Researches On Human Gait Recognition

Gait recognition, as a new kind of biometrics recognition method, refers to automatic identification of an individual based on his/her style of walking. Compared with face recognition, fingerprint recognition and so on, it has prominent characteristics which are identification of an individual from a distance, non-contact, difficult to camouflage and difficult to hid, so it is the most potential biometric recognition in the field of remote identity. In addition, it has a broad application prospect at security system, human ID management, and digital surveillance, and has become a research hotspot in recent years.Existing gait recognition algorithms are almost based on side view sequences. It's rare to find researches on other walking directions, especially the front-view direction, so gait recognition of considering walking direction has become a challenging problem. As a result, focusing on front-view gait sequences, this paper completes the two major tasks:first, improving the gait recognition algorithms; second, building a gait recognition system.On aspect of algorithms research, the work of this paper includes:Firstly, when background subtraction method is used in the front-view gait motion detection to extract the silhouettes, there is always an empty region in every silhouette. So this paper proposed two solutions which are very effective:one is improved background subtraction algorithm, and the other is hole-filled method.Secondly, considering that existing methods of gait cycle estimation are almost based on side view gait but cannot be used for front-view gait, this paper proposed a new method called "W-U" gait cycle estimation, and it uses the number of pixels difference between the left region and the right region which are all from the low quarter of silhouette to estimate the front-view gait cycle.Thirdly, silhouettes in front-view gait sequences change by a small gradual changes one after another, and the situation is against feature extraction, so this paper proposed an adaptive normalization method of image size.Fourthly, a new sampling method, in which the number of sampling points can be set flexibly in different regions, is proposed in this paper.Finally, each boundary point's central moment has been calculated in the new sampling, this paper made use of this convenience to improve the algorithm of Fourier Descriptor based gait contour feature representation. And the Fourier Descriptor calculated from central moment sequence was as gait characteristics for feature match and classfication.On aspect of system construction, we also construct a gait recognition system based on CASIA gait database which has been already built by Institute of Automation Chinese Academy of Sciences. System is designed to be clear, simple, fast and effective. The front-view gait recognition algorithm proposed in this paper is programmed in this system and experimental results show that the correct recognition rate of the proposed approach on small sample dataset is 5% higher than that of the related reference.