Gait analysis using ISOMAP

ISOMAP is an excellent nonlinear dimensionality reduction method. This paper uses ISOMAP for view-independent gait cycle detection and dynamic feature extraction in a gait cycle. Experiment results show that 1-d ISOMAP embedding of gait sequences has almost the same information as that of 2-d ISOMAP embedding, and is sufficient for gait analysis. Zero crossing of 1-d or 2-d ISOMAP embedding shows gait cycle and is view-independent, and the shape of 1-d or 2-d ISOMAP embedding shows the dynamic features in a gait cycle.