Abstract

The present work has the objective of developing a system for the automatic recognition of a playing card on a table, as part of a more general project to create a device to assist the blind in the chance game called “cuarenta”. The aim of this device will be to inform the user about the cards being played, via audio. For this phase of the project the algorithm used was k-NN, trained with a set of alphanumeric synthetic characters. The test set contained photographs taken in controlled lighting conditions, with the card positioned in arbitrary orientations. The parameterization of the algorithm gave a value of 1 as the optimal k, with which a classification error of 5% was obtained in the test set. Only two characters were confused by the classifier, the “A” and the “J”, with 20% and 40% errors each one. The algorithm was implemented in an embedded Raspberry Pi 3 system, obtaining a response time of 5 seconds, including the conversion to audio, and a memory occupation no greater than 60% of the total capacity of the system. These results suggest its applicability in portable devices.