Research
Objective
Short Term : To create a system that can distinguish one sign from another without requiring intrusive equipment such as gloves or head gear.
Long Term : To expand on the above to create a system that can interpret British Sign Language user independantly regardless of situation providing an invaluable asset to signers and those who work with them.
Why is it difficult?
- BSL is a complex language and has many thousands of signs each differing from the next by minor changes in hand motion, shape or position. It also has a complex grammar completely differing from English grammar for example adverbs are often shown by manipulating the action sign rather than compounding signs together creating many variations on a theme.
- Different users will vary in many different ways depending not only on their physical form but on their fluency in sign as well. A new signer will be slower, more deliberate and make larger signs as opposed to a proficient signer who will sign quickly, often combining signs together and who uses a smaller sign space.
- It is impossible to envisage all environments in which the system could be used, therefore any previous expertise that required a known background can be dismissed. It is not even certain that a background will be static, for example if a news reporter were filing from in front of a busy street the background could be a mass of moving traffic and people.
Approaches
Most recently work has focused on working with large weakly supervised data sets from the television to counteract the lack of accurately labelled sign data. Previously the focus was on Large Lexicon Detection of Sign which concentrates on detection at the viseme level and before this Boosted Spatio-Temporal Features were investigated as a method for word level recognition. Details are available throughout this website and related papers are on the publications page.