RELIX: A Lightweight Augmented Reality Heritage Guide using Image Recognition

Abstract Cultural heritage sites often struggle to communicate historical knowledge to visitors due to static information boards, limited signage, and language barriers. This creates a disconnected experience, especially for tourists, children, and visually impaired individuals. To address these gaps, this paper proposes RELIX, a lightweight mobile-based Augmented Reality (AR) system designed to deliver accessible cultural information in real time. RELIX integrates a MobileNetV2-based image recognition pipeline with a Flutter-based frontend and FastAPI backend to identify monuments, statues, and temple features through live camera input. Instead of computationally expensive 3D rendering, the system overlays lightweight 2D visual, textual, and audio explanations to improve accessibility and user engagement. Experimental testing demonstrates efficient inference time ranging between 400–700 ms, smooth AR overlay performance on mid-range smartphones, and positive user feedback regarding engagement, educational value, and accessibility. The findings suggest that combining AR with optimized machine learning provides an effective solution for enhancing cultural learning in heritage environments. Future improvements include multilingual support, dataset expansion, and offline inference using TensorFlow Lite. Keywords Augmented Reality, Cultural Heritage, MobileNetV2, Image Recognition, Accessibility, Mobile Computing, FastAPI, Flutter