Graphical Models for Fusion of Higher and Lower Level Information in Tracking and Gesture Recognition
For gesture recognition, in most cases it is required to track the body or hand of the user and then compute visual features from this information and perform classification in a subsequent step.
This talk will present an approach to perform integrated tracking and gesture recognition in one single step using Graphical Models (GM). The first part of the talk will present a brief introduction into GMs, starting with a very simple classification scheme and scaling this up to several popular paradigms, such as Gaussian Mixtures, Neural Networks, Hidden Markov Models and finally Graphical Models and thus can show that GMs have the expressive power to unite all these paradigms. This will be the basis for the second part of the talk, where it will be shown that GMs can be also used to model a Particle Filter and the combination of this with a GM-implemented classifier will eventually result into the combined tracking-gesture recognition scheme.
Ort: INB Seminarraum
