CN101251894A - Gait feature extraction method and gait recognition method based on infrared thermal imaging - Google Patents

Abstract

The invention belongs to the technical field of computer image processing, relates to a gait feature extraction method and a gait recognition method using the feature extraction method. The present invention uses an infrared thermal imager to collect gait image sequences and segment human targets, then normalizes and superimposes them to obtain gait feature maps, and then uses wavelet decomposition, invariant moments and combined with skeleton theory to extract gait feature parameters to obtain This is used as a gait feature recognition parameter input to the support vector machine for classification and recognition. The gait feature extraction method and recognition method provided by the invention can effectively detect human objects, eliminate the influence of changing factors such as background and illumination, and improve the correct recognition rate of gait.

Description

Gait feature extraction method and gait recognition method based on infrared thermal imaging

technical field

The invention belongs to the technical field of computer image processing, and in particular relates to a gait feature extraction method and a gait recognition method using the feature extraction method.

Background technique

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 recognition has unique advantages that other biometric authentication technologies do not have, that is, the recognition potential in long-distance or low video quality situations, and the gait is difficult to hide or camouflage, etc. However, in practical applications, due to the existence of many objective factors, it brings many difficulties to the final recognition of gait. How to recognize gait features more accurately is a difficult problem in the field of gait recognition. The uncertainty in the acquisition process of the gait image sequence makes the gait recognition process bound to be disturbed by various external factors, which makes it very difficult to detect human objects in complex backgrounds or under different lighting conditions. How to eliminate the influence of these factors and accurately extract the target features of the moving human body becomes the key to gait feature extraction and subsequent processing.

Contents of the invention

The gist of the present invention is to overcome the above-mentioned defects of the prior art, propose a method that can eliminate the interference of external factors such as complex backgrounds, and more accurately extract the characteristics of moving human body targets, and based on this method, further propose a method that can A gait recognition method that improves the correct gait recognition rate. By adopting the feature extraction method and gait recognition method provided by the present invention, the identification of personnel entering and leaving an important and sensitive place can be carried out, and day and night monitoring can be realized, so that the physical channel control and management in this area can reach a higher security level, and at the same time provide It brings safe and convenient identity authentication function to the person, and prevents the unauthorized access of the imposter to the greatest extent.

The method for extracting gait characteristic parameters based on infrared thermal imaging proposed by the present invention comprises the following steps:

(1) Using the human body as a radiation source, using an infrared thermal imager to collect human body gait image sequences;

(2) Segmenting human body contours to the collected gait image sequence;

(3) Normalize and superimpose the human body contour sequence to obtain a gait feature map containing the overall analysis model information;

(4) extracting the boundary moment feature parameters of the gait feature map;

(5) Carry out wavelet decomposition to the gait feature map to extract horizontal, vertical and diagonal moment invariant feature parameters respectively;

(6) Skeletonize the gait feature map, and extract the skeleton feature parameters including the limb proportion and angle information of the simplified human body model information;

As a preferred embodiment, threshold segmentation and dilation and erosion methods of binary images can be used to segment human body contours; mathematical morphology methods can be used to perform skeleton processing on gait feature maps to extract skeleton feature parameters.

Based on the above-mentioned feature extraction method, the present invention also proposes a gait recognition method based on infrared thermal imaging, which includes the following steps:

(1) Use an infrared thermal imager to collect infrared gait image sequences of persons to be registered;

(2) the infrared gait image sequence of collecting is stored in the gait sequence database;

(3) Perform the following feature extraction processing on the collected infrared gait image sequence: segment the human body contour, normalize it and superimpose it, obtain the gait feature map containing the overall analysis model information; extract the boundary moments of the gait feature map Feature parameters; perform wavelet decomposition on the gait feature map to extract horizontal, vertical, and diagonal moment invariant feature parameters; process the gait feature map into a skeleton, and extract the skeleton containing the limb proportion and angle information of the simplified human model information Characteristic Parameters;

(4) input the combination of characteristic parameters of each person who needs to be registered into the sample training library for the training of the support vector machine;

(5) Install the thermal imaging camera in the place where the monitoring personnel need to enter and exit;

(6) Utilize the infrared thermal imager to collect the gait image sequence of the personnel entering and leaving the place;

(7) adopting the method of step (3) to carry out feature extraction process respectively to the gait image sequence of the gathered personnel entering and leaving this occasion, obtain the invariant moment parameter and skeleton characteristic parameter of personnel entering and exiting this occasion;

(8) Input the combination of characteristic parameters of the personnel entering and leaving the place into the support vector machine for gait classification and recognition, and determine whether they are registered personnel.

As a preferred embodiment, in step (3), threshold segmentation and dilation and erosion methods of binary images can be used to segment human body contours; mathematical morphology methods can be used to skeletonize the gait feature map to extract skeleton feature parameters.

Based on the principle of infrared radiation, the present invention uses the human body as the radiation source, uses infrared thermal imaging technology to obtain human body gait images, and converts invisible body surface temperature into visible and quantifiable infrared thermal images, which can effectively eliminate complex backgrounds, The impact of external interference factors such as illumination changes can more accurately extract the essential structural features of moving human targets, and realize more accurate identification; the present invention also organically combines the overall model and the simplified model, and then uses wavelet decomposition (Wavelet Decomposition, WD), skeleton theory (Skeleton Theory) and invariant moments (InvariantMoments, IMs) combined method to extract gait characteristic parameters, input to Support Vector Machines (Support VectorMachines, SVM) for gait classification recognition, the recognition method Associating gait recognition with human movement patterns in sequence images reflects current changes and enables estimation of past and future changes. The identification method of the present invention can be applied to the security access control system to carry out all-weather and long-distance reliable and accurate identification of important and sensitive places, so that the physical channel control and management of the monitored area can achieve a higher security level and create a more secure environment. Harmonious social living environment.

Description of drawings

Fig. 1 is a general flowchart of the gait recognition method of the present invention.

Figure 2 is an infrared gait image collected by an infrared thermal imager.

Figure 3 realizes the infrared gait image of human body contour segmentation.

Figure 4 is the gait feature map obtained after processing the gait image sequence.

Fig. 5 Schematic diagram of the key points for skeletonizing the gait feature map.

Detailed ways

The human body is a natural source of biological infrared radiation. Generally, the infrared light emitted by the human body is strong, and the infrared radiation from non-heating objects such as the background is relatively weak. Therefore, it is easier to detect human targets in the gait images obtained by using infrared thermal imaging technology, and the surrounding environment is not easy to affect the gait images. And can realize day and night monitoring. The invention is based on the principle of infrared radiation, uses the human body as the radiation source, uses an advanced infrared thermal imager to collect gait image sequences, converts the invisible body surface temperature into a visible and quantifiable infrared thermal image, and proposes a step State information processing and pattern recognition methods.

Two kinds of human body models are usually used in gait recognition: the whole model and the simplified model, but the single model has many defects. Combining the two models organically, gait recognition can be associated with the movement patterns of people in sequence images, which can reflect current changes and estimate past and future changes. Therefore, the present invention organically combines the overall model with the simplified model, and then uses the method of combining wavelet decomposition (Wavelet Decomposition, WD), skeleton theory (Skeleton Theory) and invariant moments (Invariant Moments, IMs) to extract characteristic parameters, and input them to Support vector machines (Support Vector Machines, SVM) are used for gait classification and recognition.

The present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

FIG. 1 is a general flowchart of the gait recognition method of the present invention. In this embodiment, the gait image sequence used to establish the gait sequence database is a gait image sequence collected by an infrared camera. In practical applications, digital video cameras can also be used to collect gait image sequences. In the present invention, the extraction of the feature parameters of the gait image sequence stored in the gait sequence database is the same as the feature parameter extraction method of the infrared gait image sequence collected in the application, that is, the following methods are used: first, the human body target is segmented , and then normalized and superimposed to obtain the gait feature map, and then use wavelet decomposition, invariant moment and combined with skeleton theory to extract gait feature parameters.

The temperature of the human target is usually higher than that of the surrounding environment, and there are differences in the distribution of the radiation energy spectrum. The target information carried by this radiation difference is converted into a corresponding electrical signal by an infrared detector (the core device of an infrared thermal imager), and then after signal processing, a thermal image of the temperature distribution on the surface of the measured human body is displayed, thereby realizing the human body temperature distribution. Accurate quantification of target thermal radiation. In the present invention, after the infrared gait image sequence is collected by the infrared thermal imager, the outline of the human body in the video is segmented through target detection, and then normalized and overlaid to obtain the gait feature map including the overall analysis model. In view of the fact that the wavelet decomposition has multi-resolution and multi-direction, and the invariant moment has the invariance of the translation, rotation and size ratio of the target, the present invention decomposes the gait feature map to one layer of wavelet to extract horizontal, vertical and diagonal angles respectively. The moment invariant parameter for the direction. Since the skeleton is an effective way to describe the shape of the object, it combines the outline and area information of the target, and reflects the important visual clues of the target. Therefore, the mathematical morphology algorithm is used to skeletonize the gait feature map and extract the simplified model information of the human body. The characteristic parameters of the limb proportion and angle information. The combination of feature parameters including the overall model information and the simplified model information is input into the support vector machine for gait classification and recognition. For those who need to register, the image data collected by the infrared thermal imager is directly stored in the infrared gait sequence database, and the gait characteristic parameters are extracted using the above feature extraction method, and then the code is stored in the sample training library for use in the support vector machine. train. The feature extraction method and gait recognition method of the present invention will be described in more detail below.

1. Acquisition of Infrared Gait Image Sequence

When the human target passes through the "ecological tunnel", the infrared thermal imaging camera can collect the infrared gait image sequence, as shown in Figure 2. Since the infrared light emitted by non-heating objects such as the background is very weak, and the temperature difference with the human body is large, the gray value of the target and the background is different in the image. Therefore, the interference information of the external background in the infrared gait image can be filtered out by setting a threshold.

2. Infrared human body contour segmentation

Image segmentation is a fundamental part of computer vision and image understanding. In gait recognition, image segmentation is an important means to obtain target features and is the basis of recognition. In the infrared gait data acquisition, the temperature of the background is lower than the temperature of the human body, so there is an obvious grayscale difference between the background and the target in the image. In this way, the infrared human target can be extracted by setting the global threshold, and the calculation formula is:

${\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& <span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; \&tau;</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& <span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; \&tau;</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

In the formula, P _i (x, y) is the gray value at the pixel point (x, y), and τ is the threshold. Effective segmentation of infrared human targets can be achieved through the above formula, as shown in Figure 3. It can be seen from the figure that the infrared gait image sequence can detect the moving target very well, and can accurately segment the human body contour from the grayscale image, while maintaining good edges. Experiments show that the threshold segmentation method is simple to calculate, and the amount of calculation is small, and it meets the requirements of real-time performance. Considering the influence of voids, gaps and other interference factors that may exist in the human body contour, the expansion and erosion method of the binary image in morphology is used to filter out the noise and tiny interference areas in the human body contour image. While maintaining good image edges.

Image normalization is a common approach in image understanding systems. The purpose of image normalization is to adjust the position and size of the image to a fixed level, so as to eliminate the error information caused by the distance between the infrared thermal imager and the target or the shaking of the instrument and equipment, and provide a more unified image for subsequent processing. image specifications. After obtaining the outline of the human body, the outline of the human body is translated and centered, and then stretched according to the specified image size, and the image size is 150×150 pixels. In this way, human body contour images with the same position and uniform pixels can be obtained. These 2D image sequences are superimposed to obtain a gait feature map, as shown in Figure 4.

3 Extraction of Gait Feature Parameters

The key to gait recognition is how to find suitable gait features and corresponding effective classification methods. Wavelet Decomposition (Wavelet Decomposition, WD) has multi-resolution and multi-direction, and invariant moments (Invariant Moments, IMs) have the invariance of the translation, rotation and size ratio of the target, so the present invention regards the gait sequence as composed of A group of "static poses" consists of a pattern, the human body contour in the video is segmented through target detection, and then normalized and superimposed to obtain the gait feature map containing the overall analysis model, and then the gait feature map is wavelet One layer of decomposition extracts horizontal, vertical and diagonal invariant moment parameters and 7 boundary moment parameters of the gait feature map, so 4×7 moment parameters can be obtained. Because the resulting invariant moment has a large dynamic range, the logarithm logφ _i is taken for the above invariant moment parameter; in addition, a positive value is taken for convenience.

In order to extract the characteristic parameters containing the information of the simplified model of the human body, the present invention uses a mathematical morphology algorithm to perform skeletonization processing on the gait characteristic map, and extracts the characteristic parameters containing the proportion and angle information of the human limbs, as shown in FIG. 5 . According to the coordinate positions of the key points shown in Figure 5, five characteristic parameters containing the simplified model information of the human body can be calculated:

${\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\frac{\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; =</span>\frac{\sqrt{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; kr</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; kr</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}{\mathrm{<span\; class="notranslate">\; L</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; kr</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; kr</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

Combining the feature parameters containing two types of human body model information, a total of 33 moment parameters and skeleton parameters can be obtained, which are used as gait recognition parameters.

4 Classifier training and recognition

The choice of classifier is also a key link in gait recognition. 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 (SVM) show many unique advantages in solving small sample, nonlinear and high-dimensional pattern recognition problems, so support vector machines are used as classifiers.

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-category pattern recognition problems, the support vector machine method can be realized through the combination of two types of problems. The invention adopts a "one-to-one" strategy, that is, one classifier completes two choices at a time. Two combinations, build $C_N^2=N(N-1)/2$ a support vector machine. In the final classification, the "voting" method is adopted to determine the classification result. Assuming that m types of gaits have been registered in the sample library, denoted as S ₁ , S ₂ , ..., S _m , these samples are input into the support vector machine for training, and the corresponding output is 1-m. When classifying and identifying, input the obtained samples into the trained classifier, if the output value is between 1-m, then it is determined that the person is a registered person, otherwise it is determined as an unauthorized person.

5 recognition effect

23 subjects carried out experiments using the gait recognition method designed by the present invention. The subjects were all young students (13 males and 10 females, aged 19-30 years old), wearing soft-soled shoes, divided into four states: natural state, holding a ball, backpack, wearing down jacket, each person collected infrared steps There are 4 sequences of state images, a total of 92 sequences. To evaluate the recognition and classification results, the Probability of Correct Recognition (PCR) is used as an evaluation index. During the test, input T _i into the trained classifier, if the output is i, it is considered that the recognition is correct, and the correct recognition rate is:

PCR＝T/N×100% (7)

In the formula, PCR is the correct recognition rate, T is the number of correctly recognized samples, and N is the total number of test samples. If the output corresponding to the input sample T _i is j (i≠j), it is determined that the recognition is wrong this time.

The form of the kernel function determines the spatial structure of the training samples to be mapped and pattern classification, and forms different SVM algorithms. To this end, it is necessary to try different forms of kernel functions for optimization. There are three main types of commonly used SVM kernel functions:

1) Linear kernel function (Linear kernel), K(x, x _i )=x x _i (8)

2) Polynomial kernel function (Polynomial kernel), K(x, x _i )=[(x x _i +1)] ^q (9)

3) Radial Basis Function kernel

K(x, x _i )＝exp{-|xx _i | ² /2σ ² } (10)

The above-mentioned three kinds of kernel functions were used to repeat the experiment for 5 times, and the average value of the correct recognition rate is shown in Table 1. From the recognition results, the recognition rate is low when the support vector machine uses the linear kernel function, and the correct recognition rate is below 80%; while the recognition rate is improved when the polynomial kernel function is used, but the recognition rate is the highest when the radial basis kernel is used. good. It shows that support vector machine recognition can deal with the matching problem of multivariate time-varying data well.

The experimental data of 23 subjects shows that the correct recognition rate of gait recognition using infrared imaging technology is not significantly affected by foreign objects carried by the human body (such as backpacks, holding balls), and the invention can effectively eliminate complex backgrounds, The influence of external disturbance factors such as light changes. It is expected to be integrated into the security access control system for all-weather and long-distance reliable and accurate identification in important and sensitive places.

Claims (6)

1. A method for extracting gait feature parameters based on infrared thermal imaging, comprising the following steps: (1) Using the human body as a radiation source, using an infrared thermal imager to collect human body gait image sequences; (2) Segmenting human body contours to the collected gait image sequence; (3) Normalize and superimpose the human body contour sequence to obtain a gait feature map containing the overall analysis model information; (4) extracting the boundary moment feature parameters of the gait feature map; (5) Carry out wavelet decomposition to the gait feature map to extract horizontal, vertical and diagonal moment invariant feature parameters respectively; (6) Skeletonize the gait feature map, and extract the skeleton feature parameters including the limb proportion and angle information of the simplified human body model information; 2. The method for extracting gait characteristic parameters based on infrared thermal imaging according to claim 1, characterized in that, adopting threshold segmentation and expansion and erosion methods of binary images to segment human body contours. 3. The method for extracting gait characteristic parameters based on infrared thermal imaging according to claim 1, characterized in that, the gait characteristic map is skeletonized to extract the skeleton characteristic parameters using a mathematical morphology method. 4. A gait recognition method based on infrared thermal imaging, comprising the following steps: (1) Use an infrared thermal imager to collect infrared gait image sequences of persons to be registered; (2) the infrared gait image sequence of collecting is stored in the gait sequence database; (3) Perform the following feature extraction processing on the collected infrared gait image sequence: segment the human body contour, normalize it and superimpose it, obtain the gait feature map containing the overall analysis model information; extract the boundary moments of the gait feature map Feature parameters; perform wavelet decomposition on the gait feature map to extract horizontal, vertical, and diagonal moment invariant feature parameters; process the gait feature map into a skeleton, and extract the skeleton containing the limb proportion and angle information of the simplified human model information Characteristic Parameters; (4) input the combination of characteristic parameters of each person who needs to be registered into the sample training library for the training of the support vector machine; (5) Install the thermal imaging camera in the place where the monitoring personnel need to enter and exit; (6) Utilize the infrared thermal imager to collect the gait image sequence of the personnel entering and leaving the place; (7) adopting the method of step (3) to carry out feature extraction process respectively to the gait image sequence of the gathered personnel entering and leaving this occasion, obtain the invariant moment parameter and skeleton characteristic parameter of personnel entering and exiting this occasion; (8) Input the combination of characteristic parameters of the personnel entering and leaving the place into the support vector machine for gait classification and recognition, and determine whether they are registered personnel. 5. The gait recognition method based on infrared thermal imaging according to claim 4, characterized in that, threshold segmentation and dilation and erosion methods of binary images are used to segment human body contours. 6. The gait recognition method based on infrared thermal imaging according to claim 4, characterized in that, a mathematical morphology method is used to perform skeletonization processing on the gait feature map to extract skeleton feature parameters.

Images