A Web-Based AR-Powered Virtual Eyewear Try-On System

Abstract

This research presents a web-based virtual try-on system for eyewear that aims to enhance the online shopping experience through computer vision and augmented reality techniques. The platform allows users to analyze their face shape and receive personalized eyewear recommendations, along with the ability to try on virtual glasses in real time. The system employs MTCNN to accurately retrieve facial regions from images, which are then processed by the VGG-16 model to classify face shapes into 5 distinct categories: oval, round, square, heart-shaped, and oblong. MediaPipe is used for facial feature localization, enabling calculation of pupil distance (PD) calculations and alignment of virtual glasses, and Three.js provides immersive 3D visualization for realistic try-on experiences. Performance evaluation proves a face shape classification accuracy of 92%, with occasional misclassifications for similar facial types. The augmented reality module achieves an average Intersection over Union (IoU) of 81% and a width error margin of approximately 5%, ensuring correct and visually appealing overlay alignment. By integrating face shape analysis and virtual try-on capabilities, this research contributes to advancing interactive and personalized solutions in the e-commerce domain, bridging the gap between traditional in-store and digital shopping experiences.