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

Abstract

The invention discloses a gender recognition method based on continuous gait images, which includes acquiring pedestrian gait images; extracting foreground images from images of pedestrian gait images; performing erosion and re-expansion processing on the images to remove noise and smooth the images, Then normalize the image; perform continuous gait image processing on the image; denoise the image to remove noise; extract features from the processed image; use the support vector machine SVM classifier to classify the image; The feature is extracted from the measured image, and then the obtained feature is compared with the feature trained in the SVM. The present invention extracts the image features of people by using continuous gait images, and integrates the vertical pixel features and horizontal pixel features of the images when extracting, effectively improving the accuracy of gender recognition, and can also adapt to human body gait photographed at different angles. Gender recognition of static images.

Description

Gender recognition method based on continuous gait images

technical field

The present invention relates to a gender recognition method, in particular to a gender recognition method based on continuous gait images

Background technique

In some monitoring environments, due to environmental constraints, the identity of the target cannot be accurately identified, or it is not necessary to identify the specific identity of the target, but some category attributes of the target are more interested, such as: gender, age, carrying status, whether the walking posture is Normal and so on.

In terms of social security, gender identification is an important research direction nowadays. Most of the early gender studies were based on the features of faces or contours. reduce. And those who plan to commit crimes will deliberately wear secret clothes to hide facial features, making gender identification difficult. In addition, face images are of limited help in long-distance monitoring of gender, which makes methods based on faces unsuitable for our application. After discussing and analyzing the data of the actors, it is found that there are indeed gender differences in the body swing and step ratio when people walk. Men usually have much larger shoulder swings and strides than women. Females are distinguished by differences in hair length and the back of the chest. Based on these characteristics, it is very helpful for future gender identification.

Gender identification research is used in stores, which can reduce manpower and is convenient. A fixed camera is installed at the entrance of the store. When the customer enters the store, the gender is identified. The store can provide the special price of the gender and the location of the product. This can reduce the time for customers to search for products, and can also know the special price of the store. Commodities can also reduce the cost of printing store catalogs in stores.

Most of the existing methods for gender recognition using images use static, single human body images as the object of training and judgment. Static body shape parameters are used as the judgment benchmark, and there is no strict boundary benchmark for the body shape between men and women, so large errors will occur, thereby reducing the accuracy of judgment.

Contents of the invention

The technical problem to be solved by the present invention is to provide a gender recognition method based on continuous gait images, comprising the following steps:

S1: Obtain the gait image of the pedestrian;

S2: extracting the foreground image from the image of the pedestrian's gait image;

The manner of extracting the foreground image may be binarization or background subtraction.

S3: Perform erosion and re-expansion processing on the image to remove noise and smooth the image, and then normalize the image;

The difference in gait of this sequence can be obtained by performing GEI (gait energy image, continuous gait image) processing after normalizing the extracted images.

S4: Execute the processing of capturing continuous gait images on the image of S3;

S5: performing noise reduction processing on the image to remove noise;

S6: extracting features from the image processed in S5;

S7: Utilize the support vector machine SVM classifier to perform classification training on the image;

S8: Extract features from the image to be tested according to the steps of S1-S6, and then compare the obtained features with the features trained in S7.

Further, the method for extracting the foreground image in step S2 is the background subtraction method, in which adjacent image data are subtracted to obtain the foreground image.

Further, the method for extracting the foreground image in step S2 may also be binarizing the image.

Further, after extracting the foreground image of the gait image of pedestrians, the steps of scanning image pixels horizontally and scanning image pixels vertically are also included.

Further, the continuous gait image processing is to carry out weighted average of the continuous foreground image data in a unit time period.

Further, step S5 sets a threshold value, compares each pixel value of the image with the threshold value, and removes data smaller than the threshold value as noise, thereby denoising the image.

Further, the threshold is obtained by multiplying the maximum pixel coefficient in the image by a scale coefficient.

Furthermore, the proportional coefficient is 0.6-0.9.

Further, when extracting image features, the horizontal feature parameters and vertical feature parameters of the image are extracted respectively.

Furthermore, the pixel parameters of the horizontal feature and the pixel parameters of the vertical feature of the image are divided into 9 regions, and each region block corresponds to a certain number of pixels.

Implement the present invention, have following beneficial effect:

The present invention extracts the image features of people by using continuous gait images, integrates the vertical pixel features and horizontal pixel features of the images during extraction, and extracts features from continuous images of people walking, taking the body state changes of men and women as walking The difference between them is used as a judgment parameter, which effectively improves the accuracy of gender recognition, and can also adapt to gender recognition of human gait images taken from different angles.

Description of drawings

Fig. 1 is a schematic flow chart of the present invention;

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

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

Fig. 4 is the schematic diagram of the principle of DEI processing of the present invention;

Fig. 5 is a schematic diagram of horizontal extraction features of the present invention;

FIG. 6 is a schematic diagram of the vertical extraction feature of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings.

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

S1: Obtain the gait image of the pedestrian;

S2: extracting the foreground image from the image of the pedestrian's gait image;

S3: Perform erosion and re-expansion processing on the image to remove noise and smooth the image, and then normalize the image;

The gait sequence image captured in the video is processed by binarization or background subtraction to obtain the foreground image, and then the image is normalized and then performed GEI (gait energy image) processing to obtain the difference of this sequence of gait.

S4 executes image processing for capturing gait energy on the image of S3;

S5: performing noise reduction processing on the image to remove noise;

S6: extracting features from the image processed in S5;

S7: Utilize the support vector machine SVM classifier to perform classification training on the image;

S8: Extract features from the image to be tested according to the steps of S1-S6, and then compare the obtained features with the features trained in S7.

Fig. 2 is a schematic diagram of the GEI image processing of the present invention, as shown in the figure, the gait frame picture obtained from the camera shooting, after binarization or background subtraction processing, the foreground image is taken out and normalized, and the normalized gait frame of the same sequence The GEI image is obtained through formula (1).

${\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; Ac</span>}}\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Among them, G _c (x, y) is the GEI image, N _c is the total number of gait sequence images, a gait sequence time length A _c , B(x, y, t) represents the image at time t, where x and y is the coordinate of the pixel in the image.

The principle example of its processing is shown in Figure 3. Assuming two normalized 3×3 images, the white block is the gait foreground image, and the black block is the background image. The total average of the two images can be converted into a GEI image . In the example, the pixel value of the GEI image is 1, which means there is no change between the first image and the second image, and the pixel value is 0.5, which means that there is a change between the first image and the second image, so you can see the characteristics of each person walking .

DEI is mainly used to remove the noise of GEI images, and at the same time, it can get the difference of gait between men and women. DEI is shown in Equation (2).

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; if</span>}{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; otherwise</span>}<span\; class="notranslate">\; ,</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

D _c (x, y) is the image after noise removal, G _c (x, y) is the GEI image, and U is a threshold value. Figure 4 is an example of DEI processing for a GEI image. For a 3×3 GEI image, the threshold value is the maximum pixel value in the GEI image (1 in the example) multiplied by the scaling factor 0.8. After the threshold value U=0.8 is obtained, the GEI image Each pixel value of is compared with U, when the GEI pixel value is greater than or equal to U, the DEI image value is equal to 1, and when it is less than U, the DEI image value is equal to 0. As shown in Figure 4, the DEI image on the right can be obtained after denoising the upper GEI image. The processed DEI image only retains the more important parts of the original GEI.

Referring to Fig. 5 and Fig. 6, the horizontal and vertical feature extraction examples are shown. Let _Xi represent the i-th feature value extracted from the DEI image, and the extraction method is as formula (3).

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 9</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; ik</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

When k=1～8, the c _ik formula is as follows:

c _ik ={b _{i((k-1)×18+l)} |l=1～18}

When k=9, the c _ik formula is as follows:

c _ik ={b _{i((k-1)×18+l)} |l=1～6}

${\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 9</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; kj</span>}}$

When k=1～8, the formula of d _kj is as follows:

d _kj ＝{b _{((k-1)×18+l)j} |l＝1～18}

When k=9, the formula of d _kj is as follows:

d _kj ＝{b _{((k-1)×18+l)j} |l＝1～6}

Where i represents the column of the image, j represents the row of the image, and k represents the block number. X _i represents the computed feature of each column, Y _j represents the computed feature of each row, c _ik represents the segmented block of each column, and d _kj represents the segmented block of each row. b _ij represents the image pixel position.

X _i is calculated from the pixels in the i-th column of the DEI image. The size of the DEI image in this paper is 150×150 pixels, so 1≤i≤150. X _i is calculated from the 150 pixels in the i-th column and 150 in the j row of the DEI Pixels are obtained by dividing each column and each row of pixels into 9 blocks. Among them, each block of c _i1 to c _i8 and d _1j to d _8j corresponds to 18 bits, and c _i9 and d _9j correspond to the last remaining 6 bits of the column, and the values of these 9 blocks are summed into the column and row The eigenvalues X _i , Y _j of . Therefore, a DEI image has 300 feature parameters as gender identification features.

The SVM used in the scheme of the present invention has three core functions for classification: Linear, Polynomial, and RadialBasis Function. The scheme uses the image data of the CASIA database as an experimental sample, and uses LIBSVM to train and test. The training data has 18 males and 14 females, according to the training model, tested the remaining 18 males and 13 females, and tested with two image sizes of 75×75 and 150×150. The test results are shown in Table 1 and Table 2.

Example 1

It can be seen from the data in Table 1 and Table 2 that the accuracy of vertical feature extraction combined with horizontal feature extraction will be much higher than that of vertical feature extraction or horizontal feature extraction alone.

Table 1

Table 2

This method can achieve 100% accuracy in gender identification when people walk at 90°. However, in daily life, people walk at different angles. The method in this paper is also tested at different angles, with 75× Taking an image with a size of 75 as an example, the average accuracy rate of vertical feature extraction is 63.64%, the average accuracy rate of horizontal feature extraction is 72.21%, and the average accuracy rate of horizontal + vertical feature extraction is 70.97%; For example, the average accuracy rate of vertical feature extraction is 72.43%, the average accuracy rate of horizontal feature extraction is 82.11%, and the average accuracy rate of horizontal + vertical feature extraction is 87.98%.

Example 2

The training data when the camera shooting angle is 90°, 0° and 180°, and the data to be tested take the deviation of 18° as an example, the measured accuracy is shown in Table 3

Table 3 Accuracy rate of gender identification in three angles of gait

The above description is a preferred embodiment of the present invention, and it should be pointed out that for those skilled in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications are also considered Be the 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