This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Exploring the Space of a Human Action Yaser Sheikh Mumtaz Sheikh Mubarak Shah Computer Vision Laboratory, School of Computer Science, University of Central Florida, Orlando, FL 32826 Abstract One of the fundamental challenges of recognizing actions is ac- counting for the variability that arises when arbitrary cameras capture humans performing actions. In this paper, we explicitly identify three important sources of variability: (1) viewpoint, (2) execution rate, and (3) anthropometry of actors, and propose a model of human actions that allows us to investigate all three. Our hypothesis is that the variability associated with the execution of an action can be closely approximated by a linear combination of action bases in joint spatio-temporal space. We demonstrate that such a model bounds the rank of a matrix of image measure- ments and that this bound can be used to achieve recognition of actions based only on imaged data. A test employing principal angles between subspaces that is robust to statistical fluctuations in measurement data is presented to find the membership of an in- stance of an action. The algorithm is applied to recognize several actions, and promising results have been obtained. 1. Introduction Developing algorithms to recognize humans actions has proven to be an immense challenge since it is a problem that combines the uncertainty associated with computational vision with the added whimsy of human behavior. Even without these two sources of variability, the human body has no less than 244 degrees of free- dom () and modeling the dynamics of an object with such non- rigidity is no mean feat. Further compounding the problem, recent research into anthropology has revealed that body dynamics are far more complicated than was earlier thought, affected by age, ethnicity, class, family tradition, gender, sexual orientation, skill, circumstance and choice, . Human actions are not merely func- tions of joint angles and anatomical landmark positions, but bring with them traces of the psychology, the society and culture of the actor. Thus, the sheer range and complexity of human actions makes developing action recognition algorithms a daunting task. So how does one appropriately model the non-rigidity of human motion? How do we account for the personal styles (or motion signatures, ) while recognizing actions? How do we account for the diverse shapes and sizes of different people? In this paper, we consider some of these questions while developing a model of human actions that approaches these issues. To begin with, it is important to identify properties that are expected to vary with each observation of an action, but which should not affect recognition: Viewpoint The relationship of action recognition to object recog- nition was observed by Rao and Shah in , and developed fur- ther by Parameswaran and Chellappa in ,  and Gritai et al in . In these papers, the importance of view invariant recognition has been stressed, highlighting the fact that, as in object recogni-...
View Full Document
This note was uploaded on 06/13/2011 for the course CAP 6412 taught by Professor Staff during the Spring '08 term at University of Central Florida.
- Spring '08