Biological motion contains information about several different emotions, intentions, personality traits and biological attributes of the agent. The human visual system is highly sensitive to biological motion and capable of extracting this information from it. We investigate the question of how such information is encoded in biological motion patterns and how it can be retrieved from it. We developed a framework that transforms biological motion into a linearized representation which enables us to apply linear methods from statistics and pattern recognition to its analysis. Initially, we used gender classification as an example and created a simple classifier, whose performance was compared to psychophysical data from human observers. However, the same approach can be used to extract information about other walker attributes as well.

The proposed framework can also be used to synthesize new motion patterns with particular attributes. The approach is data-driven rather than hypothesis-driven and reveals the complex orchestra of correlated movements between the different body parts.

Here are some demos:

GENDER CLASSIFICATION: ANALYSIS AND SYNTHESIS

Analysis of sex-specific differences in walking style reveals that the dynamic part of the motion contains more information about gender than motion-mediated structural cues. In this demonstration differences in the two components can be visualized and exaggerated separately.

A simple motion modeler

A linear discriminant function can be computed either on the basis of known attributes such as the sex or the weight of a walker, or it can be computed on the basis of ratings that have been attributed to each walker by one or several observers. This example shows motion axes that reflect gender, weight, sadness (vs. happyness), and relaxedness (vs. nervousness).

Extract your own motion axis!

You can use this demonstration to extract your own motion axis from our database. It will take you no longer than 10 minutes. You can freely choose any attribute you might be interested in. You will then be shown a number of walkers and you will have to rate them according to the attribute of your choice. Once you are done, the system will compute a discriminant function based on your ratings and will visualize the corresponding walking axis in terms of an interactive point-light animation.

Publications:

• Troje, N. F. (2002). Decomposing biological motion: A framework for analysis and synthesis of human gait patterns.Journal of Vision, 2:371-387, Full Text DOI 10.1167/2.5.2.
• Troje N. F. (2002) The little difference: Fourier based gender classification from biological motion.  Dynamic Perception, R. P. Würtz and M. Lappe (eds), Aka Press, Berlin, pp. 115-120. Full Text
• Troje, N. F.  (2008) Retrieving information from human movement patterns.  In: Shipley, T. F. and Zacks, J. M. (eds.) Understanding Events: How Humans See, Represent, and Act on Events. Oxford University Press, pp. 308-334. Full Text