TL;DR
This paper introduces PaRUS, a novel VR shopping method that reconstructs a user's real-world space to allow virtual furniture selection within that context. This approach significantly reduces the discrepancy between user expectations and actual product fit, leading to increased trust and satisfaction in online shopping.
What Problem Does This Paper Address?
Online shopping, especially for large items like furniture, often suffers from a disconnect between how a product looks in a virtual catalog or a generic virtual mall and how it will actually fit and look in a user's home. Existing VR shopping experiences typically place products in virtual malls or abstract metaphorical scenes, failing to provide the crucial context of the user's own living space. This leads to a "discrepancy" between pre-purchase expectations and post-purchase reality, often resulting in dissatisfaction and returns. The paper aims to bridge this gap by enabling users to visualize and select products within a virtual replica of their actual usage environment.
Methods
The core of PaRUS is a semantic scene reconstruction pipeline. This pipeline has two main stages:
- Scene Understanding: This involves capturing the structure and layout of a real-world room. The authors use Apple's RoomPlan API (leveraging LiDAR and camera data) to create a structured 3D model of the room, representing objects as bounding boxes without detailed textures or shapes. This structured scene is then imported into Unity for further processing.
- Object Reconstruction: This stage focuses on capturing the detailed shape and texture of individual objects. Users take multi-view photographs of objects within the real scene. These images are then processed using PhotoCatch (an ARKit-based photogrammetry tool) to generate raw 3D models. These models are then optimized using MeshLab to remove extraneous geometry and reduce polygon count for better performance in Unity.
Finally, these reconstructed 3D object models are integrated into the structured scene within Unity, replacing the bounding boxes. Lighting and post-processing are applied to create a realistic virtual representation of the user's space, ready for product selection.
Datasets
Not discussed in the paper. The method relies on capturing data from the user's own environment using mobile devices (iPad Pro) and processing it with tools like RoomPlan, PhotoCatch, and Unity.
Key Insights and Ideas
- Context is King: The most significant insight is that the real usage scene is a critical, yet often overlooked, context for product selection in VR shopping. Simply placing products in a virtual mall isn't enough; users need to see them in their own space.
- Reconstruction Enables Trust: By virtually reconstructing a user's environment, PaRUS significantly reduces perceived performance risk and builds trust. Users feel more confident in their choices because they can visualize the product's fit and aesthetic in situ.
- VR Outperforms AR for Contextual Fit: While AR can overlay products onto the real world, VR, when combined with scene reconstruction, offers a more seamless and integrated experience for visualizing product fit within a complete environment.
- User Experience Trade-offs: While VR shopping in a reconstructed scene (RV) offers superior immersion and reality perception, it can also increase user workload and decision time compared to simpler web-based methods.
Key Results
The paper presents findings from two user studies:
User Study 1 (Comparing Presentation Methods: Web, AR, VR): * Reality & Immersion: VR (RV) and AR (RA) significantly outperformed Web (RW) in terms of felt sense of reality and immersion. RV achieved the highest scores. * Workload: Web (RW) was the least demanding. RV and RA had comparable workloads, with RV slightly higher in physical load and RA in mental load. * User Experience: RV was rated highest for attractiveness, stimulation, and novelty, indicating a more engaging and trustworthy experience. However, it did not significantly improve efficiency. * Qualitative Feedback: Users preferred RV for its intuitive product-environment correlation and reduced sense of reality discrepancy. However, VR beginners experienced dizziness and a higher learning curve.
User Study 2 (Comparing VR Scenes: Metaphorical vs. Reconstructed): * Perceived Performance Risk: Reconstructed virtual scenes (RV) significantly reduced perceived performance risk by 39.15% compared to metaphorical scenes (VV). * Trust & Satisfaction: Participants favored RV, reporting higher trust in their decisions and a greater likelihood of satisfaction due to the intuitive context. * Objective Measures: While RV took longer for selection, more jumps between products, and longer average viewing time per product, these differences were not statistically significant. The authors suggest this longer engagement might indicate more thorough consideration rather than inefficiency.
Limitations and Future Work
- Lack of Haptic Feedback: The studies primarily focused on visual perception. The authors acknowledge that tactile feedback devices could further enhance the sense of reality and immersion.
- 2D UI in VR: The use of 2D interfaces within VR environments might detract from the overall immersion. Future work could explore 3D or other UI paradigms.
- Single Item Preview: Participants could only preview one furniture piece at a time. Supporting multi-item previews would be beneficial.
- User Discomfort: Dizziness and discomfort associated with VR equipment remain a challenge for widespread adoption, especially for new users.
- User Control & Customization: Users expressed a desire to freely redesign and customize the reconstructed environment (e.g., adjust lighting, move objects). Future work could incorporate these interactive features.
- Search & Filtering Efficiency: Enhancing search and filtering capabilities within VR scenes is suggested as a way to improve the user experience.
Summary updated