4-IJAEST-Volume-No-3-Issue-No-1-Human-Gait-Recognition-Using-Gaussian-Membership-Function-020-023 by ISERP ISERP

Pratibha Mishra et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 3, Issue No. 1, 020 - 023

Human Gait Recognition Using Gaussian Membership Function Pratibha Mishra

Shweta Ezra

Samrat Ashok Technological Institute Vidisha, (M.P.) India Mpratibha9@gmail.com Pratibha_mishra_satna@yahoo.co.in

Dhar polytechnic college Dhar (M.P.) India shwetaezra@gmail.com

Keywords- Gait recognition; Biometrics; Gaussian membership function ; Control points; Database; Mean; Variance.

1. INTRODUCTION

Gait recognition is a kind of biometrics using the manner of walking to recognize an individual. More formal definition of biometrics is given by [1], “Gait recognition refers to automatic ndividual based on the style of identification of an iindividual walking”.Gait is treated as a sequence of holistic binary patterns Gait recognition Approaches can be broadly categorized into del-based ap approach, proach, where human body structure is the model-based an d the model-free approach, where explicitly modeled, and studies have now shown that it is possible (silhouettes). Many studies to recognize people by the way they walk. It is well-known that biometrics is a powerful tool for reliable automated person identification, but at present, none of the conventional biometrics like fingerprints recognition recognition. Iris recognition can work well from a large distance. In visual surveillance, the distances between the cameras and the people under surveillance are often large. In these situations, it is almost impossible to acquire the detailed conventional biometric information. Unlike other biometrics, gait can be

ISSN: 2230-7818

istant camera, without drawing the attention of captured from a distant the observed subject. image-based gait recognition is not very The performance of image-based good because Features extracted from image sequences have a th e original information included in the little difference with the gait. There is no efficient algorithm proposed yet to minimize gait. th e three three dimensional dimensional information and the difference between the features extracted from projected images [3]. features

Abstract- Human gait is a spatio-temporal phenomenon and typifies the motion characteristics of an individual. The gait of a person is easily recognizable when extracted from side view of the person. Accordingly, gait-recognition algorithms work best when presented with images where the person walks parallel to the camera (i.e. the image plane). Gait Recognition refers to identification of an individual based on the style of walking. This paper proposed a new algorithm which is based on Gaussian membership function. The Gaussian Membership function was chosen because of its popularity and simplicity. Gaussian membership functions are generated according to person’s walk and recognition is achieved by matching the curves by calculating the mean and variance, and used for recognition. Only the sideprovide ovide view of the person is considered, since this viewing angle pr the richest information of the gait of the waking person.

2. RELATED WORKS

pre Chan-Su Lee [7] presented an approach “Identification of people using using silhouette gait image”. He used a bilinear model to people separate two independent factors, gait style and phase. Nnormalized gait poses is defined and generated by embedding gait image sequences to a standard lower dimensional manifold gait and learning mapping from the manifold to every pixel. This normalized gait phase is used to collect aligned gait poses from diffe di ffe different speed walking image sequence. He identified gait style-vectors, which represent factors invariant to gait pose. Using a boosted gait content vector, he got a better human identification accuracy than when using the original phase vector before identifying gait content vector. Hong, Lee, Oh, Park, and Kim [4] have proposed a new feature vector, sampled point vector, for gait recognition based on model-free method. The mean and variance of value of pixels are chosen which are sampled along to central axis of silhouette image for several frames. Yanmei Chai Jinchang Ren, Rongchun Zhao and Jingping Jia [5] proposed a statistical approach for dynamic gait signature extraction. The DVS on each of the pixel position for a full gait sequence is extracted firstly, and then compute their variance features respectively to construct a dynamic variance matrix as gait signature for identification. Alam and Hama [6] presented an approach to typify

Page 23

Pratibha Mishra et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 3, Issue No. 1, 020 - 023

3. PROPOSED ALGORITHM In this paper, we have proposed a new algorithm for recognizing gait system. This algorithm is based on Gaussian membership function. The proposed gait recognition system consists of three units: (I) Image Preprocessing. (II) Feature Extraction. (III) Gait Recognition. 3.1 Image Preprocessing

A simple background extraction method is to subtract a background model from the current frame. This method is based on pixel level processing. The extracted foreground is used for recognition and tracking. This is a very simple and convenient method in motion detection. The difficulty of this method is not the subtraction computation, but maintaining the background model. There are several classic background subtraction methods. The following methods have self-adaptive ability. 1). Mean & threshold method: first compute the mean of new biometrics recognition background pixels. It is a new aimed essentially to recognize technology. Gait recognition aimed rson by automatically extracting movement characteristic of person Foreground reground pixels are those that walking person in the video. Fo differr by more than a threshold. ean & variance method update the mean and variance 2). Mean continuously, and then compute the distance. If the distance is larger than the threshold, set the pixels to be the foreground. The gait gait of a person is best brought brought out out in the side-view [5]. Video of a walking individual iiss captured by camera and sequence frames are extracted ffrom rom that video. Each frame is converted into grayscale if it is a color image.

In our experiments, there are two assumptions for the human walking sequences: (1) the camera is static and the body in the field of view is not occluded. (2) the image sequence of side-view is used.

Background subtraction

object contours in a database by a reduced number of data points and to match object shapes in occluded conditions. For simplicity, contours are approximated by a set of membership function. and all control points are stored in the database. Distance matrix is introduced, which is constructed from curve to curve distance measurement between the test and database contours.

Grayscale images are used in this work because these images are entirely sufficient for our tasks and so there is no need ne ed to use more complicated and harder-to-process color images.

Fig 1. Proposed Algorithm

ISSN: 2230-7818

Fig. 2 Producing a silhouette from an image. (a) Original Image (b) Silhouette. Gait Feature Selection The definition of Gait is defined as â&#x20AC;&#x153;A particular way or manner of moving on foot â&#x20AC;&#x153;Using gait as a biometric is a relatively new area of study, within the realms of computer vision. It has been receiving growing interest within the

Page 24

Pratibha Mishra et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 3, Issue No. 1, 020 - 023

computer vision community and a number of gait metrics have been developed. An important issue in gait is the extraction of appropriate salient features that will effectively capture the gait characteristics. The features must be reasonably robust to operating conditions and should yield good discriminability across individuals. A fast and efficient method is adopted to select only most discriminative features. Fig 4. Key Frames 3.2.2 Computation of Gaussian membership function

[2]:The formula is given below [2]:μ A(x) = -exp (x – m)2/2σ2

Where Where X represents represents the crisp data, repr μ represents the membership membership function of x, m represents the mean of all the crisp data x in the distribution and σ represents the variance of all the crisp data in the distribution. di stribution.

Fig 3. a) Sequence of selecting control points.

The Gaussian Membership function was chosen because of its popularity and simplicity.

1. ankle 2. Toe 3. Knee 4. Palm 5. Shoulder

The variance σ can be represented mathematically as

σ = √ (Σ(x – m) 2/n)

Fig 3. b) Skeleton model of human frame

3.2.1 Key Frames Generation

We determine the key frames of a walking gait by observing the different phases of a human walk cycle as shown in Figure. th e pose where front leg is standing The first key is defined at the raight while the back leg is bend and slight above the ground. straight the location where the front leg’s foot is The second key is at the flat on the ground and back leg’s toe touches the ground. The third key is defined as the pose where the back leg’s foot if flat on the ground and front leg’s ankle touches the ground. The fourth key will return back to the first key and complete the cycle.

ISSN: 2230-7818

Where n is the number of angles in the distribution, Where x represents the crisp data and m represents the mean of all the crisp data x in the distribution.

The Gaussian membership function was used to fuzzify all the crisp data obtained. The data for each subject was stored in the knowledge base and used for the inference when a gait pattern or signature is to be tested, classified and recognized. The actual Gaussian membership function obtained for the crisp data for all the four patterns associated with the five subjects. Once the control points are given, the curve shape is determined. 3.3 Gait Recognition The experiment involves capturing subtle changes in an individual’s walk, taking into consideration the variation in angles of the various parts of the body or the amplitude of the

Page 25

Pratibha Mishra et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 3, Issue No. 1, 020 - 023

Video Sequence Each subject had a total of five markers (objects capable of reflecting light over a camera) attached to the following parts of their body: 1. The Shoulder 2. The Hip 3. The Knee 4. The Ankle 5. The Toe

Points selection

After the mark-in and mark-out points were chosen for each captured gait video, the data points were cropped between the start and finish markers. Whenever the markers exceeded the software’s default minimum and maximum outline, the setup ccommodate the excesses. The cropped data was changed 6 to accommodate points were digitized automatically by the software using the centriod of each marker. For points like those of the hips which could not be digitized automatically by the software, as a result of the obstruction caused by the arm during the gait, the cursor location was used to digitize the points instead. Points Processing

The digitized points were processed by the process wizard in the software. The 2-dimensional angles of rotation of the saved in system as database. marked parts of the body were saved Reflex angles were recorded for the hip, torso and ankle movement, while obtuse angles were recorded for the knee movement as shown in the stick diagram below. We adopted a

ISSN: 2230-7818

Advantages of Proposed Method: 1. The Gaussian Membership function is chosen because of its popularity and simplicity. 2.. It does not require silhouette images and GEI images. 3.. Computational speed becomes high due to use of simple mathematical calculations like m mean ean and variance. CONCLUSION

The membership functions associated with each mblance is aalso lso displayed we proposed a novel gait resemblance cognition method based on the Gaussian membership recognition function. First we select the the ppoints oints on sequence frames, calculate the co coordinates ordinates of of G Gaussian aussian membership function from those points, draw the ccurves urves and finally, calculate the variance and mean from Gaussian membership function variance coordinates. These variance and mean are used to fulfill the person pe rson identification. identification.

The video of each subject’s gait pattern was captured randomly using the installed camera. For each captured gait pattern, a mark-in point and mark-out point was chosen arbitrarily. The mark-in point represents the first point in the captured gait video where all the markers were visible, while the mark out point represents the last point in the video where the five markers were visible. These mark-in points and markout points were arbitrary in the sense that different points were chosen for each subjects captured gait video.

simple and straightforward way in order to test the recognition capability of our proposed method. First we calculate the variance of x- and y- coordinates of any curve of all frames of an individual separately and then finally calculate the mean of x- coordinate with its corresponding y- coordinate. In contrast to other system, proposed features are very simple and require low storages.

persons walking pattern. There are three different stages: Stage 1: Video Sequence Stage 2: Points selection. Stage 3: Points Processing

REFERENCES

[1] Xiaxi Huang and Nikolaos V. Boulgouris, “Gait Recognition Using Multiple Views”, IEEE 2008. [2] Elizabeth I. Maduko, “pattern recognition of human gait signatures”. [3] Seungdo Jeong, Su-Sun Kim, Byung-Uk Choi, “Canonical View Synthesis for Gait Recognition”, IEEE 2007. [4] Sungjun Hong, Heesung Lee, Kyongsae Oh, Mignon Park, and Euntail Kim, “Gait Recognition using Sampled Point Vectors”, IEEE 2006. [5] Yanmei Chai Jinchang Ren, Rongchun Zhao and Jingping Jia, “Automatic Gait Recognition using Dynamic Variance Features”, IEEE 2006. [6] Md. Jahangir Alam and Hiromitsu Hama, “Occluded Shape Matching for Image Database of Reduced Data Points”, IEEE. [7] Chan-Su Lee, “Identification of people using silhouette gait image”.

Page 26