs(x, y, t) is the significance of t position (x, y) place pixel, and in flame notable figure (x, y) Locate the pixel value of pixel.
The GMM flame color models that said process is mentioned refer to that for describing a certain pixel color be flame color Probability gauss hybrid models (GMM).The model is to train what is obtained by the flame pixels point largely demarcated manually in advance, Allow it to describe the probability that a certain pixel color belongs to flame color, used color space is rgb space.GMM Flame color model parameter initial value is obtained by Kmeans algorithms, then obtains optimal parameter by EM algorithm optimizations.Gauss Mixed model (GMM) training process is as follows：
1) using the gauss hybrid models of K Gaussian mixture components, to n observation data Q=[q1,q2,…,qN] application K-means algorithms try to achieve initial parameter value
2) E steps：Calculate training data and concentrate probability of i-th of data in k-th of Gaussian component
M steps：The result of calculation walked according to E, update each parameter value and weights of each Gaussian components, specific meter Calculation process is
3) repeat step (2), until likelihood function (shown in formula (1.5)) convergence.
S2：According to flame notable figure, the pixel of candidate's flame is filtered out using thresholding method, it is fiery that construction contains candidate The mask image of flame pixel.Specifically in the following way：
According to the flame Saliency maps obtained by above-mentioned formula (6), noise spot is filtered out by Threshold segmentation, obtains including candidate The mask image of flame pixels point：
Wherein, fsFor the flame notable figure of inputted video image, τfFor threshold value obtained by experiment, if Ms(x, y, t)=1, table It may be flame pixels to show the pixel, need to further be judged；If Ms(x, y, t)=0, then it represents that the point is not necessarily in fire In flame region, without subsequent treatment.
S3：Original image is handled as follows according to mask image：If the value of pixel is 1 in mask image, corresponding Pixel in the original image of position retains initial value, and otherwise the value of pixel is set to zero in original image；Will be above-mentioned treated Original image be divided into multiple sub-blocks, calculate the candidate in row block number, row block number and each sub-block that image is divided out The number of flame pixels is so as to judging whether the sub-block contains flame；
S31：Width and high respectively W according to sub-blockbAnd HbPiecemeal is carried out to image, if the width of image and high respectively Wi And Hi, then image will be separated out come row block number NrWith row block number NcTo be
Wherein,Operator represents downward rounding operation；
S32：The number of candidate's flame pixels in each sub-block is counted in the following way, by flame pixels number The few sub-block of mesh is determined as nonflame sub-block, without subsequent treatment；
Wherein, Mb,i(t) possibility that i-th of sub-block of t includes flame is represented, the value is 1 i-th of son of expression t Block may contain flame, will carry out spatial pyramid statistics with histogram to it, and the value represents that the sub-block does not contain flame for 0, no With carrying out subsequent treatment；Bi(t) region that i-th of sub-block represents, T are representedbWhole sub-block pixel count is accounted for for candidate's flame pixels point Purpose proportion threshold value.
S4：Corresponding sub-block in previous frame image is found by sub-block, and the space gold of sub-block is corresponded to using successive frame Distance Judgment image between word tower histogram whether there is flame.
In the calculating of spatial pyramid histogram, a series of grids are established by 0~L at varying resolutions, i.e., in l Wide and high two dimensions are divided into 2 respectively on class resolution ratiolSection, it is overall that image is divided into 4lSub-regions, can to each Statistics color or grey level histogram in candidate's subregion of flame can be included.Most at last under different resolution all subregions it is straight Side's figure vector joins end to end successively, forms pyramid histogram vectors, then for piece image, its spatial pyramid histogram Dimension be
Wherein, BnumFor the bins of histogram number.
S4 falls into a trap the spatial pyramid histogram of each sub-block of nomogram picture, and according to corresponding blocks in continuous multiple image Whether the Distance Judgment sub-block between pyramid histogram contains flame：
S41：R channel images are selected to calculate spatial pyramid histogram, according to the mask figure containing candidate's flame pixels point Picture, necessarily initial value will be kept constant not at the pixel R channel value zero setting of flame region, candidate pixel point, i.e.,
S42：To it is each may be M containing the sub-block of flameb,i(t) it is empty that sub-block=1 calculates it on F (x, y, t) image Between pyramid histogram H (Bi(t) piecemeal operation again, l class resolution ratios), i.e., are carried out respectively according to 0~L of resolution ratio to sub-block It is lower by sub-block width and high two dimensions are all divided into 2lSection, wherein l=1,2 ... L, so as to which sub-block is separated into some fritters, unite Color histogram of each fritter on R passages is counted, the histogram vectors of each fritter are finally multiplied by weight betalHead and the tail phase afterwards Company forms the spatial pyramid histogram of the sub-block：
S43：Sub-block to may each include flame in present frame, finds corresponding sub-block in its previous frame image, I.e. in previous frame image by this frame be presently in reason sub-block centered on original position, in the range of-R~R search for calculate The spatial pyramid histogram of all sub-blocks, calculates the distance between the histogram of its sub-block handled by with present frame, and distance is minimum Sub-block be with present frame handled by the corresponding sub-block of sub-block；
If Bi(t) sub-block represented is R (xi1:xi2,yi1:yi2, t), above-mentioned detailed process such as formula (13)
Wherein, Bc,i(t-m) be in preceding m frames with present frame handled by the corresponding sub-block of sub-block, H () represents zoning Spatial pyramid histogram, dis [] represent calculate two histograms distance function, histogram intersection letter can be applied Number[5]Calculate
Wherein, Hx(i),Hy(i) two histogram vectors i-th dimension data, D are represented respectivelyHRepresent the dimension of vector.
S44：According to the corresponding sub-block determined in S43, the space that currently processed sub-block corresponds to sub-block with former frame is calculated The distance between pyramid histogram；As final judgement foundation after the result of calculation of multiframe is carried out averagely, such as formula (15) shown in：
Wherein Bc,i(t-1) corresponding sub-block of i-th of the sub-block of t in former frame, N are representedmExpression carries out average frame Number；
S45：According to the distance calculated in S44, judge to whether there is flame in currently processed sub-block, method such as formula (16) It is shown
Wherein, TsFor threshold value, if at least one sub-block is judged as flame being present in image, then it is assumed that the two field picture is deposited In flame.
In order to verify effectiveness of the invention, computer simulation experiment has been carried out.In an experiment, experiment parameter CPU IntelR CoreTM i5 2.4GHz, 6G internal memories, video card are AMD Radeon HD 6470, and system is Window7 Ultimates, Software programming environment is Matlab2014a.
The video that the present invention tests is the color video of common CCD camera shooting, is carried out from open-minded university of South Korea etc. The experiment video that the laboratory of fire detection area research based on image procossing is issued, and for detection method performance institute The video specially shot.
Weighing a kind of good and bad standard of fire defector algorithm mainly has two, verification and measurement ratio rdWith false drop rate (false alarm rate) fa, its Definition is respectively such as formula (20) and formula (21).
Wherein, npRepresent the number of image frames for including flame, nnRepresent the number of image frames without flame, ntpRepresent algorithm detection Contain the frame number containing flame, i.e. true positives frame number really in flame and image, nfpRepresent to work as and do not contained in real image Flame frame number added by flame but algorithm at survey, i.e., pseudo- positive frame number.Verification and measurement ratio rdWith false alarm rate faReflect fire defector calculation The quality of method performance.Wherein, verification and measurement ratio represents the probability being detected in the presence of flame, and verification and measurement ratio is higher, shows the detection system The reliability of system is stronger.False alarm rate expression sends the probability of alarm when flame be present, can reflect algorithm to a certain extent Stability.False alarm rate is lower, shows that the degree that the detecting system is disturbed by other objects is lower, the operability applied to reality It is stronger.
The method that this paper is improved and document [1] and document [2] simulation comparison experiment is subjected to herein, specific emulation As a result, refer to shown in figure below and following table.Wherein experiment video comes from the open-minded university's fire detection laboratory of South Korea[6]。
The algorithm performance of table 1 compares
As it can be seen from table 1 in terms of false drop rate, set forth herein method document [1] and the method phase in document [2] Than no matter being showed in terms of verification and measurement ratio with false drop rate excellent.It is the trees of burning in Fig. 2 (a), the substantial amounts of cigarette of surrounding is to flame There is certain effect of blocking, if using common Hard clustering color model, it will there are a large amount of flame region pixels to be missed, but Flame notable figure can reduce or even avoid the generation of the above situation, obtain preferable testing result.Fig. 2 figures (b) are clapped for night Take the photograph video, ambient lighting is dark, the successful detection of the scene show it is bright with compared with dark situation, above-mentioned algorithm can obtain compared with Good Detection results.Apart from camera farther out, and flame burn area is smaller for Fig. 2 figures (c) Flame, it is more difficult to is detected. The pedestrian move in Fig. 2 (d) and Fig. 2 (f), red vehicle etc., is respectively provided with stronger interference effect in Fig. 2 (e), but based on empty Between the algorithm achievement of pyramid histogram chaff interference is excluded, obtained relatively low false drop rate.
The foregoing is only a preferred embodiment of the present invention, but protection scope of the present invention be not limited thereto, Any one skilled in the art the invention discloses technical scope in, technique according to the invention scheme and its Inventive concept is subject to equivalent substitution or change, should all be included within the scope of the present invention.
Bibliography
[1]CELIK T,DEMIREL H,OZKARAMANLI H.Automatic fire detection in video sequences.Fire Safety Journal,2006,6(3):233–240.
[2]HABIBOGLU Y H,GUNAY O,CETIN A E.Covariance matrix-based fire and flame detection method in video.Mach Vision Appl,2012,23(6):1103-1113.
[3]T REYIN B U.Fire detection algorithms using multimodal signal and image analysis；Bilkent University,2009.
[4] Wang Shu texts distribution microphone arrays localization method research Dalian University of Technology, 2013.
[5]LAZEBNIK S,SCHMID C,PONCE J.Beyond bags of features:Spatial pyramid matching for recognizing natural scene categories.Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2006.
[6]KO B,CHEONG K H,NAM J Y.Early fire detection algorithm based on irregular patterns of flames and hierarchical Bayesian Networks.Fire Safety J,2010,45(4):262-70。

Claims (4)

  1. A kind of 1. fire detection method based on flame notable figure and spatial pyramid histogram, it is characterised in that：Including following Step：
    S1：The Strength Changes value and prospect degree of each pixel in image are calculated, method and Gaussian Mixture flame are made the difference using successive frame Color model obtains flame notable figure；
    S2：According to flame notable figure, the pixel of candidate's flame is filtered out using thresholding method, constructs and contains candidate's flame picture The mask image of vegetarian refreshments；
    S3：Original image is handled as follows according to mask image：If the value of pixel is 1 in mask image, correspondence position Original image in pixel retain initial value, otherwise the value of pixel is set to zero in original image；By above-mentioned treated original Beginning image is divided into multiple sub-blocks, calculates candidate's flame in row block number, row block number and each sub-block that image is divided out The number of pixel is so as to judging whether the sub-block contains flame；
    S4：Corresponding sub-block in previous frame image is found by sub-block, and the spatial pyramid of sub-block is corresponded to using successive frame Distance Judgment image between histogram whether there is flame；
    It is further characterized in that：In S1 specifically in the following way：
    S11：Calculate the Strength Changes value of each pixel in image：
    <mrow> <msub> <mi>P</mi> <mrow> <mi>d</mi> <mi>i</mi> <mi>f</mi> <mi>f</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>,</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mo>|</mo> <mi>I</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>,</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>-</mo> <mi>I</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>,</mo> <mi>t</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>|</mo> </mrow> <mn>255</mn> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>
    Wherein：Pdiff(x, y, t) is the Strength Changes value for the pixel that t is located at (x, y) position, and I (x, y, t) is t position In the intensity level of the pixel of (x, y) position；
    S12：The prospect degree of each pixel in image is calculated, the prospect degree is to belong to prospect in foreground detection stage pixel Probable value：
    <mrow> <msub> <mi>P</mi> <mi>F</mi> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>,</mo> <mi>t</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>N</mi> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <msub> <mi>&amp;omega;</mi> <mi>i</mi> </msub> <mi>log</mi> <mi> </mi> <msub> <mi>P</mi> <mrow> <mi>d</mi> <mi>i</mi> <mi>f</mi> <mi>f</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>,</mo> <mi>t</mi> <mo>-</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>
    Wherein, PF(x, y, t) represents the prospect degree of t position (x, y) place pixel, and it is that the intensity of N frames before t is become The weighting logarithm value summation of change, describes the characteristics of flame region intensity continuous changes, wiIt is weight coefficient；
    S13：According to the Gaussian Mixture flame color model trained, each pixel color is calculated in RGB color as fire The probability of flame color：
    <mrow> <msub> <mi>P</mi> <mi>c</mi> </msub> <mrow> <mo>(</mo> <mi>q</mi> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>,</mo> <mi>t</mi> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </munderover> <msub> <mi>&amp;alpha;</mi> <mi>k</mi> </msub> <msub> <mi>&amp;eta;</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>q</mi> <mo>(</mo> <mrow> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>,</mo> <mi>t</mi> </mrow> <mo>)</mo> <mo>|</mo> <msub> <mi>&amp;mu;</mi> <mi>k</mi> </msub> <mo>,</mo> <msub> <mi>&amp;Sigma;</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow>
    Wherein, q (x, y, t) is color vector of the pixel at t position (x, y) place in RGB color, representation For
    Q (x, y, t)={ R (x, y, t), G (x, y, t), B (x, y, t) } (4)
    K is the number of unimodal Gaussian density function component in the Gaussian Mixture flame color model trained, μkAnd ΣkRespectively It is the mean vector and covariance matrix of k-th of component in the Gaussian Mixture flame color model trained, αkIt is each Gauss The weight of component；Function ηkThe expression formula of () is
    <mrow> <msub> <mi>&amp;eta;</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>q</mi> <mo>|</mo> <msub> <mi>&amp;mu;</mi> <mi>k</mi> </msub> <mo>,</mo> <msub> <mi>&amp;Sigma;</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <msup> <mrow> <mo>(</mo> <mn>2</mn> <mi>&amp;pi;</mi> <mo>)</mo> </mrow> <mrow> <mi>D</mi> <mo>/</mo> <mn>2</mn> </mrow> </msup> <mo>|</mo> <msub> <mi>&amp;Sigma;</mi> <mi>k</mi> </msub> <msup> <mo>|</mo> <mrow> <mn>1</mn> <mo>/</mo> <mn>2</mn> </mrow> </msup> </mrow> </mfrac> <mi>exp</mi> <mo>&amp;lsqb;</mo> <mo>-</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mrow> <mo>(</mo> <mi>q</mi> <mo>-</mo> <msub> <mi>&amp;mu;</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mi>T</mi> </msup> <msubsup> <mi>&amp;Sigma;</mi> <mi>k</mi> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msubsup> <mrow> <mo>(</mo> <mi>q</mi> <mo>-</mo> <msub> <mi>&amp;mu;</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow>
    Wherein, D is vectorial q dimension, wherein, D=3；
    S14：According to prospect degree and flame color probability structure flame notable figure：
    According to formula (2) calculate pixel prospect degree and by formula (3) calculate each pixel of flame color probability calculation it is notable Degree
    fs(x, y, t)=PF(x,y,t)+logPc(q(x,y,t)) (6)
    Wherein fs(x, y, t) is the significance of t position (x, y) place pixel, and (x, y) place picture in flame notable figure The pixel value of vegetarian refreshments.
  2. 2. a kind of fire detection method based on flame notable figure and spatial pyramid histogram according to claim 1, It is characterized in that：In S2 in the following way：
    According to the flame Saliency maps obtained by above-mentioned formula (6), noise spot is filtered out by Threshold segmentation, obtains including candidate's flame The mask image of pixel：
    Wherein, fsFor the flame notable figure of inputted video image, τfFor threshold value obtained by experiment, if Ms(x, y, t)=1, representing should Pixel may be flame pixels, need to further be judged；If Ms(x, y, t)=0, then it represents that the point is not necessarily in flame zone In domain, without subsequent treatment.
  3. 3. a kind of fire detection method based on flame notable figure and spatial pyramid histogram according to claim 1, It is characterized in that：S3 by sub-block to being handled in the following way：
    S31：Width and high respectively W according to sub-blockbAnd HbPiecemeal is carried out to image, if the width of image and high respectively WiAnd Hi, Then image will be separated out next row block number NrWith row block number NcTo be
    Wherein,Operator represents downward rounding operation；
    S32：The number of candidate's flame pixels in each sub-block is counted in the following way, flame pixels number is few Sub-block be determined as nonflame sub-block, without subsequent treatment：
    Wherein, Mb,i(t) possibility that i-th of sub-block of t includes flame is represented, the value is that 1 expression i-th of sub-block of t can Flame can be contained, spatial pyramid statistics with histogram will be carried out to it, the value be 0 expression the sub-block do not contain flame, do not have into Row subsequent treatment；Ms(x, y, t) represents the mask image for including candidate's flame pixels point, Bi(t) represent what i-th of sub-block represented Region, TbThe proportion threshold value of whole sub-block number of pixels is accounted for for candidate's flame pixels point.
  4. 4. a kind of fire detection method based on flame notable figure and spatial pyramid histogram according to claim 3, It is characterized in that：S4 falls into a trap the spatial pyramid histogram of each sub-block of nomogram picture, and according to corresponding in continuous multiple image Whether the Distance Judgment sub-block between the pyramid histogram of block contains flame：
    S41：R channel images are selected to calculate spatial pyramid histogram, will according to the mask image containing candidate's flame pixels point Necessarily keep initial value constant not at the pixel R channel value zero setting of flame region, candidate pixel point, i.e.,
    S42：To it is each may be M containing the sub-block of flameb,i(t) sub-block B=1i(t) it is empty that it is calculated on F (x, y, t) image Between pyramid histogram H (Bi(t) piecemeal operation again, l class resolution ratios), i.e., are carried out respectively according to 0~L of resolution ratio to sub-block It is lower by sub-block width and high two dimensions are all divided into 2lSection, wherein l=1,2 ... L, so as to which sub-block is separated into some fritters, Ms (x, y, t) represents the mask image for including candidate's flame pixels point, counts color histogram of each fritter on R passages, most The histogram vectors of each fritter are multiplied by weight beta afterwardslJoin end to end afterwards and form the spatial pyramid histogram of the sub-block：
    <mrow> <msub> <mi>&amp;beta;</mi> <mi>l</mi> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <msup> <mn>2</mn> <mrow> <mi>L</mi> <mo>-</mo> <mi>l</mi> </mrow> </msup> </mfrac> <mo>=</mo> <msup> <mn>2</mn> <mrow> <mi>l</mi> <mo>-</mo> <mi>L</mi> </mrow> </msup> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>12</mn> <mo>)</mo> </mrow> </mrow>
    S43：Sub-block to may each include flame in present frame, corresponding sub-block is found in its previous frame image, that is, is existed In previous frame image by this frame be presently in the sub-block of reason centered on original position, in the range of-R~R search calculate all The spatial pyramid histogram of sub-block, calculate the distance between the histogram of its sub-block handled by with present frame, the minimum son of distance Block be with present frame handled by the corresponding sub-block of sub-block；
    S44：According to the corresponding sub-block determined in S43, the space gold word that currently processed sub-block corresponds to sub-block with former frame is calculated The distance between tower histogram；As final judgement foundation after the result of calculation of multiframe is carried out averagely, such as formula (15) institute Show：
    <mrow> <msub> <mi>Dis</mi> <mrow> <mi>m</mi> <mi>e</mi> <mi>a</mi> <mi>n</mi> </mrow> </msub> <mrow> <mo>(</mo> <mrow> <msub> <mi>B</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mi>m</mi> </msub> </mfrac> <mrow> <mo>(</mo> <mrow> <mi>d</mi> <mi>i</mi> <mi>s</mi> <mrow> <mo>&amp;lsqb;</mo> <mrow> <mi>H</mi> <mrow> <mo>(</mo> <mrow> <msub> <mi>B</mi> <mrow> <mi>c</mi> <mo>,</mo> <mi>i</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> <mo>,</mo> <mi>H</mi> <mrow> <mo>(</mo> <mrow> <msub> <mi>B</mi> <mrow> <mi>c</mi> <mo>,</mo> <mi>i</mi> </mrow> </msub> <mrow> <mo>(</mo> <mrow> <mi>t</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> </mrow> <mo>&amp;rsqb;</mo> </mrow> <mo>+</mo> <munderover> <mi>&amp;Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mi>t</mi> <mo>-</mo> <mn>1</mn> </mrow> <mrow> <mi>t</mi> <mo>-</mo> <msub> <mi>N</mi> <mi>m</mi> </msub> <mo>+</mo> <mn>1</mn> </mrow> </munderover> <mi>d</mi> <mi>i</mi> <mi>s</mi> <mrow> <mo>&amp;lsqb;</mo> <mrow> <mi>H</mi> <mrow> <mo>(</mo> <mrow> <msub> <mi>B</mi> <mrow> <mi>c</mi> <mo>,</mo> <mi>i</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>j</mi> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> <mo>,</mo> <mi>H</mi> <mrow> <mo>(</mo> <mrow> <msub> <mi>B</mi> <mrow> <mi>c</mi> <mo>,</mo> <mi>i</mi> </mrow> </msub> <mrow> <mo>(</mo> <mrow> <mi>j</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> </mrow> <mo>&amp;rsqb;</mo> </mrow> </mrow> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>15</mn> <mo>)</mo> </mrow> </mrow>
    Wherein Bc,i(t) i-th of sub-block of t corresponding sub-block in the current frame, B are representedc,i(t-1) i-th of son of t is represented Corresponding sub-block of the block in former frame, NmExpression carries out average frame number, and dis [] represents to calculate the letter of the distance of two histograms Number, is calculated as follows
    <mrow> <mi>d</mi> <mi>i</mi> <mi>s</mi> <mrow> <mo>(</mo> <msub> <mi>H</mi> <mi>x</mi> </msub> <mo>,</mo> <msub> <mi>H</mi> <mi>y</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <msub> <mi>D</mi> <mi>H</mi> </msub> </munderover> <mi>m</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <msub> <mi>H</mi> <mi>x</mi> </msub> <mo>(</mo> <mi>i</mi> <mo>)</mo> <mo>,</mo> <msub> <mi>H</mi> <mi>y</mi> </msub> <mo>(</mo> <mi>i</mi> <mo>)</mo> <mo>)</mo> </mrow> </mrow>
    Wherein, Hx(i),Hy(i) two histogram vectors H are represented respectivelyxAnd HyI-th dimension data, DHRepresent vectorial HxAnd HyDimension Degree；
    S45：According to the distance calculated in S44, judge to whether there is flame in currently processed sub-block, shown in method such as formula (16)
    Wherein, TsFor threshold value, if at least one sub-block is judged as flame being present in image, then it is assumed that the two field picture has fire Flame.
CN201510503877.7A 2015-08-14 2015-08-14 A kind of fire detection method based on flame notable figure and spatial pyramid histogram Active CN105139429B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510503877.7A CN105139429B (en) 2015-08-14 2015-08-14 A kind of fire detection method based on flame notable figure and spatial pyramid histogram

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510503877.7A CN105139429B (en) 2015-08-14 2015-08-14 A kind of fire detection method based on flame notable figure and spatial pyramid histogram

Publications (2)

Publication Number Publication Date
CN105139429A CN105139429A (en) 2015-12-09
CN105139429B true CN105139429B (en) 2018-03-13

Family

ID=54724761

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510503877.7A Active CN105139429B (en) 2015-08-14 2015-08-14 A kind of fire detection method based on flame notable figure and spatial pyramid histogram

Country Status (1)

Country Link
CN (1) CN105139429B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105741481B (en) * 2016-04-21 2018-07-06 大连理工大学 A kind of fire monitoring positioning device and fire monitoring localization method based on binocular camera
CN106546351A (en) * 2016-11-08 2017-03-29 天津艾思科尔科技有限公司 A kind of explosion-proof detector with temperature detection function
CN107016679B (en) * 2017-02-23 2019-10-11 中国南方电网有限责任公司超高压输电公司广州局 A kind of mountain fire detection method based on single picture
CN107328392A (en) * 2017-07-05 2017-11-07 贵州大学 Flame abnormal signal extracting method based on Euler's video enhancement techniques
CN110084160B (en) * 2019-04-16 2021-08-10 东南大学 Video forest smoke and fire detection method based on motion and brightness significance characteristics
CN110334685A (en) * 2019-07-12 2019-10-15 创新奇智(北京)科技有限公司 Flame detecting method, fire defector model training method, storage medium and system
CN111898549A (en) * 2020-07-31 2020-11-06 平安国际智慧城市科技股份有限公司 Fire monitoring method and device based on artificial intelligence, computer equipment and medium
CN112949453B (en) * 2021-02-26 2023-12-26 南京恩博科技有限公司 Training method of smoke and fire detection model, smoke and fire detection method and equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2000998A2 (en) * 2007-05-31 2008-12-10 Industrial Technology Research Institute Flame detecting method and device
CN104504382A (en) * 2015-01-13 2015-04-08 东华大学 Flame identifying algorithm based on image processing technologies
CN104809463A (en) * 2015-05-13 2015-07-29 大连理工大学 High-precision fire flame detection method based on dense-scale invariant feature transform dictionary learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2000998A2 (en) * 2007-05-31 2008-12-10 Industrial Technology Research Institute Flame detecting method and device
CN104504382A (en) * 2015-01-13 2015-04-08 东华大学 Flame identifying algorithm based on image processing technologies
CN104809463A (en) * 2015-05-13 2015-07-29 大连理工大学 High-precision fire flame detection method based on dense-scale invariant feature transform dictionary learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Research on Evaluation of Fire Detection Algorithms;LI Jian等;《火灾科学》;20050731;第14卷(第3期);第144-149页 *
应用GMM的快速火焰检测;唐岩岩 等;《计算机科学》;20121130;第39卷(第11期);第283-285,297页 *

Also Published As

Publication number Publication date
CN105139429A (en) 2015-12-09

Similar Documents

Publication Publication Date Title
CN105139429B (en) A kind of fire detection method based on flame notable figure and spatial pyramid histogram
CN109816024B (en) Real-time vehicle logo detection method based on multi-scale feature fusion and DCNN
CN111444821B (en) Automatic identification method for urban road signs
CN103400111B (en) Method for detecting fire accident on expressway or in tunnel based on video detection technology
CN109409242B (en) Black smoke vehicle detection method based on cyclic convolution neural network
CN105427626B (en) A kind of statistical method of traffic flow based on video analysis
CN109871763A (en) A kind of specific objective tracking based on YOLO
CN109523476B (en) License plate motion blur removing method for video detection
CN110490043A (en) A kind of forest rocket detection method based on region division and feature extraction
CN106339657B (en) Crop straw burning monitoring method based on monitor video, device
CN111222478A (en) Construction site safety protection detection method and system
CN103745216B (en) A kind of radar image clutter suppression method based on Spatial characteristic
CN109190455B (en) Black smoke vehicle identification method based on Gaussian mixture and autoregressive moving average model
CN109241902A (en) A kind of landslide detection method based on multi-scale feature fusion
CN107273852A (en) Escalator floor plates object and passenger behavior detection algorithm based on machine vision
CN109165602A (en) A kind of black smoke vehicle detection method based on video analysis
CN109271904A (en) A kind of black smoke vehicle detection method based on pixel adaptivenon-uniform sampling and Bayesian model
CN104463104A (en) Fast detecting method and device for static vehicle target
Tao et al. Smoky vehicle detection based on multi-feature fusion and ensemble neural networks
CN109325426B (en) Black smoke vehicle detection method based on three orthogonal planes time-space characteristics
CN109446938B (en) Black smoke vehicle detection method based on multi-sequence double projection
CN111062319A (en) Driver call detection method based on active infrared image
CN112836608B (en) Forest fire source estimation model training method, estimation method and system
CN108985197B (en) Automatic detection method for taxi driver smoking behavior based on multi-algorithm fusion
CN115527118A (en) Remote sensing image target detection method fused with attention mechanism

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