CN101561932B - Method and device for detecting real-time movement target under dynamic and complicated background - Google Patents

Abstract

The invention relates to a method and a device for detecting a real-time movement target under a dynamic and complicated background. The method comprises the following proposal: acquiring a frame image from a video, determining a covariance of colors in the current frame image, and acquiring a gray image of the frame image; and acquiring a gray value and a gradient value of a pixel point in the gray image, determining a pixel point value according to the gray value, the gradient value and preset parameters of the pixel point, judging whether the pixel point value is less than a preset multipleof the covariance of the colors determined in the last frame image, if so, taking the pixel point as a background point, otherwise, taking the pixel point as a foreground point, and taking collection of foreground points as the real-time movement target. The method and the device restrain the influence of the sudden change of light on a detection result by converting the video image into the gray image so as to more precisely detect the movement target.

Description

Method for detecting real-time moving object and device under a kind of DYNAMIC COMPLEX background

Technical field

The present invention relates to video analysis and monitoring technology, be specifically related to method for detecting real-time moving object and device under a kind of DYNAMIC COMPLEX background.

Background technology

Motion target detection is the basis of digital video analysis and video monitoring system, and the accurate detection of moving target can bring very useful effect for follow-up work.At present the motion target detection method mainly contains three kinds of frame difference method, optical flow method and background subtraction methods; Wherein the background subtraction method is the most general at present also effective method, and it is to utilize scene information, sets up a background model that does not have moving target; Thereby and itself and present frame compared detect moving target; It can not adapt to the unexpected variation of illumination, if background illumination changes suddenly, thereby then the color of most of background dot can change and is identified as the foreground point; It also can't be discerned like irregular background motion targets such as branch that waves or current.

Summary of the invention

The embodiment of the invention provides method for detecting real-time moving object and the device under a kind of DYNAMIC COMPLEX background, and it can adapt to the unexpected variation of illumination, can detect moving target more accurately.

Method for detecting real-time moving object under a kind of DYNAMIC COMPLEX background that the embodiment of the invention provides comprises:

From video, obtain two field picture, confirm the covariance of color in the current frame image, and obtain the gray level image of said two field picture;

Obtain gray values of pixel points and Grad in the said gray level image; Said gray values of pixel points, Grad and preset parameter according to obtaining are confirmed the pixel point value, judge said pixel point value whether less than the prearranged multiple of the covariance of the color of confirming in the former frame image, if; Confirm that then said pixel is a background dot; If not, confirm that then said pixel is the foreground point, the set of said foreground point is the real time kinematics target.

Real time kinematics object detecting device under a kind of DYNAMIC COMPLEX background that the embodiment of the invention provides comprises:

The covariance acquisition module is used for obtaining two field picture from video, confirms the covariance of color in the current frame image;

The pixel acquisition module; Be used for converting two field picture described in the covariance acquisition module into gray level image; And obtain gray values of pixel points and Grad in the said gray level image; Gray values of pixel points and Grad in the gray level image described in the gray level image acquisition module, and confirm the pixel point value according to the said gray values of pixel points, Grad and the preset parameter that obtain;

The first dynamic object determination module; Be used for judging the prearranged multiple of the covariance of the color whether determined said pixel point value of pixel acquisition module is confirmed less than former frame image covariance acquisition module; If confirm that then said pixel is a background dot, if not; Confirm that then said pixel is the foreground point, the set of said foreground point is the real time kinematics target.

Method for detecting real-time moving object and device under a kind of DYNAMIC COMPLEX background that the embodiment of the invention provides, it suppresses the influence of the unexpected variation of illumination to testing result through video image being converted into gray level image, can detect moving target more accurately.

Description of drawings

Fig. 1 is the structural representation of the real time kinematics object detecting device under described a kind of DYNAMIC COMPLEX background of providing of the embodiment of the invention;

Fig. 2 is the structural representation that the real time kinematics object detecting device under described a kind of DYNAMIC COMPLEX background of providing of the embodiment of the invention further describes in an embodiment.

Embodiment

In the scheme of the method for detecting real-time moving object under a kind of DYNAMIC COMPLEX background that the embodiment of the invention provides, at first, from video, obtain two field picture, confirm the covariance of color in the current frame image, and obtain the gray level image of said two field picture; Then, obtain gray values of pixel points and Grad in the said gray level image, confirm the pixel point value according to the said gray values of pixel points, Grad and the preset parameter that obtain; Judge that said pixel point value is whether less than the prearranged multiple of the covariance of the color of confirming in the former frame image; If confirm that then said pixel is a background dot, if not; Confirm that then said pixel is the foreground point, the set of said foreground point is the real time kinematics target.

In such scheme; Further can comprise according to the foreground point of confirming in the former frame image foreground point of confirming in the present frame is adjusted; Concrete adjustment process can for: extract in the former frame image pixel point value for the present frame of foreground point; Confirm that according to pixel neighborhood of a point point described in the pixel point value of the said present frame that extracts and the former frame image pixel of present frame is the probability of real foreground point, if determined probability less than designated value, then is adjusted into background dot with the foreground point of confirming in the said present frame; If determined probability more than or equal to designated value, is not then adjusted the foreground point of confirming.

Comprehensive such scheme, concrete process can for:

Step S1, from video, obtain two field picture;

Step S2, confirm the weight w of color k in the said two field picture _k, average μ _kAnd covariance sigma _k ², wherein k generally chooses 5 or 3 for preset natural number;

Step S3, convert said two field picture into gray level image, obtain gray values of pixel points A in the said gray level image _Ij, according to gray values of pixel points A in the said gray level image that obtains _IjObtain the Grad of pixel $M_{\mathrm{Ij}}=\sqrt{{(A_{\mathrm{Ij}}-A_{i+1,j})}^2+{(A_{\mathrm{Ij}}-A_{i,j+1})}^2}$ And gradient direction R _Ij=ac tan 2 (A _Ij-A _{I+1, j}, A _Ij-A _{I, j+1}), wherein i is a horizontal ordinate, j is an ordinate, and according to the said gray values of pixel points A that obtains _Ij, Grad M _IjConfirm pixel point value X with preset parameter λ _Ij=(λ A _Ij, (1-λ) M _Ij), wherein preset parameter λ is the value in [0,1], it is to confirm according to the colouring information of image.Set up mixed Gauss model according to above-mentioned data:

$P\left(X_t\right)=\underset{k}{\mathrm{\Σ}}{w}_k\×\frac{1}{{\left(2\mathrm{\π}\right)}^{n/2}{\left|{\mathrm{\Σ}}_k\right|}^{1/2}}{e}^{-\frac{1}{2}{(X_{\mathrm{ij}}-{\mathrm{\μ}}_k)}^T{\mathrm{\Σ}}_k^{-1}(X_{\mathrm{ij}}-{\mathrm{\μ}}_k)},$

∑ wherein _kIt is the standard covariance sigma _kMultiply each other with unit matrix and to obtain (X _Ij-μ _k) ^TExpression (X _Ij-μ _k) transposition;

Step S4, the said current frame pixel point value X of judgement _T+1Whether with the mixed Gauss model coupling of former frame image, if, execution in step S5 then, if not, execution in step S6 then;

Step S5, confirm that said pixel is a background dot, and according to the weight w of color k in the former frame image _t, average μ _tAnd covariance sigma _t ²And the Grad M of present frame _T+1With pixel point value X _T+1Upgrade the weight w of color k in the current frame image _T+1, average μ _T+1And covariance sigma _T+1 ², concrete renewal process can comprise: $\begin{array}{c}{w}_{t+1}=(1-\mathrm{\&alpha;})w_t+\mathrm{\&alpha;}{M}_{t+1}\\ {\mathrm{\&mu;}}_{t+1}=(1-\mathrm{\&rho;}){\mathrm{\&mu;}}_t+\mathrm{\&rho;}{X}_{t+1}\\ {{\mathrm{\&sigma;}}_{t+1}}^2=(1-\mathrm{\&rho;}){{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="H2009100840075E00021.png"\; state="\{\{state\}\}">t</figure-callout>}}}^2+\mathrm{\&rho;}{(X_{t+1}+{\mathrm{\&mu;}}_{t+1})}^T(X_{t+1}-{\mathrm{\&mu;}}_{t+1})\end{array},$ Wherein α is a predetermined value, and it is to confirm ρ=α/w according to the intensity of variation of former frame image and current frame image _T+1, and according to above-mentioned Data Update mixed Gauss model P (X _T+1);

Step S6, confirm that said pixel is the foreground point, and upgrade the weight w of color k in the current frame image according to predetermined value _T+1', average μ _T+1' and the standard covariance sigma _T+1', detailed process can comprise: make weight w _T+1' the value that equals to be scheduled to generally chooses 0.05, makes the standard covariance sigma _T+1' the value that equals to be scheduled to generally chooses 30, makes average μ _T+1'=X _T+1, and according to above-mentioned Data Update mixed Gauss model P (X _T+1);

Step S7, according to foreground point and background dot that step S5 in the former frame image and step S6 confirm, the foreground point of confirming in the present frame is adjusted, concrete adjustment process comprises: extract that pixel is the pixel X of the present frame of foreground point in the former frame image _T+1, according to pixel neighborhood of a point point Y in the former frame image _IjConfirm the pixel X of present frame _T+1Probability for real foreground point

Wherein a is a positive number, and f and 1 is respectively the variable quantity of horizontal ordinate, and f and l are taken as integer; D gets odd number, the expression width neighborhood, and b is the distance between pixel described in the former frame image and the said pixel neighborhood of a point point; It can replace with Gauss's template of the symmetric matrix of a d * d; The element value of said symmetric matrix meets Gaussian distribution, can calculate with gaussian kernel function, and for example Gauss's template of n=3 is: $\left(\begin{array}{ccc}1& 2& 1\\ 2& 4& 2\\ 1& 2& 1\end{array}\right),$ If with the color of 256 grades gray-scale value remarked pixels, so

Can be one is 256 array in advance by the good size of calculating, if determined probability

Less than designated value, if execution in step S8 then is determined probability

More than or equal to designated value, execution in step S9 then, described designated value is to confirm according to the scene of video image, if scene is complicated, then designated value can be chosen the value in [0.5,0.7], if scene is simple, then designated value can be chosen the value in [0.3,0.5];

Step S8, the foreground point of confirming in the said present frame is adjusted into background dot, and execution in step S10;

Step S9, the foreground point of confirming is not adjusted, and execution in step S10;

Step S10, current frame image data replacement former frame view data, and the last frame image of execution in step S1 in video again.

The embodiment of the invention also provides the real time kinematics object detecting device under a kind of DYNAMIC COMPLEX background, and is as shown in Figure 1, comprising:

Covariance acquisition module 1 is used for obtaining two field picture from video, confirms the covariance of color in the current frame image;

Pixel acquisition module 2; Be used for converting two field picture described in the covariance acquisition module 1 into gray level image; And obtain gray values of pixel points and Grad in the said gray level image; Gray values of pixel points and Grad in the gray level image described in the gray level image acquisition module, and confirm the pixel point value according to the said gray values of pixel points, Grad and the preset parameter that obtain;

The first dynamic object determination module 3; Be used to judge pixel acquisition module 2 determined said pixel point values whether with the mixed Gauss model coupling of former frame image; If confirm that then said pixel is a background dot, if not; Confirm that then said pixel is the foreground point, the set of said foreground point is the real time kinematics target.

In the said apparatus, confirm that according to the said gray values of pixel points, Grad and the preset parameter that obtain the process of pixel point value can comprise in the pixel acquisition module 2:

Pixel point value X _Ij=(λ A _Ij, (1-λ) M _Ij), wherein, A _IjAnd M _IjBe respectively said gray values of pixel points and Grad, wherein i is a horizontal ordinate, and j is an ordinate, and λ is a preset parameter.

Set up mixed Gauss model according to above-mentioned data:

$P\left(X_t\right)=\underset{k}{\mathrm{\&Sigma;}}{w}_k\&times;\frac{1}{{\left(2\mathrm{\&pi;}\right)}^{n/2}{\left|{\mathrm{\&Sigma;}}_k\right|}^{1/2}}{e}^{-\frac{1}{2}{(X_{\mathrm{ij}}-{\mathrm{\&mu;}}_k)}^T{{\mathrm{\&Sigma;}}_k}^{-1}(X_{\mathrm{ij}}-{\mathrm{\&mu;}}_k)},$

∑ wherein _kIt is the standard covariance sigma _kMultiply each other with unit matrix and to obtain (X _Ij-μ _k) ^TExpression (X _Ij-μ _k) transposition, the weight table of color k is shown w in the two field picture _k, average is expressed as μ _kAnd covariance is expressed as σ _k ²

In the said apparatus, the first dynamic object determination module 3 can also comprise:

The first covariance update module, be used for confirming that said pixel is background dot after, upgrade the covariance of color in the current frame image, detailed process can comprise:

Said covariance sigma _T+1 ²=(1-ρ) σ _t ²+ ρ (X _T+1-μ _T+1) ^T(X _T+1-μ _T+1), wherein α is a predetermined value, ρ=α/w _T+1, w _T+1And μ _T+1Be respectively weight and the average of color k in the current frame image.

The second covariance update module, be used for confirming that said pixel is the foreground point after, upgrade the standard covariance sigma of color k in the current frame image _T+1' the value that equals to be scheduled to.

Based on the foreground point of confirming in the former frame image foreground point of confirming in the present frame is adjusted, as shown in Figure 2, in the said apparatus, can also be comprised:

Extraction module 4 is used for foreground point and the background dot confirmed according to the former frame image first dynamic object determination module 3, extracts in the former frame image pixel point value X for the present frame of foreground point _T+1

Probability generation module 5 is used for the pixel point value X according to the said present frame of extraction module 4 acquisitions _T+1With the Y of pixel neighborhood of a point point described in the former frame image _IjThe pixel of confirming present frame is the probability of real foreground point

Wherein a is a positive number, and f and 1 is respectively the variable quantity of horizontal ordinate, and f and l are taken as integer; D gets odd number, the expression width neighborhood, and b is the distance between pixel described in the former frame image and the said pixel neighborhood of a point point; It can replace with Gauss's template of the symmetric matrix of a d * d; The element value of said symmetric matrix meets Gaussian distribution, can calculate with gaussian kernel function, and for example Gauss's template of n=3 is: $\left(\begin{array}{ccc}1& 2& 1\\ 2& 4& 2\\ 1& 2& 1\end{array}\right),$ If with the color of 256 grades gray-scale value remarked pixels, so

Can be one is 256 array in advance by the good size of calculating;

The second dynamic object determination module; Be used for determined probability of probability generation module and designated value are compared; If determined probability

is less than designated value; Then the foreground point of confirming in the said present frame is adjusted into background dot; If determined probability

is more than or equal to designated value; Then the foreground point of confirming is not adjusted; If scene is complicated; Then designated value can be chosen the value in [0.5,0.7], if scene is simple; Then designated value can be chosen the value in [0.3,0.5].

The method and apparatus that the embodiment of the invention provides; Suppress the influence of the unexpected variation of illumination through video image being converted into gray level image to testing result; And through extracting foreground point and the background dot that present frame is further confirmed in the foreground point of confirming in the former frame image; Some irregular background motion targets can be removed, thereby moving target can be detected more accurately.

The above; Be merely the preferable embodiment of the present invention, but protection scope of the present invention is not limited thereto, any technician who is familiar with the present technique field is in the technical scope that the present invention discloses; The variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (8)

1. the method for detecting real-time moving object under the DYNAMIC COMPLEX background is characterized in that, comprise,

From video, obtain two field picture, confirm the weight w of color k in the said two field picture _k, average μ _kAnd covariance sigma _k ², wherein k chooses 5 or 3, and obtains the gray level image of said two field picture for preset natural number;

Obtain gray values of pixel points and Grad in the said gray level image; Said gray values of pixel points, Grad and predefined parameter based on obtaining are confirmed the pixel point value, judge said pixel point value whether with the mixed Gauss model coupling of former frame image, if; Confirm that then said pixel is a background dot; If not, confirm that then said pixel is the foreground point, the set of said foreground point is the real time kinematics target;

Said gray values of pixel points, Grad and the preset parameter that said basis obtains confirms that the process of pixel point value comprises:

Pixel point value X _Ij=(λ A _Ij, (1-λ) M _Ij), wherein, A _IjAnd M _IjBe respectively said gray values of pixel points and Grad, wherein i is a horizontal ordinate, and j is an ordinate, and λ is a preset parameter,

Said mixed Gauss model is:

∑ wherein _kIt is the standard covariance sigma _kMultiply each other with unit matrix and to obtain (X _Ij-μ _k) ^TExpression (X _Ij-μ _k) transposition.

2. method according to claim 1 is characterized in that, confirm that said pixel is background dot after, upgrade the covariance of color in the current frame image, detailed process comprises:

Said covariance sigma _T+1 ²=(1-ρ) σ _t ²+ ρ (X _T+1-μ _T+1) ^T(X _T+1-μ _T+1), wherein α is a predetermined value, ρ=α/w _T+1, w _T+1And μ _T+1Be respectively the weight and the average of color in the current frame image, σ _t ²The covariance of color in the expression former frame image, X _T+1Expression current frame pixel point value.

3. method according to claim 1 is characterized in that, confirm that said pixel is the foreground point after, upgrade the standard covariance sigma of color in the current frame image _T+1' the value that equals to be scheduled to.

4. according to each described method in the claim 1 to 3, it is characterized in that, comprise also according to the foreground point of confirming in the former frame image foreground point of confirming in the present frame is adjusted that concrete adjustment process comprises:

Extract in the former frame image is the pixel point value X of the present frame of foreground point _T+1', according to the pixel point value X of the said present frame that extracts _T+1' with the former frame image described in prospect neighborhood of a point point Y _IjThe pixel of confirming present frame is the probability of real foreground point:

wherein a is positive number; F and 1 is respectively the variable quantity of horizontal ordinate; D is a width neighborhood; B is the distance between foreground point described in the former frame image and the said prospect neighborhood of a point point; If determined probability

is less than designated value; Then the foreground point of confirming in the said present frame is adjusted into background dot; If determined probability more than or equal to designated value, is not then adjusted the foreground point of confirming.

5. the real time kinematics object detecting device under the DYNAMIC COMPLEX background is characterized in that, comprising:

The covariance acquisition module is used for obtaining two field picture from video, confirms the weight w of color k in the said two field picture _k, average μ _kAnd covariance sigma _k ², wherein k chooses 5 or 3 for preset natural number;

The pixel acquisition module; Be used for converting two field picture described in the covariance acquisition module into gray level image; And obtain gray values of pixel points and Grad in the said gray level image, and confirm the pixel point value according to the said gray values of pixel points, Grad and the preset parameter that obtain; Said gray values of pixel points, Grad and the preset parameter that said basis obtains confirms that the process of pixel point value comprises:

Pixel point value X _Ij=(λ A _Ij, (1-λ) M _Ij), wherein, A _IjAnd M _IjBe respectively said gray values of pixel points and Grad, wherein i is a horizontal ordinate, and j is an ordinate, and λ is a preset parameter,

Said mixed Gauss model is:

∑ wherein _kIt is the standard covariance sigma _kMultiply each other with unit matrix and to obtain (X _Ij-μ _k) ^TExpression (X _Ij-μ _k) transposition;

The first dynamic object determination module; Be used to judge the determined said pixel point value of pixel acquisition module whether with the mixed Gauss model coupling of former frame image; If confirm that then said pixel is a background dot, if not; Confirm that then said pixel is the foreground point, the set of said foreground point is the real time kinematics target.

6. device according to claim 5; It is characterized in that the said first dynamic object determination module also comprises: the first covariance update module, be used for confirming that said pixel is background dot after; Upgrade the covariance of color in the current frame image, detailed process comprises:

Said covariance sigma _T+1 ²=(1-ρ) σ _t ²+ ρ (X _T+1-μ _T+1) ^T(X _T+1-μ _T+1), wherein α is a predetermined value, ρ=α/w _T+1, w _T+1And μ _T+1Be respectively the weight and the average of color in the current frame image, σ _t ²The covariance of color in the expression former frame image, X _T+1Expression current frame pixel point value.

7. device according to claim 5 is characterized in that, the said first dynamic object determination module also comprises: the second covariance update module, be used for confirming that said pixel is the foreground point after, upgrade the standard covariance sigma of color in the current frame image _T+1' the value that equals to be scheduled to.

8. according to each described device in the claim 5 to 7, it is characterized in that, also comprise

Extraction module is used for the foreground point definite according to the former frame image first dynamic object determination module, and extracting in the former frame image is the pixel point value X of the present frame of foreground point _T+1';

The probability generation module, the pixel point value X of the said present frame that is used for obtaining according to extraction module _T+1' with the former frame image described in prospect neighborhood of a point point Y _IjThe pixel of confirming present frame is the probability of real foreground point

Wherein a is a positive number, and f and 1 is respectively the variable quantity of horizontal ordinate, and d is a width neighborhood, and b is the distance between foreground point described in the former frame image and the said prospect neighborhood of a point point;

The second dynamic object determination module; Be used for determined probability of probability generation module

and designated value are compared; If determined probability

is less than designated value; Then the foreground point of confirming in the said present frame is adjusted into background dot; If determined probability more than or equal to designated value, is not then adjusted the foreground point of confirming.

Images