Edge-Centric Augmented Reality Framework for Realtime Wristwatch Try-On

Abstract

The rapid expansion of online retail has intensified the need for realistic and interactive product visualization. Virtual try-on technologies have emerged as a critical tool for enhancing user confidence and reducing product return rates. Most existing research has focused on apparel-oriented solutions that rely on computationally intensive algorithms. In contrast, comparatively little attention has been given to smaller accessories such as wristwatches and jewelry, which present unique modeling challenges due to their scale and placement. Furthermore, the deployment of such systems on resource-constrained edge devices remains largely underexplored. In this work, we present a markerless augmented reality framework for wristwatch try-on, optimized for execution on smartphones and web browsers to enable real-time, privacy-preserving operation without reliance on cloud processing. The framework incorporates hand pose estimation, local 3D rendering, and buffer-based geometric parameter smoothing. Our approach integrates a hand landmark detection algorithm capable of estimating the watch model’s 3D position from three key hand landmarks, and introduces a buffer-based method for smoothing geometric parameters during movement. Features such as photorealistic reflections and physics-based materials are outside the current modeling scope. Our primary contribution is a lightweight, edge-executable pipeline for small-accessory try-on that achieves interactive frame rates (>30 fps) and a high level of visual quality. Evaluations using smartphone and web cameras demonstrate competitive rendering stability, with a mean opinion score of 4.35 on an introduced dataset, indicating that the augmented frames were generally perceived as highly realistic. These results demonstrate the feasibility of delivering immersive AR try-on for small accessories on edge-devices, offering a viable alternative to cloud-based solutions in online retail.