Learning Shape Models for a Vision Based Human-Computer Interface

Contents

Shape interpretation methods based on a stochastic graph can recognize nonrigid shapes, even if they are partially occluded, and interpret scenes composed of overlapping nonrigid shapes. These methods also identify most parts of each shape. This paper is an overview of work on learning stochastic graph models and their symbolic primitives from examples, for 2-D and 3-D shapes. A practical application of this work is a trainable real-time vision system, that allows users to control computer applications with hand gestures.