Sign recognition with an electronic glove driven by a neural network programmed on an 8-bit microcontroller

This paper compares the general functionality of two algorithms used to recognize ten selected signs from the Peruvian Sign Language captured at their final spatial posture through an electronic glove. The first algorithm uses thresholds that separate accepted values from rejected ones, therefore being able to identify predefined signs. The second algorithm consists in a neural network trained with backpropagation. Both algorithms were programmed on an 8-bit microcontroller. After analyzing the measured performance data, we conclude that the neural network achieves higher recognition accuracy at the expense of more memory usage and a small increment in execution time.