Neural Network Implementation of Divers Sign Language Recognition based on Eight Hu-Moment Parameters

Abstract

Improvement in the aspects of human-to-human and human-to-machine (and vice-versa) communication is still needed amidst the rapid development of technology. Divers sign language, a type of communication usually done underwater is the primary focus of this paper. Human divers are always at risk due to the unpredictable and unstable condition of water. With the help of image processing and artificial neural network, recognition of 13 commonly used hand signals is implemented. The significance of this study is with regards to the extension of the capabilities of a machine to interpret commands or meanings of signals. This adds to the probability of assurance of safety of divers especially when their voice equipment fails. The aim is to show the conformity and effectivity of relating underwater communications, image processing utilizing Hu-Moments as feature extraction method, and neural network. The results are shown through graphical representations of correlation coefficients, errors and success rates of pattern recognition. This research serves as a solution, although indirect, to present technologies such that people may consider the possibility of incorporating a neural network attribute.