A Feature Fusion Based Gait Recognition Algorithm

With the development of information technology, using the biometric technology to identify and judge the behavior of human and identify a person's identity has become a social problem. Biometrics recognition technology uses biological characteristics of human to identify people' identify. Because of the exclusive and unchangeable characteristic, biometric identification has aroused more and more attention in the field of recognizing individuals. It is not only simple and fast, but also safe and reliable. Compared with other biometrics, such as face recognition, fingerprint recognition, iris recognition and so on, gait recognition requires the low quality of the images. The identity of human can be identified even by the blurred images. However, images can be transmitted with high compression ratio without affecting the accuracy of recognition, which can make full use of transmission bandwidth.Gait recognition is a new biometric technology. It aims to recognize individuals by extracting the change characteristics of individuals from the different ways they walk. Compared to the first generation of biological characteristics, gait characteristic is non-contact, difficult to camouflage and concealment, easily acquired at a distance and can be carried out to detect and identify by the low-resolution image sequences.Therefore, from the view of video surveillance, gait recognition is the biometric with the most potential in the case of long-distance. Gait recognition has caused many researchers'interest.Gait recognition includes moving target detection, feature extraction and classification and recognition. Based on the study of the various of gait recognition algorithms, detailed study about feature extraction and classification has done and a new algorithm has proposed in this paper. The main achievements of this thesis include:1. Extraction of effective gait characteristics.Because of the impact of external factors, the recognition rate based on a single feature is unsatisfactory. In this context, we have proposed a new representation for human gait recognition based on features fusion. Intuitively, gait recognition based on human vision depends on the change of the shape of the body contours over time to a large extent. The gait recognition algorithm based on the contour of human body is simple and easy to implement. But the contour can only reflect human's movement characteristics indirectly. Medical studies have shown that people's walking behavior includes hundreds of simultaneous movement of limbs and these muscles and bone structure can reflect the movement characteristics and gait pattern showed in the process of walking accurately. Therefore, limbs angles are the best features for identification. However, because of restrictions for the video quality, it is difficult to obtain accurate information about the angle of the joints. Taking into account the symmetry of gait, reflective symmetry feature, which is easy to extract and fuse with other characteristics, is introduced here. Based on the analysis above, an efficient gait recognition algorithm based on the fusion of body contour, limbs angles and reflective symmetry feature is presented in this paper.2. Research about classification and recognition algorithm.This text aims to use three types of features to identify people's identity. But different features have different data types and scales. A nearest neighbor fuzzy classifier is introduced, which is actually a simple fusion of multiple classifiers. First, combine the three kinds of gait characteristics extracted above into a simple joint feature vector to be the input of the classifier. Then calculate the membership degree at the same dimensional feature. So a membership degree vector can be got which means the matching between the joint feature vector and all the samples of templates. Considering that each feature has different contributions on the result in the gait recognition process, we distribute different weights to different membership degree in order to improve the recognition rate. Finally, a kind of decision algorithm is used to get the recognition result.Experiment results show that the proposed algorithm has better recognition performance.