Study On Gait Feature Extraction And Human Identification Based On Computer Vision

Gait is a relatively new and emergent behavioral biometric, which pertains to the use of an individual's walking style to determine or validate identity. In particular, gait can be used for human identification at a distance. Gait recognition has recently received growing interest within the computer vision community, and become a hot area in the field of computer vision. In this paper, the technique of gait feature extraction and human identification based on computer vision is researched. Especially, problems such as human detection in complicated background, gait feature extraction, and human identification, etc. are studied. The main contributions of this thesis include:(1) A novel background subtraction method is presented. To cope with background change problems such as gradual illumination changes and moved background objects, an efficient background updating algorithm based on Dynamic Information Window (DIW) is proposed. Updating decisions are made according to the pixel-wise Dynamic Information Window. To cope with shadow problem, a new computational color model is presented. Chromaticity distortion is measured in an effective way. Experiments have been performed on a surveillance system in indoor environments as well as outdoor environments. The experimental results demonstrate that using the proposed approach the background can be updated adaptively, and shadows can be suppressed with a better performance than other methods.(2) A new gait recognition method based on hidden Markov models (HMMs) and Fourier descriptors (FD) is put forward. The body contours are processed by Fourier descriptors. K-means clustering method is used to analyse the image sequence within a gait cycle, and gait is represented by key stances. The hidden Markov models are applied to model the gait, where the key stances are considered as analogues of states of the HMMs while the distance vector sequence is considered as the observed process. The proposed gait recognition method is evaluated on the gait database. The experimental results demonstrate that our approach using Markov models and Fourier descriptors has a good recognition performance. (3) A new gait recognition method based on support vector machine (SVM) and linear discriminant analysis (LDA) is proposed. The body silhouette is projected along columns and rows to obtain horizontal and vertical projection vectors, which are then combined into one dimensional data vector. Linear discriminant analysis is applied to reduce data dimension and extract gait features. As support vector machine can solve small sample learning problems effectively, gait recognition is performed by support vector machine finally. The proposed gait recognition method is evaluated on the gait database. The experimental results demonstrate that our approach using linear discriminant analysis and support vector machine has a better recognition performance than other similar methods.