Recognition of Two-Handed Posture Finger Turkish Sign Language Alphabet

Abstract

The importance of hearing-impaired individuals to communicate easily with other person of the society is indisputable. In this study, the recognition of two-handed posture finger alphabet of Turkish sign language was studied by using a device based on the sensor called Leap Motion. In addition to collecting data, the study consists of four stages: pretreatment, feature extraction, dimension reduction and classification. Recognition success was analyzed using traditional classifier which is one of the machine learning methods. In this analysis, the recognition model was tested using k-fold cross-validation and the results obtained were compared. In the analyses, performance obtained with the features selected by using the PCA (Principal Component Analysis) and LDA (Linear Discriminant Analysis) feature selection algorithms independently and in a hybrid structure, and all the features of the original data were compared in terms of recognition success. The results of detailed analyses showed that a more successful recognition is achieved with the features selected using the PCA and LDA algorithms in a hybrid structure.