CN104899895A - Detection method of trace complexity of mobile targets of fire video in channel of power transmission line - Google Patents

Abstract

The invention provides a detection method of trace complexity of mobile targets of a fire video in a channel of a power transmission line. Trace complexity of motion of mass center of a moving target is regarded as one of characteristics of fire movement. The detection method comprises steps of obtaining foreground images with the background subtraction method; obtaining difference images after binarization processing, wherein the pixel values of the binaryzation difference image only include 0 and 1; averaging row coordinates and line coordinates of all pixels with pixel value of 1 so as to obtain the coordinate value of the current frame of the mass center of the mobile target; and obtaining movement curves of the mass center of the mobile target after calculation of multi-frame image coordinate values. According to the invention, image-based fire disaster detection technology is adopted, so judgment and detection of fire are achieved; lower rate of false alarm and of missing report of detection results is ensured; and image signals acquired by a camera head is converted into digital signals via an acquisition card and then transmitted into a computer.

Description

A kind of electric transmission line channel pyrotechnics video moving target track complexity detection method

Technical field

A kind of electric transmission line channel pyrotechnics of the present invention video moving target track complexity detection method, relates to transmission line of electricity safety monitoring field.

Background technology

Electric power facility outside destroy refers to the artificial destruction comprising theft, traffic hazard, building operation etc.Statistics finds that a large amount of outside destroy is due to careless and inadvertent, premeditated and cause.In the example of electric power facility outside destroy, transmission line of electricity be suffer outside destroy the most frequently, the most serious electric power facility.Often occur that therefore, in passage, pyrotechnics becomes the principal element causing transmission line of electricity outside destroy and threaten transmission line of electricity safe operation by burning away the refuse or stalk and the flame that produces and flue dust in power transmission line corridor.

Sensor, it is more tight that the progress of microelectronics and infotech makes fire detection be combined with artificial intelligence, signal processing technology, and the novel fire detection technology of many physically based deformation and concept start to occur.For the fire monitoring early warning technology in the wide visual field on a large scale, the main fire detection technology using image recognition at present.This technology can flame detection and smog, and effectively reduce rate of false alarm and rate of failing to report, realizes the security monitoring to important operation area and guarantee.The picture signal of camera collection is converted to after digital signal through capture card imports computing machine into, computer background carries out treatment and analysis to the multiple characteristics of image of flame smog of fire by using computer vision and pattern-recognition scheduling theory, and whether final differentiation reports to the police.But transmission line of electricity running environment is complicated, greatly, polling transmission line background comprises mountain forest, river, farmland, road, sleet etc., and background appearance also can change along with the change in the four seasons in image background change, and this makes to identify that Objective extraction is very difficult; In pyrotechnics video detection technology, particularly tree shade is not taken into full account on the outdoor pyrotechnics video detection technology of long-distance large-range, Yun Ying moves impact on pyrotechnics testing result, the situation that testing result exists certain wrong report or fails to report; Pyrotechnics testing cost based on satellite remote sensing and particle imaging technique is higher, detects practice be very limited at electric transmission line channel pyrotechnics.Image processing algorithm aspect, the Smoke Detection algorithm of the somatotype coding that the people such as Japanese scholars Nobuyuki propose uses mapping calculation, and lack real-time, algorithm is comparatively consuming time; The method that B UgurToreyin employs based on wavelet transformation detects video smoke, when video camera is static, background thickens in smog, image medium-high frequency energy step-down, but this algorithm has higher requirements to the edge of background image and texture, versatility is affected; Xueming Shu etc. propose based on light-scattering system, smoke path system, particle imaging system, and detect smog by the shape of particle obtaining smog, but need by hardware implementing, cost is higher; Khananian A etc. detect the smog of forest fire generation in conjunction with neural network and remote sensing images, need the support of satellite remote sensing, lack economy equally.Therefore, on existing Research foundation, need to explore the stronger image processing algorithm of a kind of applicability, versatility and economy and fire monitoring early warning technology solves these problems.

Summary of the invention

For above problem, the present invention proposes a kind of electric transmission line channel pyrotechnics video moving target track complexity detection method, the method adopts the fire detection technology based on image, image procossing aspect adopts the average gray to image sequence, time series complexity, moving target motion track complexity, the multiple features fusion disposal routes such as moving target profile variation coefficient and prospect gray scale, realize the judgement to pyrotechnics and detection, and guarantee the rate of false alarm that testing result is lower and rate of failing to report, camera collection to picture signal to be converted to after digital signal through capture card and to import computing machine into, background program carries out treatment and analysis to the multiple characteristics of image comprised in digital signal, and judge whether to need to report to the police with this.

The technical scheme that the present invention takes is:

A kind of electric transmission line channel pyrotechnics video moving target track complexity detection method, using the intimate movement locus complexity of moving target as one of pyrotechnics motion characteristics, foreground image is obtained by Background difference, the difference image obtained after binary conversion treatment, the pixel value of binaryzation difference image only comprises " 0 " and " 1 " two values, by pixel value be all pixel column coordinates of " 1 " and row coordinate respectively averaged obtain moving target barycenter present frame place coordinate figure, and calculate through multiple image coordinate figure, obtain the curve movement of moving target barycenter.

Just can weigh moving target footprint characteristic by the property variable coefficient and equivalent diameter that calculate moving target, and in continuous dynamic image, calculate its profile variation coefficient; The image outline of pyrotechnics motion can constantly change in the process of pyrotechnics ascension, by calculating the equivalent diameter of profile and the form parameter with moving target, by the form parameter of moving target and the change severe degree of equivalent diameter, weighing moving target is rigidity or flexibility.

Using the data of the second-order statistics parameter relevant to foreground image gray scale as foreground image gray-scale statistical, by the former figure of pyrotechnics, the foreground image drawn out, and the second-order statistics curve of pyrotechnics foreground image sequence, and contrast the second-order statistics curve of the mobile image sequences such as vehicle, pedestrian, tree shade, reach the object of pyrotechnics identification and monitoring.

A kind of electric transmission line channel pyrotechnics of the present invention video moving target track complexity detection method, technique effect is as follows:

1), multiple features Data fusion technique, improve the accuracy of pyrotechnics identification.

2), adopt moving target center of mass motion curve Leah Pu Nuofu exponentiation algorithm pursuit movement target trajectory, better get rid of the disturbing factors such as cloud layer, leaf and sunlight angular intensities.

3) rescaled range (R/S) analytic approach, is adopted to carry out calculating the computation process of average gray time series entropy simplification without inclined calculated amount.

Accompanying drawing explanation

Fig. 1 is multiple features fusion pyrotechnics video Cleaning Principle figure of the present invention.

Fig. 2 (a) is pedestrian of the present invention, vehicle and pyrotechnics collection site lab diagram.

Fig. 2 (b) is tree shade of the present invention, Yun Ying and pyrotechnics collection site lab diagram.

Embodiment

Power transmission line corridor fire has certain seasonal characteristic, and early stage effective early warning and intervention can reduce the risk led to a disaster greatly.There is the disturbing factors such as such as cloud layer, leaf and sunlight angular intensities in video testing process, analyze static feature and the behavioral characteristics of various disturbing factor, a kind of electric transmission line channel pyrotechnics of the present invention video moving target track complexity detection method, proposes multiple features fusion methods such as adopting the average gray time series complexity of image sequence, moving target motion track complexity, moving target profile variation coefficient and prospect gray scale and realizes pyrotechnics early warning.As shown in Figure 1, collection site as shown in Figure 2 for its principle.

A kind of electric transmission line channel pyrotechnics video moving target track complexity detection method, monitoring objective is in motion process, its movement locus of different moving targets is different, vehicle and pedestrian mainly move with the form of near linear in field range, it is to-and-fro movement that tree shade rocks main manifestations, but it is a lot of that the image centroid movement locus of pyrotechnics is disturbed factor, movement locus is very complicated, this patent is using the intimate movement locus complexity of moving target as one of pyrotechnics motion characteristics, foreground image is obtained by Background difference, the difference image obtained after binary conversion treatment, the pixel value of binaryzation difference image only comprises " 0 " and " 1 " two values, by pixel value be all pixel column coordinates of " 1 " and row coordinate respectively averaged obtain moving target barycenter present frame place coordinate figure, and calculate through multiple image coordinate figure, obtain the curve movement of moving target barycenter.Motion target tracking and track adopt moving target center of mass motion curve Leah Pu Nuofu index calculation method, and calculation procedure comprises:

(1) phase space reconfiguration is carried out to moving target center of mass motion curve, Takens embedding theorems can be used to remove phase space reconstruction Rm for N point center of mass motion scalar time sequence

{v(t ₀+k·△t):k＝0,1,…N-1},

X _i＝(x(t _i),x(t _i+p·△t),…,x(t _i+(m-1)p·△t)) i＝1,2,3,…,M；

x(t _r)＝v(t ₀+(r-1)·△t) r＝1,2,…N-1.

Wherein, Xi is i-th point in phase space reconstruction Rm in M point reconstruct track, and M=N-(m-1) p, m embeds dimension, and τ=p. Δ t is time delay, and τ w=(m-1) τ is time window, wherein Δ t is the sampling period, τ w=(m-1) τ;

(2) maximum Leah Pu Nuofu index λ 1 is calculated: in order to λ 1 (i) can be calculated on the maximum direction of diffusion after phase space reconfiguration:

${\mathrm{\λ}}_i\left(i\right)=\frac{1}{\mathrm{jh}\·\mathrm{\Δt}}\log \frac{\left|\right|X_{\mathrm{ip}+\mathrm{jh}}-X_{i+\mathrm{jh}}\left|\right|}{\left|\right|X_{\mathrm{ip}}-X_i\left|\right|}=\underset{k}{\max }\left\{\frac{1}{h\·\mathrm{\Δt}}\log \frac{\left|\right|X_{\mathrm{ip}+\mathrm{jh}}-X_{i+\mathrm{jh}}\left|\right|}{\left|\right|X_{\mathrm{ip}}-X_i\left|\right|}\right\}$

Above formula comprises two subsidiary condition:

(i) $\begin{array}{cc}1<h<\frac{a{\mathrm{\&tau;}}_{\mathrm{\&omega;}}}{\mathrm{\&Delta;t}}& 0<a<1\end{array};$

(ii)

Finally, can obtain

${\mathrm{\&lambda;}}_1=\frac{1}{M-1}\underset{i=1}{\overset{M-1}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_1\left(i\right).$

Thus obtain the maximum Leah Pu Nuofu index of moving target image centroid curve movement.

Profile variation coefficient is the important evidence distinguishing monitoring objective, and the image outline coefficient of variation is also one of pyrotechnics video features.The present invention just can weigh moving target footprint characteristic by the property variable coefficient and equivalent diameter calculating moving target, and in continuous dynamic image, calculate its profile variation coefficient.The image outline of pyrotechnics motion can constantly change in the process of pyrotechnics ascension, calculates the equivalent diameter of profile and the form parameter with moving target by following formula:

$D_b=\sqrt{\frac{A_b}{\mathrm{\&pi;}}}$

${\mathrm{\&rho;}}_b=\frac{L_b^2}{4\mathrm{\&pi;}{A}_b}$

Wherein in formula, Db represents moving target equivalent diameter, ρ _brepresent the form parameter of described moving target, Lb represents the moving target profile length of side, and Ab is the area of moving target.

Weighing moving target by the form parameter of moving target and the change severe degree of equivalent diameter is rigidity or flexibility.

The foreground image intensity profile of monitoring moving target shows different features according to different objects, therefore carries out to the foreground image gray scale of moving target the important step that Information Statistics are monitoring and identification moving target.This patent is using the data of the second-order statistics parameter relevant to foreground image gray scale as foreground image gray-scale statistical, by the former figure of pyrotechnics, the foreground image drawn out, and the second-order statistics curve of pyrotechnics foreground image sequence, and contrast the second-order statistics curve of the mobile image sequences such as vehicle, pedestrian, tree shade, reach the object of pyrotechnics identification and monitoring.If SAR image is a stochastic process, the Two-dimensional Statistical parameter mainly used is as follows:

(1) histogram.For a width gray level image, if (i, j) the two dimensional gray value as initial point at place is g (i, j) and gc (i, j), then the density of gradation of image joint distribution is: P (a, b)=P{g (i, j)=a, gc (i, j)=b}, wherein a and b is the gray shade scale between 0 to L-1.

(2) auto-correlation.Can be expressed as: $B_A=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0d}{\overset{L-1d}{\mathrm{\&Sigma;}}}\mathrm{abP}(a,b)$

(3) energy.Can be expressed as: $B_N=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0d}{\overset{L-1d}{\mathrm{\&Sigma;}}}{\left[P(a,b)\right]}^{2d}$

(4) covariance.Can be expressed as: $B_c=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0d}{\overset{L-1d}{\mathrm{\&Sigma;}}}(a-\stackrel{\&OverBar;}{a})(b-\stackrel{\&OverBar;}{b})P(a,b)$

(5) moment of inertia.Can be expressed as: $B_I=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0d}{\overset{L-1d}{\mathrm{\&Sigma;}}}{(a-\stackrel{\&OverBar;}{a})}^{2d}P(a,b)$

(6) entropy.Can be expressed as: $B_I=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0d}{\overset{L-1d}{\mathrm{\&Sigma;}}}P(a,b)\log \left[p(a,b)\right]$

In formula (1) ~ (6), wherein a and b is the gray shade scale between 0 to L-1.L represents the gray shade scale of image, and represent mean gradation.

The characteristic quantity of gray feature is also applicable to subimage above.

Average gray time series entropy adopts rescaled range (R/S) analytic approach to calculate.Hurst Exponent (H) is set up as judging that time series data defers to the index that random walk still has inclined random walk process by rescaled range (R/S) analytical approach.If Xi=X1 ... Xn is a seasonal effect in time series n successive value, takes the logarithm and Data Placement after carrying out first difference is length is the adjacent sub-range A of H, i.e. A*H=n.Then: the average in each sub-range is:

Xm＝(X1+…+Xh)/H

Standard deviation is:

$S_h=\sqrt{\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{(x_i-x_m)}^2/h}$

The accumulation transfer (XKA) of average is:

$X_{T,A}=\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}(x_{i,A}-x_m)$

In group, extreme difference is:

Rh＝max(Xr,A)-mix(Xr,A)

Hurst Exponent (H) is:

$R_n/S_n-(1/A)*\underset{h=1}{\overset{A}{\mathrm{\&Sigma;}}}(R_n/S_n)$

The pass that Hurst releases is:

R _n/S _n＝c*n ^H

Wherein c is constant, and n is the number of observed value, and H is Hurst Exponent.

Pyrotechnics identification of the present invention adopts multiple features Data fusion technique, data fusion is the most important step of multi-biological characteristic identification, the identity recognizing technology of Feature-level fusion requires to associate the proper vector extracted from different sensors. and sensor output data type often differs very large, thus carries out non-linear correlation to dissimilar data and forms one to merge vector ratio more difficult.Neural network can realize special nonlinear transformation, the input space transform to hidden layer the space of opening, classification problem thereafter can be made in this space to become than being easier to.This conversion maximizes special feature extraction criterion, can regard a kind of special feature extractor as.The data fusion of this patent be the information from multiple sensor or multi-source carried out comprehensively, be correlated with, the process such as filtration.More weak in Infrared Targets characteristic, out of true, incomplete target information can be produced when ground unrest more complicated, adopt separately a certain feature to associate, usually can produce larger error.Therefore need to adopt each degree of association of mode to candidate target of multiple features fusion to consider, obtain the Synthesis Relational Grade of candidate target, to improve the Stability and veracity of association.For eliminating the feature non-equilibrium numerically participating in merging, be first normalized various feature, target correlated characteristic gray scale is 255, and target area is 50, and for barycenter deviation, normalization formula is:

$d=\frac{\sqrt{{(x-T_x)}^2+{(y-T_y)}^2}}{\sqrt{w^2+h^2}}$

W and H is respectively the length of image and wide, T _xand T _yit is the locations of real targets that former frame is judged.

Need after normalized to determine weight factor, definition base is the fail-safe program of each eigenwert.Because target area is comparatively large all the time for noise spot, in image, some noise can repeat in some positions, and the weight coefficient now distributing to target area should be slightly larger.Weight coefficient should meet:

$\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{a}_i=1$

By multiple features Data fusion technique, overall treatment is carried out to the average gray of image sequence, time series complexity, moving target motion track complexity, moving target profile variation coefficient and prospect gray scale etc., needing to realize complex background further when in real time process video information obtains early warning result and camera itself rocks the kinematic error compensation caused, moving object monitoring and the accuracy identified can be improved.

Claims (4)

1. an electric transmission line channel pyrotechnics video moving target track complexity detection method, it is characterized in that, using the intimate movement locus complexity of moving target as one of pyrotechnics motion characteristics, foreground image is obtained by Background difference, the difference image obtained after binary conversion treatment, the pixel value of binaryzation difference image only comprises " 0 " and " 1 " two values, by pixel value be all pixel column coordinates of " 1 " and row coordinate respectively averaged obtain moving target barycenter present frame place coordinate figure, and calculate through multiple image coordinate figure, obtain the curve movement of moving target barycenter,

Just can weigh moving target footprint characteristic by the property variable coefficient and equivalent diameter that calculate moving target, and in continuous dynamic image, calculate its profile variation coefficient; The image outline of pyrotechnics motion can constantly change in the process of pyrotechnics ascension, by calculating the equivalent diameter of profile and the form parameter with moving target, by the form parameter of moving target and the change severe degree of equivalent diameter, weighing moving target is rigidity or flexibility;

Using the data of the second-order statistics parameter relevant to foreground image gray scale as foreground image gray-scale statistical, by the former figure of pyrotechnics, the foreground image drawn out, and the second-order statistics curve of pyrotechnics foreground image sequence, and contrast the second-order statistics curve of the mobile image sequences such as vehicle, pedestrian, tree shade, reach the object of pyrotechnics identification and monitoring.

2. a kind of electric transmission line channel pyrotechnics video moving target track complexity detection method according to claim 1, it is characterized in that, the curve movement calculation procedure of moving target barycenter comprises:

Step 1: carry out phase space reconfiguration to moving target center of mass motion curve, can use Takens embedding theorems to remove phase space reconstruction Rm for N point center of mass motion scalar time sequence

{v(t ₀+k·Δt):k＝0,1,…N-1},

X _i＝(x(t _i),x(t _i+p·Δt),…,x(t _i+(m-1)p·Δt))i＝1,2,3,…,M；

x(t _r)＝v(t ₀+(r-1)·Δt)r＝1,2,…N-1.

Wherein, Xi is i-th point in phase space reconstruction Rm in M point reconstruct track, and M=N-(m-1) p, m embeds dimension, and τ=p. Δ t is time delay, and τ w=(m-1) τ is time window, wherein Δ t is the sampling period, τ w=(m-1) τ;

Step 2: calculate maximum Leah Pu Nuofu index λ 1: in order to λ 1 (i) can be calculated on the maximum direction of diffusion after phase space reconfiguration:

${\mathrm{\&lambda;}}_i\left(i\right)=\frac{1}{\mathrm{jh}\&CenterDot;\mathrm{\&Delta;t}}\log \frac{\left|\right|X_{\mathrm{ip}+\mathrm{jh}}-X_{i+\mathrm{jh}}\left|\right|}{\left|\right|X_{\mathrm{ip}}-X_i\left|\right|}=\underset{k}{\max }\left\{\frac{1}{h\&CenterDot;\mathrm{\&Delta;t}}\log \frac{\left|\right|X_{\mathrm{ip}+\mathrm{jh}}-X_{i+\mathrm{jh}}\left|\right|}{\left|\right|X_{\mathrm{ip}}-X_i\left|\right|}\right\}$

Above formula comprises two subsidiary condition:

(i)0<a<1；

(ii) 0<b<1,b∝m.

Finally, can obtain

${\mathrm{\&lambda;}}_1=\frac{1}{M-1}\underset{i=1}{\overset{M-1}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_1\left(i\right).$

Thus obtain the maximum Leah Pu Nuofu index of moving target image centroid curve movement.

3. a kind of electric transmission line channel pyrotechnics video moving target track complexity detection method according to claim 1, is characterized in that, the form parameter of described moving target and equivalent diameter calculate as follows:

$D_b=\sqrt{\frac{A_b}{\mathrm{\&pi;}}}$

${\mathrm{\&rho;}}_b=\frac{{L:}_b^2}{4\mathrm{\&pi;}{A}_b}$

Wherein in formula, Db represents moving target equivalent diameter, ρ _brepresent the form parameter of described moving target, Lb represents the moving target profile length of side, and Ab is the area of moving target.

4. a kind of electric transmission line channel pyrotechnics video moving target track complexity detection method according to claim 1, it is characterized in that, the second-order statistics curve computing method of mobile image sequence are as follows:

(1) histogram: for a width gray level image, if the two dimensional gray value as initial point at (I, j) place is g (I, j) and g ^'(I, j), then the density of gradation of image joint distribution is: P (a, b)=P _r{ g (I, j)=a, g ^'(I, j)=b};

(2) auto-correlation:: be expressed as: $B_A=\underset{a=0}{\overset{l-1}{\mathrm{\&Sigma;}}}\underset{b=0}{\overset{l-1}{\mathrm{\&Sigma;}}}\mathrm{abP}(a,b);$

(3) energy: be expressed as: $B_N=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0}{\overset{l-1}{\mathrm{\&Sigma;}}}{\left[P(a,b)\right]}^2;$

(4) covariance: be expressed as: $B_c=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}(a-\stackrel{\&OverBar;}{a})(b-\stackrel{\&OverBar;}{b})P(a,b);$

(5) moment of inertia: be expressed as: $B_I=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}{(a-\stackrel{\&OverBar;}{a})}^{2}P(a,b);$

(6) entropy: be expressed as: $B_I=\underset{a=0}{\overset{L-1}{\mathrm{\&Sigma;}}}\underset{b=0}{\overset{L-1}{\mathrm{\&Sigma;}}}P(a,b)\log \left[P(a,b)\right];$

In formula (1) ~ (6), wherein a and b is the gray shade scale between 0 to L-1, and L represents the gray shade scale of image, and represent mean gradation;

The characteristic quantity of gray feature is also applicable to subimage above;

Average gray time series entropy adopts rescaled range (R/S) analytic approach to calculate, Hurst Exponent (H) is set up as judging that time series data defers to the index that random walk still has inclined random walk process by rescaled range (R/S) analytical approach, if Xi=X1, Xn is a seasonal effect in time series n successive value, to take the logarithm and Data Placement after carrying out first difference is length is the adjacent sub-range A of H, i.e. A*H=n.Then: the average in each sub-range is:

Xm＝(X1+…+Xh)/H

Standard deviation is: $S_h=\sqrt{\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{(x_i-x_m)}^2/h}$

The accumulation transfer (XKA) of average is: $X_{T,A}=\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}(x_{i,A}-x_m)$

In group, extreme difference is:

Rh＝max(Xr,A)-mix(Xr,A)

Hurst Exponent (H) is: $R_n/S_n-(1/A)*\underset{h=1}{\overset{A}{\mathrm{\&Sigma;}}}(R_n/S_n)$

The pass that Hurst releases is:

R _n/S _n＝c*n ^H

Wherein c is constant, and n is the number of observed value, and H is Hurst Exponent.