CN101630364A - Method for gait information processing and identity identification based on fusion feature - Google Patents

Abstract

The invention relates to the extraction, processing and identification of gait feature information when people are walking. In order to reduce the interference of external factors such as complex backgrounds, achieve better adaptability to actual conditions, and more accurately extract effective information that can reflect the walking characteristics of a moving human body, thereby improving the accuracy of gait recognition, the present invention adopts The technical solution is that the gait information processing and identity recognition method based on fusion features includes the following steps: the gait information processing and identity recognition method based on fusion features includes the following steps: input a video sequence, and segment the video image through target detection. The contour information of the human target, and then use the boundary center distance and Radon transform to extract the gait feature parameters at the same time, and then perform corresponding post-processing on the obtained feature parameters, and finally select the support vector machine as the classifier for classification and recognition, and recognize The effect is evaluated. The invention is mainly applied to identity recognition based on gait feature information.

Description

Gait Information Processing and Identification Method Based on Fusion Feature

technical field

The invention relates to the extraction and processing of gait feature information and identity recognition based on gait feature information when people are walking, in particular to the gait information processing and identity recognition method based on fusion features.

Background technique

Biometric identification is the identification of personal identity through various high-tech information detection methods and the use of the inherent physiological or behavioral characteristics of the human body. Biological characteristics mainly include physiological characteristics and behavioral characteristics: physiological characteristics refer to innate and congenital physical characteristics of the human body, such as fingerprints, irises, faces, etc.; behavioral characteristics refer to the characteristics extracted from the movements performed by people The characteristics that come out are mostly acquired, such as gait, handwriting and so on. In recent years, biometric authentication technology has experienced a leap-forward development and has become a hot spot widely concerned by industry, academia, research and management. In the MIT Technology Review magazine in 2001, biometric technology was listed as one of the 10 most likely technologies to change the world. It is expected that in the next 10 years or so, biometric authentication technology will penetrate into every aspect of our lives, and its comprehensive influence will be no less than that of the Internet.

Gait Recognition is one of the emerging fields in biometric technology. It aims to realize the identification of personal identity or the detection of physiological, pathological and psychological characteristics according to people's walking posture, and has broad application prospects. Gait is a complex behavioral feature, and it is a comprehensive reflection of human physiology, psychology and reaction to the outside world. Gaits vary due to inter-individual variability, these differences are a function of the overall musculature and skeleton (body weight, limb length, bone structure, etc.) and are entirely determined by hundreds of kinematic parameters. Early medical research shows that there are 24 different components in human gait. If these components are taken into account, gait is unique to the individual, which makes it possible to use gait for identification. Compared with other biometric authentication technologies, gait recognition has unique advantages such as non-invasiveness, long-distance recognition, simplified details, and difficulty in camouflage.

The existing gait recognition algorithms can be roughly divided into two categories: model-based and non-model-based. The model-based method refers to building a model of the human body structure or the obvious walking characteristics of the human body in the gait sequence images, and using the parameters derived from the model to extract the gait features. Its characteristic is that it can more accurately describe the gait characteristics and greatly reduce the sensitivity to changes in external conditions, but the huge amount of calculation is a difficult problem for practical applications that emphasize real-time performance. Based on the non-model method, it refers to the features extracted by directly analyzing the shape or movement of the human body during walking. Its characteristic is that the amount of calculation is relatively small, which is helpful to achieve the purpose of real-time calculation in practical links, but it is sensitive to changes in background and light signals. Once occlusion occurs in the scene, the recognition ability will be greatly affected.

Gait recognition combines multiple technologies such as computer vision, pattern recognition and video/image sequence processing. With the rapid development of biometric authentication technology, the identification technology based on gait features has increasingly shown its advantages, especially in the fields of access control system, security monitoring, human-computer interaction, medical diagnosis and other fields, which has broad application prospects and economic value. Therefore, it has aroused the strong interest of many researchers at home and abroad, and has become a frontier direction that has attracted much attention in the field of biomedical information detection in recent years.

However, no technology is perfect. Capturing images from the environment under realistic conditions will be affected by various factors (such as changes in climate conditions, changes in lighting conditions, background chaos, shadows of moving objects, and the distance between objects and the environment. Occurrence or occlusion between objects, camera movement, etc.), which brings many difficulties to the final recognition of gait. How to eliminate the influence of these factors and extract the effective gait features of the moving human body more accurately is a difficult problem in the field of gait recognition.

Contents of the invention

In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a gait information processing and identification method based on fusion features, to reduce the interference of external factors such as complex backgrounds, and to have better adaptability to actual conditions. More accurately extract effective information that can reflect the walking characteristics of a moving human body, so as to improve the accuracy of gait recognition.

The technical scheme adopted by the present invention is that the gait information processing and identity recognition method based on the fusion feature includes the following steps: input video sequence, segment the outline information of the human body target in the video image through target detection, and then calculate the boundary center distance, Radon Transformation is also used to extract gait feature parameters, and then corresponding post-processing is performed on the obtained feature parameters. Finally, support vector machine is selected as a classifier for classification and recognition, and the recognition effect is evaluated.

The described target detection segmented the contour information of the human target in the video image, i.e. moving target detection and key frame extraction, including the following steps:

(1) Minimum median variance method background modeling:

If let I _{(x, y)} ^t represent the collected N frame sequence images, where $(x,\mathrm{the\; y})\∈I_{(x,\mathrm{the\; y})}^t,$ t represents the frame index value (t=1, 2, ..., N), then the background B _{(x, y)} is:

${\mathrm{<span\; class="notranslate">\; B</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; min</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; med</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span>$

In the formula, P is the gray value to be determined at the pixel position (x, y), and the three components of R, G, and B are modeled separately, and a color background image in RGB format is obtained through synthesis;

(2) Motion Segmentation:

Use the indirect difference function to perform the difference operation:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \&times;</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}$

Where a and b represent the grayscale (intensity) levels of the current image and the background image at the same pixel (x, y) respectively, 0≤f(a,b)≤1, 0≤a, b≤255. The sensitivity of the difference function can be changed automatically with the gray level of the background. This adaptability improves the accuracy of image segmentation. After the difference, the binarized image of the moving target can be obtained by threshold segmentation:

${\mathrm{<span\; class="notranslate">\; S</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; T</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; ;</span>$

(3) Morphological processing and connected domain analysis:

Perform post-processing combining morphological processing and connected domain analysis on the binarized image of the moving target to remove residual noise and obtain better segmentation results;

(4) Division of gait cycle and key frame extraction:

Taking advantage of the feature that the width of the human body contour changes synchronously and periodically over time, the gait cycle is divided by the width change signal of the human body contour, and two maximum points in a gait cycle are extracted as key frames.

Described gait feature extraction comprises the following steps:

Firstly, the boundary tracking algorithm is used to extract the contour of the human body, and the necessary normalization and resampling are performed to obtain the contour of the moving human body. Then the contour-based boundary center distance feature and Radon transform feature are extracted. The boundary center distance feature refers to:

The original two-dimensional contour shape is indirectly represented by a one-dimensional distance signal D=(d ₁ , d ₂ ,...d ₂₅₆ ), and has translation invariance and rotation invariance, where,

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>$

The Radon transform feature refers to:

The essence of Radon transformation is the projection of the image matrix in the specified direction. The projection can be carried out along any angle. Generally, the Radon transformation of f(x, y) is the line integral in the direction parallel to the y-axis in the rotating coordinate system. , of the form:

${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \&Integral;</span>}_{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; \&infin;</span>}^{<span\; class="notranslate">\; \&infin;</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; dy</span>}}^{<span\; class="notranslate">\; \&prime;</span>}$

in: $\left[\begin{array}{c}{x}^{\&prime;}\\ {\mathrm{the\; y}}^{\&prime;}\end{array}\right]=\left[\begin{array}{cc}\cos \mathrm{\&theta;}& \sin \mathrm{\&theta;}\\ -\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}\end{array}\right]\left[\begin{array}{c}x\\ \mathrm{the\; y}\end{array}\right].$

The corresponding post-processing of the obtained feature parameters refers to the strategy of data dimensionality reduction and feature fusion, that is, principal component analysis (PCA) is applied to data dimensionality reduction.

When selecting the support vector machine as the classifier for classification and recognition, the "one-to-one" strategy is adopted, that is, one classifier completes two choices each time. This method combines two pairs of N-type training data to construct ${\mathrm{<figure-callout\; id="2"\; label="C\; N"\; filenames="A2009100701740014H3.png"\; state="\{\{state\}\}">C</figure-callout>}}_N^{}=N(N-1)/2$ a support vector machine. In the final classification, the voting method is adopted to determine the classification result: suppose there are m categories of gaits to be recognized, which are recorded as S ₁ , S ₂ , ..., S _m , and one of the samples S _ij is randomly selected from each category, where i is the category, j is the serial number of the sample in this class for training, and other samples S _it (j≠t) are used for testing. During the test, the test sample S _it is input into the trained classifier. If the output is i, then the The sample is judged as the i-th class; if the output is j, it is judged as a recognition error.

The present invention provides can bring following effect:

The present invention simultaneously uses boundary center distance and Radon transform to extract gait characteristic parameters, which can reflect most of the energy information of the original contour, including both appearance information of gait and dynamic information, and can effectively reduce various disturbances, such as Influenced by factors such as shadows, lighting, occlusion, and background chaos, and through feature fusion, it effectively makes up for the lack of a single feature, more accurately extracts effective information that can reflect the walking characteristics of a moving human body, and improves the accuracy of gait recognition.

The present invention applies the idea of Principal Component Analysis (PCA) to data dimensionality reduction, and can effectively reduce the dimensionality of data on the basis of retaining most of the original information, thus greatly reducing the amount of computation in the present invention, improving efficiency, reducing costs.

Description of drawings

Fig. 1 technical flow chart of the present invention.

Fig. 2 is a flow chart of moving target detection in the present invention.

Fig. 3 is a curve diagram of the normalized boundary center distance of the present invention.

Fig. 4 is a schematic diagram of Radon transformation coordinates of the present invention.

Fig. 5 is a Radon transformation eigenvector diagram of the present invention.

Figure 6 Cumulative matching score curves for different features.

Detailed ways

The invention cleverly introduces the idea of Radon transformation into it. The Radon transform is widely used in the line segment detection in the image, which just fits the characteristics that the leg is approximated as a line segment in a certain direction in the image contour, and a large angle change will occur relative to the horizontal axis during walking. This means that the characteristic parameters obtained by Radon transformation can reflect most of the energy information of the original contour, including both the appearance information of the gait and the dynamic information, which can effectively reduce the influence of self-occlusion and shadows. In addition, the contour-based boundary center distance feature is extracted and fused with the Radon transform feature, which effectively makes up for the lack of a single feature. The fusion features are finally input to Support Vector Machines (SVM) for gait classification and recognition.

The method for gait recognition using dual-feature fusion technology of the present invention involves key technologies including: video processing, image processing, pattern recognition, and the like. The technical process is as follows: for the input video sequence, the moving target in the video image is segmented through target detection, and then the contour of the moving human body is extracted by using the boundary tracking algorithm, and the contour is resampled and normalized; then the boundary center distance, Radon transform and other theories are used to extract gait feature parameters at the same time, and realize the fusion of the two features; finally, the support vector machine is selected as the classifier for classification and recognition. Compared with similar technologies, this method is more robust to changes in clothing, carrying items, occlusions, etc., and can effectively suppress the interference of external factors such as complex backgrounds. It provides a new method for exploring more reliable gait identification methods in the future. train of thought.

Below in conjunction with accompanying drawing and embodiment, further illustrate the present invention.

Gait recognition is mainly for the analysis and processing of moving image sequences containing people, usually including: human body detection, gait feature extraction, and identity recognition. Fig. 1 shows the technical flow chart of the present invention: the overall goal is to input a video sequence, segment the contour information of the human target in the video image through target detection, and then apply theories such as boundary center distance and Radon transformation to the gait characteristic parameters at the same time In order to find a gait recognition method with strong robustness to changes in clothing, carrying items, occlusions, etc., and perform corresponding post-processing on the feature data, finally select the support vector machine as a classifier for classification and recognition, and Evaluate the recognition effect.

1 Moving target detection and key frame extraction

To extract gait feature information, it involves the extraction of moving objects in complex backgrounds, which is the pre-processing of gait recognition. Because there are many kinds of interference in the actual application environment, such as shadow, light, occlusion and background chaos, etc., this puts forward higher requirements for the real-time and reliability of the algorithm. The specific process of moving target detection adopted by the present invention is shown in FIG. 2 .

(1) Minimum median variance method background modeling

The least median variance method (LmedS) is an algorithm proposed based on the theory of robust statistics.

If let I _{(x, y)} ^t represent the collected N frame sequence images, where $(x,\mathrm{the\; y})\&Element;I_{(x,\mathrm{the\; y})}^t,$ t represents the frame index value (t=1, 2, ..., N), then the background B _{(x, y)} is:

${\mathrm{<span\; class="notranslate">\; B</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; min</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; med</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

In the formula, P is the gray value to be determined at the pixel position (x, y), where med means to take the middle value, and min means to take the minimum value. The three components of R, G, and B are modeled separately, and a color background image in RGB format is obtained through synthesis.

The specific process of the algorithm is:

step[1] selects the pixel point position (x, y);

step[2] Let P=0;

step[3] sequentially calculate (I ¹ _{(x, y)} -P) ² , (I ² _{(x, y)} -P) ² ,..., (I ^N _{(x, y)} -P) ² ;

Step[4] sorts the calculation results. If N is an even number, take the average of the N/2th and (N+1)/2 numbers after sorting. If N is an odd number, take the N/2th number. The result Save to the array med, namely med ₀ ;

step[5]P=P+1, when P<=255, return to step[3], repeat step[3], [4], [5], save the result as med _P , otherwise execute step[6];

step[6] Find out the minimum value among med ₀ , med ₁ , ..., med ₂₅₅ , and the corresponding P is the background gray level of the pixel position;

Step[7] reselects the pixel position, returns to step[2] and repeats until all the pixels in the image are calculated.

(2) Motion Segmentation

Use the indirect difference function to perform the difference operation:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \&times;</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</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">\; 2</span>}<span\; class="notranslate">\; )</span>$

Where a and b represent the grayscale (intensity) levels of the current image and the background image at the same pixel point (x, y) respectively, 0≤f(a, b)≤1, 0≤a, b≤255. The sensitivity of the difference function can change automatically with the gray level of the background, and this adaptability improves the accuracy of image segmentation.

After the difference, the binarized image of the moving target can be obtained by threshold segmentation:

${\mathrm{<span\; class="notranslate">\; S</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; T</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

(3) Morphological processing and connected domain analysis

Due to the influence of weather, light, shadow and other external factors, there will inevitably be noise in the image after motion segmentation, and at the same time, a small number of points in the moving target will be misjudged as the background, so the image needs to be further processed to obtain the best Good segmentation effect. The present invention uses morphological filtering to eliminate noise in binary images and fill in the absence of moving objects. As a commonly used image noise filtering method, the most basic operation of morphology used in image filtering is dilation and erosion, and the combination of dilation and erosion derives two other operations: opening operation and closing operation. The open operation can smooth the convex contour of the object, disconnect the narrow connection, and remove the small protrusions; the closed operation can smooth the concave contour of the object, and connect the narrow and long gaps into slender bends. These properties can be used to achieve the purpose of filtering and filling holes.

After morphological filtering, there may still be some stray noise forming blocks of different sizes, and the real moving target is often the largest of these blocks. Therefore, the connected domain analysis is further carried out on the image, with the purpose of retaining only the moving objects in the image.

(4) Division of gait cycle and key frame extraction

Human walking is a periodic behavior. The gait cycle is defined as: the time from the heel strike to the heel strike of the same side leg again, including two stance periods and two swing periods. In order to improve efficiency, the present invention utilizes the feature that the width of the human body outline changes synchronously and periodically over time, divides the gait cycle through the width change signal of the human body outline, and extracts two maximum points in a gait cycle as key frames , thus simplifying the research process.

2 Gait feature extraction

How to extract effective gait feature parameters is the key to gait recognition. Since gait identification largely depends on the change of human body contour shape over time, the two gait features selected in the present invention are both contour-based. The essence of moving human body contour extraction is boundary tracking, and necessary normalization and resampling to obtain the moving human body contour.

2.1 Boundary Center Distance Features

The boundary center distance is defined as the distance from the boundary point to the centroid. In this way, the original two-dimensional contour shape is represented indirectly by the one-dimensional distance signal D=(d ₁ , d ₂ , . . . d ₂₅₅ ), and has translation invariance and rotation invariance.

in,

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ;</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

Figure 3 shows the normalized boundary center distance curve. The peaks and troughs reflect the protrusion and depression of the contour. The amplitude of the peaks and troughs has a certain relationship with the anatomical structure of the human body and walking posture. Numbers are closely related to human posture. It is because of these properties that we can use it to distinguish different objects.

2.2 Radon Transform Features

Radon transform has superposition, linearity, scaling, delay and rotation invariance, and is widely used for line segment detection in images. The essence of Radon transformation is the projection of the image matrix in the specified direction. The projection can be carried out along any angle. Generally, the Radon transformation of f(x, y) is the line integral in the direction parallel to the y-axis in the rotating coordinate system. , of the form:

${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \&Integral;</span>}_{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; \&infin;</span>}^{<span\; class="notranslate">\; \&infin;</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; dy</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

in: $\left[\begin{array}{c}{x}^{\&prime;}\\ {\mathrm{the\; y}}^{\&prime;}\end{array}\right]=\left[\begin{array}{cc}\cos \mathrm{\&theta;}& \sin \mathrm{\&theta;}\\ -\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}\end{array}\right]\left[\begin{array}{c}x\\ \mathrm{the\; y}\end{array}\right].---\left(6\right)$

The leg is approximated as a line segment in a certain direction in the image outline, and its value in the vertical direction is larger after Radon transformation; in the process of walking, the leg will have a large angle change relative to the horizontal axis. It means that the characteristic parameters obtained by Radon transformation can reflect most of the energy information of the original contour, and these parameters change significantly over time, which is manifested as the swing of the lower limbs. Therefore, by learning and analyzing these parameters, important information about individual morphology and gait can be obtained.

The features extracted by image Radon transform include not only appearance information of gait, but also dynamic information, which can effectively reduce the influence of self-occlusion and shadow. Another significant advantage of this algorithm is that each Radon transformation parameter includes the collective contribution of many pixels, so it is not easily affected by the pseudo-pixel interference of the original contour image.

Figure 4 depicts the relationship between the angle information after Radon transformation and the angle of the lower limbs. When the integral direction is perpendicular to the line segment of the thigh, the integral value is the largest. Since the angle between the thigh and the vertical direction changes in the range of 0° to 60° during the walking process of a person, and the pixels in other directions in the silhouette image are mostly noises such as shadows, so the present invention only detects the image in the range of 0° to 60°. Radon transformation is performed in the directions of 60°, 120°~180°, and the peak values of the angles on the two intervals are obtained respectively, and the two are combined to obtain the feature vector, as shown in Figure 5. The feature information extracted in this way is not affected by the self-occlusion of the body, and effectively reduces the amount of calculation. Compared with other modeling methods, it is simple and fast.

3 Data Dimensionality Reduction and Feature Fusion Strategy

In order to reduce the amount of computation and eliminate redundant information, the present invention applies the idea of Principal Component Analysis (PCA) to data dimensionality reduction, which can effectively reduce the dimensionality of data on the basis of retaining most of the original information. The specific steps of the PCA dimensionality reduction process can be summarized as follows:

(1) Raw data standardization:

In order to eliminate the impact of different dimensions and orders of magnitude among data, it is necessary to standardize the original data to make them comparable. The standardized method of the present invention is: each element in the matrix subtracts the mean value of the column in which it is located, and then divides it by the standard deviation of the column in which it is located, so that each variable can be standardized as a matrix X with a mean value of 0 and a variance of 1, namely:

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; ]</span>}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

in,

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; p</span>}$

$\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\sqrt{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; A</span>}}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

(2) Calculate the correlation coefficient matrix:

$\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccc}{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}& {\mathrm{<span\; class="notranslate">\; <figure-callout\; id="12"\; label="r"\; filenames="A2009100701740014H3.png"\; state="\{\{state\}\}">r</figure-callout></span>}}_{\mathrm{<span\; class="notranslate">\; 12</span>}}& <span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span>& {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; p</span>}}\\ {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; one</span>}}& {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; twenty\; two</span>}}& <span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span>& {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; p</span>}}\\ <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>\\ <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>\\ <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>\\ {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}& {\mathrm{<span\; class="notranslate">\; <figure-callout\; id="2"\; label="r\; p"\; filenames="A2009100701740014H3.png"\; state="\{\{state\}\}">r</figure-callout></span><figure-callout\; id="2"\; label="r\; p"\; filenames="A2009100701740014H3.png"\; state="\{\{state\}\}">\; </figure-callout>}}_{\mathrm{<span\; class="notranslate">\; </span>}}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span>& {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; pp</span>}}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$

R is a real symmetric matrix (ie r _ij =r _ji ), where r _ij (i, j=1, 2, . . . , p) is the correlation coefficient of standardized variables X _i , X _j . It is defined as the variable covariance divided by the standard deviation (variance) of the variable, calculated as:

${\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; the\; ki</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; kj</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; )</span>}{\sqrt{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; the\; ki</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; kj</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

In the formula: $\stackrel{\&OverBar;}{x_i}=\frac{1}{\mathrm{no}}{\mathrm{\&Sigma;}}_{k=1}^{\mathrm{no}}{x}_{\mathrm{the\; ki}},$ $\stackrel{\&OverBar;}{x_j}=\frac{1}{\mathrm{no}}{\mathrm{\&Sigma;}}_{k=1}^{\mathrm{no}}{x}_{\mathrm{kj}}$ Respectively represent the mean value of each vector in the X _i and X _j columns in the original matrix.

(3) Eigen decomposition, finding eigenvalues and eigenvectors:

Solve the characteristic equation |R-λE|=0, find the eigenvalue _λi (i=1, 2,...p) of the correlation coefficient matrix R, and arrange it in descending order, That is, λ ₁ ≥λ ₂ ≥...≥λ _P ; then obtain the eigenvectors U _i (i=1, 2,...p) corresponding to each eigenvalue λ _i , respectively.

(4) Determine the principal component through the cumulative contribution rate:

The formula for calculating the cumulative contribution rate is:

${\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; l</span>}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; p</span>}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

When the cumulative contribution rate reaches a certain threshold (the present invention takes 85%), all the first m eigenvalues λ ₁ , λ ₂ ,...≥λ _m (m≤p) and their corresponding eigenvectors are retained down, as the main component, and the rest are discarded.

(5) Calculate the score matrix

The eigenvectors U=U ₁ , U ₂ , _. Calculate the scoring matrix using the following formula:

F _(n×m) = X _(n×p) U _(p×m) (12)

Among them, X is the original data matrix, U is the principal component load, and the score matrix F is the final result obtained after PCA dimensionality reduction. Each row of it is equivalent to the projection of all the rows of the original data matrix (that is, the vector composed of the original variables) on the principal component coordinate axis (load axis), and the vector formed by these new projections is the principal component score vector.

It can be seen from the above steps that the PCA algorithm approximately reflects all the information of the original data array through several maximum principal component scores. This not only achieves the purpose of dimension reduction, but also greatly reduces the correlation between data, so that the data can be optimized and reorganized.

Single gait features are often not stable enough and not robust enough to provide enough information for recognition. In view of this, the present invention tries the idea of multi-gait feature fusion, and fuses the two kinds of gait features extracted above—the boundary center feature and the Radon transformation feature at the feature layer. For the same gait sequence, different algorithms will be used respectively The extracted boundary center distance feature is fused with the Radon transformation parameter feature, which belongs to the fusion of the feature layer. The process of fusion is essentially splicing the above two extracted features together and merging them into a feature vector, so as to obtain more accurate, complete and meaningful information than any single feature, and then send this fusion feature into the classifier for classification. Classification recognition to obtain improvement of recognition effect.

4 Gait recognition based on support vector machine

Traditional statistical pattern recognition is carried out on the premise that the number of samples is sufficient, and its performance can be guaranteed theoretically only when the number of samples tends to infinity. In the application of gait recognition, the number of samples is limited, and it is difficult for many methods to achieve ideal results. Support vector machines (Support Vector Machines, SVM) model through the principle of structural risk minimization, which minimizes the expected risk and significantly improves its model recognition. This recognition method shows many unique advantages in solving small sample, nonlinear and high-dimensional pattern recognition problems. The main idea of the support vector machine is to find an optimal classification hyperplane that meets the classification requirements, so that the hyperplane can maximize the interval of the hyperplane while ensuring the classification accuracy. In theory, support vector machines can achieve the optimal classification of linearly separable data.

Gait recognition is a multi-category classification problem, and the support vector machine method is proposed for the classification of two categories, which cannot be directly applied to multi-class classification problems. For multi-class pattern recognition problems, the support vector machine method can be realized through the combination of two kinds of problems. The present invention adopts a "one-to-one" strategy, that is, one classifier completes the selection of two each time, and the method combines two pairs of N types of training data to construct ${\mathrm{<figure-callout\; id="2"\; label="C\; N"\; filenames="A2009100701740014H3.png"\; state="\{\{state\}\}">C</figure-callout>}}_N^{}=N(N-1)/2$ a support vector machine. In the final classification, the "voting" method is adopted to determine the classification results. Assuming that there are m categories of gaits to be recognized, denoted as S ₁ , S ₂ , ..., S _m , one of the samples S _ij is randomly selected from each category (where i is the category, and j is the sample number in this category) for For training, other samples S _it (j≠t) are used for testing. During the test, the test sample S _{it is} input into the trained classifier, if the output is i, the sample is judged as the i-th class; if the output is j, it is judged as a recognition error.

Beneficial effect

The experimental data of the present invention comes from Dataset B in the CASIA gait database, from which 20 people are extracted, only considering the viewing angle of 90°, divided into two states: natural walking and backpack walking. Each person is regarded as a class, and each class contains multiple samples, including 6 gait cycles of normal walking and 4 gait cycles of backpacking.

In order to evaluate the recognition results more comprehensively and effectively, two evaluation methods, Probability of Correct Recognition (PCR) and Cumulative Match Scores (CMS), are used together as evaluation indicators.

The correct recognition rate is defined as:

PCR＝T/N×100% (7)

In the formula, T is the number of correctly identified samples, and N is the total number of test samples. During the test, input T _i into the trained classifier, if the output is i, it is determined that this recognition is correct; if the output is j (i≠j), it is determined that this recognition is wrong. The statistical results of the correct recognition rate are shown in Table 1.

The cumulative matching score is defined as the cumulative probability p(k) of a test metric's actual class among its top k matching values. Through the evaluation of the accumulated matching scores, not only the recognition rate with the highest matching degree can be obtained, but also the data recognition rate with the top n places in the matching degree can be obtained, so that a more comprehensive evaluation of the recognition system can be obtained. Figure 6 shows the cumulative matching score curves of different gait features in the normal walking state, which also reflects the convergence speed of the algorithm from another angle.

Table 1 Statistics of the recognition results of 20 subjects

Experiments show that both the boundary center distance feature and the Radon transform feature are more effective gait features, which contain rich contour information, can comprehensively reflect human gait information, and show a good recognition effect;

Using the fusion features after the fusion of the two for recognition can significantly improve the recognition performance and expand new ideas for the exploration of multi-feature fusion gait recognition.

The invention proposes a new gait recognition method, which effectively fuses the contour-based boundary center distance feature and Radon transformation feature to reduce the interference of external factors such as complex backgrounds, and has better adaptability to actual conditions, and is more In order to accurately extract effective information that can reflect the walking characteristics of a moving human body, and improve the accuracy of gait recognition.

The invention can provide help for the effective use of the monitoring system and the reliable evaluation of the monitoring effect, and can obtain considerable social benefits and the improvement of public security services. And it can be integrated and applied in the security access control system, so that the physical channel control management of the monitored area can reach a higher security level and create a safer and more harmonious social living environment. The best implementation plan is to use patent transfer, technical cooperation or product development.

Claims (5)

1. a kind of gait information processing and identity recognition method based on fusion feature, it is characterized in that, comprise the following steps: input video sequence, segment out the outline information of human target in video image by target detection, then boundary center distance, Radon Transformation is also used to extract gait feature parameters, and then corresponding post-processing is performed on the obtained feature parameters. Finally, support vector machine is selected as a classifier for classification and recognition, and the recognition effect is evaluated. 2. a kind of gait information processing and identification method based on fusion feature according to claim 1, it is characterized in that, described object detection is segmented out the contour information of human target in video image, namely moving object detection and key Frame extraction, including the following steps: (1) Minimum median variance method background modeling: If let I _{(x, y)} ^t represent the collected N frame sequence images, where $(x,\mathrm{the\; y})\&Element;I_{(x,\mathrm{the\; y})}^t,$ t represents the frame index value t=1, 2, ..., N, then the background B _{(x, y)} is: ${\mathrm{<span\; class="notranslate">\; B</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; min</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; med</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span>$ In the formula, P is the gray value to be determined at the pixel position (x, y), med means to take the middle value, and min means to take the minimum value. The three components of R, G, and B are modeled separately, and the RGB format is obtained through synthesis. colored background image; (2) Motion Segmentation: Use the indirect difference function to perform the difference operation: $\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \&times;</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 256</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}$ Among them, a and b respectively represent the gray level of the current image and the background image at the same pixel point (x, y), that is, the intensity level, 0≤f(a, b)≤1, 0≤a, b≤255, the difference function The sensitivity can be changed automatically with the gray level of the background. After the difference, the binarized image of the moving target can be obtained by threshold segmentation: ${\mathrm{<span\; class="notranslate">\; S</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; T</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; otherwise</span>}\end{array}\right.<span\; class="notranslate">\; ;</span>$ (3) Morphological processing and connected domain analysis: Perform post-processing combining morphological processing and connected domain analysis on the binarized images of moving targets to remove residual noise and obtain better segmentation results; (4) Division of gait cycle and key frame extraction: Taking advantage of the feature that the width of the human body contour changes synchronously and periodically over time, the gait cycle is divided by the width change signal of the human body contour, and two maximum points in a gait cycle are extracted as key frames. 3. a kind of gait information processing and identification method based on fusion feature according to claim 1, it is characterized in that, described gait feature extraction comprises the following steps: adopt boundary center distance feature, Radon transformation feature pair simultaneously The contour line of the moving human body is extracted, that is, boundary tracking, and necessary normalization and resampling are performed to obtain the contour of the moving human body. The feature of the center distance of the boundary refers to: The original two-dimensional contour shape is represented indirectly by a one-dimensional distance signal D=(d ₁ , d ₂ ,...,d ₂₅₆ ), and has translation invariance and rotation invariance, where, $d_i=\sqrt{{(x_i-x_c)}^2+{({\mathrm{the\; y}}_i-{\mathrm{the\; y}}_c)}^2}$ (i=1, 2, ... 256), The Radon transform feature refers to: The essence of Radon transformation is the projection of the image matrix in the specified direction. The projection can be carried out along any angle. Generally, the Radon transformation of f(x, y) is the line integral in the direction parallel to the y-axis in the rotating coordinate system. , of the form: ${\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; \&Integral;</span>}_{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; \&infin;</span>}^{<span\; class="notranslate">\; \&infin;</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; dy</span>}}^{<span\; class="notranslate">\; \&prime;</span>}$ in: $\left[\begin{array}{c}{x}^{\&prime;}\\ {\mathrm{the\; y}}^{\&prime;}\end{array}\right]=\left[\begin{array}{cc}\cos \mathrm{\&theta;}& \sin \mathrm{\&theta;}\\ -\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}\end{array}\right]\left[\begin{array}{c}x\\ \mathrm{the\; y}\end{array}\right].$ 4. a kind of gait information processing and identification method based on fusion feature according to claim 1, it is characterized in that, described then carrying out corresponding post-processing to the feature parameter that obtains refers to data dimensionality reduction and feature fusion The strategy is to apply principal component analysis (PCA) to data dimensionality reduction. 5. a kind of gait information processing and identification method based on fusion feature according to claim 1, it is characterized in that, select support vector machine for use as classifier to carry out classification identification, adopt " one to one " strategy, i.e. one The classifier selects one of the two each time, and this method combines two pairs of N types of training data to construct $C_N^2=N(N-1)/2$ In the final classification, voting is used to determine the classification result: Suppose there are m categories of gaits to be recognized, which are denoted as S ₁ , S ₂ ,..., S _m , and one of the samples S _ij is randomly selected from each category, Where i is the category, j is the serial number of the sample in this class, for training, other samples S _it (j≠t) are used for testing, when testing, input the test sample S _it into the classifier obtained after training, if the output If it is i, the sample is judged as the i-th class; if the output is j, it is judged as a recognition error.

Images