Malaysia currency recognizer mobile application for visual impairment

Lee, Kyai Lun (2021) Malaysia currency recognizer mobile application for visual impairment. Final Year Project, UTAR.

Preview

PDF Download (3039Kb) | Preview

Abstract

In Malaysia, the prevalence of blindness for all ages has an average of 1.2%. People with vision impairments have difficulty recognizing objects through a vision in daily life, especially for some principal daily used objects such as a banknote. The braille feature on top of the banknotes becomes unusable in its tactile form after a brief usage circulation. Nevertheless, the existing currency recognizer is inefficient and not able to identify Malaysia currency. In this project, a mobile application based on recognizing currency has been developed using deep learning transfer learning techniques. The pre-trained deep learning model, MobileNet V2 utilized to transfer learning on our detection model. One hundred images of each value of Malaysia currency were collected as the dataset for training the model. The output lightweight model from the training process deployed on the mobile application. The mobile application is designed with vision-friendly features embedded into the mobile application through Android studio. Embedded features include vibration notification, real-time scanning on a phone camera, and real-time result voice feedback. Our system reached a mean average accuracy of 97% and average inference time of 0.06 second on detecting the currency. In conclusion, we bring a Malaysia banknotes recognizer mobile application in more accessible and more accurate ways.

Actions (login required)