Abstarct: In this thesis, we address the problem of human recognition using gait. First we review well-known approaches reported in literature. There are two approaches for gait recognition, model baxsed and appearance baxsed approach. One of recent methods with good reported result, was introduced by Lee et al [22]. In this method difference between human silhouette frxames in a gait cycle and the key frxame is computed. From this four feature vectors are extracted. We address two problems of this approach: using the key frxame and redundant features. Since the detection of key frxame is subject to error we propose to use difference of succestive silhouettes instead of using key frxame. We also remove the redundant features from the list of features used in the Lee et al method[22]. As we will show through the experiments the proposed modifications improve the recognition performance. The gait recognition result significantly deteriorates when person changes clothes. This is one of the main problems addressed for gait recognition in the literature. We noticed that legs movement during walking cycle less depends to clothes style in compare to other part of human body. baxsed on this observation we model leg dynamics using linear grouping analysis (LGA), then we extract discriminate features from this model. As the experimental result shows using the proposed approach we achieve a significant improvement for gait recognition in presence of cloths change.