Identity Recognition

Can you tell which actor who is doing jumping jacks is the actor who is moving sideways?

Human's ability to recognize someone solely based on the way that he/she moves is fascinating. Research has shown that people can recognize themselves and their friends through simple point-light displays (e.g. Cutting & Kozlowski, 1977; Loula et al., 2005).

As deep neural network became increasingly popular, machine vision's capability of recognizing human's identity has become ever so powerful that it may have already exceeded that of human's. However, to me, using deep neural network to perform this kind of task could be useful in practical scenarios, but it does not provide a lot of insights into how it actually accomplishes this difficult feat, or does it inform us of how humans do this.

Studies have suggested that there is a dynamic identity signature embedded in human movement (Knight & Johnston, 1997; Yovel & O'Toole, 2016). However, this so far has only been a theoretical construct, instead of being a quantifiable measure. In this project, I used various dimensionality reduction techniques from machine learning to identify how we can extract such a dynamic identity signature through human movement.