Wearable device for Malaysian ringgit banknotes recognition based on embedded decision tree classifier

Abstract

The purpose of this project is to develop a wearable Malaysian Ringgit banknotes recognition device for assisting the visually impaired people to recognize the value of Malaysian banknotes. In this project, the RGB values in six different classes of the banknotes (MYR 1, MYR 5, MYR 10, MYR 20, MYR 50 and MYR 100) were taken at 12 different points (6 upside, 6 downside) using colour sensor (TCS 34725) before three features called RB, RG, and GB were extracted from the RGB values. After that, these features are used to model the embedded Decision Tree Classifier (DTC) in Matlab for recognizing each classes of banknote. Cross validation with 10-fold was used to select the optimize DTC which is based on the smallest cross validation loss. The performance of optimize DTC model is presented in confusion matrix and compared with Naïve Bayesian and k-Nearest Neighbour classifier before this model is implemented in Lilypad Arduino. The performance of the device in term of accuracy is evaluated by asking 10 subjects to use the device. Result shows that the proposed embedded optimize DTC model managed to achieve 84.7% accuracy which outperforms other classifier. In conclusion, proposed device is successfully developed and it should be possible, therefore, to integrate other features (instead of colour) in recognizing the ringgit banknote.