Abstract
In this thesis, we combine the dynamic and static information extracted from gait to identify the walking human object. First we utilize the periodicity of swing distances to estimate the gait period for each gait sequence and divide them to sub-cycles. For each gait cycle, we extract the static information by proceeding intersecting operation and dynamic information by analyzing the statistic histogram of motion vectors. The extracted information is transformed into low dimensional embedding space by dimensionality reduction process. The low-dimensional feature vector is used to represent the subject. Then, we use a set of discriminant functions to determine the decision regions for normal data distribution, and then we can recognize the human walking path. Given a test feature vector, the nearest neighbor classifier is applied to compare with the feature vectors established from a gait database for subject identification. The proposed algorithm is evaluated on the CASIA gait database, and the experimental results demonstrate that own system achieves a high recognition rate.