Boosted Spatio-Temporal Features for Sign Language Recognition

We have been investigating the use of Spatio-Temporal Features to accuarately describe the motion of sign, using boosting to combine the most effective features into a single strong classifier per sign.

A brief look at this system is given over the following pages:

The Integral Volume - An introduction to the basis of our method.

Spatio-Temporal Features - The features that make use of the Integral Volume.

Image Buffer - Our solution to numerical precision over long video streams.

Learning - Some thoughts on Boosting algorithms and their suitability for this problem.

Results - Results of experimentation