CN103839232A - Pedestrian shadow restraining method based on block mass model - Google Patents

Abstract

The invention provides a pedestrian shadow restraining method based on a block mass model. The method comprises the steps of firstly, constructing a pedestrian block mass model, obtaining a preliminary shaded area by calculating the torque feature of a block mass and the perpendicular column diagram of the block mass and by means of a geometrical method, then taking the gray level, the direction and the central position of the shadow area as parameters to conduct shadow modeling on an overall pedestrian and the shadow area, and trimming the shadow area obtained by preliminarily segmentation. The method can detect and restrain shadows of pedestrians in different directions and of different numbers, the video processing speed is about 7.5 frame/second, and a motion target and the shadow of the motion target can be efficiently and accurately separated out.

Description

A kind of pedestrian's shade inhibition method based on agglomerate model

Technical field

The invention belongs to image processing, video monitoring and traffic safety technology field, specifically refer to a kind of pedestrian's shade inhibition method based on agglomerate model.

Background technology

Video Supervision Technique is exactly the erroneous judgement in order to prevent Security Personnel, utilizes automatic analysis technology to carry out the technology of video monitoring.Current research method all supposes not exist in video sequence shade, and in well-lighted scene, and mobile shade is by by the wrong foreground object that is divided into.This will cause error and the difficulty of the subsequent treatment such as target location estimation, goal behavior analysis and target identification.

At present, the method of moving shadow detection and inhibition is roughly divided into following three classes: based on the method for color model, select a suitable color space, utilize the chromatic characteristic in color space of shadows pixels value to carry out shade inhibition, such as HSV space, color character invariant C1C2C3 space and normalization rgb space, but these methods are easily affected by noise and to light intensity sensitive; Based on the method for physical model, the specific appearance feature of the Method Modeling by physics or study shadows pixels, such as two light source dichromatic reflection model BIDR, but it cannot the processing target tone situation identical with background; Based on the method for texture model, obtain candidate region by shade spectral signature, come differentiation prospect and shade according to the correlativity of these zone-textures again, such as normalized crosscorrelation, Gabor filtering, orthogonal transformations etc., owing to will multiple neighborhoods of a pixel being calculated, therefore operand is larger, real-time is poor.

Therefore, how efficiently and accurately moving target and its shade to be separated, become the hot issue of current research.

Summary of the invention

Technical matters to be solved by this invention is in order to overcome the deficiencies in the prior art, proposes a kind of pedestrian's shade inhibition method based on agglomerate model.First the inventive method builds pedestrian's agglomerate model, then by calculating moment characteristics and the agglomerate vertical histogram of agglomerate, obtains preliminary shadow region by the method for how much.Then, the gray scale of this shadow region, direction, center are carried out to shade modeling as parameter to whole pedestrian and shadow region, former shadow region is pruned, make algorithm herein can difference be detected and be suppressed towards, varying number pedestrian's shade; The video processing speed of the inventive method is about 7.5 frames/s.

In order to solve the problems of the technologies described above, the technical solution adopted in the present invention is: a kind of pedestrian's shade inhibition method based on agglomerate model, comprises the steps:

Steps A, builds pedestrian's agglomerate, and each pedestrian is represented with corresponding agglomerate, and its concrete steps are as follows:

Steps A-1, obtains current frame image, utilizes agglomerate extracting method, and the scene color of current frame image is carried out to cluster, obtains image agglomerate;

Steps A-2, according to described image agglomerate, utilize mixed Gaussian background modeling method, obtain the prospect masterplate of current frame image, obtain foreground moving agglomerate;

Steps A-3, utilize fuzzy clustering method to merge foreground moving agglomerate, obtain pedestrian's agglomerate;

Step B, for pedestrian's agglomerate, utilizes shadow detection method that pedestrian and its shade are tentatively cut apart, and obtains the shadow region of pedestrian's agglomerate:

Step B-1, calculates the center square of pedestrian agglomerate, so the shade that obtains pedestrian's agglomerate towards;

Step B-2, calculates pedestrian agglomerate vertical histogram, obtains the cut-point of pedestrian and shade in pedestrian's agglomerate;

Step B-3, according to the shade of pedestrian's agglomerate towards with the cut-point of pedestrian and shade, determine respectively color average threshold values and the variance threshold values of pedestrian's agglomerate pedestrian and shadow region, select the region of color average and color variance minimum, as the shadow region of pedestrian's agglomerate;

Step C, according to the gray scale of shadow region, size and Orientation, increases and decreases the shadow region of Preliminary detection, obtains pedestrian shadow region; Its detailed process is as follows:

Step C-1, according to the color average of pedestrian's agglomerate shadow region, reject the part that color and this average in pedestrian's agglomerate shadow region differ by more than shadow color threshold values, and the pixel that pedestrian's agglomerate non-hatched area color and this average are less than shadow color threshold values is made as to candidate's shadows pixels;

Step C-2, take pedestrian and Shadow segmentation point as initial point, sets shadow region length and is the N of pedestrian's height doubly; In rejecting pedestrian's agglomerate shadow region, pixel and initial point distance are greater than the part of this shadow region length, and pixel in pedestrian's agglomerate non-hatched area and initial point are made as to candidate's shadows pixels apart from the part that is less than shadow region length;

Step C-3, take pedestrian and Shadow segmentation point as initial point, according to pedestrian's shade towards, pixel in pedestrian's agglomerate region and initial point distance are projected to elliptical coordinate system, reject the part that the distance of pixel and initial point in pedestrian's agglomerate shadow region is greater than threshold values, and the part that the distance of pixel and initial point in pedestrian's agglomerate non-hatched area is less than threshold value is made as to candidate's shadows pixels;

Step C-4, gets common factor by candidate's shadows pixels of above-mentioned steps, the pedestrian shadow region after composition increase and decrease;

Step D, returns to steps A, until video finishes.

In step C-2, described shadow region length is N times of pedestrian's height, and the span of N is: 0<N<0.8.

Described agglomerate refers to that image pixel is spatially communicated with and has the region of identical image feature, and the feature of described agglomerate comprises: agglomerate area coordinate, context marker, shade mark, center agglomerate mark, agglomerate numbering, agglomerate centre coordinate, agglomerate color average.

The invention has the beneficial effects as follows: the present invention proposes a kind of pedestrian's shade inhibition method based on agglomerate model, first described method builds pedestrian's agglomerate model, by calculating moment characteristics and the agglomerate vertical histogram of agglomerate, obtain preliminary shadow region by the method for how much again; Then, the gray scale of this shadow region, direction, center are carried out to shade modeling as parameter to whole pedestrian and shadow region, prune tentatively cutting apart the shadow region obtaining.The inventive method detects and suppresses towards, varying number pedestrian's shade difference, and video processing speed is about 7.5 frames/s, can efficiently and accurately moving target and its shade be separated.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of a kind of pedestrian's shade inhibition method based on agglomerate model of the present invention.

Embodiment

Below in conjunction with accompanying drawing, a kind of pedestrian's shade inhibition method based on agglomerate model that the present invention is proposed is elaborated:

As shown in Figure 1, a kind of pedestrian's shade inhibition method based on agglomerate model of the present invention, its step is as follows: the modeling of step 1 pedestrian agglomerate, comprises the steps:

1.1 to each two field picture, is divided into the agglomerate of N*N and calculates pixel color average and the centre coordinate in each agglomerate;

1.2 calculate difference and the centre coordinate distance of the color average between each agglomerates, by the difference of color be less than color threshold values and and centre coordinate distance be less than apart from the agglomerate of threshold values and merge, revise center agglomerate mark, characterize the color distribution of whole scene;

The 1.3 prospect templates of utilizing mixed Gaussian background modeling algorithm to extract, the agglomerate of marker motion;

1.4 centre coordinates take each motion agglomerate are node, utilize fuzzy clustering algorithm Jiang Qi center coordinate distance to be less than apart from the agglomerate of threshold values and again merge, and revise center agglomerate mark, obtain pedestrian's agglomerate.

Step 2 shadow Detection, comprises the steps:

2.1 according to pedestrian's agglomerate array R, calculates agglomerate vertical histogram H _rand the angle theta of pedestrian and shade (x) _r;

Described agglomerate vertical histogram H _r(x), can obtain by the quantity of traversal array R along continuous straight runs statistics vertical direction agglomerate.

Described angle theta _r, can calculate by the center square of pedestrian's agglomerate array R (x):

${\left({\mathrm{\&mu;}}_{p,q}\right)}_R=\underset{(x,y)\&Element;R}{\mathrm{\&Sigma;}}{(x-\stackrel{\&OverBar;}{x})}^p{(y-\stackrel{\&OverBar;}{y})}^q,(\stackrel{\&OverBar;}{x},\stackrel{\&OverBar;}{y})=(\frac{1}{n}\underset{(x,y)\&Element;R}{\mathrm{\&Sigma;}}x,\frac{1}{n}\underset{(x,y)\&Element;R}{\mathrm{\&Sigma;}}y)$

${\mathrm{\&theta;}}_R=\frac{1}{2}{\tan }^{-1}\left(\frac{2{\mathrm{\&mu;}}_{\mathrm{1,1}}}{{\mathrm{\&mu;}}_{\mathrm{2,0}}-{\mathrm{\&mu;}}_{\mathrm{0,2}}}\right)$

Wherein, (μ _p,q) _rfor the square of pedestrian's agglomerate array R, for the geometric center of pedestrian's agglomerate array R, (x, y) is pedestrian's agglomerate array element R[i] geometric center, n is the length of array R, p, q are respectively x, the exponent number of y square.

2.2 according to agglomerate vertical histogram H _r(x) calculate agglomerate histogram of difference dH _r(x);

dH _R(x)=|H _R(x)-max(H _R)|

Wherein, H _r(x) be agglomerate vertical histogram H _rx row component

2.3 calculate pedestrian and Shadow segmentation point P _rhorizontal ordinate x _r,

Wherein, X _rbottombe the horizontal ordinate span of agglomerate histogram of difference, the width of pedestrian's agglomerate array overlay area, works as x _rwhile having multiple value, retain the point nearest apart from agglomerate histogram crest.

2.4 scanning agglomerate model array R, obtain agglomerate centre coordinate set C _r(x);

2.5 pass through P _rhorizontal ordinate and C _r(x), obtain P _rordinate y _r, y _r=C _r(x _r)

2.6 utilize angle theta _rand P _robtain the cut-off rule of pedestrian and shade.Color average and the variance of computed segmentation line both sides picture, select the equal minimum region of color average and variance as preliminary shadow region R ₂;

y=mx+c,m=tanθ _R,c=y _R-x _Rtanθ _R

Wherein, m is straight slope, and c is straight line intercept, (x _r, y _r) be cut-point P _rcoordinate.

The modeling of step 3 shade, further judges from 3 angles such as the color of shade, size, directions each pixel in the shadow region of tentatively cutting apart, and comprises following steps:

3.1 shade gray scale criterions, μ _rtentatively to cut apart shadow region R ₂gray average, T _hthreshold value, I _k(x, y) is the gray-scale value that pedestrian's agglomerate area pixel point (x, y) is located.

3.2 shade size criterion,

(x _r, y _r) be P _rpoint coordinate, (x, y) controls area pixel point coordinate, T for pedestrian rolls into a ball _dfor threshold value, be no more than 0.8 times of pedestrian's height.

3.3 shade direction criterions, $\left(\begin{array}{c}s\\ t\end{array}\right)=\left(\begin{array}{cc}\cos {\mathrm{\&theta;}}_R& -\sin {\mathrm{\&theta;}}_R\\ \sin {\mathrm{\&theta;}}_R& \cos {\mathrm{\&theta;}}_R\end{array}\right)\left(\begin{array}{c}x-x_R\\ y-y_R\end{array}\right),$

θ _rshade direction when tentatively cutting apart, (s, t) is pixel and Shadow segmentation point composition distance vector (x-x _r, y-y _r) according to shade direction θ _r, project the expression behind elliptic coordinates space, T by original coordinates space _oritfor T _din the expression in elliptic coordinates space.

3.4 merge three accurate sides, build the discrimination formula of increase and decrease shadow region:

$d(x,y)=w_1\sqrt{s^2+t^2}+w_2|I_k(x,y)-{\mathrm{\&mu;}}_R|$

w ₁=αw ₁+(1-α)R _dist，w ₂=αw ₂+(1-α)R _grey

$w_1=\frac{w_1}{w_1+w_2},w_2=\frac{w_2}{w_1+w_2}$

Wherein, α is weight renewal rate, gets 0.1, w herein ₁shade length weight, w ₂it is shade gray scale weight.Weights can be adjusted automatically according to the matching degree of each factor.

Step 4 shade mark and inhibition, carry out shade modeling to each pixel of foreground area in present frame, if d (x, y) is less than threshold value T _s, this pixel can be divided into shadows pixels so.Here T _sget shadow region R while tentatively cutting apart ₂the average of interior all pixel d values, n is shadow region R ₂the quantity of interior pixel, T _sexpression formula be:

$T_s=\frac{1}{n}\underset{(x,y)\&Element;R_2}{\mathrm{\&Sigma;}}d(x,y).$

Claims (3)

1. the pedestrian's shade inhibition method based on agglomerate model, is characterized in that, comprises the steps:

Steps A, builds pedestrian's agglomerate, and each pedestrian is represented with corresponding agglomerate, and its concrete steps are as follows:

Steps A-1, obtains current frame image, utilizes agglomerate extracting method, and the scene color of current frame image is carried out to cluster, obtains image agglomerate;

Steps A-2, according to described image agglomerate, utilize mixed Gaussian background modeling method, obtain the prospect masterplate of current frame image, obtain foreground moving agglomerate;

Steps A-3, utilize fuzzy clustering method to merge foreground moving agglomerate, obtain pedestrian's agglomerate;

Step B, for pedestrian's agglomerate, utilizes shadow detection method that pedestrian and its shade are tentatively cut apart, and obtains the shadow region of pedestrian's agglomerate:

Step B-1, calculates the center square of pedestrian agglomerate, so the shade that obtains pedestrian's agglomerate towards;

Step B-2, calculates pedestrian agglomerate vertical histogram, obtains the cut-point of pedestrian and shade in pedestrian's agglomerate;

Step B-3, according to the shade of pedestrian's agglomerate towards with the cut-point of pedestrian and shade, determine respectively color average threshold values and the variance threshold values of pedestrian's agglomerate pedestrian and shadow region, select the region of color average and color variance minimum, as the shadow region of pedestrian's agglomerate;

Step C, according to the gray scale of shadow region, size and Orientation, increases and decreases the shadow region of Preliminary detection, obtains pedestrian shadow region; Its detailed process is as follows:

Step C-1, according to the color average of pedestrian's agglomerate shadow region, reject the part that color and this average in pedestrian's agglomerate shadow region differ by more than shadow color threshold values, and the pixel that pedestrian's agglomerate non-hatched area color and this average are less than shadow color threshold values is made as to candidate's shadows pixels;

Step C-2, take pedestrian and Shadow segmentation point as initial point, sets shadow region length and is the N of pedestrian's height doubly; In rejecting pedestrian's agglomerate shadow region, pixel and initial point distance are greater than the part of this shadow region length, and pixel in pedestrian's agglomerate non-hatched area and initial point are made as to candidate's shadows pixels apart from the part that is less than shadow region length;

Step C-3, take pedestrian and Shadow segmentation point as initial point, according to pedestrian's shade towards, pixel in pedestrian's agglomerate region and initial point distance are projected to elliptical coordinate system, reject the part that the distance of pixel and initial point in pedestrian's agglomerate shadow region is greater than threshold values, and the part that the distance of pixel and initial point in pedestrian's agglomerate non-hatched area is less than threshold value is made as to candidate's shadows pixels;

Step C-4, gets common factor by candidate's shadows pixels of above-mentioned steps, the pedestrian shadow region after composition increase and decrease;

Step D, returns to steps A, until video finishes.

2. a kind of pedestrian's shade inhibition method based on agglomerate model according to claim 1, is characterized in that, in step C-2, described shadow region length is N times of pedestrian's height, and the span of N is: 0<N<0.8.

3. a kind of pedestrian's shade inhibition method based on agglomerate model according to claim 1, it is characterized in that, described agglomerate refers to that image pixel is spatially communicated with and has the region of similar image feature, and agglomerate feature comprises: agglomerate area coordinate, context marker, shade mark, center agglomerate mark, agglomerate numbering, agglomerate centre coordinate, agglomerate color average.

Images