CN102622763A - Method for detecting and eliminating shadow - Google Patents

Abstract

The invention relates to a method for detecting and eliminating shadow in monitored videos. The method includes setting up a background model; performing foreground/background segmentation for a current frame and obtaining an objective region; transforming an RGB (red, green, blue) color space of the objective region into an HSV (hue, saturation and value) color space; using three components including hue H, saturation S and value V as shadow detecting operators, and preliminarily eliminating parts of shadows; using gray levels of a background of the current frame as features and dividing the objective region into a plurality of feature sub-blocks; describing gray level distribution probability by a Gaussian model and determining objective sub-blocks; and eliminating isolated objective sub-blocks, filling discontinuous objective sub-blocks and finishing objective detection and shadow elimination. The shadow detecting and eliminating method excellently resolves the problem that a threshold value cannot be adaptively adjusted along environments, shadow false detecting rate is reduced, both shadow detection completeness and accuracy are good, cavities and breakages of a moving object are obviously reduced, and shadow detecting rate is improved.

Description

Shadow Detection and removing method

Technical field

The present invention relates to digital video and image processing field, particularly, relate to a kind of shadow Detection and removing method.

Background technology

The existence of shade has caused the degeneration of image in detection process of moving target; Follow-up graphical analysis and understanding are brought great interference; Therefore the detection of shade is the prerequisite of graphical analysis and understanding with eliminating, and the shadow Detection of moving target has important application in intelligent video monitoring.

Existing shadow Detection algorithm roughly can be divided three classes: based on the method for model, based on the method for texture with based on the method for shade attribute.These methods are used separately and are all had shortcoming separately: based on time of position, direction and the shade generation of the real-time detection light source of method needs of model etc.; But these conditions all be difficult to confirm and change with different application environments, and be difficult to satisfy the needs that detect in real time; Method based on texture is to compare analysis to image texture characteristic with the background texture characteristic, if the not obvious mistake that causes easily of some textural characteristics is surveyed; Usually need define the threshold value of some attributes based on the method for shade attribute, and these threshold values usually and environment certain relation is arranged, and alter a great deal and be difficult to confirm.

Summary of the invention

The object of the present invention is to provide a kind of shadow Detection and removing method based on color space and the associating of gray scale ratio; Adopt the hsv color model tentatively to judge the shadow region to the foreground target that extracts, through Gauss model gray scale is described the shade that accurately detects target than characteristic again.Method of the present invention has solved threshold value can not reduce the shade false drop rate with the problem of environment self-adaption adjustment, has improved shadow Detection integrality and accuracy.

According to a main aspect of the present invention, a kind of shadow Detection and removing method are provided, it may further comprise the steps:

(a) set up background model;

(b) whether satisfy background model according to current frame pixel, present frame is carried out foreground/background segmentation, obtain to have the target of shade thus;

(c) with this target area that has a shade by the RGB color space conversion to the hsv color space;

(d) with tone H, saturation degree S and the brightness V three-component in hsv color space as the shadow Detection operator, tentatively eliminate the part shade;

(e) be converted into gray scale and represent eliminating target area behind the part shade, and to liken to the gray scale of itself and background model respective pixel be characteristic, and this target area is divided into the sub-piece of a plurality of characteristics;

(f) with each characteristic sub-block as a sequence of random variables, and gray scale described with Gauss model than distribution probability, thereby obtains the average and the variance of the Gaussian distribution of each block of pixels, calculate average and the variance of each gray scale than characteristic sub-block; If the variance of Gauss model is greater than setting threshold, then this sub-piece is the sub-piece of target, otherwise is the sub-piece of shade;

(g) eliminate the isolated sub-piece of target, fill the sub-piece of discontinuous target, accomplish target detection and shade and eliminate.

According to an aspect of the present invention, in step (a), utilize the Density Estimator background modeling method that background is described, with sequence of frames of video according to the modeling of Density Estimator background model.

According to an aspect of the present invention, in step (a), sequence of frames of video comprises the 30-50 frame.

According to an aspect of the present invention, in step (a), constantly upgrade the Density Estimator background model, and the prospect part is not participated in the renewal of background model, thereby obtain background model accurately.

According to an aspect of the present invention, in step (d), utilize formula tentatively to eliminate the part shade:

In the formula, I representes present frame, and B representes background frames, and α, β are the threshold value of brightness ratio, T ^HAnd T ^SBe respectively tone threshold value and saturation degree threshold value.

According to an aspect of the present invention, in step (e), this target area is divided into m * n the sub-piece of characteristic; And the size of each sub-piece is 12 * 12, if certain sub-block size is then cast out this sub-piece less than 12 * 12; Perhaps replenish the row or the row of this sub-piece, satisfy 12 * 12 up to its size.

According to an aspect of the present invention, in step (f), calculate each gray scale and be than the average of the sub-piece of characteristic and the formula of variance:

$E{\left(\mathrm{gray}\right)}_i=\frac{1}{T_{D_i}}\underset{p(x,y)\∈D_i}{\mathrm{\Σ}}\frac{\mathrm{gray}-{\mathrm{gray}}_{\mathrm{shadow}}}{\mathrm{gray}}$

$D{\left(\mathrm{gray}\right)}_i=\underset{p(x,y)\∈D_i}{\mathrm{\Σ}}{[\frac{\mathrm{gray}-{\mathrm{gray}}_{\mathrm{shadow}}}{\mathrm{gray}}-E\left(\mathrm{gray}\right)]}^2$

D wherein _iBe the i block of pixels, E (gray) _i, D (gray) _iBe respectively the average and the variance of i block of pixels.

Based on one aspect of the present invention, in step (g), utilize morphological operator to eliminate the isolated sub-piece of target.

Will be appreciated that the characteristic in the above each side of the present invention is independent assortment within the scope of the invention, and do not receive the restriction of its order---as long as the technical scheme after the combination drops in the connotation of the present invention.

Description of drawings

In order to be illustrated more clearly in the technical scheme among the present invention, will do to introduce simply to accompanying drawing of the present invention below, wherein:

Fig. 1 has shown the process flow diagram of the inventive method;

Embodiment

Hereinafter will combine the preferred embodiments of the present invention that technical scheme of the present invention is elaborated.

Need to understand that the description of hereinafter (comprising accompanying drawing) only is exemplary, but not the description of limitation of the present invention property.Can relate to the concrete quantity of parts in the following description, yet also need should be appreciated that, these quantity also only are exemplary, and those skilled in the art can choose the parts of right quantity with reference to the present invention arbitrarily.And,, only make the difference name of parts and be referred to as to use in the present invention if wordings such as mentioned " first ", " second " are not the ordering of expression to parts importance.

Referring to Fig. 1, the flow process of the inventive method has been described wherein.

In one embodiment of the invention, comprise that key step is following:

(a) establish background model, so that from present frame, be partitioned into foreground target.Cuclear density background modeling method according to present embodiment adopted obtains background image, and Density Estimator is estimated unknown Density Distribution through the local function that the weighted mean central point is positioned at sampled value.A sample set S={x in given certain pixel characteristic space ₁, x ₂..., x _N, observed reading x _tDensity Distribution with p (x _t) estimate:

$p\left(x_t\right)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_i{K}_h(x_t-x_i)---\left(1\right)$

Weights α in the formula (1) _i=1/N; K _h=(1/h) K (t/h) is a kernel function; H is the width of kernel function.A pixel is considered m characteristic, kernel function K _hSelect Gaussian function, σ _mThe kernel function bandwidth of representing each characteristic, formula (1) can further be expressed as:

$p\left(x_t\right)={\mathrm{\α}}_i\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{m=1}{\overset{d}{\mathrm{\Π}}}\frac{1}{\sqrt{{2\mathrm{\π\σ}}_m^2}}\exp (-\frac{1}{2}[\frac{{(x_{\mathrm{tm}}-x_{\mathrm{im}})}^2}{{\mathrm{\σ}}_m^2}\left]\right)---\left(2\right)$

Suppose x _iNormal Distribution N (μ, σ ²), (x then _i-x _I+1) Normal Distribution N (0,2 σ ²), m is the absolute value intermediate value of adjacent two frame pixel characteristic value differences, promptly m=median (| x _i-x _I+1|), obtain by the symmetry and the MEAN VALUE PROPERTY of normal distribution:

P[N(0，2σ ²＞m]＝0.25 (3)

Know by normal distribution:

$\mathrm{\σ}=\frac{m}{0.68\sqrt{2}}$

In other embodiments of the invention,, calculate the time of the intermediate value of sample frame sequence pixel,, also can the intermediate value of sample sequence be used mean approximation in order to improve real-time with labor if it is longer to be used for carrying out the sample frame sequence of Density Estimator.

(b) with prospect and background segment such as moving targets.If certain pixel of current frame image does not satisfy the background model that formula (2) is described, then this pixel is foreground target pixel (possibly contain shade), otherwise is background pixel.Accomplish the target area that the back obtains to have shade of cutting apart according to this to present frame;

(c) color space conversion, the foreground area after step (b) cut apart from the RGB color space conversion to the hsv color space.Formula is following:

$H\left\{\begin{array}{ccc}\mathrm{\&theta;}& \mathrm{if}& B<G\\ 360-\mathrm{\&theta;}& \mathrm{if}& B\&GreaterEqual;G\end{array}\right.---\left(4\right)$

$S=1-\frac{3\min (R,G,B)}{(R+G+B)}---\left(5\right)$

$V=\frac{1}{3}(R+G+B)---\left(6\right)$

Wherein $\mathrm{\&theta;}={\mathrm{Cos}}^{-1}\left\{\frac{[(R-G)+(R-B)]/2}{{[{(R-G)}^2+(R-B)(G-B)}^{1/2}}\right\}$

(d) carrying out the hsv color model detects with preliminary elimination dash area., as the shadow Detection operator each pixel of image is detected according to tone H, saturation degree S and the brightness V in hsv color space.

(i) because background is reduced by the pixel intensity of shade occlusion area,, compare the V color component of the component of the V color of each pixel in the target image that has shade that obtains and same position background pixel so can reduce in the V in hsv color space channel value.

$\mathrm{\&alpha;}\&le;\frac{I^V(x,y)}{B^V(x,y)}\&le;\mathrm{\&beta;}---\left(7\right)$

In the formula, I representes present frame, and B representes background frames, and α, β are the threshold value of brightness ratio, and they are all less than 1, and α reflects intensity of illumination, and β then is for fear of being shade with the background disturbance treatment, reduces the shadow Detection scope through formula (7).

(ii),, in the hsv color space, the image utilization shadow Detection operator that obtains in (i) is detected at H, S passage so must further handle because the brightness of object pixel also maybe be lower than the brightness of background pixel.

|I ^H(x，y)-B ^H(x，y)|≤T ^H (8)

|I ^S(x，y)-B ^S(x，y)|≤T ^S (9)

In the formula, I representes present frame, and B representes background frames, T ^HAnd T ^SBe the threshold value of color harmony saturation degree, its interval is (0.01,0.02), and this expression color component can have little variation, is the pixel assignment that satisfies following formula simultaneously 0, tentatively eliminates dash area.

(e) carrying out gray scale detects than model.Under certain brightness conditions, same object is in the shadow region with not in the shadow region, and its tone is almost constant, and the gray scale and the gray scale of the original that promptly are positioned at the object of shadow region have differed a coefficient k, and following mathematic(al) representation is then arranged:

$\mathrm{SG}(x,y)=\frac{\mathrm{gray}-{\mathrm{gray}}_{\mathrm{shadow}}}{\mathrm{gray}}=\frac{\mathrm{gray}-k\&times;\mathrm{gray}}{\mathrm{gray}}=1-k---\left(10\right)$

With the preliminary area dividing of eliminating dash area in the step (d) is m * n the sub-piece of characteristic, and the size of each sub-piece is 12 * 12.If certain sub-block size is then cast out this sub-piece less than 12 * 12.In other embodiments, also can replenish the row or the row of this sub-piece, satisfy 12 * 12 up to its size.

(f) utilize Gauss model to judge the sub-piece of target.The gray scale of utilizing Gauss model to describe the sub-piece of characteristic compares characteristic; Thereby obtain the average and the variance of the sub-piece Gaussian distribution of each characteristic; The variance of gray scale ratio that its respective pixel piece is described if Gaussian distribution has precipitous crest is very little, otherwise explains that then the variance of corresponding blocks gray scale ratio is bigger.When the variance of certain sub-piece during less than threshold value, this sub-piece is a shade, otherwise then is foreground target.It is that robustness has been improved in shadow region or target area that the piecemeal in suspected target zone is handled differentiation, and each gray scale is calculated through formula than the average and the variance of characteristic block:

$E{\left(\mathrm{gray}\right)}_i=\frac{1}{T_{D_i}}\underset{p(x,y)\&Element;D_i}{\mathrm{\&Sigma;}}\frac{\mathrm{gray}-{\mathrm{gray}}_{\mathrm{shadow}}}{\mathrm{gray}}---\left(11\right)$

$D{\left(\mathrm{gray}\right)}_i=\underset{p(x,y)\&Element;D_i}{\mathrm{\&Sigma;}}{[\frac{\mathrm{gray}-{\mathrm{gray}}_{\mathrm{shadow}}}{\mathrm{gray}}-E\left(\mathrm{gray}\right)]}^2---\left(12\right)$

D wherein _iBe the i block of pixels, E (gray) _i, D (gray) _iBe respectively the average and the variance of i block of pixels.

(g) utilize morphological operator, earlier image is carried out " expansion " computing, and then carry out " corrosion " computing; In order to eliminate " cavity " that possibly exist in institute's acquisition prospect; Thereby eliminate the isolated sub-piece of target, fill the sub-piece of discontinuous target, accomplish target detection and shade and eliminate.

Above basis has preferred embodiment been done detailed description to the present invention; But it will be appreciated that; Scope of the present invention is not limited to these concrete embodiments, but comprises that those skilled in the art are according to any change and the change that openly can make of the present invention.

Claims (8)

1. shadow Detection and removing method, it may further comprise the steps:

(a) set up background model;

(b) whether satisfy background model according to current frame pixel, present frame is carried out foreground/background segmentation, obtain to have the target of shade thus;

(c) with this target area that has a shade by the RGB color space conversion to the hsv color space;

(d) with tone H, saturation degree S and the brightness V three-component in hsv color space as the shadow Detection operator, tentatively eliminate the part shade;

(e) be converted into gray scale and represent eliminating target area behind the part shade, and to liken to the gray scale of itself and background model respective pixel be characteristic, and this target area is divided into the sub-piece of a plurality of characteristics;

(f) with each characteristic sub-block as a sequence of random variables, and gray scale described with Gauss model than distribution probability, thereby obtains the average and the variance of the Gaussian distribution of each block of pixels, calculate average and the variance of each gray scale than characteristic sub-block; If the variance of Gauss model is greater than setting threshold, then this sub-piece is the sub-piece of target, otherwise is the sub-piece of shade;

(g) eliminate the isolated sub-piece of target, fill the sub-piece of discontinuous target, accomplish target detection and shade and eliminate.

2. shadow Detection according to claim 1 and removing method is characterized in that, in step (a), utilize the Density Estimator background modeling method that background is described, with sequence of frames of video according to the modeling of Density Estimator background model.

3. shadow Detection according to claim 2 and removing method is characterized in that, in step (a), sequence of frames of video comprises the 30-50 frame.

4. shadow Detection according to claim 2 and removing method is characterized in that, in step (a), constantly upgrade the Density Estimator background model, and the prospect part is not participated in the renewal of background model, thereby obtain background model accurately.

5. shadow Detection according to claim 1 and removing method is characterized in that, in step (d), utilize formula tentatively to eliminate the part shade:

In the formula, I representes present frame, and B representes background frames, and α, β are the threshold value of brightness ratio, T ^HAnd T ^SBe respectively tone threshold value and saturation degree threshold value.

6. shadow Detection according to claim 1 and removing method is characterized in that, in step (e); This target area is divided into m * n the sub-piece of characteristic; And the size of each sub-piece is 12 * 12, if certain sub-block size is then cast out this sub-piece less than 12 * 12; Perhaps replenish the row or the row of this sub-piece, satisfy 12 * 12 up to its size.

7. shadow Detection according to claim 1 and removing method is characterized in that, in step (f), calculate each gray scale and than the average of the sub-piece of characteristic and the formula of variance are:

$E{\left(\mathrm{gray}\right)}_i=\frac{1}{T_{D_i}}\underset{p(x,y)\&Element;D_i}{\mathrm{\&Sigma;}}\frac{\mathrm{gray}-{\mathrm{gray}}_{\mathrm{shadow}}}{\mathrm{gray}}$

$D{\left(\mathrm{gray}\right)}_i=\underset{p(x,y)\&Element;D_i}{\mathrm{\&Sigma;}}{[\frac{\mathrm{gray}-{\mathrm{gray}}_{\mathrm{shadow}}}{\mathrm{gray}}-E\left(\mathrm{gray}\right)]}^2$

D wherein _iBe the i block of pixels, E (gray) _i, D (gray) _iBe respectively the average and the variance of i block of pixels.

8. shadow Detection according to claim 1 and removing method is characterized in that, in step (g), utilize morphological operator to eliminate the isolated sub-piece of target.