earticle

영어

In our society based on communication, the barrier between the spoken and gestural languages remains a main problem for the deaf-mute. To make this communication possible, we propose developing an interface using mobile phone camera with a module of recognition allowing the recognition and the restitution of the sign language in the form of synthesized words. This module is based on a spectral comparison between the image to recognize and a correlation multidecision segmented filter resulting from images reference coming from a predetermined database: correlation technique. The latter has been used for a long time to recognize a target image. However an adaptation of this technique is necessary to take into account particular characteristics of gestures (for which the target image to recognize evolves in time) and to be able to send it with a high transmission rate. Thus to obtain a reliable decision, the module of recognition must carry out a very large number of correlations, which slows down the translation process. In order to reduce this processing time, we propose implementing our recognition module by using multipost technique “PVM”.
This technique is based on the distribution of calculation on various stations in parallel. This
technique enables us to distribute the required computing effort on various parallel stations
using (PVM) a Parallel Virtual Machine. It showed significant gains in computing time but
not sufficiently for a real-time translation (mobile phone communication). To further reduce
processing time, we propose adding a Comprehension Pre-selection module to the initial
recognition module. The former is based on the division of images (resulting from the gesture
target) in various zones, in order to reduce and eliminate all the improbable gestures, according to the positions of hands. Thus, we reduce the needed identification time.