Diseño y construcción de un dispositivo portátil para la identificación de billetes orientado a personas con discapacidad visual mediante el uso de visión artificial.

Abstract

In the current degree, work was developed in a portable device capable to detect bills and its denomination, guiding its use to people with visual impairment, employs artificial vision and computational learning. For the implementation of this prototype, the deductive, heuristic and experimental methods were applied. The same counts with a programming stage and other of design related to the hardware. The programming was carried out in Python, into it was developed a lineal classificator, logistic regression, obtaining to differentiate a bill from the bottom of an image. This was achieved with the extraction of characteristics of a set of training of bills and bottom, applying a variation of the descriptor Local Binary Pattern that do it invariant to the rotation, then these results were trained using logistic regression for obtaining the model with lower error cost of all the elements of training set. This model is multiplied by each window of applied image the descriptor in order to compare this value with certain threshold and classify them like bill or bottom. Once segmented the bill of 1,5,10 or 20 dollars, is applied the correlation for determining the better matching between the segmented bill and the templates informing in this way the bill denomination. Regarding to the hardware, a casing that contains a Raspberry pi 3 was designed, a Raspberry pi camera, a battery of 3000mAh, a loudspeaker of 0,25W with its respective circuit, a switch and button. The range of distance to get a good number of correct identifications is between 16 and 29 cm from the bill to analyse the device, taking a processing time of 2,41 minutes approximately. Is recommendable to reduce the employed time for the bills identification without affect its management such as to add other techniques or mechanisms to improve the identification results.