# Design of Embedded Automatic Gait Recognition System Based on Chip SH7709S and Memory

“The core of an embedded system consists of one or more microprocessors or microcontrollers that are pre-programmed to perform a few tasks. Embedded system is application-centric, based on computer technology, software and hardware can be tailored, and is suitable for special computer systems that have strict requirements on functions, reliability, cost, size, and power consumption of application systems. Gait recognition is an emerging biometric recognition technology, which aims to identify people through their walking posture. Compared with other biometric technologies, gait recognition has the advantages of non-contact long distance and not easy to camouflage.In the field of intelligent video surveillance, it is more important than face recognition

“

Author: Liu Haitao, Guan Shengxiao

The core of an embedded system consists of one or more microprocessors or microcontrollers that are pre-programmed to perform a few tasks. Embedded system is application-centric, based on computer technology, software and hardware can be tailored, and is suitable for special computer systems that have strict requirements on functions, reliability, cost, size, and power consumption of application systems. Gait recognition is an emerging biometric recognition technology, which aims to identify people through their walking posture. Compared with other biometric technologies, gait recognition has the advantages of non-contact long distance and not easy to camouflage. In the field of intelligent video surveillance, it has more advantages than face recognition.

So far, almost all researches on gait recognition are based on PC, but in many cases, non-PC environment is required, so the research on gait recognition system based on embedded platform has certain engineering significance. The function of this system is to perform image processing on the collected gait video sequence to obtain the human gait information in the video sequence, and then use the gait algorithm to perform gait recognition according to the obtained gait information.

**1. System structure**

The embedded automatic gait recognition system mainly includes a CCD camera, an image acquisition card, an embedded system, and a Display screen. The core of which is the embedded system part, which includes Renesas 32-bit embedded chip SH7709S, memory, peripheral circuits, keyboard, mouse and so on. It mainly completes functions such as preprocessing, processing, gait recognition, and display output of video sequence signals. The schematic diagram of the structure of the system is shown in Figure 1.

**2. The basic principle of gait recognition**

2.1 Binocular Stereo Vision

Binocular stereo vision is a new technology developed in the field of image measurement this year. Compared with monocular vision, binocular vision has the following advantages: it can obtain parallax or depth information that is not available in monocular vision; When occlusion occurs, binocular stereo vision can handle occlusion very well. Because the scene of gait recognition is inevitably occluded, in order to better obtain gait video sequences from all directions, so as to pave the way for correct gait recognition, binocular stereo vision is used to obtain human gait video sequences.

In this experiment, two CCD cameras were fixed on both sides of a tripod to form binocular stereo vision.

**2.2 Optical flow field in gait image sequence**

Optical flow refers to the speed at which patterns move in an image. An optical flow field is a two-dimensional (2D) instantaneous velocity field, where the 2D velocity vector is the projection of the three-dimensional (3D) velocity vector of the visible points in the scene onto the imaging surface. Optical flow not only contains the motion information of the observed object, but also carries rich information about the 3D structure of the scene. The optical flow method assumes that the interval between adjacent moments is small (usually tens of ms), so the image difference between adjacent moments is also relatively small.

**2.2.1 Basic equations of optical flow**

**2.2.2 Calculations related to optical flow**

For each point (xi,yi) on the image, the optical flow field equation (2) is solved, and the solution expressed in iterative form is:

**2.3 Extraction of motion features in optical flow field**

Features extracted from optical flow include motion point T, weighted motion point |(u,v)|, |u|, |v|, and centroid features of optical flow distribution, etc. Through the optical flow field, the moving point (white) and the non-moving point (black) are distinguished by T(u, v), which is expressed by the following formula:

**2.4 Data fusion of gait features**

**2.5 Identification**

The feature space transformation based on PCA is performed on the features obtained by data fusion. Suppose the initial training sample set is T={pi-j}, i=1, 2,…, C, j=1, 2,…, Ni; the jth gait sample vector of the i-th person is Xij, and the sample vector The total is NT=N1+N2+…+Nc.

Find the population mean vector μ and covariance matrix ∑ of the sample set,

If the rank of the covariance matrix ∑ is N, the N eigenvalues λ1, λ2, λ3, …, λN of the matrix ∑ are obtained by det|λI-∑|=0, and the matrix equation λiI-∑=0, i= 0, 1, 2, …, N; find N eigenvectors e1, e2, e3, …, eN corresponding to N eigenvalues λ1, λ2, λ3, …, λN.Pick the top K eigenvectors corresponding to the top K largest eigenvalues, and make

where α represents the percentage of the energy of the sample set on the top K axes to the total energy. Usually the value of α is taken close to 1, so that the energy of the sample set on the first K axes is almost close to the whole energy.

Each sample in the initial sample set is reconstructed with the K eigenvectors obtained in formula (2). The algorithm is as follows:

In this way, a K-dimensional weight vector Ωi, j is obtained for identification.

Select the nearest neighbor classification method for gait pattern classification. Assuming that after feature extraction and projection into feature space, the obtained feature vector is Ω, and the Euclidean distance between Ω and the average vector Ω i,j of each mode class is obtained.

in

According to the decision criterion of the nearest neighbor classification method, when the value of εi(x) is the smallest, then x ∈ εi; otherwise, x ∈ εi.

2.6 Effectiveness and error rate of recognition

According to the principle of pattern recognition, when there are two types of gaits, the error rate of gait recognition is given by:

in

The integral interval R1 is the misjudgment interval when w2 is misjudged as w1, and the integral interval R2 is the misjudgment interval when w1 is misjudged as w2. When p(e) is the smallest, the recognition is more effective, and when p(e) is large, the recognition performance is worse. When there are multiple types of gait, and so on.

**3. System implementation**

3.1 Hardware Implementation

The block diagram of the system hardware connection is shown in Figure 2.

**3.2 Software Implementation**

The system software flow chart is shown in Figure 3.

**4 Conclusion**

Gait recognition has become a new research direction in the field of computer vision in recent years. In this paper, a simple automatic gait recognition method is proposed, and an automatic gait recognition system based on Renesas embedded chip is presented, which will be widely used in the long run.

The Links: **BSM100GB60DLC** **EPM3032ALC44-10N**