Desing and application of a sensory glove for learning and classification of Turkish sing language for disabled people

In this study, a sign language recognition glove design and application was made that could help people with hearing or speech disabilities be able to communicate. In the design gloves, 12 flexibility sensors, 2 inertial sensors, 10 magnetic field sensors and enhancement cards were used. In total, 34 signals were obtained from the sensors used and a signal Matrix was created. Matrices consisting of sign language movements were then transformed into input vectors consisting of 272 attributes by subjecting them to statistical operations such as variance, rudeness, skewness and standard deviation. Machine learning algorithms are able to recognize signs made by the glove user by learning a database previously constructed from attributes of sign language movements. Logistic Regression, Artificial Neural Networks, Support Vector Machines, Naive Bayes, decision trees, random forest, and Nearest Neighbor techniques were used to recognize selected sign language movements. With the system created, 86.4% Test success was achieved by training for 32 words with 320 sample sign language movements. Ardından örnek sayısı arttırılıp 960 örnek ile eğitim yapılarak 96,9% test başarısı elde edilmiştir.