Técnicas de Clustering basada en Redes Neuronales para reconocer gestos estáticos del alfabeto dactilológico de la Lengua de Señas Peruana

Hoyos C.A.E.; Huamaní-Navarrete P. · Proceedings of the LACCEI international Multi-conference for Engineering, Education and Technology (2021) · DOI: 10.18687/laccei2021.1.1.544

Resumen

The Peruvian Sign Language (LSP) is the language used by people with hearing disabilities throughout the national territory [1]. At this point, in the social reality of our country there is still an isolation from the deaf community due to the lack of knowledge about this section of population and its needs, this situation has as result a social problem, turning them into an isolated and vulnerable community. Faced with this lack, the possibility of applying current technologies for the deaf community is envisaged using Artificial Neural Networks (RNA) in the field of pattern recognition and classification (clustering) as many authors have proven. The rationale behind this article was to recognize 24 static gestures of the LSP dactylological alphabet, acquired of an electronic glove and using two clustering techniques based on the neural networks of Generalize Regression and Learning by Vector Quantization with the support of the Toolbox Deep Learning del Matlab. Moreover, three different neural network topologies were applied for each technique: for the gestures, for spatial orientation sensors (MPU) a conjugate so-called for grouping results from the anterior topologies. The recognition of static gestures was assessed using the Contingency Table and Precision, Recall and Magnitude-F metrics. The results revealed the Neural Network of Generalize Regression gave better results in the recognition of gestures, obtaining precision and recall measurements equal to 100% and a Magnitude F in the same way, along with a noise resistance of 1%; by contrast, the Neural Network of Learning by Vector Quantization, reached just a Precision measurement equal to 81.25% and Recall to 87.5%, obtaining a Magnitude F equal to 84.26% and nearly not a thing of noise tolerance.

Instituciones: Universidad Ricardo Palma