Abstract:
Sign language are used worldwide by many of individuals. They are mostly used by the
deaf communities and their teachers, or people associated with them by ties of friendship
or family. Speakers are a minority of citizens, often isolated, technically they are lack of
attention in such forms of communication. Actually, there is some, but limited research and
development in computer vision. In particular case of Ethiopian Sign language there is no
an efficient system to perform the real time recognition of Ethiopian sign language. With
advance of technology, there are new possibilities to find or solve this problem.
In this thesis, we develop a prototype and contribute to the problem of gesture recognition
apply to Ethiopian sign language. In this work we convert the hand gesture into text and
voice that is Ethiopian sign language to Amharic text and voice.
In this work computer vision technique is adapted, and capturing, pre-processing, feature
extraction, classification/identification also present. Open-source computer vision supports
us to implement our protype. Typically, Open-Source computer vision (OpenCV) provides
with better image or video processing function that helps the hand to process images using
different approaches including recognition, detection and reconstruction. These approach
helps in identification and detection of hand signs and obtain the related text that people
(lack of sign language) can be understand.
In this study we have done a real time prototype that convert Ethiopian Sign Language
(ETHSL) to text for six Amharic letters and for six Numbers (0-5) using web camera and
computer vision techniques.
When we evaluate the efficiency sign for number zero and sign for number 1 has highest
efficiency (higher value of efficiency is better) followed by sign 2 and the sign 4 while
sign for number 5 has the least efficiency.
On Amharic letter side, sign for ምልክት አ has highest efficiency (higher value of efficiency
is better) followed by ምልክት ሀ while sign for ምልክት ሐ has the least efficiency.
Accuracy of the prototype for Realtime conversion of sign language for Amharic letter into
text is 83.3% and sign language for Number into text is 76% for recognition. Limitations
are due to lower web camera resolution or physical strain of training dataset. Overall, the
work Realtime conversion system provides an easy to use and accurate Sign language input
and modality without placing restriction on users.