CN103577805A - Gender identification method based on continuous gait images - Google Patents

Abstract

The invention discloses a gender identification method based on continuous gait images. The gender identification method comprises the steps that the gait images of a person are obtained; foreground image extracting is carried out on pictures of the gait images of the person; the pictures are subjected to erosion re-expanding processing, noise in the pictures is removed, smoothing is carried out, then the images are normalized; the images are subjected to continuous gait image capturing processing; the pictures are subjected to noise reduction processing, noise is removed; the processed pictures are subjected to characteristic extracting; a support vector machine (SVM) classifier is used for carrying out classifying training on the pictures; the images to be tested are subjected to characteristic extracting, and then the obtained characteristics are compared with the characteristics trained in an SVM. According to the method, the continuous gait images are used for extracting the image characteristics of the person, during extracting, the vertical pixel characteristics and the horizontal pixel characteristics of the pictures are synthesized, the accuracy rate of gender identification is effectively improved, and meanwhile the method can be suitable for gender identification of human body gait image shot at different angles.

Description

Gender identification method based on continuous gait image

Technical field

The present invention relates to a kind of gender identification method, relate in particular to a kind of gender identification method based on continuous gait image

Background technology

In some monitoring environments, because environmental restraint can not accurately identify the identity of target, or do not need to identify concrete target identities, and interested in some category attributes of target, whether normal etc. such as: sex, age, carrying status, walking.

Aspect social security, sex is recognized as important now research direction, and early stage sex research is the basis that is characterized as with people's face or profile mostly, but people's face image often can be because resolution is low or other factors reduce discrimination power under supervisory system.And the people that plan is committed a crime deliberately dress is hidden close, and hiding face characteristic causes sex identification difficulty.In addition, people's face image is also very limited to the help of remote monitor sex, causes take the method that people's face is feature and is not suitable in our application.After inquiring into the data of analyzing actor, find, people's body swing and step ratio when walking exists gender differences really, and the common shoulder of the male sex swings and the long-range what women of size who strides.Women distinguishes with the difference at hair length and chest back.According to these characteristics, quite helpful to sex identification backward.

Sex Research on Identification is used in shop, sales field, can alleviate manpower and convenient again.With fixed camera mounting, be located at shop door mouth, identification sex when client enters sales field, sales field can provide the bargain goods of this sex and the placement location of commodity, one can reduce the time that client searches commodity, and carry out also can learn the bargain goods of sales field, also can reduce the cost of printing sales field, sales field catalogue.

The existing image that utilizes carries out sex knowledge method for distinguishing, mostly adopt static, single body image as the object of training and judgement, the variance factor that when this mode cannot be walked men and women's property, condition changes is as the parameter of judgement, only using static body shape parameter as judgment standard, and body shape between a men and women's property strict boundary benchmark not, therefore can produce larger error, thereby reduce the accuracy of judgement.

Summary of the invention

Technical matters to be solved by this invention is, a kind of gender identification method based on continuous gait image is provided, and comprises the steps:

S1: the gait image that obtains pedestrian;

S2: the image of pedestrian's gait image is extracted to prospect image;

The mode of described extraction prospect image can be to adopt binaryzation or background subtracting.

S3: image is done and corroded expansion process again image is removed to noise smoothing, then by image normalization;

It is continuous gait image that the image normalization of extracting is carried out to GEI(gait energygram looks like afterwards) process the otherness that can obtain this sequence gait.

S4: the image of S3 is carried out to the continuous gait image processing of acquisition;

S5: image is carried out to noise reduction process and remove noise;

S6: the image after S5 is processed extracts feature;

S7: utilize support vector machines sorter to carry out classification based training to image;

S8: image to be measured is extracted to feature according to the step of S1～S6, then the feature of training in the feature of acquisition and S7 is compared.

Further, the method that step S2 extracts prospect image is background subtracting method, thereby adjacent image data is subtracted each other and obtains prospect image.

Further, step S2 extract the method for prospect image can also be for by image binaryzation.

Further, the step that also comprises horizontal scanning image pixel and vertical scanning image pixel after the foreground image that extracts pedestrian's gait image.

Further, gait image processing is that continuous prospect image data in unit interval section is weighted on average continuously.

Further, step S5 arranges a threshold value, and each pixel value of image and this threshold value are compared, and the data that are less than described threshold value is removed as noise, thereby image is carried out to noise reduction.

Further, the maximum pixel coefficient in image is multiplied by scale-up factor and obtains described threshold value.

Further, described scale-up factor is 0.6～0.9.

While further, extracting characteristics of image, the horizontal property parameters of image and vertical features parameter are extracted respectively.

Further, the pixel parameter of the horizontal properties of described image and the pixel parameter of vertical features are all divided into 9 regions, and each region unit is to there being the pixel of some.

Implement the present invention, there is following beneficial effect:

The present invention is by utilizing continuous gait image to extract people's image feature, during extraction, combine vertical pixel feature and the horizontal pixel feature of image, continuous image while walking by people carries out feature extraction, difference during the men and women's property of usining walking between condition variation is as the parameter of judgement, effectively improve the accuracy rate of sex identification, can also adapt to the sex identification of the body gait image of different angles shooting simultaneously.

Accompanying drawing explanation

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

Fig. 2 is GEI image processing schematic diagram of the present invention;

Fig. 3 is the principle schematic of GEI image processing of the present invention;

Fig. 4 is the principle schematic that DEI of the present invention processes;

Fig. 5 is level acquisition feature schematic diagram of the present invention;

Fig. 6 is that the present invention vertically captures feature schematic diagram.

Embodiment

For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing, the present invention is described in further detail.

As shown in Figure 1, flow process of the present invention comprises:

S1: the gait image that obtains pedestrian;

S2: the image of pedestrian's gait image is extracted to prospect image;

S3: image is done and corroded expansion process again image is removed to noise smoothing, then by image normalization;

By capturing gait sequence image in film, through binaryzation or background subtracting, processes and obtains prospect image, then image normalization is carried out to GEI(gait energygram picture afterwards) processing can obtain the otherness of this sequence gait.

S4 carries out acquisition gait energygram picture to the image of S3 and processes;

S5: image is carried out to noise reduction process and remove noise;

S6: the image after S5 is processed extracts feature;

S7: utilize support vector machines sorter to carry out classification based training to image;

S8: image to be measured is extracted to feature according to the step of S1～S6, then the feature of training in the feature of acquisition and S7 is compared.

Fig. 2 is GEI image processing schematic diagram of the present invention, as shown in the figure, from video camera, take the gait shadow lattice picture obtaining, after binaryzation or background subtracting processing, take out prospect image normalization, the regular gait shadow lattice of same sequence calculate GEI image via formula (1).

$G_c(x,y)=\frac{1}{N_c}{\mathrm{\Σ}}_{t\∈\mathrm{Ac}}B(x,y,t),---\left(1\right)$

Wherein, G _c(x, y) is GEI image, N _cfor gait sequence image sum, a gait sequence time span A _c, B (x, y, t) representative is at the image of t time, and wherein x is the coordinate of pixel in image with y.

The principle example of its processing as shown in Figure 3, is supposed 3 * 3 images after two normalizations, and white block is gait prospect image, and black block is background video, and two image overall averages are become and can obtain GEI image.In example, GEI image pixel value is that first image of 1 expression and the 2nd image do not change, and the 1st image of the expression that pixel value is 0.5 and the 2nd image change, so just can find out feature when everyone walks.

DEI is mainly with the noise that removes GEI image, can obtain men and women's gait otherness simultaneously.DEI as shown in Equation (2).

$D_c(x,y)=\left\{\begin{array}{c}1,\mathrm{if}{G}_c(x,y)\≥U,\\ 0,\mathrm{otherwise},\end{array}\right.---\left(2\right)$

D _c(x, y) is for removing the image after noise, G _c(x, y) is GEI image, and U is a threshold value.Fig. 4 is that GEI image is made DEI processing example, the GEI image of one 3 * 3, threshold value is that GEI image the inside max pixel value (example is 1) is multiplied by scale-up factor 0.8, obtain after threshold value U=0.8, each pixel value and the U of GEI image are made comparisons, the words that are more than or equal to U when this pixel value of GEI DEI image value equal 1, and when being less than U, DEI image value equals 0.As Fig. 4 GEI image up can obtain right-hand DEI image after impurity elimination news are processed, process DEI image later and only retain in former GEI compared with part and parcel.

With reference to Fig. 5, Figure 6 shows that level and vertical features acquisition example, make X _iexpression is by i eigenwert of DEI image capture, and its acquisition method is as formula (3).

$X_i=\underset{k=1}{\overset{9}{\mathrm{\Σ}}}{c}_{\mathrm{ik}}---\left(3\right)$

When k=1～8, c _ikformula is as follows:

c _ik＝{b _{i((k-1)×18+l)}|l＝1～18}

Work as k=9, c _ikformula is as follows:

c _ik＝{b _{i((k-1)×18+l)}|l＝1～6}

$Y_j=\underset{k=1}{\overset{9}{\mathrm{\Σ}}}{d}_{\mathrm{kj}}$

When k=1～8, d _kjformula is as follows:

d _kj＝{b _{((k-1)×18+l)j}|l＝1～18}

Work as k=9, d _kjformula is as follows:

d _kj＝{b _{((k-1)×18+l)j}|l＝1～6}

Wherein i represents the row of image, and j represents the row of image, k Representative Region block number.X _irepresent feature after each column count, Y _jrepresent the rear feature of every a line calculating, c _ikrepresent each column split block, d _kjrepresent every a line cut section piece.B _ijrepresent image pixel position.

X _ibe that i row pixel is calculated and obtained in DEI image, this paper method DEI image size is 150 * 150 pixels event 1≤i≤150, X _iin DEI, 150 pixels of i row and 150 capable pixels of j are tried to achieve, and through experiment, try to achieve every row and every row pixel are divided into 9 blocks.C wherein _i1to c _i8with d _1jto d _8jcorresponding 18 bits of each block, c _i9with d _9jto being listed as last remaining 6 bits, these 9 block numerical value are added up and become these row and the eigenwert X going _i, Y _j.Therefore a DEI image has 300 characteristic parameters as sex identification feature.

The SVM that the present invention program adopts has tri-kinds of Linear, Polynomial, Radial Basis Function for the core function of classifying, this programme adopts the image data of CASIA database as experiment sample, and adopt LIBSVM to train and test, the data of training have 18 male sex and 14 women, according to training pattern, remaining 18 male sex and 13 women are tested, and test with 75 * 75 and 150 * 150 two kinds of image sizes, test result is as shown in table 1, table 2.

Embodiment 1

From table 1, table 2 data, take vertical features can be much higher than independent vertical features acquisition or horizontal properties acquisition in conjunction with the accuracy rate of horizontal properties acquisition.

Table 1

Table 2

This method is when people walks at 90 °, sex identification in can reach 100% accuracy rate, but in daily life, people's walking is in different angles, this paper method has also been done experiment in different angles situation, and the image of 75 * 75 sizes of take is example, and the vertical features of take acquisition Average Accuracy is 63.64%, the horizontal properties of take acquisition Average Accuracy is 72.21%, and the Average Accuracy of horizontal+vertical feature extraction of take is 70.97%; The image of 150 * 150 sizes of take is example, and the vertical features of take acquisition Average Accuracy is 72.43%, and the horizontal properties of take acquisition Average Accuracy is 82.11%, and the Average Accuracy of horizontal+vertical feature extraction of take is 87.98%.

Embodiment 2

When video camera shooting angle is 90 °, the 0 ° training data during with 180 °, and data to be measured to take the deviation of 18 ° be example, the accuracy rate recording is as shown in table 3

Three kinds of angle gait sex identification accuracys rate of table 3

The above is the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications are also considered as protection scope of the present invention.

Claims (10)

1. the gender identification method based on continuous gait image, is characterized in that, comprises the steps:

S1: the gait image that obtains pedestrian;

S2: the image of pedestrian's gait image is extracted to prospect image;

S3: image is done and corroded expansion process again image is removed to noise smoothing, then by image normalization;

S4: the image of S3 is carried out to the continuous gait image processing of acquisition;

S5: image is carried out to noise reduction process and remove noise;

S6: the image after S5 is processed extracts feature;

S7: utilize support vector machines sorter to carry out classification based training to image;

S8: image to be measured is extracted to feature according to the step of S1～S6, then the feature of training in the feature of acquisition and S7 is compared.

2. recognition methods according to claim 1, is characterized in that, the method that step S2 extracts prospect image is background subtracting method.

3. recognition methods according to claim 1, is characterized in that, the method that step S2 extracts prospect image is by image binaryzation.

4. recognition methods according to claim 1, is characterized in that, also comprises the step of horizontal scanning image pixel and vertical scanning image pixel after the foreground image that extracts pedestrian's gait image.

5. recognition methods according to claim 1, is characterized in that, gait image processing is that continuous prospect image data in unit interval section is weighted on average continuously.

6. recognition methods according to claim 1, is characterized in that, step S5 arranges a threshold value, the data that are less than described threshold value is removed as noise, thereby image is carried out to noise reduction.

7. recognition methods according to claim 6, is characterized in that, the maximum pixel coefficient in image is multiplied by scale-up factor and obtains described threshold value.

8. recognition methods according to claim 7, is characterized in that, described scale-up factor is 0.6～0.9.

9. recognition methods according to claim 1, is characterized in that, while extracting characteristics of image, the horizontal property parameters of image and vertical features parameter is extracted respectively.

10. according to the recognition methods described in claim 1 or 9, it is characterized in that, the pixel parameter of the horizontal properties of described image and the pixel parameter of vertical features are divided into 9 regions.

Images