TECNOLOGIA ASSISTIVA PARA RECONHECIMENTO DE CARTAS DE BARALHO UTILIZANDO APRENDIZADO PROFUNDO

Abstract

We present in this paper an inclusive applicationthat allows visually impaired people to interact socially with otherindividuals in leisure activities such as card games. For this, webuilt a generalizable model that was trained for the detectionand recognition of playing cards by using convolutional neuralnetwork through deep learning. A system called Smart Assistantwas designed and implemented based on TensorFlow’s objectdetection API. At a predefined location, a digital camera is usedto detect cards in real-time. Then, these detected cards are sentto the classifier. After the classification, the SAPI Text to SpeechSynthesis API (TTS) is used to convert the labels of recognizedcards (in text format) to speech output. Experiments show thatin real game situations, the application can identify and classifycards with high assertiveness.