Real-time sign languages recognition by artificial neural networks

In this thesis, universal sign languages recognition system is proposed using two different approaches. Although a variety of sign languages exist in use, the proposed system recognizes three widely used sign languages as the American Sign Language, British Sign Language and Turkish Sign Language to make the recognition universal. One of the proposed approaches is primarily based on Hausdorff distance and Hu invariants as the feature vectors and the system first focuses on how to process the hand movements and then it recognizes the different letters by the help of Hausdorff distance and Hu invariants measurements. The second approach is implemented by a feed forward neural network structure and recognition is provided with training data as three different sign language alphabets. It is shown that both approaches operate successfully while experimental results demonstrate that neural network based method provides superior performance. These two systems are useful and handy for hearing impaired people as they can easily communicate and no longer face difficulties in so many aspects of their lives.