GPU implementation for Arabic Sign Language real time recognition using Multi-level Multiplicative Neural Networks

Sign Language (SL) recognition has been explored for a long time now. Two main aspects of successful SL recognition systems are required: High recognition accuracy and real-time response. This paper shows a contribution in these issues, the first contribution describes a real-time response recognition for Arabic Sign Language (ArSL) based on a Graphics Processing Unit (GPU) implantation. The second contribution exploits Multi-level Multiplicative Neural Network(MMNN) for hand gesture classification. The system architecture mainly depends on two consequent layers of (MMNN), the first layer determines if the signer uses one hand or two hands and the second determines the final class. The experiment was conducted on 200signs and the resultreaches83% recognition accuracy for test data confirming objects dataset offline extendibility. The recognition system is being accelerated using NVIDIA GPU and programming in CUDA.