Human Identification by Spatio-Temporal Symmetry

We describe spatio-temporal symmetry and its extraction via a Generalised Symmetry Operator. Its use in gait recognition is reinforced by the view from psychology that human gait is a symmetrical pattern of motion. We show that by including temporal information in our symmetry calculations we are not recognizing people by their body shape but also by their motion. Here, the new technique is applied to a database of 28 subjects, which equals in size the largest contemporaneous gait databases. The results of the new approach agree with earlier results that the symmetrical properties of human gait appear to be unique and can indeed be used for analysis and for recognition. The results achieved so far give promising performance and higher recognition rates than those of an earlier spatial approach. Performance analyses suggest that symmetry enjoys practical advantages such as ability to handle noise and occlusion, and especially when resolution is too low for other biometrics to be deployed.