CN107832723B - Smoke identification method and system based on LBP Gaussian pyramid - Google Patents

Smoke identification method and system based on LBP Gaussian pyramid Download PDF

Info

Publication number
CN107832723B
CN107832723B CN201711144817.6A CN201711144817A CN107832723B CN 107832723 B CN107832723 B CN 107832723B CN 201711144817 A CN201711144817 A CN 201711144817A CN 107832723 B CN107832723 B CN 107832723B
Authority
CN
China
Prior art keywords
lbp
image
gray
smoke
pyramid
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.)
Expired - Fee Related
Application number
CN201711144817.6A
Other languages
Chinese (zh)
Other versions
CN107832723A (en
Inventor
王琳
雷丹
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 Maritime University
Original Assignee
Dalian Maritime University
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 Maritime University filed Critical Dalian Maritime University
Publication of CN107832723A publication Critical patent/CN107832723A/en
Application granted granted Critical
Publication of CN107832723B publication Critical patent/CN107832723B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]

Abstract

The invention discloses a smoke identification method and system based on an LBP Gaussian pyramid. Graying an image of a suspected smoke area, performing Gaussian smoothing and sampling twice to obtain gray level images with the sizes of 1/4 and 1/16, and forming n-layer pyramid images with the gray level images of an original image; respectively adopting an LBP operator with the value of P being 8 and R being 1 to calculate n layers of Gaussian pyramid gray-scale images to obtain binary LBP codes of the n layers of Gaussian pyramid gray-scale images, adopting a rotation invariant mode and an equivalent mode to carry out dimension reduction on the LBP codes of each layer to obtain 9 LBP coding modes, counting the number of each LBP code and taking the number as a characteristic value; and forming AdaBoost input vectors by using 9 multiplied by n characteristic values of the n layers of LBP Gaussian pyramids, and using the AdaBoost input vectors for judging the interference of smoke and pseudo smoke. The method provided by the invention has good robustness to illumination and high recognition rate.

Description

Smoke identification method and system based on LBP Gaussian pyramid
Technical Field
The invention relates to the technical field of image processing and image classification, in particular to a smoke identification method and system based on an LBP Gaussian pyramid.
Background
Serious fire accidents often result in serious casualties and significant property damage. If the early stage of fire can be found and alarmed in time, the loss can be reduced to the minimum, so that the early detection of the fire is very important. Under the general condition, the smoke is the fire start, the smoke is the accompanying product from the fire to the second stage, and if the smoke of the fire can be detected in time, the fire rescue measures can be taken as soon as possible, so that the personal casualties and the property loss are reduced.
Lbp (local binary patterns) is an effective texture description operator, is a binary description for expressing the size relationship between a certain pixel point of a gray image and surrounding pixel points, has the advantages of gray invariance and the like, and has good robustness to illumination. In recent years, LBP operators are continuously developed and evolved, and are widely applied to the fields of texture classification, texture segmentation and the like.
The smoke will appear differently when viewed at different scales, and therefore the description of the image at multiple scales needs to be considered simultaneously. Burt P.J and adelisin E.H proposed an image pyramid algorithm in 1983, which is a multi-resolution, multi-scale method. The image pyramid is a collection of different scales of the original image, obtained by continuously downsampling the original image, with the bottom of the image pyramid being a high resolution representation of the original image and the top being a low resolution approximation of the original image.
At present, a common smoke detector mainly monitors the concentration of smoke to realize fire prevention, and generally adopts an ion smoke sensor as a core component. But the application range is limited to the active area of the detector installation, and the installation and maintenance cost of the device is high.
Disclosure of Invention
In view of the defects in the prior art, the invention aims to provide the smoke identification method based on the LBP Gaussian pyramid, the identification algorithm has the advantages of small illumination influence, high response speed, high identification rate and the like, whether smoke exists in a video can be accurately identified in real time, and early fire alarm can be realized.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a smoke identification algorithm based on an LBP Gaussian pyramid comprises the following steps:
s100, acquiring a suspected smoke area image to be distinguished, wherein the step of extracting the suspected smoke area image comprises the following steps: s110, extracting a moving area target in a video frame by adopting a motion detection algorithm; s120, processing the moving area target by adopting a morphological method of median filtering, expansion and corrosion to obtain a moving area image; s130, extracting a suspected smoke area image by adopting a color segmentation method;
s200, gray processing the suspected smoke area image, and performing gray processing on the suspected smoke area gray image l1Performing Gaussian smoothing and sampling to obtain
Figure BDA0001472236530000021
Gray scale map of suspected smoke area of size2Then go right again
Figure BDA0001472236530000022
Gray scale map of suspected smoke area of size1Performing Gaussian smoothing and sampling to obtain
Figure BDA0001472236530000023
Gray scale map of suspected smoke area of size3(ii) a Repeatedly executing the above processes to obtain
Figure BDA0001472236530000024
Gray scale map of suspected smoke area of sizenN is more than or equal to 2, and n suspected smoke area gray level images l with different sizes are used1、l2…lnForming a pyramid gray scale image;
s300, respectively calculating to obtain three layers of images l in the pyramid gray-scale image by adopting an LBP operator with P being 8 and R being 11、l2And l3Carrying out corresponding binary LBP coding, and adopting a rotation invariant mode and an equivalent mode to carry out dimension reduction on the LBP coding corresponding to each layer of image; any layer of image corresponds to 9 LBP coding modes, the frequency of occurrence of each LBP coding mode is counted to obtain the LBP characteristic value of each layer of image, and any layer corresponds to 9 characteristic values;
s400, forming an input vector of the AdaBoost model by using a total of 27 characteristic values corresponding to three layers of images of the pyramid gray level image, and judging the authenticity of the smoke according to the input vector.
Another objective of the present invention is to provide a smoke identification system based on LBP gaussian pyramid, which includes:
the extraction unit is used for acquiring a suspected smoke area image to be distinguished and extracting a motion area target in a video frame by adopting a motion detection algorithm; processing the moving area target by adopting a morphological method of median filtering, expansion and corrosion to obtain a moving area image; extracting a suspected smoke area image by adopting a color segmentation method;
a processing unit for processing the suspected smoke area image in a gray scale manner and processing the suspected smoke area gray scale image l1Performing Gaussian smoothing and sampling to obtain
Figure BDA0001472236530000025
Gray scale map of suspected smoke area of size2Then go right again
Figure BDA0001472236530000026
Gray scale map of suspected smoke area of size2Performing Gaussian smoothing and sampling to obtain
Figure BDA0001472236530000027
Gray scale map of suspected smoke area of size3(ii) a Gray level map l of 3 suspected smoke areas with different sizes1、l2And l3Forming a pyramid gray scale image;
a calculating unit, which calculates three layers of images l in the pyramid grayscale image by using an LBP operator with P being 8 and R being 1 respectively1、l2And l3Carrying out corresponding binary LBP coding, and adopting a rotation invariant mode and an equivalent mode to carry out dimension reduction on the LBP coding corresponding to each layer of image; any layer of image corresponds to 9 LBP coding modes, the frequency of occurrence of each LBP coding mode is counted to obtain the LBP characteristic value of each layer of image, and any layer corresponds to 9 characteristic values;
and the judging unit is used for forming an input vector of the AdaBoost model by using 9 multiplied by n characteristic values corresponding to the three layers of images of the Gaussian pyramid, and judging the authenticity of the smoke according to the input vector.
Compared with the prior art, the invention has the beneficial effects that:
(1) after the LBP codes are processed by the rotation invariant mode and the equivalent mode, the number of the binary LBP codes is reduced from the original 256 to 9, and on the basis of reserving most information, the dimension of the characteristic vector is reduced and the influence of high-frequency noise is reduced;
(2) the method combines a multi-scale space theory and constructs the LBP Gaussian pyramid by changing the resolution of the image. Firstly, decomposing an original image into multi-scale pyramid images, then solving an LBP feature binary pattern for each image in the pyramid images by using the same LBP operator, and finally combining the LBP binary patterns of all scales as identification features for classification and identification, and the method has the advantages of rich and accurate feature extraction, small calculated amount and the like;
(3) the method adopts AdaBoost to train the extracted LBP characteristic value, and uses the obtained learning model to classify and detect the test sample video. The method has good real-time performance and accuracy, can quickly detect whether fire smoke exists in the video, and can be used in the fields of security monitoring and the like.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of a smoke recognition algorithm of the present invention;
FIG. 2 is a schematic view of a pyramid image of the present invention;
FIG. 3 is a three-level pyramid grayscale diagram of the present invention;
FIG. 4 is a basic LBP operator of the present invention;
FIG. 5 is a schematic view of a rotation invariant LBP in accordance with the present invention;
FIG. 6 shows an embodiment of the present invention1The distribution diagram of 9 kinds of characteristic values is arranged on the layer;
FIG. 7 shows an embodiment of the present invention2The distribution diagram of 9 kinds of characteristic values is arranged on the layer;
FIG. 8 shows an embodiment of the present invention3Distribution of 9 kinds of characteristic values of layerA drawing;
FIG. 9 is a graph of smoke video detection results of the present invention;
FIG. 10 is a diagram illustrating the results of the interference video detection according to the present invention;
FIG. 11 is a schematic diagram of the smoke recognition system of the present invention;
fig. 12 is a diagram illustrating an example of the operation flow of the smoke recognition system of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention discloses a smoke recognition algorithm based on an LBP Gaussian pyramid, which is further explained by combining the following drawings and specific embodiments:
as shown in fig. 1, a smoke identification method based on LBP gaussian pyramid includes the following steps:
s100, acquiring a suspected smoke area image to be distinguished, wherein the step of extracting the suspected smoke area image comprises the following steps:
s110, extracting a moving area target in a video frame by adopting a motion detection algorithm;
s120, processing the moving area target by adopting a morphological method of median filtering, expansion and corrosion to obtain a moving area image;
s130, extracting a suspected smoke area image by adopting a color segmentation method;
s200, gray processing the suspected smoke area image, and performing gray processing on the suspected smoke area gray image l1Performing Gaussian smoothing and sampling to obtain
Figure BDA0001472236530000041
Gray scale map of suspected smoke area of sizel2Then go right again
Figure BDA0001472236530000042
Gray scale map of suspected smoke area of size1Performing Gaussian smoothing and sampling to obtain
Figure BDA0001472236530000043
Gray scale map of suspected smoke area of size3(ii) a Repeatedly executing the above processes to obtain
Figure BDA0001472236530000044
Gray scale map of suspected smoke area of sizenN is more than or equal to 2, and n suspected smoke area gray level images l with different sizes are used1、l2…lnForming a pyramid gray scale image; in order to make the judgment accuracy and the program operation efficiency compatible, in this embodiment, n is preferably 3, that is, 3 gray-scale maps l with different sizes of suspected smoke areas are preferred1、l2And l3Forming a pyramid gray scale image;
to obtain
Figure BDA0001472236530000051
Gray scale map of suspected smoke area of size2The specific process is as follows:
s211, using a 5 multiplied by 5 Gaussian low-pass filter to carry out gray-scale image l on suspected smoke area1Carrying out smoothing treatment;
s212, sampling the image after Gaussian smoothing, and taking pixel points of even rows and even columns to form
Figure BDA0001472236530000052
Gray scale map of suspected smoke area of size2；
In the same way, extract
Figure BDA0001472236530000053
Gray scale map of suspected smoke area of size3Comprises the following steps:
s221. use a 5 × 5 Gaussian low pass filter pair
Figure BDA0001472236530000054
Gray scale map l of suspected smoke area of size2Carrying out smoothing treatment;
s222, sampling the image after Gaussian smoothing, and taking the pixel point composition size of even lines and even lines as an original image
Figure BDA0001472236530000055
Gray scale map l of suspected smoke area3。
Gray level map l of 3 suspected smoke areas with different sizes1、l2And l3And forming a pyramid gray scale image. Fig. 2 shows a pyramid grayscale.
The gaussian low-pass filter is selected to be a 5 x 5 window function with low-pass characteristics, wherein,
Figure BDA0001472236530000056
the pair l1、l2The sampling process is as follows:
Gk(x, y) is a k-th layer gaussian pyramid image, where k is 1,2,3, G1(x, y) is the original image as the lowest layer of the Gaussian pyramid, where
Figure BDA0001472236530000057
FIG. 3 is a schematic representation of the layers of a pyramid image of an embodiment.
S300, respectively calculating to obtain three layers of images l in the pyramid gray-scale image by adopting an LBP operator with P being 8 and R being 10、l1And l2Carrying out corresponding binary LBP coding, and adopting a rotation invariant mode and an equivalent mode to carry out dimension reduction on the LBP coding corresponding to each layer of image; any layer of image corresponds to 9 LBP coding modes, the number of times of occurrence of each LBP coding mode is counted to obtain the LBP characteristic value of each layer of image, and any layer corresponds to 9 characteristic values. FIG. 5 is a schematic diagram showing the LBP operator rotation invariant pattern.
Wherein, the step of calculating the LBP code comprises:
s311, calculating LBP characteristics by adopting an LBP operator with the scale of 3 multiplied by 3, taking the gray value of the current central pixel point as a threshold value, if the gray value of the neighborhood pixel point is larger than or equal to the threshold value, marking the gray value as 1, and if the gray value is smaller than the threshold value, marking the gray value as 0;
s312, the eight bits are connected in series according to the clockwise direction to obtain an LBP binary code;
s313, adopting an LBP operator with rotation invariance, namely continuously rotating the neighborhood clockwise for 7 times to obtain 8 basic LBP codes, and taking the minimum value of the basic LBP codes as the LBP codes of the neighborhood;
s314, dimension reduction is carried out by adopting an LBP equivalent mode, most LBP codes in the image only contain two jumps from 1 to 0 or from 0 to 1 at most, namely an equivalent mode class, and if the jumps are more than two times, the LBP codes are classified as a mixed mode class;
s315, after dimension reduction is carried out on the LBP coding through the rotation invariant mode and the equivalent mode, the type of the LBP coding mode is 28The number of LBP coding patterns is counted as a characteristic value.
The step of extracting the LBP characteristic value of each layer of pyramid gray-scale map comprises the following steps:
s321, calculating LBP codes of pixel points in each layer of image of the pyramid gray-scale image;
s322, counting each image of each layer after dimension reduction processing in a rotation invariant mode and an equivalent mode
Figure BDA0001472236530000061
Figure BDA0001472236530000062
Wherein
Figure BDA0001472236530000063
Fig. 6-8 show the calculation process of LBP characteristic values of each layer of image in pyramid gray scale.
The specific process of obtaining the LBP characteristic value of the three-layer Gaussian pyramid gray level image is as follows: respectively extracting the first, second and third layers of Gauss pyramidsTower LBP eigenvalue n11,n12,…,n19}、{n21,n22,…,n29}、{n31,n32,…,n39Total 27 LBP feature values.
S400, forming an input vector of the AdaBoost model by using a total of 27 characteristic values corresponding to the three layers of images of the pyramid, and judging the authenticity of the smoke according to the input vector.
The specific process for obtaining the AdaBoost learning model is as follows:
s411, respectively selecting 600 frames of images from the smoke and pseudo-smoke interference video as samples, and extracting 27 LBP characteristic values of each pyramid image from the samples;
s412 takes 1200 groups of pyramid image LBP characteristic values as input vectors, and an AdaBoost learning model is obtained through training.
The specific process for obtaining the judgment result of the smoke and the pseudo-smoke interference video is as follows:
s421, respectively taking the video containing smoke and the video interfered by pseudo-smoke as test samples, and extracting 27 characteristic values of an LBP Gaussian pyramid from the test samples;
s422 takes the LBP Gaussian pyramid characteristic value extracted from each frame of image in the video as an input vector, and detects the classification accuracy of the AdaBoost learning model. Fig. 9 is a graph showing a smoke video detection result, and fig. 10 is a graph showing an interference video detection result.
As shown in fig. 11, the present invention further provides a smoke recognition system based on LBP gaussian pyramid, and fig. 12 is an example diagram of the working process of the system. The system comprises:
the extracting unit 501 is configured to obtain a suspected smoke area image to be distinguished, and extract a moving area target in a video frame by using a motion detection algorithm; processing the moving area target by adopting a morphological method of median filtering, expansion and corrosion to obtain a moving area image; and extracting a suspected smoke area image by adopting a color segmentation method.
A processing unit 502 for processing the suspected smoke area image in a gray scale manner and processing the suspected smoke area gray scale image l1Performing Gaussian smoothing and sampling to obtain
Figure BDA0001472236530000071
Gray scale map of suspected smoke area of size2Then go right again
Figure BDA0001472236530000072
Gray scale map of suspected smoke area of size2Performing Gaussian smoothing and sampling to obtain
Figure BDA0001472236530000073
Gray scale map of suspected smoke area of size3(ii) a Gray level map l of 3 suspected smoke areas with different sizes1、l2And l3And forming a pyramid gray scale image.
To obtain
Figure BDA0001472236530000074
Gray scale map of suspected smoke area of size2The specific process is as follows:
gray scale map l of suspected smoke area using a 5 x 5 Gaussian low pass filter1Carrying out smoothing treatment;
sampling the image after Gaussian smoothing, and taking pixel points of even rows and even columns to form
Figure BDA0001472236530000075
Gray scale map of suspected smoke area of size2；
In the same way, extract
Figure BDA0001472236530000076
Gray scale map of suspected smoke area of size3Comprises the following steps:
using a 5 x 5 gaussian low-pass filter pair
Figure BDA0001472236530000077
Gray scale map l of suspected smoke area of size2Carrying out smoothing treatment;
sampling the image after Gaussian smoothing, and taking pixel points of even rows and even columns to form a size ofOriginal drawing
Figure BDA0001472236530000078
Gray scale map l of suspected smoke area3。
Gray level map l of 3 suspected smoke areas with different sizes1、l2And l3And forming a pyramid gray scale image.
The gaussian low-pass filter is selected to be a 5 x 5 window function with low-pass characteristics, wherein,
Figure BDA0001472236530000081
the pair l1、l2The sampling process is as follows:
Gk(x, y) is a k-th layer gaussian pyramid image, where k is 1,2,3, G1(x, y) is the original image as the lowest layer of the Gaussian pyramid, where
Figure BDA0001472236530000082
The calculating unit 503 calculates three layers of images l in the pyramid grayscale map by using LBP operators with P being 8 and R being 1 respectively1、l2And l3Carrying out corresponding binary LBP coding, and adopting a rotation invariant mode and an equivalent mode to carry out dimension reduction on the LBP coding corresponding to each layer of image; any layer of image corresponds to 9 LBP coding modes, the number of times of occurrence of each LBP coding mode is counted to obtain the LBP characteristic value of each layer of image, and any layer corresponds to 9 characteristic values.
Wherein, the step of calculating the LBP code comprises:
calculating LBP characteristics by adopting an LBP operator with the scale of 3 multiplied by 3, taking the gray value of the current central pixel point as a threshold value, marking the gray value as 1 if the gray value of the neighborhood pixel point is larger than or equal to the threshold value, and marking the gray value as 0 if the gray value is smaller than the threshold value;
the eight bits are connected in series according to the clockwise direction to obtain LBP binary code;
adopting an LBP operator with rotational invariance, namely continuously rotating the neighborhood clockwise for 7 times to obtain 8 basic LBP codes, and taking the minimum value of the basic LBP codes as the LBP codes of the neighborhood;
adopting an LBP equivalent mode to reduce dimension, wherein most LBP codes in the image only comprise two jumps from 1 to 0 or from 0 to 1 at most, namely the LBP codes are equivalent mode classes, and if the LBP codes comprise jumps more than two times, the LBP codes are classified as mixed mode classes;
after the LBP coding is subjected to dimension reduction through the rotation invariant mode and the equivalent mode, the type of the LBP coding mode is 28The number of LBP coding patterns is counted as a characteristic value.
The step of extracting the LBP characteristic value of each layer of pyramid gray-scale map comprises the following steps:
calculating LBP codes of pixel points in each layer of image of the pyramid gray-scale image;
counting each LBP code of each layer image after dimension reduction processing of a rotation invariant mode and an equivalent mode
Figure BDA0001472236530000083
Figure BDA0001472236530000091
Wherein
Figure BDA0001472236530000092
The specific process of obtaining the LBP characteristic value of the three-layer Gaussian pyramid gray level image is as follows: respectively extracting first, second and third layers of Gaussian pyramid LBP characteristic values { n }11,n12,…,n19}、{n21,n22,…,n29}、{n31,n32,…,n39Total 27 LBP feature values.
The determining unit 504 forms an input vector of the AdaBoost model by using a total of 27 feature values corresponding to the pyramid images, and determines the authenticity of the smoke according to the input vector.
The specific process for obtaining the AdaBoost learning model is as follows:
selecting 600 frames of images from the smoke and pseudo-smoke interference video as samples, and extracting 27 LBP characteristic values of each pyramid image;
and training 1200 groups of pyramid image LBP characteristic values as input vectors to obtain an AdaBoost learning model.
The specific process for obtaining the judgment result of the smoke and the pseudo-smoke interference video is as follows:
respectively taking a video containing smoke and a video interfered by pseudo-smoke as test samples, and extracting 27 characteristic values of an LBP Gaussian pyramid from the test samples;
and detecting the classification accuracy of the AdaBoost learning model by taking the LBP Gaussian pyramid characteristic value extracted from each frame of image in the video as an input vector. Fig. 9 is a graph showing a smoke video detection result, and fig. 10 is a graph showing an interference video detection result.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A smoke identification method based on an LBP Gaussian pyramid is characterized by comprising the following steps:
s100, acquiring a suspected smoke area image to be distinguished, wherein the step of extracting the suspected smoke area image comprises the following steps:
s110, extracting a moving area target in a video frame by adopting a motion detection algorithm;
s120, processing the moving area target by adopting a morphological method of median filtering, expansion and corrosion to obtain a moving area image;
s130, extracting a suspected smoke area image by adopting a color segmentation method;
s200, gray processing the suspected smoke area image, and performing gray processing on the suspected smoke area gray image l1Performing Gaussian smoothing and sampling to obtain
Figure FDA0002335633150000011
Gray scale map of suspected smoke area of size2Then go right again
Figure FDA0002335633150000012
Gray scale map of suspected smoke area of size1Performing Gaussian smoothing and sampling to obtain
Figure FDA0002335633150000013
Gray scale map of suspected smoke area of size3Repeatedly executing the above-mentioned Gaussian smoothing and sampling process to obtain
Figure FDA0002335633150000014
Gray scale map of suspected smoke area of sizenN is more than or equal to 2, and n suspected smoke area gray level images l with different sizes are used1、l2…lnForming a pyramid gray scale image;
s300, respectively calculating n layers of images l in the pyramid gray-scale image by adopting an LBP operator with P being 8 and R being 11、l2…lnCarrying out corresponding binary LBP coding, and adopting a rotation invariant mode and an equivalent mode to carry out dimension reduction on the LBP coding corresponding to each layer of image; any layer of image corresponds to 9 LBP coding modes, the frequency of occurrence of each LBP coding mode is counted to obtain the LBP characteristic value of each layer of image, and any layer corresponds to 9 characteristic values;
s400, forming an input vector of the AdaBoost model by using 9 multiplied by n characteristic values corresponding to n layers of images of the pyramid gray level image, and judging the authenticity of the smoke according to the input vector.
2. The method for smoke identification based on LBP Gaussian pyramid as claimed in claim 1, wherein in step S200, the extraction is performed
Figure FDA0002335633150000015
Gray scale map of suspected smoke area of sizenComprises the following steps:
s211, using a 5 multiplied by 5 Gaussian low-pass filter to carry out gray-scale image l on suspected smoke arean-1Carrying out smoothing treatment;
s212, sampling the image after Gaussian smoothing, and taking pixel points of even rows and even columns to form
Figure FDA0002335633150000016
Gray scale map of suspected smoke area of sizen。
3. The method for smoke identification based on LBP gaussian pyramid as claimed in claim 1, wherein in step S300, the process of calculating LBP coding comprises:
s311, calculating LBP characteristics by adopting an LBP operator with the scale of 3 multiplied by 3, taking the gray value of the current central pixel point as a threshold value, if the gray value of the neighborhood pixel point is larger than or equal to the threshold value, marking the gray value as 1, and if the gray value is smaller than the threshold value, marking the gray value as 0;
s312, the eight bits are connected in series according to the clockwise direction to obtain an LBP binary code;
s313, adopting an LBP operator with rotation invariance, namely continuously rotating the neighborhood clockwise for 7 times to obtain 8 basic LBP codes, and taking the minimum value of the basic LBP codes as the LBP codes of the neighborhood;
s314, adopting an LBP equivalent mode to reduce the dimension, wherein if the LBP code only contains two jumps from 1 to 0 or from 0 to 1 at most, the LBP code is an equivalent mode class, and if the LBP code contains jumps more than two times, the LBP code is a mixed mode class;
s315, after dimension reduction is carried out on the LBP coding through the rotation invariant mode and the equivalent mode, the type of the LBP coding mode is 28The number of the LBP coding patterns is reduced to 9, and the occurrence frequency of each LBP coding pattern is counted.
4. The LBP gaussian pyramid-based smoke identification method according to claim 3, wherein in the step S300, the extracting LBP feature values of each layer of pyramid gray-scale map comprises:
s321, calculating LBP codes of pixel points in each layer of image of the pyramid gray-scale image;
s322, counting the number of times of each LBP code of each layer of image after dimension reduction processing of the rotation invariant mode and the equivalent mode, and recording the number of times as
Figure FDA0002335633150000021
Counting the total number of 9 LBP codes
Figure FDA0002335633150000022
Wherein
Figure FDA0002335633150000023
S323, calculating the ratio of each LBP code of each layer image as the LBP characteristic value, and recording as the LBP characteristic value
Figure FDA0002335633150000024
Wherein
Figure FDA0002335633150000025
5. The LBP gaussian pyramid-based smoke identification method according to claim 1, wherein in the step S400, the training method of the AdaBoost model is as follows:
s411, respectively selecting 600 frames of images from the smoke and pseudo-smoke interference video as samples, and extracting 9 x n LBP characteristic values of each pyramid image from the samples;
s412 takes 1200 groups of pyramid image LBP characteristic values as input vectors, and an AdaBoost learning model is obtained through training.
6. A smoke identification system based on an LBP Gaussian pyramid, comprising:
the extraction unit is used for acquiring a suspected smoke area image to be distinguished and extracting a motion area target in a video frame by adopting a motion detection algorithm; processing the moving area target by adopting a morphological method of median filtering, expansion and corrosion to obtain a moving area image; extracting a suspected smoke area image by adopting a color segmentation method;
a processing unit for processing the suspected smoke area image in a gray scale manner and processing the suspected smoke area gray scale image l1Performing Gaussian smoothing and sampling to obtain
Figure FDA0002335633150000031
Gray scale map of suspected smoke area of size2Then go right again
Figure FDA0002335633150000032
Gray scale map of suspected smoke area of size1Performing Gaussian smoothing and sampling to obtain
Figure FDA0002335633150000033
Gray scale map of suspected smoke area of size3(ii) a Repeating the above Gaussian smoothing and sampling process to obtain
Figure FDA0002335633150000034
Gray scale map of suspected smoke area of sizenN is more than or equal to 2, and n suspected smoke area gray level images l with different sizes are used1、l2…lnForming a pyramid gray scale image;
a calculating unit, which calculates n-layer images l in the pyramid gray-scale image by using LBP operator with P being 8 and R being 1 respectively1、l2…lnCarrying out corresponding binary LBP coding, and adopting a rotation invariant mode and an equivalent mode to carry out dimension reduction on the LBP coding corresponding to each layer of image; any layer of image corresponds to 9 LBP coding modes, the frequency of occurrence of each LBP coding mode is counted to obtain the LBP characteristic value of each layer of image, and any layer corresponds to 9 characteristic values;
and the judging unit is used for forming an input vector of the AdaBoost model by using 9 multiplied by n characteristic values corresponding to the n-layer image of the Gaussian pyramid, and judging the authenticity of the smoke according to the input vector.
7. The LBP gaussian pyramid-based smoke signature of claim 6A classification system characterized by extraction
Figure FDA0002335633150000035
Gray scale map of suspected smoke area of sizenComprises the following steps:
s211, using a 5 multiplied by 5 Gaussian low-pass filter to carry out gray-scale image l on suspected smoke arean-1Carrying out smoothing treatment;
s212, sampling the image after Gaussian smoothing, and taking pixel points of even rows and even columns to form
Figure FDA0002335633150000036
Gray scale map of suspected smoke area of sizen。
8. The smoke identification system based on the LBP gaussian pyramid as claimed in claim 6, wherein said calculating unit calculates LBP features by using LBP operator with dimension of 3 x 3, taking the gray value of the current central pixel as a threshold, if the gray value of the neighborhood pixel is greater than or equal to the threshold, then marking as 1, if less than the threshold, then marking as 0; the eight bits are connected in series according to the clockwise direction to obtain LBP binary code; adopting an LBP operator with rotational invariance, namely continuously rotating the neighborhood clockwise for 7 times to obtain 8 basic LBP codes, and taking the minimum value of the basic LBP codes as the LBP codes of the neighborhood; adopting an LBP equivalent mode to reduce dimension, wherein most LBP codes in the image only comprise two jumps from 1 to 0 or from 0 to 1 at most, namely an equivalent mode class, and if the jumps are more than two times, the LBP codes are a mixed mode class; after the LBP coding is subjected to dimension reduction through the rotation invariant mode and the equivalent mode, the type of the LBP coding mode is 28The number of the LBP coding patterns is reduced to 9, and the occurrence frequency of each LBP coding pattern is counted.
9. The system for smoke identification based on the LBP gaussian pyramid as claimed in claim 6, wherein said computing unit extracts LBP feature values of each layer of pyramid gray-scale map, first computing LBP codes of pixel points in each layer of image of pyramid gray-scale map; second statisticsThe number of each LBP code of each layer image after the dimension reduction of the rotation invariant pattern and the equivalent pattern is recorded as
Figure FDA0002335633150000041
Counting the total number of 9 LBP codes
Figure FDA0002335633150000042
Wherein
Figure FDA0002335633150000043
Calculating the ratio of each LBP code of each layer image as LBP characteristic value, and recording as
Figure FDA0002335633150000044
Wherein
Figure FDA0002335633150000045
10. The smoke identification system based on the LBP gaussian pyramid as claimed in claim 6, wherein said determining unit selects 600 frames of images from the smoke and pseudo-smoke interference video as samples, and extracts 9 × n LBP feature values of the pyramid image from each sample; and training 1200 groups of pyramid image LBP characteristic values as input vectors to obtain an AdaBoost learning model.
CN201711144817.6A 2017-03-22 2017-11-17 Smoke identification method and system based on LBP Gaussian pyramid Expired - Fee Related CN107832723B (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2017101755036 2017-03-22
CN201710175503 2017-03-22

Publications (2)

Publication Number Publication Date
CN107832723A CN107832723A (en) 2018-03-23
CN107832723B true CN107832723B (en) 2020-04-17

Family

ID=61652858

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711144817.6A Expired - Fee Related CN107832723B (en) 2017-03-22 2017-11-17 Smoke identification method and system based on LBP Gaussian pyramid

Country Status (1)

Country Link
CN (1) CN107832723B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110991458B (en) * 2019-11-25 2023-05-23 创新奇智(北京)科技有限公司 Image feature-based artificial intelligent recognition result sampling system and sampling method
CN113658238B (en) * 2021-08-23 2023-08-08 重庆大学 Near infrared vein image high-precision matching method based on improved feature detection

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101727573A (en) * 2008-10-13 2010-06-09 汉王科技股份有限公司 Method and device for estimating crowd density in video image
CN103020628A (en) * 2012-11-30 2013-04-03 北京理工大学 Smoke detection method based on red, green and blue (RGB) contrast image and target shape
CN103218831A (en) * 2013-04-21 2013-07-24 北京航空航天大学 Video moving target classification and identification method based on outline constraint
CN103870818A (en) * 2014-03-31 2014-06-18 中安消技术有限公司 Smog detection method and device
CN104268562A (en) * 2014-09-15 2015-01-07 武汉大学 Effective multiscale texture recognition method
CN105469105A (en) * 2015-11-13 2016-04-06 燕山大学 Cigarette smoke detection method based on video monitoring

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8712505B2 (en) * 2010-11-11 2014-04-29 University Of Pittsburgh-Of The Commonwealth System Of Higher Education Automated macular pathology diagnosis in three-dimensional (3D) spectral domain optical coherence tomography (SD-OCT) images
CN106778582B (en) * 2016-12-07 2020-06-09 哈尔滨工业大学 RGB reconstruction based flame/smoke recognition method after forest image cutting

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101727573A (en) * 2008-10-13 2010-06-09 汉王科技股份有限公司 Method and device for estimating crowd density in video image
CN103020628A (en) * 2012-11-30 2013-04-03 北京理工大学 Smoke detection method based on red, green and blue (RGB) contrast image and target shape
CN103218831A (en) * 2013-04-21 2013-07-24 北京航空航天大学 Video moving target classification and identification method based on outline constraint
CN103870818A (en) * 2014-03-31 2014-06-18 中安消技术有限公司 Smog detection method and device
CN104268562A (en) * 2014-09-15 2015-01-07 武汉大学 Effective multiscale texture recognition method
CN105469105A (en) * 2015-11-13 2016-04-06 燕山大学 Cigarette smoke detection method based on video monitoring

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Video-based smoke detection with histogram sequence of LBP and LBPV pyramids;Feiniu Yuan;《Fire Safety Journal》;20111231;全文 *

Also Published As

Publication number Publication date
CN107832723A (en) 2018-03-23

Similar Documents

Publication Publication Date Title
CN107423690B (en) Face recognition method and device
CN107527009B (en) Remnant detection method based on YOLO target detection
CN106874894B (en) Human body target detection method based on regional full convolution neural network
Zhang et al. Learning-based license plate detection using global and local features
Jahedsaravani et al. An image segmentation algorithm for measurement of flotation froth bubble size distributions
Rezazadegan Tavakoli et al. Fast and efficient saliency detection using sparse sampling and kernel density estimation
Promdaen et al. Automated microalgae image classification
CN110298297B (en) Flame identification method and device
CN107038416B (en) Pedestrian detection method based on binary image improved HOG characteristics
CN109918971B (en) Method and device for detecting number of people in monitoring video
Espinal et al. Wavelet-based fractal signature analysis for automatic target recognition
CN103324951B (en) A kind of polynary isomery monitoring water environment data evaluation and method for early warning
CN107832723B (en) Smoke identification method and system based on LBP Gaussian pyramid
CN104143102A (en) Online image data processing method
CN116563768B (en) Intelligent detection method and system for microplastic pollutants
CN109978916B (en) Vibe moving target detection method based on gray level image feature matching
CN110321808B (en) Method, apparatus and storage medium for detecting carry-over and stolen object
Teutsch et al. Noise resistant gradient calculation and edge detection using local binary patterns
CN112836669B (en) Driver distraction driving detection method
CN111191575B (en) Naked flame detection method and system based on flame jumping modeling
Joshi et al. A simple scheme for contour detection.
CN111597973A (en) Method for detecting adulteration of qualified salted sea cucumbers by using machine vision
Al-Amaren et al. Edge Map Extraction of an Image Based on the Gradient of its Binary Versions
Hernández Structural analysis of textures based on LAW´ s filters
Rahmaniar et al. A novel object detection method based on Fuzzy sets theory and SURF

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200417

Termination date: 20211117

CF01 Termination of patent right due to non-payment of annual fee