Gait Recognition based on Measurements of Moving Human Legs Angles

Abstract

Biometrics plays a crucial role in security systems. Gait recognition as an aspect of behavioral biometrics deals with identifying human based on the walking. In this research, we proposed a new method to identify people according to their gait characteristics. This approach is performed via several steps; a) the video frames of a walking subject is recorded. b) each video is decomposed into eight phases, wherein each phase, there are several frames. One frame per phase is selected. c) the extracted frame is segmented to separate the subject from the background. d) the image is converted to a binary image, and the skeletonization is applied to obtain the silhouette of the image. e) the fast Fourier transform is applied to each frame, and several statistical features are calculated. The average of the statistical measures for eight frames of a video is computed as a feature vector for the subject. Finally, the correlation is used to identify the subject based on the feature vector. The accuracy of 100% is achieved for the recognition of the subjects in the CASIA dataset.