Human Identification Using Motion Information and Digital Video Processing

Motion information as a biometric has the advantage of identification at distance and low resolution images. This paper proposes a simple and effective approach for the human identification based on the statistical shape analysis. We use the overlapped image of the contours of thinned silhouettes for each image sequence as the feature for identification. To reduce the dimension of the feature space, we apply eigenspace transformation based on PCA. For improvement of the efficiency, we add a new feature based on periodicity of the motion. For an image sequence, this feature is extracted by the correlation of each frame with the first frame where PCA algorithm is applied to reduce its dimension. Experimental results on the CASTA database showed the efficiency of the proposed approach.