High level models for sign language analysis by a vision system

Sign language processing is often performed by processing each individual sign and most of existing sign language learning systems focus on lexical level. Such approaches rely on an exhaustive description of the signs and do not take in account the spatial structure of the sentence. We present a high level model of sign language that uses the construction of the signing space as a representation of both (part of) the meaning and the realization of a sentence. We propose a computational model of this construction and explain how it can be attached to a sign language grammar model to help analysis of sign language utterances and to link lexical level to higher levels. We describe the architecture of an image analysis system that performs sign language analysis by means of a prediction/verification approach. A graphical representation can be used to explain sentence construction.