Arabic sign language recognition algorithm based on deep learning for smart cities

The deaf community relies on sign language as a primary means of communication. Automatic finger spelling recognition transcribes sign language into text. Such recognition will facilitate communication between deaf-mutes and ordinary citizens. The present work proposes a new method for static Arabic sign language recognition. Our method is based on the deep learning Alexnet model and the categorical cross-entropy loss function. It recognizes 32 classes of alphabets. The advantages of our method have been demonstrated experimentally on ArSL2018 dataset. The experiments show that our method offers an average F-measure of 97.38%.