Oxford - Sign Recognition

Our final goal is to perform sign recognition for signers which were not observed during training.

This is complicated by the fact that different people perform signs differently. These variations in signing are referred to as phonetic differences and depend on: region, formal/informal settings, co-articulation effects, if deafness was acquired early in life, etc.

To train a classifier which can generalise between signers, not only a large training set is required (which we can obtain automatically), but also features which support this transfer of knowledge.