CN104134075A - Mineral content monitoring system based on videos - Google Patents

Abstract

The invention belongs to the field of computer vision, mode recognition and monitoring, and particularly relates to a mineral monitoring system based on videos. Nowadays, with the rapid science and technology development, coal is still the most economic and most reliable energy source and energy strategic resource in the world. In a new period, the further increase of the coal demand is caused by rapid increase of national economy, and mining industry attracts more attention. In the mining industry technology, mine reserves are calculated and checked; dynamic conditions of the reserves are measured and counted; the reserve loss is analyzed; a mine reserve table is compiled; the basis is provided for improving the mine reserve level and increasing the mine reserves; and the service is provided for normal production replacement and reasonable resource utilization. Phenomena of mining content nonuniformity, mineral content reduction or mineral content loss often occur in the mining industry, so that in order to realize real-time monitoring on the mining quantity on one hand and to realize real-time detection of mine environment on the other hand for ensuring the smooth completion of underground operation, the system is specially developed.

Description

A kind of mineral products content monitoring system based on video

Affiliated technical field

The invention belongs to computer vision, pattern-recognition and monitoring field, be specifically related to a kind of mineral products monitoring system based on video.

Background technology

Today of science and technology fast development, coal is still the most reliable most economical energy and energy strategy resource in the world.In the new period, national economy rapid growth causes coal demand further to increase, and mining industry receives much concern.In mining industry technology, calculate and examine reserves on mine, measure and statistics reserves dynamic, analyze reserves losses, establishment reserves on mine table, provides foundation for improving reserves on mine rank and expanding reserves on mine, takes over, resource rational utilization service for producing normally.Owing to usually occurring that in mining industry mining content is inhomogeneous, mineral content reduces, or the phenomenon of losing, and for one side monitoring mining in real time amount, also can detect in real time on the other hand Minepit environment, guarantees that borehole operation completes smoothly.Special this system of research and development.

Summary of the invention

The present invention proposes a kind of mineral products monitoring system based on video, and the problem that cause mineral run off inaccurate with mineral content monitoring in solving in prior art in the process of miner operation mine.

Brief description of the drawings

Fig. 1 does training set according to the different mine car picture of a large amount of mineral contents of the choosing of the embodiment of the present invention, extracts feature, uses svm classifier process flow diagram.

Fig. 2 cuts apart according to the image of the embodiment of the present invention process flow diagram that extracts proper vector.

Fig. 3 is the process flow diagram generating based on video mineral products monitoring system according to the embodiment of the present invention.

specific embodiments:

It should be noted that, if do not conflicted, each feature in the embodiment of the present invention and embodiment can mutually combine, all within protection scope of the present invention.In addition, can in the computer system such as one group of computer executable instructions, carry out in the step shown in the process flow diagram of accompanying drawing, and, although there is shown logical order in flow process, but in some cases, can carry out shown or described step with the order being different from herein.

Describe below with reference to the accompanying drawings and in conjunction with the embodiments the present invention in detail.

Embodiment mono-, extraction prospect specifically comprises: use mixed Gaussian Background Algorithm to calculate the image of described target frame, obtain the mixed Gauss model of described target frame image, take out foreground image.

Embodiment mono-: cut apart mine car picture and comprise: the objective contour in detected image, traversal boundary rectangle, judges the rectangular area at people place and the rectangular area at mine car place based on experience value, further to Image Segmentation Using, obtains mine car figure.

The characteristic parameter that embodiment mono-: SVM trains appropriate gray shade litre matrix comprises:

First to the different mine car picture of a large amount of mineral products content, these picture yardsticks are to be unanimously in the same sizely 128 × 128, and direction is unanimously the position consistency of mine car in every width picture, trains, and calculates the gray scale symbiosis square of figure four direction.

Wherein gray level co-occurrence matrixes for (being designated as P battle array) be on statistical space, there is certain position relationship a pair of pixel gray scale to the frequency occurring. (position is x to the pixel that to its essence is from gradation of image be i, y) set out, statistics is d with its distance, gray scale is the pixel (x+Dx of j, y+Dy) the frequency P (i simultaneously occurring, j, d, θ).

Mathematic(al) representation is:

P (i, j, d, θ)=[(x, y), (x+Dx, y+Dy) | f (x, y)=i; F (x+Dx, y+Dy)=j] }. the gray level co-occurrence matrixes of generation can finely be portrayed the textural characteristics of figure image.

In formula: x, y=0,1,2 ..., N-1 is the cell coordinate of image; I, j=0,1 ... L-1 is gray level; Dx, Dy is that position offset .d is the generation step-length of W battle array; θ is the generation direction of W battle array, gets 0 °, 45 °, 90 °, 135 ° these 4 directions. get d=4 according to empirical value.

Conventional characteristic parameter has:

Angle second moment:

$\mathrm{ASM}=\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{P}_{\mathrm{ij}}^2$

Entropy:

$\mathrm{ENT}=-\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{P}_{\mathrm{ij}}\log P_{\mathrm{ij}}$

Isotropism: $\mathrm{HOM}=\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{P}_{\mathrm{ij}}/{[1+{(i-j)}^2]}^2$

Dissimilarity: $\mathrm{DIS}=\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}|i-j|P_{\mathrm{ij}}$

Contrast:

$\mathrm{CON}=\underset{\mathrm{ij}}{\mathrm{\Σ}}{(i-j)}^2{P}_{\mathrm{ij}}$

Correlativity: $\mathrm{COR}=\frac{1}{{\mathrm{\σ}}_1{\mathrm{\σ}}_2}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}(i-u_1)(j-u_2)P_{\mathrm{ij}}$

Wherein: $u_1={\mathrm{\Σ}}_{i=0}^{N-1}i{\mathrm{\Σ}}_{j=0}^{N-1}{P}_{\mathrm{ij}}$ $u_2={\mathrm{\Σ}}_{i=0}^{N-1}j{\mathrm{\Σ}}_{j=0}^{N-1}{P}_{\mathrm{ij}}$

${\mathrm{\σ}}_1^2=\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{(i-u_1)}^2\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{P}_{\mathrm{ij}}$ ${\mathrm{\σ}}_{12}^2=\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{(j-u_1)}^2\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{P}_{\mathrm{ij}}$

Calculate 6 typical characteristic parameters, each gets the mean value of four gray level co-occurrence matrixes, constitutive characteristic vector is in order to reduce dimension so that flow process afterwards well completes, and we train 4 characteristic parameters that classifying quality is best with SVM, with these 4 characteristic parameter constitutive characteristic vector V ₁.

Embodiment mono-: the mine-containing amount in mine car picture is classified and specifically comprised:

First to disturb in order reducing to greatest extent the identification effect improving mine car content, to carry out pre-service to the mine car picture being partitioned into, pretreatment process comprises: (1) carries out ashing processing to the image being partitioned into; (2) look for and see the maximum boundary rectangle that comprises mine car, and be partitioned into this rectangle; (3) size to 128 × 128 of specification mine car picture; The mine car picture principal direction obtaining is inconsistent, thus to obtain pre-service image further cut apart to make the proper vector obtaining not to be subject to as far as possible the restriction of direction.Cutting apart flow process is: proposing even number line odd column, to obtain size be 64 × 64 picture A ₁and odd-numbered line even column to obtain size be 64 × 64 picture A ₂; Again to P ₁continue to cut apart that to obtain size be 32 × 32B ₁and B _2,to P ₂continue to cut apart that to obtain size be 32 × 32B ₃and B ₄; Carry out again decomposing for three times and obtain C ₁, C ₂, C ₃, C ₄, C ₅, C ₆, C ₇, C ₈, size is respectively 16 × 16, and this eight sub-picture is calculated respectively to its gray level co-occurrence matrixes, then extracts 8 characteristic parameters that train, be aligned to the proper vector that obtains together 32 dimensions, then classify with SVM, draw affiliated classification, mine-containing amount without, less, or more.

Claims (6)

1. based on a video mineral products content monitoring system, it is characterized in that, comprising:

Target frame image in original video is carried out to background modeling, obtain background image;

Utilize described background image to extract moving target people and the mine car in described target frame image;

Image is cut apart and obtained mine car picture;

Utilize the criteria for classification that SVM trains to classify to the mine-containing amount in mine car picture.

2. according to claim 1ly it is characterized in that based on video mineral products content monitoring system, the image of the target frame in original video carried out to background extracting and comprise:

Use mixed Gaussian Background Algorithm to calculate the image of described target frame, obtain the mixed Gauss model of described target frame image.

3. according to claim 1ly it is characterized in that based on video mineral products monitoring system, extract foreground image.

4. according to claim 1 based on video mineral products content monitoring system, it is characterized in that, image is cut apart and obtained mine car and comprise: the objective contour in detected image, traversal boundary rectangle, judge based on experience value the rectangular area at people place and the rectangular area at mine car place, further, to Image Segmentation Using, obtain mine car figure.

5. according to claim 4ly it is characterized in that based on video mineral products content monitoring system, utilize the criteria for classification that SVM trains to comprise:

First to the different mine car picture of a large amount of mineral products content, these picture yardsticks are to be unanimously in the same sizely 128 × 128, and direction is unanimously the position consistency of mine car in every width picture, trains, and calculates the gray scale symbiosis square of figure four direction.Calculate 6 typical characteristic parameters, each gets the mean value of four gray level co-occurrence matrixes, constitutive characteristic vector is in order to reduce dimension so that flow process afterwards well completes, and we train 4 characteristic parameters that classifying quality is best with SVM, with these 4 characteristic parameter constitutive characteristic vector V1.

Wherein gray level co-occurrence matrixes for (being designated as P battle array) be on statistical space, there is certain position relationship a pair of pixel gray scale to the frequency occurring. (position is x to the pixel that to its essence is from gradation of image be i, y) set out, statistics is d with its distance, gray scale is the pixel (x+Dx of j, y+Dy) the frequency P (i simultaneously occurring, j, d, θ).

Mathematic(al) representation is:

P (i, j, d, θ)=[(x, y), (x+Dx, y+Dy) | f (x, y)=i; F (x+Dx, y+Dy)=j] }. the gray level co-occurrence matrixes of generation can finely be portrayed the textural characteristics of figure image.

In formula: x, y=0,1,2 ..., N-1 is the cell coordinate of image; I, j=0,1 ... L-1 is gray level; Dx, Dy is that position offset .d is the generation step-length of W battle array; θ is the generation direction of W battle array, gets 0 °, 45 °, 90 °, 135 ° these 4 directions.

According to above-mentioned formula, get d=4 according to empirical value, each sub-picture all generates four gray level co-occurrence matrixes.

Conventional characteristic parameter has:

Angle second moment:

Entropy:

Isotropism:

Dissimilarity:

Contrast:

Correlativity:

Wherein:

6. according to claim 4ly it is characterized in that based on video mineral products content monitoring system, utilize the criteria for classification that SVM trains that the mine-containing amount in mine car picture is classified and comprised:

First to disturb in order reducing to greatest extent the identification effect improving mine car content, to carry out pre-service to the mine car picture being partitioned into, pretreatment process comprises: (1) carries out ashing processing to the image being partitioned into; (2) look for and see the maximum boundary rectangle that comprises mine car, and be partitioned into this rectangle; (3) size to 128 × 128 of specification mine car picture;

Secondly due to the mine car picture principal direction arriving inconsistent, so to obtain pre-service image further cut apart to make the proper vector obtaining not to be subject to as far as possible the restriction of direction.Cutting apart flow process is: proposing even number line odd column, to obtain size be 64 × 64 picture A ₁and odd-numbered line even column to obtain size be 64 × 64 picture A ₂; Again to P ₁continue to cut apart that to obtain size be 32 × 32B ₁and B _2,to P ₂continue to cut apart that to obtain size be 32 × 32B ₃and B ₄; Carry out again decomposing for three times and obtain C ₁, C ₂, C ₃, C ₄, C ₅, C ₆, C ₇, C ₈, size is respectively 16 × 16, and this eight sub-picture is calculated respectively to its gray level co-occurrence matrixes, then extracts 8 characteristic parameters that train, be aligned to the proper vector that obtains together 32 dimensions, then classify with SVM, draw affiliated classification, mine-containing amount without, less, or more.