Gait Recognition Algorithm Research Based On Hidden Markov Model

With the development of the times and social progress in recent years, Because of its special security, stability and convenience, Biometrics Technology is widely used in the security, identification and other areas of certification. The usual biological characteristics include face, iris, fingerprint, palm-print, voiceprint, etc. Compared with above characteristics, Gait is the external, dynamic description, and it is close with the temporal and spatial variation. Compared with other biometric technologies based on the static characteristics, gait recognition has some significant advantages, such as non-contact, non-invasive and difficult to hide. Moreover, a series of psychology experiments and biology researches shows that, if considered all the factors in the process of walking, gait is unique for different people. Therefore, gait analysis plays an important role in the visual monitoring, control, identification, etc. and it draws extensive attention of computer vision researchers.This paper analyses the home and abroad research status of the gait recognition, and studies the main methods and problems in gait recognition. On this basis, the paper proposes a gait recognition algorithm based on Hidden Markov. This paper is mainly divided into three parts.Firstly, the paper introduces the basic theory of the Hidden Markov Model. Started with the Markov chain, the article explains the basic concepts and model parameters of a Hidden Markov model. Then we analysis three basic issues and discuss three basic algorithms for Hidden Markov Model. At the same time, we analysis the problems of the Hidden Markov Model in the practical application, and we do the corresponding improvements for the basic algorithms. This part is the theoretical foundation of the paper.Then, we introduce the preprocessing technology for gait sequences. The preprocessing technology is divided into following parts to discuss: motion detection, motion segmentation, interested region extraction and processing, gait cycle detection. In the motion detection, we analysis the usual motion detection algorithms, and use background subtraction method to detect motion. We construct background model using a median filter. In motion segmentation part, we discuss the binarization impacts using different fixed threshold values, and use iterative method to calculate optimization threshold for each foreground image. And this method can greatly increase the binarization impact. Then using morphological filters to real with the binary image and reduce the noise impact. In interested region extraction and processing part, using traversing search method to find the smallest human rectangular box, and doing regulation and centered to human rectangular box. In gait cycle detection, based on the derivative of extreme points is zero, we propose a new gait cycle detecting method. It uses the walking human's feet distance as a processed signal, and can accurately detect gait cycle in most cases.Finally, the article introduces gait feature extraction and gait recognition. Feature extraction is the key technology for gait recognition. At first, we use human rectangular boxes to generate gait energy images, and then we extract gait characters from gait energy images. In the feature extraction, using fuzzy approach reduces the volume of data and the noise impact. When training model parameters, we propose a way to build virtual test samples and well solve the problem of inadequate training samples.In simulation tests, we made a lot of tests in the UCSD gait database and CASIA Dataset A gait database (that is, the original NLPR database), and the results are analyzed in detail, we make CCR, ROS curve, ROC curve respectively. Test analysis shows that the gait recognition algorithm we proposed achieves satisfactory results.