CoTeSys Seminar: Invited Talk by Daniel Weinland about Action Representation and Recognition

Daniel Weinland is postdoctoral fellow at the Ecole Polytechnique Federale de Lausanne. He will give an invited talk about his work on action representation and recognition.

Time and Location

Thursday, December 11, 2008; 13:00 - 14:00; Room 03.13.010

Abstract

Recognizing human actions is an important and challenging topic in computer vision, with many important applications including video surveillance, video indexing and understanding of social interaction. From a computational perspective, actions can be defined as four-dimensional patterns, in space and in time. Such patterns can be modeled using several representations which differ from each other with respect to, among others, the visual information used, e.g. shape or appearance, the representation of dynamics, e.g. implicit or explicit, and the amount of invariance that the representation exhibits, e.g. a viewpoint invariance allowing to learn and recognize using different camera configurations.

In this talk I will present two such representations: "Motion History Volumes" are a free-viewpoint representation based on 3D reconstructions computed form multiple calibrated, and background-subtracted, video cameras. "Action recognition using 3D exemplars" is an approach for recognizing actions from arbitrary views, even from a single cameras. In experiments we demonstrate that the two representations can be used to learn and recognize basic human action classes, such as sitting-down, punching, or reaching, independently of gender, body size and viewpoint.