CN106339657A - Straw incineration monitoring method and device based on monitoring video - Google Patents
Straw incineration monitoring method and device based on monitoring video Download PDFInfo
- Publication number
- CN106339657A CN106339657A CN201510400389.3A CN201510400389A CN106339657A CN 106339657 A CN106339657 A CN 106339657A CN 201510400389 A CN201510400389 A CN 201510400389A CN 106339657 A CN106339657 A CN 106339657A
- Authority
- CN
- China
- Prior art keywords
- image
- crop straw
- straw burning
- pixel
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a straw incineration monitoring method based on a monitoring video. The method is based on the real-time and easy-to-acquire monitoring video, and utilizes a video image processing technology and a pattern identification technology to analyze and process the video image. A straw incineration incident in a video monitoring area is intelligently identified, and an alarm is given. The method provided by the invention has the advantages that a classifier is trained; a down sampling pre-processing method is used to overcome the problem of imbalanced positive and negative samples; the classifier is used to preliminarily identify the straw incineration incident; and the color gradient magnitude value which reflects smoke image gradation edge information is used to correct the preliminary identification result of the classifier, so as to acquire a more accurate monitoring result. The invention further discloses a straw incineration monitoring device based on the monitoring video. Compared with the prior art, the method and device have the advantages that the monitoring video is used to carry out full intelligent monitoring on straw incineration, and have the characteristics of more accurate monitoring result and better instantaneity.
Description
Technical field
The present invention relates to video image intelligent technology of identification field, more particularly, to a kind of straw based on monitor video
Burn monitoring method, device.
Background technology
In recent years, agricultural crop straw becomes the new source of Rural Plane Source Pollution.The annual summer harvest and when the autumn and winter, always
There are the straws such as substantial amounts of Semen Tritici aestivi, Semen Maydiss to burn in field, create a large amount of heavy smogs, not only become rural area
The bottleneck problem of environmental conservation, or even become the arch-criminal bringing disaster to urban environment.Instantly enjoy the haze denouncing
Weather, one of may inducement be straw burning.Therefore, administration sections at different levels and environmental protection unit are for straw
The strick precaution demand of burning also becomes increasingly urgently.
Currently for the anti-flaming application scenarios of straw, common method is to be analyzed and processed by satellite remote sensing images.
However, satellite remote sensing has that poor in timeliness and spatial resolution are low, parameter comparison such as following table.
In addition, for basic unit's environmental protection unit, being difficult to obtain remote sensing satellite image.Therefore satellite remote sensing technology is many
Used with above environmental administration of provinces and cities, major function is to solve the problems, such as fix duty afterwards, rather than prior Precautions.
With the rapid application of technology of Internet of things and correlation technique, straw burning precautionary technology is dashed forward
Broken, for example, by unmanned aerial photography, the method such as video monitoring, to straw burning key area real-time image acquisition,
And monitor the purpose anti-flaming to reach straw.But the method for this kind of video monitoring still has a problem that
It is exactly that video scene still needs special messenger uninterruptedly on duty, still labor intensive material resources.
At present in terms of the wooded mountain fire protection, occur in that the intelligent identification technology being based partially on video monitoring image, can
Intelligently brushfire is monitored according to monitor video.However, due to there is difference using scene,
These technology cannot be applied to the intellectual monitoring of crop straw burning.For example, a Chinese patent application (Application No.
Cn201410043121, a kind of entitled " mountain fire image recognition side of electric power line pole tower image monitoring system
Method ") disclose a kind of mountain fire image-recognizing method of electric power line pole tower image monitoring system, this invention adopts
Color characteristic threshold value and otsu image partition method extract flame region, extract smog using wavelet transformation and obscure
Change feature, and be analyzed using another group of color characteristic threshold value；Because mountain fire is easily sent out in dry weather, this
Bright need to carry out auxiliary judgment using humidity detection data；And field straw burns monitoring and cannot obtain these auxiliary
Monitoring Data, and all possible someone's stealthily crop straw burning, fixing color characteristic threshold under substantially any weather condition
Value parameter obviously can not be suitable for.
Crop straw burning, due to being artificially to cause, is round-the-clock, and has village local-style dwelling houses, at a high speed public affairs near farmland
The complex scenes such as road, the woods, rivers and lakes are it is therefore desirable to one kind more robust, effective intelligent monitoring method.
Content of the invention
The technical problem to be solved is to overcome prior art not enough, provides one kind to be based on monitor video
Crop straw burning monitoring method, available monitor video carries out full intellectual monitoring to crop straw burning, and monitoring result
Accuracy is higher, and real-time is more preferable.
The crop straw burning monitoring method based on monitor video for the present invention, comprises the following steps:
There is the image positive sample of crop straw burning and there is not the image negative sample of crop straw burning in step a, acquisition, and profit
There is not the sample size of the image negative sample of crop straw burning with Downsapling method reduction, obtain training sample set；
Then extract the characteristics of image that training sample concentrates each training sample, two classification device model is trained, obtains
To grader；
Step b, obtain the absolute difference image with respect to reference frame image for the current frame image of monitor video, and to absolute
Difference image carries out binary conversion treatment: by below skyline in absolute difference image partly middle pixel value be more than predetermined threshold value
Pixel be set to target, rest of pixels is disposed as background, obtains the absolute difference image of two-value；
Step c, partly scanned pixel-by-pixel using below skyline in sliding window absolute difference image to described two-value,
As in current window, object pixel proportion exceedes preset ratio threshold value, then by current frame image correspond to work as
The characteristics of image of the video in window of front window position inputs described grader and is classified, the taking of described proportion threshold value
Value scope is 10%～70%；If classification results are to exist to correspond to current window in crop straw burning and current frame image
The color gradient range value of the video in window of mouth position is less than predetermined gradient threshold value, then judge to deposit in current frame image
In a crop straw burning event.
Preferably, described Downsapling method is serial clustering method, specific as follows: with first image negative sample
As a sample cluster；From the beginning of second image negative sample, judge this image negative sample in default feature successively
In space, whether the minimum Eustachian distance and each sample cluster center having existed between is more than predeterminable range threshold value,
In this way, then using this image negative sample as a new sample cluster, such as no, then by this image negative sample put under away from
Exist in sample cluster from nearest；After the completion of all image negative samples are processed, by currently each sample cluster
The heart is as new image negative sample.Preferably, described default feature space is the normalization of hsl color space
The feature space that histogram feature is constituted.
Preferably, described image is characterized as histogram feature and the local binary patterns feature of hsl color space
Combination.
Existing various horizons line detecting method can be adopted in technique scheme, multiple in order to reduce algorithm further
Miscellaneous degree, improves monitoring real-time it is preferable that described skyline is obtained using following methods detection:
Step 1, using symmetrical exponential function wave filter in Shen Jun edge detection operator, current frame image is carried out vertically
Direction Fast Recursive filtering, obtains filtering image；
Step 2, the error image to filtering image and current frame image carry out binary conversion treatment: by all of on the occasion of picture
Element is all entered as 1, and other pixel is all 0；
Step 3, expansion process is carried out to obtained bianry image；
Step 4, monochrome pixels that the image after expansion process is carried out are anti-phase, then carry out label processing, and note goes up most
Portion background area is sky areas；
Step 5, each pixel belonging to sky areas in the image after expansion process is respectively processed: as with current
There is the pixel that more than one value is 1, then by current picture in the neighborhood of 3 × 3 pixel sizes centered on pixel
Element is labeled as object pixel, is otherwise labeled as background, thus obtaining new image；
Step 6, using Hough transformation, straight-line detection is carried out to new image obtained by step 5, and from detected
Most one of object pixel thereon is taken as skyline in straight line.
Preferably, the color gradient range value of image obtains by the following method: is somebody's turn to do in rgb color space
Three color component images of image, respectively with the sobel gradient magnitude of all pixels in each color component images
Sum as the sobel gradient magnitude of this color component images, then from the sobel of three color component images
Maximum is selected as the color gradient range value of this image in gradient magnitude.
Preferably, described two classification device model is support vector cassification model.
Preferably, described monitor video is obtained by being arranged on the video monitoring equipment on mobile communication base station tower body top
Obtain.
Technical scheme below can also be obtained according to identical invention thinking:
Crop straw burning monitoring device based on monitor video, comprising:
Video acquisition unit, for the real-time monitor video obtaining region to be monitored；
Two-value absolute difference image acquisition unit, for obtaining the current frame image of monitor video with respect to reference frame image
Definitely difference image, and absolute difference image is carried out with binary conversion treatment: by part below skyline in absolute difference image
The pixel that middle pixel value is more than predetermined threshold value is set to target, and rest of pixels is disposed as background, obtains two-value exhausted
To difference image；
Crop straw burning event identifying unit, including grader, color gradient range value computing module, scan module；Its
In, training in advance obtains described grader in accordance with the following methods: obtain exist crop straw burning image positive sample and
There is not the image negative sample of crop straw burning, and born using the image that Downsapling method reduction does not have crop straw burning
The sample size of sample, obtains training sample set；Then extracting training sample concentrates the image of each training sample special
Levy, two classification device model is trained；Described color gradient range value computing module is used for calculating present frame
The color gradient range value of the video in window of current window position is corresponded in image；Described scan module is used for profit
With partly being scanned pixel-by-pixel below skyline in sliding window absolute difference image to described two-value, such as current window
In mouthful, object pixel proportion exceedes preset ratio threshold value, then will correspond to current window position in current frame image
The characteristics of image of the video in window put inputs described grader and is classified, and the span of described proportion threshold value is
10%～70%；If classification results are to exist to correspond to current window position in crop straw burning and current frame image
The color gradient range value of video in window is less than predetermined gradient threshold value, then judge to exist straw in current frame image
Stalk burns event.
Preferably, described Downsapling method is serial clustering method, specific as follows: with first image negative sample
As a sample cluster；From the beginning of second image negative sample, judge this image negative sample in default feature successively
In space, whether the minimum Eustachian distance and each sample cluster center having existed between is more than predeterminable range threshold value,
In this way, then using this image negative sample as a new sample cluster, such as no, then by this image negative sample put under away from
Exist in sample cluster from nearest；After the completion of all image negative samples are processed, by currently each sample cluster
The heart is as new image negative sample.Preferably, described default feature space is the normalization of hsl color space
The feature space that histogram feature is constituted.
Preferably, described image is characterized as histogram feature and the local binary patterns feature of hsl color space
Combination.
Preferably, described skyline is obtained using following methods detection:
Step 1, using symmetrical exponential function wave filter in Shen Jun edge detection operator, current frame image is carried out vertically
Direction Fast Recursive filtering, obtains filtering image；
Step 2, the error image to filtering image and current frame image carry out binary conversion treatment: by all of on the occasion of picture
Element is all entered as 1, and other pixel is all 0；
Step 3, expansion process is carried out to obtained bianry image；
Step 4, monochrome pixels that the image after expansion process is carried out are anti-phase, then carry out label processing, and note goes up most
Portion background area is sky areas；
Step 5, each pixel belonging to sky areas in the image after expansion process is respectively processed: as with current
There is the pixel that more than one value is 1, then by current picture in the neighborhood of 3 × 3 pixel sizes centered on pixel
Element is labeled as object pixel, is otherwise labeled as background, thus obtaining new image；
Step 6, using Hough transformation, straight-line detection is carried out to new image obtained by step 5, and from detected
Most one of object pixel thereon is taken as skyline in straight line.
Preferably, described color gradient range value computing module calculates the color gradient width of image by the following method
Angle value: three color component images to this image in rgb color space, respectively with each color component images
The sobel gradient magnitude sum of middle all pixels as the sobel gradient magnitude of this color component images, then
Select maximum as the color gradient amplitude of this image from the sobel gradient magnitude of three color component images
Value.
Preferably, described two classification device model is support vector cassification model.
Preferably, described video acquisition unit is to be arranged on the video monitoring equipment on mobile communication base station tower body top.
Further, described crop straw burning event count is also included based on the crop straw burning monitoring device of monitor video
Device and alarm unit；Crop straw burning event counter is used for counting that crop straw burning event identifying unit judged works as
Crop straw burning event number in the presence of prior image frame；Alarm unit is used in the presence of current frame image
Crop straw burning event number is compared with preset alarm threshold value, such as larger than alarm threshold value, then reported to the police.
Compared to existing technology, the method have the advantages that
Present invention support image processing and analyzing technology, the technological trend of environmental protection industry (epi) of combining closely, for user
The picture signal that some video monitorings obtain analyzes and processes in real time, to realize 7*24 hour, in real time, automatically to supervise
Survey and quick early warning is made to straw burning scene；
The present invention is directed to training sample imbalance problem, and mainly positive sample number is far less than negative sample number, if
Do not carry out pretreatment, be then easy to occur training precision 99%, and the situation of verification and measurement ratio almost 0, profit is proposed
Reduce Negative training sample number with serial clustering method, make the positive sample accuracy of identification that grader burns to pyrotechnics significantly
Improve；
Actual scene that the present invention monitors according to crop straw burning is it is proposed that a kind of brand-new quick skyline detection side
Method, on the one hand can reduce process range, improve recognition speed and precision, on the other hand reduce due to sky cloud
The error detection that color motion causes；
The present invention, on the basis of the classification results of grader, carries out comprehensively sentencing further combined with color gradient range value
Disconnected, effectively improve monitoring precision, reduce rate of false alarm.
Brief description
Fig. 1 is the schematic flow sheet of the inventive method；
Fig. 2 is the current frame image in specific embodiment；
Fig. 3 is the filtering image after current frame image is carried out with the filtering of vertical direction Fast Recursive binaryzation；
Fig. 4 is the filtering image after expansive working；
Fig. 5 is the sky areas image obtaining after label processing；
Fig. 6 be each pixel belonging to sky areas is respectively processed after the new image that obtains；
Fig. 7 is the straight line being detected using Hough transformation；
Fig. 8 is the current frame image being superimposed skyline；
Fig. 9 is the crop straw burning event in current frame image using the preliminary identification of grader.
Specific embodiment
Below in conjunction with the accompanying drawings technical scheme is described in detail:
The thinking of the present invention is monitor video that is good based on real-time and being easily obtained, using Computer Vision skill
Art and mode identification technology are analyzed to video image processing, to the crop straw burning thing in video monitoring regional
Part carries out Intelligent Recognition and reports to the police.The inventive method trains grader first, then using grader, straw is burnt
Burning event is tentatively identified, recycle reflection pyrotechnics image gradient marginal information color gradient range value to point
The preliminary recognition result of class device is modified, thus available more accurately monitoring result.
The basic procedure of monitoring method of the present invention as shown in figure 1, mainly including training stage and real-time monitoring stage,
It is described in detail separately below.
First, the training stage:
There is the image positive sample of crop straw burning and there is not the image negative sample of crop straw burning in step 1, acquisition,
And pretreatment is carried out to it.
More difficult due to collecting pyrotechnics burnup image pattern, the positive example sample number of acquisition is typically less, and non-burnup sample
This (negative example sample) is then a lot, therefore occurs in that training sample imbalance problem, mainly positive sample number much
Less than negative sample number, if not carrying out pretreatment, being easy to training precision 99%, and verification and measurement ratio is almost
Situation for 0.Training sample in conventional model training is by artificial selection, does not consider imbalance
Sample classification problem, and artificial selection's sample acquires a certain degree of difficulty for the research worker in this field.Therefore having must
Pretreatment is carried out to overcome training sample imbalance problem to sample.
In order that positive negative data reaches balance, machine learning data excavation applications generally adopt the side of resampling
Method, including the down-sampling to many numerical examples and the over-sampling to a few sample.If our over-sampling pyrotechnics burn
Sample, due to excessively approaching to training sample, the probability of over-fitting will increase.But, random under
Sampling also has shortcoming, and while we abandon mass efficient sample, some information useful to classification are also therewith
Lose.Viewpoint [chawla n v, hall l o, joshi a. according to chawla et al.
wrapper-based computation and evaluation of sampling methods for imbalanced
datasets[c]//proceedings of the 1st international workshop on utility-based data
mining(chicago,illinois,august 21-21,2005).ubdm'05.new york:acm press,
2005:24-33.], in resampling, the over-sampling to a few sample is better than to the down-sampling of many numerical examples.Cause
This, the present invention to solve the problems, such as the imbalanced data classification issue in crop straw burning monitoring using Downsapling method.
Conventional Downsapling method includes the methods such as random down-sampling and cluster, more favourable to classifying in order to retain
Information, the present invention preferentially selects serial clustering method to reduce Negative training sample number, it is to avoid general clustering method needs
The problem of clusters number to be set, makes grader that the positive sample accuracy of identification that pyrotechnics burns is greatly improved.Serial
Clustering method is specific as follows: using first image negative sample as a sample cluster；From second image negative sample
Start, judge successively this image negative sample in default feature space and each sample cluster center having existed between
Minimum Eustachian distance whether be more than predeterminable range threshold value, in this way, then this image negative sample is new as one
Sample cluster, such as no, then this image negative sample is put under closest existing in sample cluster；In all images
After the completion of negative sample is processed, using the center of currently each sample cluster as new image negative sample.
When using serial clustering method reduction Negative training sample number, existing all kinds of characteristics of image can be adopted,
Present invention preferably employs the normalization histogram feature of hsl color space.For each image negative sample,
If it is to the center of the sample cluster having existed in the normalization histogram feature space of hsl color space
Minimum Euclideam distance be more than threshold value set in advance (preferred value be 0.6) if, be increased by one new
Cluster；Otherwise, it just belongs to the cluster being located from its nearest center.After the completion of all of negative sample is processed, with
Each cluster central sample is as new image negative sample, composing training collection together with original positive image sample.
Step 2, extraction training sample concentrate the characteristics of image of each training sample, and two classification device model is carried out
Training, obtains grader.
For different application scene, the reasonable selection of characteristics of image can directly affect final result.For
For crop straw burning monitoring, the combination of preferred colors feature of the present invention and textural characteristics.In crop straw burning image,
There is the smog that burning produces, or there being naked light, or both occur simultaneously；And in combustion,
The shape of smog and flame can change in time.Therefore, the color characteristic that the present invention extracts is hsl face
The histogram feature of the colour space；Conventional textural characteristics have gabor filtering textural characteristics and lbp (local
Binary patterns, local binary patterns) feature, due to lbp feature extraction speed, disclosure satisfy that
Ageing requirement in crop straw burning monitoring, and to grey scale change, there is robustness, present invention preferably employs
Textural characteristics are lbp feature.Both the above characteristics of image is prior art, below to its concrete extraction side
Method is briefly described:
(1) histogram feature of hsl color space:
For one of image pixel, according to its h, the value of l, s component quantifies in 8 bin, so respectively
It is combined into 8*8*8=512 bin by this 8 bin afterwards, and counts total number of pixels, be finally normalized.
(2) lbp feature:
Local binary patterns are a kind of effective texture description operators, and it has rotational invariance and gray scale invariance
Etc. significant advantage, it is used widely in fields such as recognition of face, remote sensing in recent years.The present invention adopts
Lbp operator definitions are the window 3 × 3, with window center pixel as threshold value, by 8 adjacent pixels
Gray value is compared with it, if surrounding pixel values are more than the value of central point, this location of pixels is labeled 1,
It is otherwise 0.So, 8 points in 3 × 3 neighborhoods can produce the unsigned number of a 8-bit, then assigns by its position
Sued for peace to obtain an integer with different weights, that is, obtain the lbp value of this window, last statistical picture (window subgraph
Picture) interior all pixels lbp value formed lbp rectangular histogram textural characteristics.Content can be found in document [ojala in more detail
t,pietikinen m,menp t.gray scale and rotation invariant texture classification with
local binary patterns[c]//proceedings of ieee european conference on computer
vision,lecture notes in computer science.berlin heidelberg:springer,2000,1842:
404-420.].
The present invention can adopt existing various two classification device models, such as support vector machine (support vector
Machine, abbreviation svm), bayesian grader, bp neutral net, k- nearest neighbor classifier etc., in view of
The good capacity that svm shows on processing small sample problem, present invention preferably employs it is as grader.
The theoretical comparative maturity and being widely used at present of svm, here no longer more to be introduced.
2nd, the real-time monitoring stage:
Step 1, the monitoring video of acquisition target area:
Monitoring video can be obtained it is also possible to be adopted using existing video by the video capture device voluntarily setting up
The resource of collection equipment, such as power department are erected at the video image acquisition equipment on electric power line pole tower, or
Communication common carrier is erected at the video monitoring equipment on mobile communication base station tower body top.In order to obtain bigger monitoring model
Enclose, present invention preferably employs being erected at the monitoring acquired in video monitoring equipment on mobile communication base station tower body top
Video, such as Jiangsu commmunication company newly developed go out " blue sky bodyguard " system, its utilize base station iron tower resource,
On field steel tower top, network high definition ball machine photographic head is installed, the monitor video of large coverage can be obtained.Fig. 2
It is using the two field picture in the monitor video of " blue sky bodyguard " system acquisition.
Step 2, obtain the absolute difference image with respect to reference frame image for the current frame image of monitor video, and right
Definitely difference image carries out binary conversion treatment: by below skyline in absolute difference image partly middle pixel value be more than default
The pixel of threshold value is set to target, and rest of pixels is disposed as background, obtains the absolute difference image of two-value:
The purpose of this step is by detecting the scene between current frame image and reference frame image in monitor video
Change carrys out the scope of downscaled images identification.Existing scene-change detecting method is more, such as Background difference, frame difference method,
Optical flow method etc., the camera lens due to crop straw burning monitoring photographic head can be with push-and-pull control, and its background is not often fixed,
And optical flow method computation complexity is higher, be not suitable for fast monitored early warning system, the therefore present invention preferentially adopts frame poor
Method carries out Scene change detection.Simultaneously as present invention solution is that straw from village burns monitoring and warning, shooting
Be erected at as first on the signal base station frame of common carrier, monitoring range is very wide, image top quite a few
It is sky.In order to avoid being smog to the mobile error detection of cloud, reduce the amount of calculation of subsequent treatment simultaneously, this
Bright first detect skyline, be then partly identified to below skyline, so it can be avoided that to sky medium cloud
Movement error detection, and improve the whole real-time monitoring process.
The present invention can adopt existing various horizons line detecting method, in order to reduce algorithm complex further, carries
High monitoring real-time, the present invention be directed to crop straw burning monitoring special scenes feature it is proposed that one kind faster
Horizon line detecting method, specific as follows:
Step (1), (Shen Jun, based on the rim detection of multiple barrier model, pattern recognition using Shen Jun edge detection operator
With artificial intelligence, the 2nd phase 1-10 in 1989) in symmetrical exponential function wave filter current frame image is hung down
Nogata filters to Fast Recursive, obtains filtering image；
Assume to use f0(i, j) represents the corresponding gray level image of current frame image, then the mathematical expression of above-mentioned filtering is as follows:
f1(i, j)=f1(i-1,j)+a0×[f0(i,j)-f1(i-1, j)] i=1,2 ..., n
(1)
f2(i, j)=f2(i+1,j)+a0×[f1(i,j)-f2(i+1, j)] i=n, n-1 ..., 1
(2)
f2(i, j) is obtained filtering image, wherein, a0For a coefficient between (0,1) for the span,
Preferably value is 0.5.
Step (2), to filtering image f2(i, j) and current frame image f0The error image of (i, j) carries out binary conversion treatment:
All it is entered as 1 by all of on the occasion of pixel, and other pixel is all 0, obtains bianry image s；Fig. 3 is
Current frame image (Fig. 2) is carried out with the filtering image after the filtering of vertical direction Fast Recursive binaryzation.
Step (3), expansion process is carried out to obtained bianry image, obtain image s1；Fig. 4 shows Fig. 3
Result after expansion process.
Step (4), monochrome pixels that the image s1 after expansion process is carried out are anti-phase, then carry out label processing,
Note topmost background area is sky areas；In Fig. 5, largest connected background area is what label processing obtained
Sky areas.
Step (5), each pixel belonging to sky areas in image s1 obtained by step (3) is respectively processed:
There is the pixel that more than one value is 1 as in the neighborhood of 3 × 3 pixel sizes centered on current pixel, then
Current pixel is labeled as object pixel (setting pixel value is 1), is otherwise labeled as background (setting pixel value
For 0), thus obtaining new image s2；Fig. 6 is and each pixel belonging to sky areas is respectively processed
The image (boundary pixel between sky areas and ground) obtaining afterwards.
Step (6), using Hough (hough) conversion straight-line detection is carried out to new image s2 obtained by step 5,
And take most one of object pixel thereon from detected straight line as skyline.Fig. 7 is to utilize Hough
The straight line that change detection goes out, Fig. 8 is the current frame image being superimposed skyline.
Assume with a certain frame of monitor video for reference frame image i1, (occurrence can be according to actual needs for time t
Select, such as 5 minutes, 10 minutes etc.) after two field picture be current frame image i2, calculate two first
Absolute difference image i3=| i2-i1 | of two field picture, then partly carries out threshold to absolute difference image i3 below skyline
Value segmentation, the pixel that every pixel value is more than certain threshold value is all set to 1, and other pixels are set to 0；In skyline
Above section all pixels are set to 0；Thus obtaining image b.
Step 3, partly carry out pixel-by-pixel using below skyline in sliding window absolute difference image to described two-value
In scanning, such as current window, object pixel proportion exceedes preset ratio threshold value, then will be right in current frame image
Described grader should be inputted in the characteristics of image of the video in window of current window position to be classified:
For example, scanned pixel-by-pixel with sliding window w in image b, if pixel value in current window
Number of pixels for 1 more than proportion threshold value p, then to corresponding in the current frame image i2 of same position
Window subimage extract hsl histogram feature and lbp feature, classified using the grader training,
Tentatively judge whether it burns.Wherein the big I of sliding window w is according to the image of concrete monitor video
Frame sign chooses 16*16,32*32,48*48 equidimension.The scope of proportion threshold value p is 10%～70%, the present invention
Preferably 25%.Fig. 9 shows the crop straw burning event utilizing the preliminary identification of grader in current frame image, in figure
Each box indicating be tentatively identified as pyrotechnics burning image subwindow.
Step 4, it is the window subimage that there is crop straw burning for classification results, determine whether its color ladder
Whether degree range value is less than predetermined gradient threshold value, in this way, then judges to exist a crop straw burning in current frame image
Event:
Although svm grader has good recognition performance, sometimes also can by travel red car,
White moving object is mistakenly identified as pyrotechnics target, and the edge based on the subimage block that there is pyrotechnics for the present invention is weaker,
The features such as blurred background, propose the tri- passage subimages of rgb in coloured image and carry out gradient edge information and carry
Take, exclude the significant sub-image area of edge feature from preliminary classification result, reduce false alarm rate further.
Wherein color gradient range value obtains by the following method: three face to this image in rgb color space
Colouring component image, respectively using the sobel gradient magnitude sum of all pixels in each color component images as this face
The sobel gradient magnitude of colouring component image, then selects from the sobel gradient magnitude of three color component images
Maximum is as the color gradient range value of this image.Circular is as follows:
Respectively sobel gradient amplitude is calculated to three color component rgb channel image of this subimage window, point
It is not designated as gr,gg,gb, then this subimage window corresponding color gradient range value is maximum, that is, among three
max(gr,gg,gb).The sobel gradient amplitude of each color component images is each pixel sobel gradient magnitude
With, taking r channel image f as a example,
Wherein m is subimage window size, | |. | | represent vector length, as gradient magnitude.
Wherein Grads threshold is preferably 10000.
A crop straw burning event is often detected, then by crop straw burning event counter count (initial value is 0)
Plus 1, after the completion of entire image scanning, the straw that the value of crop straw burning event counter count as monitors
Stalk burns event.
Step 5, judge whether the value of crop straw burning event counter count reaches default alarm threshold value, such as
It is then to be reported to the police using means such as sound, light, the ejection of message window, short messages, point out this photographic head of this moment
Someone's crop straw burning in monitoring range.Because the pyrotechnics that crop straw burning produces shows as regional event in duration in the picture,
It is not isolated generation, therefore the present invention arranges an alarm threshold value (being 3 in the present embodiment), when some sons
When image window is identified as crop straw burning, just reported to the police simultaneously, so can be reduced rate of false alarm.
Step 6, current frame image i2 is set to reference frame image, after elapsed time t, starts next round monitoring.
The present invention is through in Nanjing somewhere practical probation, verification and measurement ratio reaches 99%, wrong report (false-alarm number) a day
Have 10 times about, be fully able to meet the actual demand of crop straw burning monitoring and warning.
Claims (17)
1. the crop straw burning monitoring method based on monitor video is it is characterised in that comprise the following steps:
There is the image positive sample of crop straw burning and there is not the image negative sample of crop straw burning in step a, acquisition, and there is not the sample size of the image negative sample of crop straw burning using Downsapling method reduction, obtains training sample set；Then extract the characteristics of image that training sample concentrates each training sample, two classification device model is trained, obtains grader；
Step b, obtain monitor video current frame image with respect to reference frame image absolute difference image, and absolute difference image is carried out with binary conversion treatment: by below skyline in absolute difference image partly middle pixel value be more than the pixel of predetermined threshold value and be set to target, rest of pixels is disposed as background, obtains the absolute difference image of two-value；
Step c, partly scanned pixel-by-pixel using below skyline in sliding window absolute difference image to described two-value, as in current window, object pixel proportion exceedes preset ratio threshold value, then the characteristics of image corresponding to the video in window of current window position in current frame image is inputted described grader to be classified, the span of described proportion threshold value is 10% ~ 70%；If classification results are that the color gradient range value that there is the video in window corresponding to current window position in crop straw burning and current frame image is less than predetermined gradient threshold value, then judge to exist a crop straw burning event in current frame image.
2. the crop straw burning monitoring method based on monitor video is it is characterised in that described Downsapling method is serial clustering method, specific as follows: using first image negative sample as a sample cluster as claimed in claim 1；From the beginning of second image negative sample, judge whether minimum Eustachian distance in default feature space and each sample cluster center having existed between for this image negative sample is more than predeterminable range threshold value successively, in this way, then using this image negative sample as a new sample cluster, as no, then this image negative sample is put under closest existing in sample cluster；After the completion of all image negative samples are processed, using the center of currently each sample cluster as new image negative sample.
3. as claimed in claim 2 the crop straw burning monitoring method based on monitor video it is characterised in that the feature space that constituted by the normalization histogram feature of hsl color space of described default feature space.
4. the crop straw burning monitoring method based on monitor video as claimed in claim 1 is it is characterised in that described image is characterized as the histogram feature of hsl color space and combining of local binary patterns feature.
5. the crop straw burning monitoring method based on monitor video as claimed in claim 1 is it is characterised in that described skyline is obtained using following methods detection:
Step 1, using symmetrical exponential function wave filter in Shen Jun edge detection operator, vertical direction Fast Recursive filtering is carried out to current frame image, obtain filtering image；
Step 2, the error image to filtering image and current frame image carry out binary conversion treatment: be all entered as 1 by all of on the occasion of pixel, and other pixel is all 0；
Step 3, expansion process is carried out to obtained bianry image；
Step 4, monochrome pixels that the image after expansion process is carried out are anti-phase, then carry out label processing, and note topmost background area is sky areas；
Step 5, each pixel belonging to sky areas in the image after expansion process is respectively processed: there is, as in the neighborhood of 3 × 3 pixel sizes centered on current pixel, the pixel that more than one value is 1, then current pixel is labeled as object pixel, otherwise it is labeled as background, thus obtaining new image；
Step 6, the straight-line detection that new image obtained by step 5 carried out using Hough transformation, and take most one of object pixel thereon from detected straight line as skyline.
6. the crop straw burning monitoring method based on monitor video as claimed in claim 1, it is characterized in that, the color gradient range value of image obtains by the following method: three color component images to this image in rgb color space, respectively using the sobel gradient magnitude sum of all pixels in each color component images as the sobel gradient magnitude of this color component images, from the sobel gradient magnitude of three color component images, then select maximum as the color gradient range value of this image.
7. the crop straw burning monitoring method based on monitor video as claimed in claim 1 is it is characterised in that described two classification device model is support vector cassification model.
8. the crop straw burning monitoring method based on monitor video as claimed in claim 1 is it is characterised in that described monitor video is acquired by being arranged on the video monitoring equipment on mobile communication base station tower body top.
9. the crop straw burning monitoring device based on monitor video is it is characterised in that include:
Video acquisition unit, for the real-time monitor video obtaining region to be monitored；
Two-value absolute difference image acquisition unit, for obtaining the absolute difference image with respect to reference frame image for the current frame image of monitor video, and absolute difference image is carried out with binary conversion treatment: by below skyline in absolute difference image partly middle pixel value be more than the pixel of predetermined threshold value and be set to target, rest of pixels is disposed as background, obtains the absolute difference image of two-value；
Crop straw burning event identifying unit, including grader, color gradient range value computing module, scan module；Wherein, training in advance obtains described grader in accordance with the following methods: obtains and there is the image positive sample of crop straw burning and there is not the image negative sample of crop straw burning, and using Downsapling method reduction do not exist crop straw burning image negative sample sample size, obtain training sample set；Then extract the characteristics of image that training sample concentrates each training sample, two classification device model is trained；Described color gradient range value computing module is used for calculating the color gradient range value of the video in window corresponding to current window position in current frame image；Described scan module is used for partly being scanned pixel-by-pixel using below skyline in sliding window absolute difference image to described two-value, as in current window, object pixel proportion exceedes preset ratio threshold value, then the characteristics of image corresponding to the video in window of current window position in current frame image is inputted described grader to be classified, the span of described proportion threshold value is 10% ~ 70%；If classification results are that the color gradient range value that there is the video in window corresponding to current window position in crop straw burning and current frame image is less than predetermined gradient threshold value, then judge to exist a crop straw burning event in current frame image.
10. the crop straw burning monitoring device based on monitor video is it is characterised in that described Downsapling method is serial clustering method, specific as follows: using first image negative sample as a sample cluster as claimed in claim 9；From the beginning of second image negative sample, judge whether minimum Eustachian distance in default feature space and each sample cluster center having existed between for this image negative sample is more than predeterminable range threshold value successively, in this way, then using this image negative sample as a new sample cluster, as no, then this image negative sample is put under closest existing in sample cluster；After the completion of all image negative samples are processed, using the center of currently each sample cluster as new image negative sample.
11. as claimed in claim 10 the crop straw burning monitoring devices based on monitor video it is characterised in that the feature space that constituted by the normalization histogram feature of hsl color space of described default feature space.
The 12. crop straw burning monitoring devices based on monitor video as claimed in claim 9 are it is characterised in that described image is characterized as the histogram feature of hsl color space and combining of local binary patterns feature.
The 13. crop straw burning monitoring devices based on monitor video as claimed in claim 9 are it is characterised in that described skyline is obtained using following methods detection:
Step 1, using symmetrical exponential function wave filter in Shen Jun edge detection operator, vertical direction Fast Recursive filtering is carried out to current frame image, obtain filtering image；
Step 2, the error image to filtering image and current frame image carry out binary conversion treatment: be all entered as 1 by all of on the occasion of pixel, and other pixel is all 0；
Step 3, expansion process is carried out to obtained bianry image；
Step 4, monochrome pixels that the image after expansion process is carried out are anti-phase, then carry out label processing, and note topmost background area is sky areas；
Step 5, each pixel belonging to sky areas in the image after expansion process is respectively processed: there is, as in the neighborhood of 3 × 3 pixel sizes centered on current pixel, the pixel that more than one value is 1, then current pixel is labeled as object pixel, otherwise it is labeled as background, thus obtaining new image；
Step 6, the straight-line detection that new image obtained by step 5 carried out using Hough transformation, and take most one of object pixel thereon from detected straight line as skyline.
The 14. crop straw burning monitoring devices based on monitor video as claimed in claim 9, it is characterized in that, described color gradient range value computing module calculates the color gradient range value of image: three color component images to this image in rgb color space by the following method, respectively using the sobel gradient magnitude sum of all pixels in each color component images as the sobel gradient magnitude of this color component images, from the sobel gradient magnitude of three color component images, then select maximum as the color gradient range value of this image.
The 15. crop straw burning monitoring devices based on monitor video as claimed in claim 9 are it is characterised in that described two classification device model is support vector cassification model.
The 16. crop straw burning monitoring devices based on monitor video as claimed in claim 9 are it is characterised in that described video acquisition unit is to be arranged on the video monitoring equipment on mobile communication base station tower body top.
The 17. crop straw burning monitoring devices based on monitor video as claimed in claim 9 are it is characterised in that also include crop straw burning event counter and alarm unit；Crop straw burning event counter is used for counting the crop straw burning event number in the presence of the current frame image that crop straw burning event identifying unit is judged；Alarm unit is used for being compared the crop straw burning event number in the presence of current frame image and preset alarm threshold value, and such as larger than alarm threshold value is then reported to the police.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510400389.3A CN106339657B (en) | 2015-07-09 | 2015-07-09 | Crop straw burning monitoring method based on monitor video, device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510400389.3A CN106339657B (en) | 2015-07-09 | 2015-07-09 | Crop straw burning monitoring method based on monitor video, device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106339657A true CN106339657A (en) | 2017-01-18 |
| CN106339657B CN106339657B (en) | 2019-02-22 |
Family
ID=57826388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510400389.3A Active CN106339657B (en) | 2015-07-09 | 2015-07-09 | Crop straw burning monitoring method based on monitor video, device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106339657B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106997461A (en) * | 2017-03-28 | 2017-08-01 | 浙江大华技术股份有限公司 | A kind of firework detecting method and device |
| CN108182706A (en) * | 2017-12-08 | 2018-06-19 | 重庆广睿达科技有限公司 | The monitoring method and system of a kind of incinerated matter |
| CN108279287A (en) * | 2018-02-01 | 2018-07-13 | 李绍辉 | Smog Quick diffusing system based on the communication technology |
| CN108777777A (en) * | 2018-05-04 | 2018-11-09 | 江苏理工学院 | A kind of monitor video crop straw burning method for inspecting based on deep neural network |
| CN111131688A (en) * | 2018-10-31 | 2020-05-08 | Tcl集团股份有限公司 | Image processing method and device and mobile terminal |
| CN111462451A (en) * | 2019-11-01 | 2020-07-28 | 武汉纺织大学 | Straw burning detection alarm system based on video information |
| CN112381802A (en) * | 2020-11-17 | 2021-02-19 | 中国科学院长春光学精密机械与物理研究所 | Image change detection method and device, electronic equipment and storage medium |
| CN114202868A (en) * | 2021-12-06 | 2022-03-18 | 九江礼涞生物科技有限公司 | System for comprehensively treating typical agricultural non-point source pollution |
| CN116047546A (en) * | 2022-07-07 | 2023-05-02 | 北京玖天气象科技有限公司 | Mountain fire monitoring method based on multi-source satellite data |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050084150A1 (en) * | 2000-03-28 | 2005-04-21 | Omnivision Technologies, Inc. | Method and apparatus for color image data processing and compression |
| CN101441771A (en) * | 2008-12-19 | 2009-05-27 | 中国科学技术大学 | Video fire hazard smoke detecting method based on color saturation degree and movement mode |
| CN101626489A (en) * | 2008-07-10 | 2010-01-13 | 苏国政 | Method and system for intelligently identifying and automatically tracking objects under unattended condition |
| CN101751744A (en) * | 2008-12-10 | 2010-06-23 | 中国科学院自动化研究所 | Detection and early warning method of smoke |
| CN102116861A (en) * | 2011-02-01 | 2011-07-06 | 环境保护部卫星环境应用中心 | Method for extracting straw burning fire based on No. 1 environment satellite |
| CN203324798U (en) * | 2013-03-19 | 2013-12-04 | 南通纺织职业技术学院 | Straw-burning monitoring system |
-
2015
- 2015-07-09 CN CN201510400389.3A patent/CN106339657B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050084150A1 (en) * | 2000-03-28 | 2005-04-21 | Omnivision Technologies, Inc. | Method and apparatus for color image data processing and compression |
| CN101626489A (en) * | 2008-07-10 | 2010-01-13 | 苏国政 | Method and system for intelligently identifying and automatically tracking objects under unattended condition |
| CN101751744A (en) * | 2008-12-10 | 2010-06-23 | 中国科学院自动化研究所 | Detection and early warning method of smoke |
| CN101441771A (en) * | 2008-12-19 | 2009-05-27 | 中国科学技术大学 | Video fire hazard smoke detecting method based on color saturation degree and movement mode |
| CN102116861A (en) * | 2011-02-01 | 2011-07-06 | 环境保护部卫星环境应用中心 | Method for extracting straw burning fire based on No. 1 environment satellite |
| CN203324798U (en) * | 2013-03-19 | 2013-12-04 | 南通纺织职业技术学院 | Straw-burning monitoring system |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106997461B (en) * | 2017-03-28 | 2019-09-17 | 浙江大华技术股份有限公司 | A kind of firework detecting method and device |
| US11532156B2 (en) | 2017-03-28 | 2022-12-20 | Zhejiang Dahua Technology Co., Ltd. | Methods and systems for fire detection |
| CN106997461A (en) * | 2017-03-28 | 2017-08-01 | 浙江大华技术股份有限公司 | A kind of firework detecting method and device |
| CN108182706B (en) * | 2017-12-08 | 2021-09-28 | 重庆广睿达科技有限公司 | Method and system for monitoring incinerated substances |
| CN108182706A (en) * | 2017-12-08 | 2018-06-19 | 重庆广睿达科技有限公司 | The monitoring method and system of a kind of incinerated matter |
| CN108279287A (en) * | 2018-02-01 | 2018-07-13 | 李绍辉 | Smog Quick diffusing system based on the communication technology |
| CN108279287B (en) * | 2018-02-01 | 2020-12-18 | 嘉兴市丰成五金材料股份有限公司 | Smoke rapid emission system based on communication technology |
| CN108777777A (en) * | 2018-05-04 | 2018-11-09 | 江苏理工学院 | A kind of monitor video crop straw burning method for inspecting based on deep neural network |
| CN111131688A (en) * | 2018-10-31 | 2020-05-08 | Tcl集团股份有限公司 | Image processing method and device and mobile terminal |
| CN111131688B (en) * | 2018-10-31 | 2021-04-23 | Tcl科技集团股份有限公司 | Image processing method and device and mobile terminal |
| CN111462451B (en) * | 2019-11-01 | 2022-04-26 | 武汉纺织大学 | Straw burning detection alarm system based on video information |
| CN111462451A (en) * | 2019-11-01 | 2020-07-28 | 武汉纺织大学 | Straw burning detection alarm system based on video information |
| CN112381802A (en) * | 2020-11-17 | 2021-02-19 | 中国科学院长春光学精密机械与物理研究所 | Image change detection method and device, electronic equipment and storage medium |
| CN114202868A (en) * | 2021-12-06 | 2022-03-18 | 九江礼涞生物科技有限公司 | System for comprehensively treating typical agricultural non-point source pollution |
| CN116047546A (en) * | 2022-07-07 | 2023-05-02 | 北京玖天气象科技有限公司 | Mountain fire monitoring method based on multi-source satellite data |
| CN116047546B (en) * | 2022-07-07 | 2024-02-27 | 北京玖天气象科技有限公司 | Mountain fire monitoring method based on multi-source satellite data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106339657B (en) | 2019-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106339657A (en) | Straw incineration monitoring method and device based on monitoring video | |
| CN110956094B (en) | RGB-D multi-mode fusion personnel detection method based on asymmetric double-flow network | |
| CN109977921B (en) | Method for detecting hidden danger of power transmission line | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| CN103069434B (en) | For the method and system of multi-mode video case index | |
| CN111339883A (en) | Method for identifying and detecting abnormal behaviors in transformer substation based on artificial intelligence in complex scene | |
| AU2009295350A1 (en) | Detection of vehicles in an image | |
| CN113887412B (en) | Detection method, detection terminal, monitoring system and storage medium for pollution emission | |
| CN102867417A (en) | Taxi anti-forgery system and taxi anti-forgery method | |
| Liu et al. | A multi-classifier image based vacant parking detection system | |
| CN114399734A (en) | Forest fire early warning method based on visual information | |
| CN104463232A (en) | Density crowd counting method based on HOG characteristic and color histogram characteristic | |
| CN102663362A (en) | Moving target detection method t based on gray features | |
| CN109086682A (en) | A kind of intelligent video black smoke vehicle detection method based on multi-feature fusion | |
| CN109271904A (en) | A kind of black smoke vehicle detection method based on pixel adaptivenon-uniform sampling and Bayesian model | |
| Thomas et al. | Smart car parking system using convolutional neural network | |
| CN104463104A (en) | Fast detecting method and device for static vehicle target | |
| CN111461076A (en) | Smoke detection method and smoke detection system combining frame difference method and neural network | |
| CN111897993A (en) | Efficient target person track generation method based on pedestrian re-recognition | |
| CN109684982B (en) | Flame detection method based on video analysis and combined with miscible target elimination | |
| CN111339927A (en) | Intelligent work state identification system for personnel in electric power business hall | |
| CN109325426B (en) | Black smoke vehicle detection method based on three orthogonal planes time-space characteristics | |
| CN113920585A (en) | Behavior recognition method and device, equipment and storage medium | |
| CN118115938A (en) | Ship behavior detection method based on image recognition and AIS data | |
| CN116071711B (en) | Traffic jam condition detection method and device |
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 |