Abstract
Online furniture shopping presents an existing, persistent challenge: consumers frequently cannot accurately picture how their furniture will appear, fit, and match within their room before purchasing. The project uses a web-based augmented reality system based on computer vision, AI- driven recommendations, and accessible AR (Augmented Reality) for visualization to convert a single room photograph into an interactive design workspace. The system utilizes SAM (Segment Anything Model) to accurately segment walls, floors, and existing furniture, and deep learning object detection algorithms to detect patterns in room layout and scale. These capabilities are used to place and control 3D furniture models directly into the room images. The recommended engine is based on a multimodal recommendation system, calculating similarity between visual and text information characteristics to recommend tailored furniture choices. This application uses lighting- aware inpainting and image blending techniques to make sure that the new furniture is adjusted to match the lighting in the room.
Mobile AR visualization is made accessible using QR codes that load GLB 3D models using model-viewer and Google's Scene Viewer, enabling users to view furniture at real scale on smartphones without the requirement of downloading special apps.