Cleadr: AI-enhanced AR navigation app for seamless driving

Ting, Brandon En Junn (2025) Cleadr: AI-enhanced AR navigation app for seamless driving. Final Year Project, UTAR.

Preview

PDF Download (13Mb) | Preview

Abstract

Navigation systems have become an essential tool for drivers to navigate through journeys with ease and confidence. However, navigation systems are far from perfect as they still pose some limitations such as ambiguous directions, insufficient real-time assistance, and poor User Experience (UX). These limitations of navigation systems often lead to confusion and uncertainty for the driver. Hence, this project proposed a solution by integrating Augmented Reality (AR) and Artificial Intelligence (AI) into a navigation system. Specifically, a mobile AR navigation application integrated with AI assistance was proposed. The development of the system adopted the agile Extreme Programming (XP) methodology that allowed for quick and iterative development, as well as ample flexibility in responding to changing requirements. Core technologies involved were Flutter, Unity, PyTorch, and TensorFlow. The development of the system was structured into 3 core modules: the Maps Module, Navigation Module, and Intelligence Module. Core features of the system included the AR navigation and lane identification. AR navigation provided clearer directions by projecting them onto the real-world environment, while lane identification provided context-awareness for effective lane change instructions. A system performance evaluation was conducted with performance metrics such as response time and accuracy. The system was responsive with an overall response time of 1.80 seconds. Additionally, a lane identification model was trained from a custom Malaysian highway roads dataset that consisted of a total 22,806 images. The model managed to obtain an accuracy of 99.21%. In summary, this project successfully developed a mobile AR navigation application integrated with AI assistance to achieve the outlined project objectives. Moreover, the project contributed a lane identification model accustomed for Malaysian highway roads with reliable accuracy.

Actions (login required)