CN101231756A - Method and apparatus for detecting moving goal shade - Google Patents

Abstract

The invention discloses a moving target shadow detection method, which comprises the following steps of: A1. detecting the moving target according to video images to determine the moving target area; B1. determining whether the shadow exists in the moving target area and turning to the following steps if the shadow exists in the area; C1. obtaining the histogram of the number of the pixels of the moving target in the vertical direction; D1. calculating an optimal threshold used for dividing the histogram into two types of areas; E1. determining the shadow area. By adopting the invention, the shadow is detected without being influenced by weather. In addition, the invention can be adapt to various complex backgrounds and targets, and can detect the shadow of the moving target accurately.

Description

Moving target shadow detection method and device

[technical field]

The present invention relates to motion target detection and processing, relate in particular to when moving target has shade method and device shadow Detection.

[background technology]

Behavioural analysis is by intellectual analysis and reasoning from logic to moving target, realize such as article leave over, article lost, illegal invasion, vehicle are driven in the wrong direction, vehicle flowrate, parking offense, vehicle make a dash across the red light, the gate inhibition drives in the wrong direction detection, demographics, trail specific function such as warning.

In the urban road monitor system, the CCD camera becomes the preferred unit in this system because of its good cost performance and intuitive image.By video image can drive in the wrong direction to vehicle, parking offense, specific behavior such as lane change positions and reports to the police in violation of rules and regulations.But video image is subjected to the influence of light source bigger, when light source is subjected to blocking of opaque article on incident direction, can produce shade.In the road monitoring system, target adhesion such as the shade of target and automobile, people and be taken as the part of target, thus influence the form parameters such as length, width of target, have a strong impact on the identification of targets ability, disturbed the succeeding target classification and followed the tracks of.As shown in Figure 1, hatched example areas is represented vehicle, oblique domain representation shadow region, grid zone, and two cars is owing to the shade reason adheres to together, and influence might cause omission or erroneous judgement to the correct identification of target; Perhaps vehicle does not enter the zone of forbidding, but shade has entered the zone of forbidding, if not with shadow removal, then may cause false alarm.Therefore target and its shade are separated into the problem that must solve in the urban road monitor system.

Existing document is studied shadow Detection, mainly is divided into based on color characteristic, based on brightness with based on the method for model.Utilize the difference of shade and color of object to cut apart shade based on the method for color characteristic.U.S. CVRR is breadboard to have done careful analysis to the feature of target, shade and the relation of target and shade, and picture information is transformed into more rich H SV space of colouring information, by suitable selection parameter, the shadow pixels point can well be detected.In rgb space, adopt the multiframe picture information, think that shadow spots satisfies mixed Gauss model, stronger adaptivity is arranged.Method based on color characteristic, can well processing target and shadow color situation about differing greatly, but the situation similar to shadow color to target, this method is just powerless, and because vehicle window and shadow color have very big similarity, this method is difficult to overcome the interference of vehicle window.Method based on brightness is linear with the brightness ratio of corresponding background dot according to shadow spots, and the consistent feature of the color of shadow region and background can well detect shade and remove shade.But these two kinds of methods all have very strong dependence to the color of target and the brightness of environment.3D shape and light source attribute based on the method hypothetical target of model are known, and the shape of shade and position can accurately be calculated by model, but this is not easy to realize in monitoring in real time.

Particularly, require Detection for Moving Target to have robustness, so-called robustness is meant that algorithm can adapt to complicated background, can adapt to various atrocious weather environment, can adapt to interference, 24 hours round-the-clock operate as normal such as shade, inverted image.

[summary of the invention]

Fundamental purpose of the present invention solves the problems of the prior art exactly, and a kind of moving target shadow detection method and device are provided, and can adapt to complicated background and various weather environment, and the shade to moving target detects more accurately.

For achieving the above object, the invention provides a kind of moving target shadow detection method, may further comprise the steps:

A1, detect moving target, determine motion target area according to video image;

B1, judge in motion target area, whether to have shade to exist, if having shade to exist then carry out following steps;

C1, draw the histogram of moving target number of pixels in vertical direction;

D1, calculate the optimal threshold that is used for histogram is divided into two class zones;

E1, determine the shadow region.

Further comprising the steps of in step e 1: calculate the mean value of the number of pixels in two class zones, one side average less be the orientation at shade place; Orientation and optimal threshold according to the shade place are determined the shadow region.

Further comprising the steps of after step e 1:

F1, the shadow region is removed, draw real motion target area.

For achieving the above object, the present invention also provides a kind of moving target shadow Detection device, comprising:

The moving object detection unit is used for detecting moving target according to video image, and determines motion target area;

The shade judging unit is used for judging at motion target area whether have shade to exist;

Histogram is set up the unit, is used for drawing when the shade judgment unit judges has shade to exist the histogram of moving target number of pixels in vertical direction;

The optimal threshold computing unit is used to calculate the optimal threshold that is used for histogram is divided into two class zones;

The shade determining unit is used to determine the shadow region.

Described shade determining unit comprises: subelement is determined in the shade orientation, is used to calculate the mean value of the number of pixels in two class zones, one side average less be the orientation at shade place; Subelement is determined in the shadow region, is used for determining the shadow region according to the orientation and the optimal threshold at shade place.

In the preferred version, described device also comprises: the shadow removal unit, and it is used for the shadow region is removed, and draws real motion target area; Described shadow removal unit is when the left side of shade is positioned at real moving target, the limit, left side that changes motion target area promptly obtains real motion target area, and the limit, left side that the limit, left side of real motion target area equals motion target area adds optimal threshold; Described shadow removal unit is when the right side of shade is positioned at real moving target, and the right edge that changes motion target area promptly obtains real motion target area, and the right edge of real motion target area equals optimal threshold.

The invention has the beneficial effects as follows: because under normal photographing, object always stands vertically in image, and shadows of objects only may be in left side, right side or rear side, the front side of object, and is angled with vertical direction.So moving target number of pixels in vertical direction in the statistical picture, the number of pixels of object is necessarily more than the number of pixels of shadow, therefore the present invention adopts histogrammic mode to detect shade judging on the basis that has shade, can not be subjected to the influence of weather, simultaneously also various complicated background and all types of target can be adapted to, the shade of moving target can be detected exactly.

[description of drawings]

Fig. 1 is the moving object detection synoptic diagram of band shade;

Fig. 2 is the shade orientation synoptic diagram of target;

Fig. 3 is the process flow diagram of an embodiment of the present invention;

Fig. 4 is the moving object detection synoptic diagram of an embodiment of the present invention;

Fig. 5 is the histogram of the pixel of an embodiment of the present invention;

Fig. 6 is the frame assumption diagram of a kind of embodiment of the present invention.

[embodiment]

Feature of the present invention and advantage will be elaborated in conjunction with the accompanying drawings by embodiment.

Moving target is a kind of name of broad sense among the present invention, promptly is different from the foreground object of background.

Please refer to Fig. 2, different the relative position of shade and target has a lot of difference constantly, but shade always may be summarized to be upper left, left, lower-left, preceding, bottom right, the right side, upper right, back 8 kinds with respect to target direction.Shade can be divided into left and right two big classes and just preceding, just back two kinds of specific types from the position.As Fig. 2, label 1 is a target, and label 2 is a shade.To the road of specific road section, camera faces road and installs, and the possible position that shade occurs in a day just can be determined.

Present embodiment may further comprise the steps for the detection of shade as shown in Figure 3:

At step S2, detect moving target.Motion target detection is the basis that video intelligent is analyzed, and bottom target detection result's quality directly has influence on intellectual analysis such as behavioural analysis and identification are carried out in the upper strata to target accuracy.Object detection method commonly used mainly contains background subtraction method, frame difference method, optical flow method etc.Present embodiment adopts the method for setting up background model to extract target, and introduces effective tracking of multiple prediction and track algorithm realization target.Substantially comprise following step:

1.1 mixed Gauss model

At any time, for specific pixel point (x ₀, y ₀) video sequence I (x, y t) can regard a time series as

{X ₁，X ₂，…，X _t}＝{I(x ₀，y ₀，i)：1≤i≤t}

And this time series can be expressed as the stack of K Gaussian distribution, and promptly mixed Gaussian distributes.The probability distribution of its current point is expressed as

$p\left(X_t\right)=\underset{i=1}{\overset{K}{\mathrm{\Σ}}}{\mathrm{\ω}}_{i,t}\mathrm{\η}(X_t,{\mathrm{\μ}}_{i,t},{\mathrm{\Σ}}_{\text{i,t}})$

In the formula, K is the Gaussian function number, ω _{I, t}Be the weight coefficient of corresponding Gaussian function, μ _{I, t}Be the mathematical expectation of i Gauss model, ∑ _{I, t}Be the covariance matrix of i Gauss model, η is a Gauss model

$\mathrm{\&eta;}(x_t,\mathrm{\&mu;},\mathrm{\&Sigma;})=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^{\mathrm{\&mu;}./2}{\left|\mathrm{\&Sigma;}\right|}^{1/2}}{\mathrm{<figure-callout\; id="1"\; label="e"\; filenames="S2008100652565E00011.png"\; state="\{\{state\}\}">e</figure-callout>}}^{\frac{1}{2}{(x_t-{\mathrm{\&mu;}}_t)}^T{\mathrm{\&Sigma;}}^{-1}(x_t-{\mathrm{\&mu;}}_t)}$

In order to simplify calculating, get K=3, and suppose that red, blue, green triple channel is separate and equation is identical, promptly covariance matrix is a diagonal matrix, and ${\mathrm{\&Sigma;}}_{k,l}={\mathrm{\&sigma;}}_k^{2}I,$ Wherein mixed Gauss model can obtain according to the EM algorithm.

1.2 background estimating

In general, the component of background parts in mixed Gauss model is that the main part and the variance of model is less, and the parameter of mixed Gauss model changes along with the variation of current frame pixel simultaneously.In order to obtain the effective constituent of background in the Gauss model.Stauffer is with ω _k/ σ _kOrdering distributes to obtain β, and promptly this ratio coefficient is big more, and gaussian component proportion in mixed Gauss model is just big more, and β can be represented by the formula

$\mathrm{\&beta;}=\underset{b}{\arg \min }(\underset{j=1}{\overset{b}{\mathrm{\&Sigma;}}}{\mathrm{\&omega;}}_j>T),$ In the formula: T is a threshold value.

When β＜T, its corresponding gaussian component is a background.

1.3 model modification

The key that background is obtained is kept for effectively carrying out background, i.e. context update.Whether each pixel need mate and determine and upgrade with K Gaussian distribution, if current pixel during with a certain gaussian component coupling wherein, upgrades according to following formula.

${\mathrm{UB}}_t=\left\{\begin{array}{c}{\mathrm{\&omega;}}_{i,t}=(1-\mathrm{\&alpha;}){\mathrm{\&omega;}}_{i,t-1}+{\mathrm{\&alpha;M}}_{i,t}\\ {\mathrm{\&mu;}}_t=(1-\mathrm{\&rho;}){\mathrm{\&mu;}}_{t-1}+{\mathrm{\&rho;X}}_t\\ {\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="S2008100652565E00011.png"\; state="\{\{state\}\}">t</figure-callout>}}^2=(1-\mathrm{\&rho;}){\mathrm{\&sigma;}}_{t-1}^2+\mathrm{\&rho;}{(X_t-{\mathrm{\&mu;}}_t)}^T(X_t-{\mathrm{\&mu;}}_t)\end{array}\right.$

If there is not the gaussian component of coupling in the mixture model, then regenerate the component that a new Gaussian distribution replaces β minimum in the mixed Gauss model.Therefore, the renewal process of background can be expressed as in the mixed Gauss model (GMM):

${\mathrm{SB}}_t=\left\{\begin{array}{ccc}{\mathrm{UB}}_t,& \mathrm{if}& I(x,y,t)\&Element;\mathrm{GMM}\\ G(\mathrm{\&mu;},\mathrm{\&sigma;},\mathrm{\&omega;}),& \mathrm{if}& I(x,y,t)\&NotElement;\mathrm{GMM}\end{array}\right.$

Prospect can be expressed as:

F(x，y，t)＝I(x，y，t)-B(x，y，t-1)

1.4 extract the target area

According to the t that obtains previously constantly prospect F (x, y, t), can obtain to represent the target area binary map BF (x, y, t)

$\mathrm{BF}(x,y,t)=\left\{\begin{array}{ccc}1& \mathrm{if}& F(x,y,t)>\mathrm{TF}\\ 0& \mathrm{if}& F(x,y,t)\&le;\mathrm{TF}\end{array}\right.$

Wherein, TF is an appropriate threshold.As shown in Figure 4, the target detection result for a certain two field picture represents background dot with white, and oblique line is partly represented the foreground point.

Can obtain the boundary rectangle of target area according to morphologic Boundary Extraction operator, obtain a plurality of target areas of this two field picture.If n target area arranged in the t time chart picture, is expressed as

{rect _i，i＝0，...n-1}

As shown in Figure 4, we come out target label with rectangle frame, thereby obtain the positional information of two targets in image, obtain motion target area.

At step S4, whether in motion target area have shade exist, if there is not shade to exist, then finish if judging, handle according to normal picture; If shade is arranged, execution in step S6 then.In this step, judged whether shade according to colouring information and monochrome information.In advance according to experiment, count the colouring information of shade and the threshold range of monochrome information, during judgement, each color of pixel and the brightness of motion target area are compared with the colouring information of setting in advance and the threshold range of monochrome information, if fall into this scope, the pixel that falls into the threshold range of colouring information and monochrome information simultaneously is linked to be the zone, then thinks to have shade.But might there be erroneous judgement in this judgement, for example the glass of automobile also may be judged into shade, therefore needs following detection further to guarantee the accuracy that detects.

At step S6, according to the judgement of shade, roughly detect the orientation of shade, then execution in step S8.

At step S8, moving target to the horizontal direction projection, is added up the number of pixels of moving target on each horizontal level, vertical direction, then execution in step S10.

At step S10, draw the histogram of horizontal level-vertical direction pixel number, execution in step S12 then according to statistics.

When the left and right sides of shade in target, because people or car all have certain height, and what take with CCD is the two-dimensional plane image, and therefore the object pixel number on the image vertical direction generally will be more than the number of pixels of shadow region.Therefore, motion target area is done projection on the horizontal direction, the number of pixels on the statistics vertical direction, the histogram of number of pixels on calculated level position-vertical direction promptly can be found out shade.

If in n the rectangular area, i rectangular area rect _i(border, upper and lower, left and right t) is respectively top, bottom, left, right for p, q.Then the width of rectangular area is designated as width=right-left, highly is designated as hight=bottom-top.

${\mathrm{rect}}_i(p,q,t)=\left\{\begin{array}{cc}1& \mathrm{BF}(p+\mathrm{top},q+\mathrm{bottom},t)==1\\ 0& \mathrm{BF}(p+\mathrm{top},q+\mathrm{bottom},t)==0\end{array}\right.,$ 0≤p＜hight wherein, 0≤q＜width

Use hist _i(p) expression rectangular area rect _i(p, q, horizontal level t)-vertical direction pixel number histogram.If rect _i(p, q, t) in, the pixel number of foreground point is designated as hist in the p row _i(p),

${\mathrm{hist}}_i\left(p\right)=\underset{q=0}{\overset{q=\mathrm{hight}-1}{\mathrm{\&Sigma;}}}{\mathrm{rect}}_i(p,q),p=0,...,\mathrm{width}-1$

At step S12, use existing histogram optimal threshold dividing method and calculate this histogrammic optimal threshold, try to achieve appropriate threshold, be made as Thresh, this threshold value is divided into two class zones with foreground area, as shown in Figure 5, execution in step S14 then.

At step S14, determine the shadow region according to the orientation and the optimal threshold of shade, and remove shade, draw real motion target area.

The determining of shade orientation in the prospect determined according to the mean value in two class zones, one side average less be shade.For solving the shade orientation problem that the direction variation changes along with sunshine in a day, system adds up a shade orientation and appears at the probability in target left side, thereby determine the next azimuth information of a period of time at set intervals.

The definite of shade orientation also can be according to determining among the step S6 in the prospect.

When known shadow is on the left of target, then change rectangular area rect _iThe limit, left side promptly obtain removing the target area of shade, be designated as newrect _i, promptly

newrect _i.left＝left+Thresh

And, change rect when shade during on the target right side _iRight edge promptly obtains removing the target area of shade, promptly

newrect _i.right＝Thresh

Thereby obtain real target area.

Wherein, step S6 also can save, and determines the orientation of shade by step S14.

According to the structural drawing of the moving target shadow Detection device of above method as shown in Figure 6, comprise that moving object detection unit, shade judging unit, histogram set up unit, optimal threshold computing unit, shade determining unit and shadow removal unit.The moving object detection unit is used for detecting moving target according to video image, and determine motion target area, whether the shade judging unit is used for judging at motion target area has shade to exist, histogram is set up the unit is used for drawing moving target number of pixels in vertical direction when the shade judgment unit judges has shade to exist histogram, the optimal threshold computing unit is used to calculate the optimal threshold that is used for histogram is divided into two class zones, and the shade determining unit is used to determine the shadow region.

Described histogram is set up the unit and is comprised that pixels statistics subelement and histogram set up subelement, the pixels statistics subelement is used to add up the number of pixels of moving target on each horizontal level, vertical direction, and histogram is set up the histogram that subelement is used to draw horizontal level-vertical direction pixel number.

Described shade determining unit comprises that the shade orientation determines that subelement and shadow region determine subelement, the shade orientation determines that subelement is used to calculate the mean value of the number of pixels in two class zones, the less one side of average is the orientation at shade place, and the shadow region determines that subelement is used for determining the shadow region according to the orientation and the optimal threshold at shade place.

The shadow removal unit is used for the shadow region is removed, and draws real motion target area; Described shadow removal unit is when the left side of shade is positioned at real moving target, the limit, left side that changes motion target area promptly obtains real motion target area, and the limit, left side that the limit, left side of real motion target area equals motion target area adds optimal threshold; Described shadow removal unit is when the right side of shade is positioned at real moving target, and the right edge that changes motion target area promptly obtains real motion target area, and the right edge of real motion target area equals optimal threshold.

Experimental result shows, the present invention can well separate target and shade, can not be subjected to the influence of weather to the detection of shade, simultaneously also can adapt to various complicated background and all types of target, and it is simple to operate, have stronger robustness and adaptivity, have good real-time performance simultaneously.

Above content be in conjunction with concrete preferred implementation to further describing that the present invention did, can not assert that concrete enforcement of the present invention is confined to these explanations.For the general technical staff of the technical field of the invention, without departing from the inventive concept of the premise, can also make some simple deduction or replace, all should be considered as belonging to protection scope of the present invention.

Claims (10)

1. moving target shadow detection method is characterized in that may further comprise the steps:

A1, detect moving target, determine motion target area according to video image;

B1, judge in motion target area, whether to have shade to exist, if having shade to exist then carry out following steps;

C1, draw the histogram of moving target number of pixels in vertical direction;

D1, calculate the optimal threshold that is used for histogram is divided into two class zones;

E1, determine the shadow region.

2. moving target shadow detection method as claimed in claim 1 is characterized in that: described step C1 may further comprise the steps:

C11, motion target area is carried out projection in the horizontal direction, the number of pixels of statistics moving target on each horizontal level, vertical direction;

C12, draw the histogram of horizontal level-vertical direction pixel number.

3. moving target shadow detection method as claimed in claim 1 is characterized in that: further comprising the steps of in step e 1: calculate the mean value of the number of pixels in two class zones, one side average less be the orientation at shade place; Orientation and optimal threshold according to the shade place are determined the shadow region.

4. moving target shadow detection method as claimed in claim 1 is characterized in that: judged whether that by each color of pixel and brightness in the detection motion target area shade exists among the described step B1.

5. as each described moving target shadow detection method in the claim 1 to 4, it is characterized in that: further comprising the steps of after step e 1:

F1, the shadow region is removed, draw real motion target area.

6. moving target shadow detection method as claimed in claim 5 is characterized in that: described step F 1 may further comprise the steps:

F11, when shade is positioned at the left side of real moving target, the limit, left side that changes motion target area promptly obtains real motion target area, the limit, left side that the limit, left side of real motion target area equals motion target area adds optimal threshold;

F12, when shade is positioned at the right side of real moving target, the right edge that changes motion target area promptly obtains real motion target area, the right edge of real motion target area equals optimal threshold.

7. moving target shadow Detection device is characterized in that comprising:

The moving object detection unit is used for detecting moving target according to video image, and determines motion target area;

The shade judging unit is used for judging at motion target area whether have shade to exist;

Histogram is set up the unit, is used for drawing when the shade judgment unit judges has shade to exist the histogram of moving target number of pixels in vertical direction;

The optimal threshold computing unit is used to calculate the optimal threshold that is used for histogram is divided into two class zones;

The shade determining unit is used to determine the shadow region.

8. moving target shadow Detection device as claimed in claim 7 is characterized in that: described histogram is set up the unit and is comprised:

The pixels statistics subelement is used to add up the number of pixels of moving target on each horizontal level, vertical direction;

Histogram is set up subelement, is used to draw the histogram of horizontal level-vertical direction pixel number.

9. moving target shadow Detection device as claimed in claim 7 is characterized in that: described shade determining unit comprises:

Subelement is determined in the shade orientation, is used to calculate the mean value of the number of pixels in two class zones, one side average less be the orientation at shade place;

Subelement is determined in the shadow region, is used for determining the shadow region according to the orientation and the optimal threshold at shade place.

10. as each described moving target shadow Detection device in the claim 7 to 9, it is characterized in that also comprising:

The shadow removal unit is used for the shadow region is removed, and draws real motion target area; Described shadow removal unit is when the left side of shade is positioned at real moving target, the limit, left side that changes motion target area promptly obtains real motion target area, and the limit, left side that the limit, left side of real motion target area equals motion target area adds optimal threshold; Described shadow removal unit is when the right side of shade is positioned at real moving target, and the right edge that changes motion target area promptly obtains real motion target area, and the right edge of real motion target area equals optimal threshold.

Images