LOVI APP DEVELOPMENT BASED ON DEEP LEARNING MODEL AS A SMART ASSISTANT FOR LOW VISION

• Main Author: Sofiyati, Mitra
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/66630
• Institution: Institut Teknologi Bandung

ii ABSTRACT LOVI APP DEVELOPMENT BASED ON DEEP LEARNING MODEL AS A SMART ASSISTANT FOR LOW VISION By Mitra Sofiyati NIM: 18118029 (Telecommunication Engineering Program) The prevalence of visually impaired in Indonesia is the highest compared to other disabilities. Low vision is one part of the visually impaired who has a partial level of vision. One of the difficulties experienced by low vision is understanding the surrounding environment, especially those not familiar to them. The convenience they need to understand the surrounding environment is to recognize objects and navigate, especially outdoors, independently. One technology that we can use to help with low vision problems is artificial intelligence, especially image classification. On the other hand, smartphone technology users are prevalent in society, including those with low vision. Smartphone technology can also help with daily life because there are no additional costs to be incurred. In this final project, an image classification system is created that helps low vision to understand the surrounding environment. The system is built for the commonly used Android platform, with the core being three TensorFlow lite models. The application is named the LoVi application, with the core model referred to as the Sherpa model. This model consists of three models for classifying sidewalks on the road, classifying indoor objects, and nominal classification of rupiah banknotes. Based on the experiments and tests, the application can meet the expected specifications. Keywords: low vision, smartphone, TensorFlow Lite