CN105069466A - Pedestrian clothing color identification method based on digital image processing - Google Patents

Abstract

The invention discloses a pedestrian clothing color identification method based on digital image processing. The method comprises the steps that (1) a pedestrian detection method which combines an HOG feature description operator and a SVM classifier is used to acquire a pedestrian image; (2) a Sobel operator is used to detect the edge contour shape of a pedestrian to acquire an image to be searched; (3) a pedestrian contour shape template is produced and matches a corresponding region in the image to be searched to acquire upper and lower body images of the pedestrian; (4) a seed filling method is used to carry out communication region labeling on clothing colors of upper and lower body regions of the pedestrian; (5) color feature extraction is carried out on a color communication region; and (6) the SVM classifier is used to carry out color classification discrimination to acquire the clothing color of the pedestrian, and a final result is output. According to the invention, the identification accuracy of the clothing color of the pedestrian is improved; standardized dress in a dangerous region is ensured; and safety risks are eliminated.

Description

Based on pedestrian's dress ornament color identification method of Digital Image Processing

Technical field

The present invention relates to a kind of color identification method, belong to technical field of image processing, what be specifically related to is a kind of pedestrian's dress ornament color identification method based on Digital Image Processing.

Background technology

Safety is the eternal theme of the special dimension such as electric power, oil gas field.In recent years, the various security incidents of the industry such as oil gas field, electric power happen occasionally, how to strengthen oil gas field, electric power enterprise production safety ability and improve its management level become related personnel must faced by matter of utmost importance.

One of the potential safety hazard in the field such as oil gas field, electric power be exactly staff in assigned work region not strict implement dress code, wear special safety clothes not according to regulation.Meanwhile, along with country more and more payes attention to safety in production operation, there is increasing video monitoring system in the industry such as oil gas field, electric power.But existing supervisory system, mostly rest on the stages such as video record, storage, query and search, thus larger error is had to the differentiation of pedestrian's dress ornament color, be difficult to accurately to identify the dressing of pedestrian in the assigned work such as oil gas field, electric power region whether meet the requirements, thus cause corresponding potential safety hazard to exist always.

Summary of the invention

The object of the present invention is to provide a kind of pedestrian's dress ornament color identification method based on Digital Image Processing, mainly solve prior art owing to there is comparatively greatly the problem of potential safety hazard to pedestrian's dress ornament colour recognition error.

To achieve these goals, the technical solution used in the present invention is as follows:

Based on pedestrian's dress ornament color identification method of Digital Image Processing, comprise the following steps:

(1) adopt HOG feature interpretation operator in conjunction with the pedestrian detection method collection pedestrian image of SVM classifier;

(2) adopt Sobel operator to detect pedestrian's edge contour shape, obtain image to be searched;

(3) make pedestrian contour shape template T according to the attitude that pedestrian is common, and the region that pedestrian contour shape template is corresponding to image to be searched is mated, obtain the upper part of the body and the lower part of the body image of pedestrian;

(4) seed filling method is adopted to carry out connected component labeling to the dress ornament color in pedestrian's upper part of the body and lower part of the body region respectively;

(5) color feature extracted is carried out to the color-connected regions obtained;

(6) according to the color characteristic extracted, utilize SVM classifier to carry out color classification differentiation, obtain pedestrian's dress ornament color, and export net result.

Further, in described step (3), the detailed process of pedestrian contour shape template and images match to be searched is as follows:

(a) by pedestrian contour shape template T on image to be searched from left to right, translation gliding successively from top to bottom, obtain representing the subgraph S that template covers image-region to be searched _i,j, wherein, i, j represent the coordinate of the upper left corner of subgraph in image to be searched;

B () utilizes the matching degree of following formula comparison pedestrian contour shape template and each subgraph:

$D(i,j)=\underset{m=1}{\overset{M}{\Σ}}\underset{n=1}{\overset{M}{\Σ}}{\[S_{i,j}(m,n)-T(m,n)\]}^2$

C () chooses the minimum value of D (i, j), the position (i, j) obtained is the position at pedestrian place in the picture, and the wide, high of pedestrian then equals the wide, high of pedestrian contour shape template T respectively.

Specifically, described step (5) comprises the following steps:

(5a) each color-connected regions of mark in step (4) is all transformed to hsv color space, YCbCr color space and Lab color space respectively from RGB color space;

(5b) extract hsv color space, YCbCr color space and Lab color space average separately, variance, energy and contrast respectively, then by it series connection, obtain color feature vector;

(5c) repeat step (5a), (5b), the color feature vector of some pedestrian's training samples is input in SVM classifier and carries out training study, obtain SVM classifier model; The color feature vector that subsequent extracted obtains, only need be fed through in SVM classifier and can be realized discriminant classification.

Compared with prior art, the present invention has following remarkable result:

(1) existing several algorithm combines by the present invention, utilize pedestrian detection, upper lower part of the body shape segmentations, the method of color-connected regions mark, and design and incorporate template matches, the mode of color feature extracted and discriminant classification, thus effectively can identify pedestrian's dress ornament color, its accuracy of identification is quite high, substantially there are not how many errors in the color identified, thus, supvr can be facilitated well to the management of dressing aspect, hazardous location, execution for specification dressing brings larger guarantee, effectively eliminate the potential safety hazard of this respect.

(2) the present invention is reasonable in design, clear process, understand, be highly suitable for the special dimension such as electric power, oil gas field aspect and apply.

Accompanying drawing explanation

Fig. 1 is schematic flow sheet of the present invention.

Fig. 2 is the pedestrian contour shape template schematic diagram of the present invention-embodiment.

Embodiment

Below in conjunction with drawings and Examples, the invention will be further described, and embodiments of the present invention include but not limited to the following example.

Embodiment

As shown in Figure 1, the invention provides one to be applicable to the special occasions safety of workers such as oil gas field, transformer station and to take dressing and detect and know method for distinguishing, it primarily of pedestrian detection, pedestrian above the waist and lower part of the body shape segmentations, upper part of the body and lower part of the body dress ornament color-connected regions mark, color feature extracted, color classification differentiate and several large step of result output forms.

One, pedestrian detection

The present invention adopts HOG feature interpretation operator in conjunction with the pedestrian detection method collection pedestrian image of SVM classifier.HOG (HistogramofOrientationGradient) feature is a kind of intensive descriptor to image local overlapping region, and it carrys out constitutive characteristic by the gradient orientation histogram calculating regional area.The edge of human body can be described well, simultaneously insensitive to the skew of illumination variation and a small amount of.HOG integrate features SVM classifier has been widely used in image recognition, especially in pedestrian detection, obtains great success.

The calculating of HOG feature needs the concept using gradient, and in image, the gradient of pixel (x, y) is:

G _x(x，y)＝H(x+1，y)-H(x-1，y)

G _y(x，y)＝H(x，y+1)-H(x，y-1)

G above in the middle of formula _x(x, y), G _y(x, y), H (x, y) represents the gradient of pixel (x, y) horizontal direction, the gradient of vertical direction and pixel value respectively.The gradient magnitude at pixel (x, y) place and gradient direction be respectively:

$G(x,y)=\sqrt{G_x{(x,y)}^2+G_y{(x,y)}^2}$

$\mathrm{\α}(x,y)={\tan }^{-1}\left(\frac{G_y(x,y)}{G_x(x,y)}\right)$

The process of HOG feature extraction: the unit (cell) Iamge Segmentation being several pixels, gradient direction is on average divided into 9 intervals (bin), inside each unit, in all directions interval, statistics with histogram is carried out to the gradient direction of all pixels, obtain the proper vector of one 9 dimension, 4 often adjacent unit form a block (block), proper vector connection in a block is got up to obtain the proper vector of 36 dimensions, scan sample image with block, scanning step is a unit.Finally the feature of all pieces is together in series, just obtains the feature of human body.Such as, for the image of 64*128, the unit (pixel of 16*16) of every 2*2 forms a block, 4*9=36 feature is had, with 8 pixels for step-length, so in each piece, horizontal direction will have 7 scanning windows, and vertical direction will have 15 scanning windows.That is, the picture of 64*128, total total 36*7*15=3780 feature.

SVM is a kind of common sorter, and it proposes from the optimal classification surface linear separability situation.So-called optimal classification, require exactly sorting track (or classifying face) not only can by faultless for two classes separately, and class interval between two classes is maximum.The former ensures empirical risk minimization, and in fact make class interval maximum be exactly make the fiducial range in generalization minimum.Be generalized to higher dimensional space, optimal classification line just becomes optimal classification surface.

Two, pedestrian's upper part of the body and lower part of the body shape segmentations

By detection and the image acquisition of previous step, obtain the image of pedestrian in monitor video picture.Such as, but typically, pedestrian is not at the center position of image, and, the upper part of the body of pedestrian may present frontal pose, also may present body inclination, partially first-class attitude.And the lower part of the body attitude of pedestrian is more complicated, such as, both legs may be existed side by side, and also may separately present various different angles.If directly the first half of pedestrian's image to be used as the upper part of the body of pedestrian, the latter half of pedestrian's image to be used as the lower part of the body of pedestrian, to make the identification decision of pedestrian's dress ornament color with this, very large error will inevitably be caused.Therefore, be necessary the upper part of the body to pedestrian, lower part of the body shape accurately splits, facilitate the follow-up colour recognition of dress ornament accurately to judge.

The present invention passes through the mode of pedestrian's edge contour SHAPE DETECTION and template matches, splits, and obtain above the waist corresponding and lower part of the body image to pedestrian's upper part of the body and lower part of the body shape.

Pedestrian's edge contour SHAPE DETECTION

Edge is the most basic feature of image, and conventional edge contour detection method has: Robert operator, Sobel operator, Prewitt operator, Canny operator etc.The present invention uses Sobel operator to carry out edge contour detection.Its calculation procedure is as follows:

Carry out gaussian filtering to original image, gaussian kernel is here as follows:

$K=\frac{1}{2\mathrm{\π}\mathrm{\σ}*\mathrm{\σ}}{e}^{-\frac{x*x+y*y}{2\mathrm{\σ}*\mathrm{\σ}}}$

Utilize amplitude and the direction of convolution mask compute gradient

The gradient convolution mask of horizontal direction is as follows:

$s_x=\left[\begin{array}{ccc}-1& 0& 1\\ -2& 0& 2\\ -1& 0& 1\end{array}\right]$

The gradient convolution mask of vertical direction is as follows:

$s_y=\left[\begin{array}{ccc}1& 2& 1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]$

Here, in order to express easily, the field point mark matrix of pending pixel (i, j) is provided, as follows:

$K=\left[\begin{array}{ccc}{a}_0& a_1& a_2\\ a_7& \[i,j\]& a_3\\ a_6& a_5& a_4\end{array}\right]$

Obviously, available mathematical formulae is expressed the gradient magnitude of its each point and is:

$G\[i,j\]=\sqrt{s_x^2+s_y^2}$

s _x＝(a ₂+2a ₃+a ₄)-(a ₀+2a ₇+a ₆)

s _y＝(a ₀+2a ₁+a ₂)-(a ₆+2a ₅+a ₄)

Gradient direction can be expressed as:

θ(x,y)＝S _y/S _x

According to the real needs of different scene, threshold value is set, and splits, obtain edge contour and image.

Template matches

The present invention has made different pedestrian contour shape template T according to the attitude that pedestrian is common, as shown in Figure 2.According to the pedestrian contour shape template made, the region that it is corresponding to image to be searched is mated, and in image to be searched, namely finds the image with same size, size and Orientation, then by the position at certain method determination target place.The process of its coupling is as follows:

(1) by pedestrian contour shape template T on image to be searched from left to right, translation gliding successively from top to bottom, obtain the subgraph S that different representative templates covers image-region to be searched _i,j, wherein, i, j represent the coordinate of the upper left corner of subgraph in image to be searched;

(2) matching degree of following formula comparison pedestrian contour shape template and each subgraph is utilized:

$D(i,j)=\underset{m=1}{\overset{M}{\Σ}}\underset{n=1}{\overset{M}{\Σ}}{\[S_{i,j}(m,n)-T(m,n)\]}^2$

(3) choose the minimum value of D (i, j), draw subgraph S corresponding with it _i,j, then pedestrian contour shape template is mated with this subgraph, the accurate location region of pedestrian can be obtained.

After having mated, just can obtain the corresponding pedestrian upper part of the body and lower part of the body shape image.

Three, the upper part of the body and lower part of the body dress ornament color-connected regions mark

Obviously, although previous step obtains the upper part of the body, the lower part of the body shape image of pedestrian, typically, the a certain color that the clothes of pedestrian, trousers are not simple, but present the mixing of multiple color, such as grid or striped clothes, or clothes there are various decorative pattern, pattern etc.Therefore, be necessary to carry out connected component labeling to the dress ornament color in pedestrian's upper part of the body obtained above, lower part of the body region.

Here color-connected regions mark is similar with the connected region detection method of bianry image, and common method for marking connected region has two-pass scan method and se ed filling algorithm two kinds.The present invention adopts seed filling method to carry out the mark of color-connected regions, and its treatment step is as follows:

Initialization mask image B, its size is identical with the image size in Color-Connected territory to be calculated, and all pixels of B are equal

Assignment is 0;

Scan image, until current mask pixel B (x, y)=1;

(1) using the pixel at changing coordinates (x, y) place as sub pixel, and give an one label, then the similar pixel in all 4 neighborhoods of this sub pixel be all pressed in storehouse; Wherein, sub pixel (x1, y1) is as follows with the calculating formula of similarity of neighborhood territory pixel (x2, y2):

$d_{12}=\sqrt{{(x_1-x_2)}^2-{(y_1-y_2)}^2}$

(2) pop-up a stack stack top pixel, gives the label that it is identical, and then is all pressed in storehouse by all foreground pixels adjacent with this stack top pixel;

Repeat step (2), until storehouse is empty;

Now, just have found a connected region in image B, the pixel value in this region is marked as label;

Repeat step (1), until the end of scan; The end of scan just can obtain the connected region of all colours component in image.

Four, color feature extracted

By each color-connected regions of marking in previous step all from RGB color notation conversion space to hsv color space, YCbCr color space and Lab color space, wherein:

RGB color space conversion is as follows to the computing formula in hsv color space:

M＝max(R，G，B)

m＝min(R，G，B)

C＝M-m

$H^{\′}=\left\{\begin{array}{ccc}undefined,& if& C=0\\ \frac{G-B}{C}\mathrm{mod}6,& if& B=R\\ \frac{B-R}{C}+2,& if& B=G\\ \frac{R-G}{C}+4,& if& M=B\end{array}\right.$

H＝60°×H′

V＝M

$S_{HSV}=\left\{\begin{array}{cc}0,& \begin{array}{cc}if& C=0\end{array}\\ \frac{C}{V},& otherwise\end{array}\right.$

RGB color space conversion is as follows to the computing formula of YCbCr color space:

Y＝0.257R+0.564G+0.098B+16

Cb＝-0.148R-0.291G+0.439B+128

Cr＝0.439R-0.368G-0.071B+128

RGB color space conversion is as follows to the computing formula of Lab color space:

RGB cannot be directly changed into Lab, needs first to convert XYZ to and converts Lab to again, that is: RGB->XYZ->Lab

Therefore conversion formula divides two parts:

RGB converts XYZ to

$\left\{\begin{array}{c}R=gamma\left(\frac{r}{255.0}\right)\\ G=gamma\left(\frac{g}{255.0}\right)\\ B=gamma\left(\frac{b}{255.0}\right)\end{array}\right.$

Wherein,

$\left[\begin{array}{c}X\\ Y\\ Z\end{array}\right]=M*\left[\begin{array}{c}R\\ G\\ B\end{array}\right]$

$\[M\]=\left[\begin{array}{ccc}0.436052025& 0.385081593& 0.143087414\\ 0.222491598& 0.716886060& 0.060621486\\ 0.013929122& 0.097097002& 0.714185470\end{array}\right].$

Gamma function above, be used to carry out non-linear tone editor to image, object improves picture contrast.

XYZ converts Lab to

L ^＊＝116f(Y/Y _n)-16

a ^＊＝500[f(X/X _n)-f(Y-Y _n)]

b ^＊＝200[f(Y/Y _n)-f(Z/Z _n)]

$f\left(t\right)=\left\{\begin{array}{cc}{t}^{1/3}& \begin{array}{cc}if& t>{\left(\frac{6}{29}\right)}^3\end{array}\\ \frac{1}{3}{\left(\frac{29}{6}\right)}^{2}t+\frac{4}{29}& otherwise\end{array}\right..$

After RGB color space conversion becomes hsv color space, YCbCr color space and Lab color space, extract hsv color space, YCbCr color space and Lab color space average separately, variance, energy and contrast respectively, then by it series connection, obtain color feature vector, wherein, the computing formula of average, variance, ability and contrast is as follows respectively:

Mean value computation formula:

$\stackrel{\‾}{H}=\underset{i=1}{\overset{M}{\Σ}}\underset{j=1}{\overset{M}{\Σ}}{H}_{ij}/MN$

Variance computing formula:

$\mathrm{\σ}=\sqrt{\underset{i=1}{\overset{M}{\Σ}}\underset{j=1}{\overset{M}{\Σ}}(H_{ij}-\stackrel{\‾}{H})/MN}$

Energy balane formula:

$E=\underset{i=1}{\overset{M}{\Σ}}\underset{j=1}{\overset{M}{\Σ}}{H}_{ij}^2$

Contrast computing formula:

$I=\underset{i=1}{\overset{M}{\Σ}}\underset{j=1}{\overset{M}{\Σ}}{(i-j)}^2{H}_{ij}$

H above in the middle of formula _ijrepresent the pixel value at coordinate (i, j) place, the width of M and N difference representative image and height.

Repeat above-mentioned steps, then some color feature vectors are input in SVM classifier and carry out classification based training, obtain SVM classifier model and be stored in SVM classifier.SVM classifier utilizes the thought of class interval to carry out training, and it depends on the pre-service to data, that is, at the space expression raw mode of more higher-dimension, by the suitable Nonlinear Mapping to enough higher-dimension the raw data belonging to two classes respectively just can be separated by a lineoid.

Five, color classification differentiates and result output

Because SVM classifier has carried out sample learning and training before the use, therefore, after obtaining SVM classifier model, the color vector feature that subsequent extracted obtains, only need be entered in SVM classifier, just can carry out discriminant classification to this color feature vector, last Output rusults.

The present invention carries out analysis by a kind of method designing the pedestrian's of identification dress ornament color to the monitor video of existing oil gas field, power industry and judges, the whether correct dressing of staff in production operation region can be identified, and then effectively improve the production safety management level to these production operation regions.The present invention is compared to existing technology, obviously progressive, has outstanding substantive distinguishing features and significant progress.

Above-described embodiment is only the present invention's preferably one of implementation; should in order to not limit the scope of the invention; all changes of under body design thought of the present invention and spirit, technical solution of the present invention being made or polishing; or carry out substitute equivalents; its technical matters solved is in fact still consistent with the present invention, all should within protection scope of the present invention.

Claims (3)

1., based on pedestrian's dress ornament color identification method of Digital Image Processing, it is characterized in that, comprise the following steps:

(1) adopt HOG feature interpretation operator in conjunction with the pedestrian detection method collection pedestrian image of SVM classifier;

(2) adopt Sobel operator to detect pedestrian's edge contour shape, obtain image to be searched;

(3) make pedestrian contour shape template T according to the attitude that pedestrian is common, and the region that pedestrian contour shape template is corresponding to image to be searched is mated, obtain the upper part of the body and the lower part of the body image of pedestrian;

(4) seed filling method is adopted to carry out connected component labeling to the dress ornament color in pedestrian's upper part of the body and lower part of the body region respectively;

(5) color feature extracted is carried out to the color-connected regions obtained;

(6) according to the color characteristic extracted, utilize SVM classifier to carry out color classification differentiation, obtain pedestrian's dress ornament color, and export net result.

2. the pedestrian's dress ornament color identification method based on Digital Image Processing according to claim 1, it is characterized in that, in described step (3), the detailed process of pedestrian contour shape template T and images match to be searched is as follows:

(a) by pedestrian contour shape template T on image to be searched from left to right, translation gliding successively from top to bottom, obtain representing the subgraph S that template covers image-region to be searched _i,j, wherein, i, j represent the coordinate of the upper left corner of subgraph in image to be searched;

B () utilizes the matching degree of following formula comparison pedestrian contour shape template and each subgraph:

001"/>

C () chooses the minimum value of D (i, j), the position (i, j) obtained is the position at pedestrian place in the picture, and the wide, high of pedestrian then equals the wide, high of pedestrian contour shape template T respectively.

3. the pedestrian's dress ornament color identification method based on Digital Image Processing according to claim 2, it is characterized in that, described step (5) comprises the following steps:

(5a) each color-connected regions of mark in step (4) is all transformed to hsv color space, YCbCr color space and Lab color space respectively from RGB color space;

(5b) extract hsv color space, YCbCr color space and Lab color space average separately, variance, energy and contrast respectively, then by it series connection, obtain color feature vector;

(5c) repeat step (5a), (5b), the color feature vector of some pedestrian's training samples is input in SVM classifier and carries out training study, obtain SVM classifier model; The color feature vector that subsequent extracted obtains, only need be fed through in SVM classifier and can be realized discriminant classification.