Local and global information information in animate motion patterns

Dr. Nikolaus F. Troje
Queen's University

Animate motion patterns are a rich source of biologically significant information. They quickly and reliably signal the presence of an animate agent and convey its identity, actions and intentions. Observing other people in motion, our visual system is able to retrieve information about sex, age and weight of a person and can even detect signatures that identify a familiar individual. The relevant information is encoded on different levels in the motion patterns ranging from purely local motion information to complex global correlation patterns.

In my talk I will present empirical and computational data from two different studies. In the first, we investigated the nature of the general saliency of biological motion. Based on findings from experiments designed to explore the cause of the inversion effect in biological motion, we propose a simple, yet reliable and fast sensory filter that can detect the presence of a living animal in the visual environment. The proposed mechanism is based on local motion trajectories and does not require any form processing.

On the other hand, I will present a computational framework that explores the complex correlation pattern of a moving body to retrieve information about a person’s sex and other attributes that define his or her identity. It is based on a morphable representation of human walking data which in turn makes it possible to formulate linear classifiers for the attributes of interest. Results on sex classification obtained from our model are being compared to behavioural data and the relevance of our framework as a model for visual information processing in the human brain will be discussed.