| Popis: |
New method of automatic gait recognition from video is proposed. The method works with a sequence of silhouettes derived from the video after background subtraction, decreasing shadows and noise. Two-dimensional silhouette shape is converted into one-dimensional signal presenting distance from center of gravity to outline of this silhouette. A set of signals extracted from a sequence of silhouettes forms a two-dimensional picture. The features extraction is performed using Gabor wavelets. Testing the method on the samples of CASIA gait database showed high recognition accuracy. |
