Electronic Thesis and Dissertation Repository

Degree

Master of Science

Program

Computer Science

Abstract

With the rapid development of mobile health, mechanisms for automatic data input are becoming increasingly important for mobile health apps. In these apps, users are often required to input data frequently, especially numbers, from medical devices such as glucometers and blood pressure meters. However, these simple tasks are tedious and prone to error. Even though some Bluetooth devices can make those input operations easier, they are not popular enough due to being expensive and requiring complicated protocol support. Therefore, we propose an automatic procedure to recognize the digits on the screen of medical devices with smartphone cameras.

The whole procedure includes several “standard” components in computer vision: image enhancement, the region-of-interest detection, and text recognition. Previous works existed for each component, but they have various weaknesses that lead to a low recognition rate. We proposed several novel enhancements in each component.

Experiment results suggest that our enhanced procedure outperforms the procedure of applying optical character recognition directly from 6.2% to 62.1%. This procedure can be adopted (with human verification) to recognize the digits on the screen of medical devices with smartphone cameras.


Share

COinS