Modelo de identificación de señas y detección de mala postura en aprendizaje del abecedario de la lengua de señas colombiana basado en inteligencia computacional

Abstract

A model of detection and correction of static signs within the LSC alphabet is proposed. The signs are captured using a device with an infrared camera called Leap Motion (LP), capable of obtaining in a graphical way the hand that is being visualized in front of it. In order to validate the sign recognition model, a database is built with a population of 38 people, where 25 samples per person are obtained. The sign recognition system uses three (3) individual classifiers from the following computational intelligence techniques: Support Vector Machines (SVM), Multilayer Perceptron (MLP) and Random Forests (BA). Subsequently, a Stacking-based Strong classifier is obtained from the individual classifiers. The Strong classifier achieves a sign recognition effectiveness with the test set of 97.41%. Subsequently, a correction of the signs is performed using a fuzzy Mandani-type fuzzy system.