Temporal and spatial analysis of local body sway movements for the identification of people

We divide the body of a standing person into local regions and analyze small sway movements over time and space. Power spectral density features extracted from these movements support person identification even when the person is not walking.

Methods that identify people by their walking patterns become less useful when someone stops walking, because large arm and leg movements disappear. However, the body still sways slightly even when a person tries to stand still. We investigate whether these small movements contain useful information for identifying people.

We divide the body in a video into several regions and measure how the small movements in each region change over time. We then describe how strongly the body moves at different rates, from slower to faster swaying. This description, called power spectral density, provides the features used for identification.

We collected three video datasets: one with many people standing upright, one covering changes over a long period, and one with different standing postures. Experiments on these datasets confirmed that small movements in individual body regions provide useful information for identifying people who are not walking.

• Takuya Kamitani, Hiroki Yoshimura, Masashi Nishiyama, Yoshio Iwai,
  Temporal and spatial analysis of local body sway movements for the identification of people,
  IEICE Transactions on Information and Systems, Vol.E102-D, No.1, pp. 165 - 174, January 2019.
• Takuya Kamitani, Hiroki Yoshimura, Masashi Nishiyama, Yoshio Iwai,
  Identifying People using Temporal and Spatial Changes in Local Movements Measured from Body Sway, Proceedings of 4th Asian Conference on Pattern Recognition (ACPR), pp. 828 - 833, November 2017.

• Body sway identification