Deep learning algorithms to recognize word based Turkish sign language

Sign language is a form of visual communication used by people with hearing problems to express themselves. The main purpose of this study is to make life easier for people with hearing problems. In this study, a data set was obtained using 3200 RGB images for 32 classes taken from three different people. Data development methods were applied to the data sets and the number of images was increased from 3200 to 19200, 600 per class. For the classification of the signs, both a 10-layer convolutional deep network model was created for the solution of the problem, and VGG166, Inception and ResNet deep network architectures, which are one of the deep learning methods, were applied by using transfer learning method. In addition, the signs are classified using the Support Vector Machines (SVM) and K- Nearest Neighbor (K-Nearest Neighbor, K-NN) methods, which are the traditional machine learning methods, with the feature vector obtained by using the feature extraction technique of deep learning. The most successful method was determined by comparing the obtained results according to time and performance ratios. In this study, stationary words belonging to Turkish Sign Language (TSL), which is a visual language, are translated into real time written language by using transfer learning with one of the deep learning methods, which is successful as a result of the analysis. In addition, with the real-time system designed, its success in recognizing the stationary words of TSL signs and printing its prediction on the computer screen were evaluated.