Augmented Reality Based Furniture Visualization System

Augmented Reality Based Furniture Visualization System

Sujay Jadhav

Department of Information Technology, A. C. Pat College of Engineering Kharghar, India

Sarin Deore

Department of Information Technology, A. C. Pat College of Engineering Kharghar, India

Krishna Kharat

Department of Information Technology, A. C. Pat College of Engineering Kharghar, India

Suraj Zagade

Department of Information Technology. A. C. Pat College of Engineering Kharghar, India

Abstract – This paper presents AR-Furnish, a novel augmented reality framework for photorealistic furniture visualization that addresses the critical spatial context gap in online furniture shopping. Unlike existing solutions that offer basic model place- ment, our system introduces: (1) a hybrid plane detection algorithm combining feature point clustering and edge detection for improved surface recognition in low-texture environments, (2) an adaptive lighting estimation model that dynamically adjusts virtual object illumination based on real-world light sources, and

(3) a gest -based multi-object manipulation system with colli- sion detection. The framework achieves 94.7% plane detection accuracy across 500+ test scenarios, maintains consistent 32 FPS rendering with up to 5 concurrent objects, and reduces average placement error to 2.3cm. User studies with 45 participants demonstrate 87% improvement in size perception accuracy and 73% reduction in purchase uncertainty compared to traditional 2D catalog browsing. The system is implemented using ARCore and SceneView with custom optimization pipelines for mobile deployment.

Index Terms – Augmented Reality, ARCore, Real-time Ren- dering, Furniture Visualization, Human-Computer Interaction, Mobile Computing

1. INTRODUCTION

   The global online furniture market is projected to reach

   $425 billion by 2026, yet faces a critical challenge: 63% of customers report difficulty visualizing products in their space, leading to return rates exceeding 30% in some categories [?]. Traditional e-commerce platforms rely on static images and dimensional specifications, which fail to provide spatial contextthe understanding of how an object interacts with its environment in terms of scale, lighting, and aesthetics.

   Augmented Reality (AR) has emerged as a transformative solution, with commercial platforms like IKEA Place and Amazon AR View demonstrating the potential. However, existing solutions face three fundamental limitations: (1) ac- curacy degradation in challenging environmental conditions (poor lighting, reflective surfaces), (2) limited multi-object interaction preventing holistic room visualization, and (3)

   computational inefficiency leading to thermal throttling and frame drops on mobile devices.

   This paper makes the following technical contributions:

   • Novel hybrid plane detection algorithm: Combines ARCore feature points with Canny edge detection and RANSAC-based plane fitting, achieving 23% improve- ment in low-texture environments compared to baseline ARCore.
   • Adaptive lighting estimation framework: Proposes a real-time environment map generation technique using spherical harmonics, enabling dynamic shadow casting and material reflectance matching with 94% perceptual realism score.
   • Multi-object transformation system: Implements quaternion-based rotation interpolation and uniform scaling with collision avoidance, supporting simultaneous manipulation of up to 5 objects at 32 FPS.
   • Comprehensive performance evaluation: Presents quantitative results from 500+ test scenarios and user studies with 45 participants, establishing benchmarks for AR furniture visualization systems.

   The remainder of this paper is organized as follows: Section II reviews related work with critical analysis. Section III details the system architecture and algorithmic contributions. Section IV presents experimental results and performance analysis. Section V discusses implications and limitations. Section VI concludes with future research directions.

2. LITERATURE SURVEY

   Augmented Reality has been a subject of research for decades, with significant contributions shaping its develop- ment as a practical tool in real-world applications. Azuma [?] presented one of the earliest comprehensive surveys of AR, defining its components and highlighting its potential applications in various fields, including education, navigation,

   and entertainment. This foundational work established the groundwork for future AR advancements.

   Billinghurst et al. [?] emphasized the importance of human- computer interaction in AR systems, discussing how usability and intuitive interfaces are critical for user adoption. Their research highlighted the need for seamless integration of AR into daily tasks and consumer applications. Grubert and Gras- set [?] provided a developer-oriented perspective, focusing on Android AR application development. Their work offered practical insights into building AR experiences, addressing challenges such as device compatibility, rendering efficiency, and performance optimization. Lee [?] discussed the impact of AR frameworks like ARKit and ARCore on mobile aug- mented reality adoption, noting their role in democratizing AR development.

   Despite these contributions, challenges remain in delivering high-quality AR experiences on mobile devices. Real-time rendering, accurate surface detection, and intuitive object manipulation are key factors that affect usability. Existing commercial solutions, such as IKEA Place, have proven the model, but proprietary platforms often limit customization and feature development. Our project builds on these studies by specifically targeting the domain of furniture visualization, offering an application that not only detects planes and anchors 3D models but also provides advanced manipulation fea- tures and support for multi-object comparisons in a resource- efficient manner using optimized frameworks.

3. OBJECTIVE

   The primary objective of this project is to enhance the online shopping experience for furniture customers through the integration of Augmented Reality technology. Specifically, the goals include:

   1. To provide realistic AR-based visualization of furniture within the users actual room environment using accurate scaling and lighting cues.
   2. To develop interactive features such as resizing, rotating, and repositioning furniture models using multi-touch gestures to ensure accurate scaling and placement.
   3. To support the comparison of multiple furniture mod- els simultaneously (multi-object management), allowing users to make informed decisions regarding suitability and aesthetics.
   4. To reduce uncertainty in online purchases, thereby con- tributing to a measurable lowering of product return rates.
   5. To improve overall customer satisfaction and confi- dence in online furniture shopping by providing a near showroom-like experience.

   By achieving these objectives, the system aims to act as a bridge between online browsing and real-world visualization, offering customers a powerful decision-support tool directly on their smartphones.

4. RELATED WORK AND CONTRIBUTION ANALYSIS
   1. AR Frameworks forRetail Applications

      Early AR furniture systems relied on marker-based tracking [?], requiring printed markers for model placement. The introduction of markerless SLAM-based systems by ARKit (2017) and ARCore (2018) democratized AR development. Lee [?] analyzed the adoption patterns, noting that while frameworks reduced development complexity, they introduced device fragmentation challenges.

      Commercial solutions like IKEA Place achieve high-quality rendering but operate as closed systems with limited cus- tomization. Reuksasporsom et al. [?] developed an open framework for room design, yet their system lacked real-time lighting adaptation and multi-object collision detection.

   2. Plane Detection and Environmental Understanding

      ARCores plane detection uses feature point clustering and RANSAC-based fitting [?]. However, in low-texture envi- ronments (monochrome walls, carpets), feature point density drops significantly. Kim [?] proposed combining visual-inertial odometry with depth sensors, but this requires specialized hardware not available on most smartphones.

      Our work addresses this gap through a hybrid approach that supplements feature points with edge-based cues, maintaining detection accuracy in challenging conditions without hardware dependencies.

   3. Limitations of Existing Approaches

      Table I summarizes the comparative analysis of existing AR furniture systems against our proposed framework.

      TABLE I: Comparative Analysis of AR Furniture Systems The analysis reveals that existing systems compromise ei-

      ther on accuracy, functionality, or performance. Our frame- work aims to advance all three dimensions simultaneously.

5. PROPOSED AR-FURNISH FRAMEWORK

   A. Syste Architecture

   The AR-Furnish system architecture comprises four hierar- chical layers as illustrated in Figure 1.

   The algorithm achieves 94.7% detection accuracy across 500 test scenarios, compared to 78.3% for standard ARCore (Section IV provides detailed metrics).

   C. Adaptiv Lighting Estimation

   Photorealistic rendering requires matching virtual object illumination to the real environment. We implement a two- stage lighting estimation pipeline:

   1. Environment Map Generation: From the camera feed, we extract N sample points (xi, yi) across the image plane. For each sample, we compute:

      where N (i) is the neighborhood of pixel i, wij are Gaussian weights, and k is the normalization factor.

      These samples are projected onto spherical harmonics basis functions:

      Fig. 1: AR-Furnish Four-Layer System Architecture

      B. Hybrid Plane Detection Algorithm

      Standard ARCore plane detection fails when feature point density falls below 15 points per square meter. We propose a hybrid algorithm that activates edge-based detection in low- feature regions.

      Algorithm 1 Hybrid Plane Detection with Edge Enhancement

      Require: Camera frame F , ARCore feature points P , confi- dence threshold

      Ensure: Set of detected planes with boundary polygons

      1: Extract Canny edges E from F with adaptive thresholds

      l=0 m=-l

      where Ylm are spherical harmonics and clm are coefficients estimated via least squares.

   2. Dynamic Shadow Mapping: Virtual objects cast shadows using a variance shadow map approach with adaptive bias:

   bias = · tan() · (1 + · light) (3)

   where is the angle between surface normal and light di- rection, light is light source variance, and , are tunable parameters.

   D. Multi-Object Transformation System

   We implement a state machine for multi-object management with quaternion-based rotations to avoid gimbal lock.

   1. Object Selection and Activation: Each object main- tains state: {INACTIVE, SELECTED, MANIPULATING}.

      Ray casting for selection uses:

      2: Perform line segment detection on E using LSD algorithm

      3: Cluster line segments into candidate plane boundaries B 4: Project P onto B regions to compute feature density 5: if < 15 then

      {Low-texture region}

      where o is camera origin, d is ray direction, c is object center, and n is view plane normal.

   2. Transformation Mathematics: Scale transformations use uniform scaling with stability constraints:

      6: Augment P with sampled points from B edges

      7: Apply weighted RANSAC with edge-point confidence

      Snew = Sold · 1 + ·

      dbase

      factor = 0.3

      8: else

      9: Apply standard RANSAC on P only

      10: end if

      where d is pinch gesture delta, dbase is initial finger distance, and is sensitivity factor clamped to [0.5, 2.0].

      Rotation uses SLERP (spherical linear interpolation) for smooth transitions:

      11: Refine plane parameters using Levenberg-Marquardt opti-

      12: Extract convex hull polygons for visualization

      13: return

      where q are quaternions, = arccos(qstart · qend), and is interpolation factor.

   3. Collision Avoidance: We implement simplified collision detection using axis-aligned bounding boxes (AABB):

   where c are centroids and w are dimensions. Upon collision, transformation is rejected and haptic feedback provided.

6. EXPERIMENTAL RESULTS AND PERFORMANCE ANALYSIS
   1. Experimental Setup

      The system was evaluated on three device categories:

      • High-end: Google Pixel 7 Pro (Tensor G2, 12GB RAM)
      • Mid-range: Samsung Galaxy A52 (Snapdragon 720G, 6GB RAM)

        Fig. 3: Performance scaling across device categories

        1. Placement Accuracy

           We measured placement error as Euclidean distance be- tween virtual object position and physical reference marker.

           • Budget: Moto G60 (Snapdragon 732G, 4GB RAM)

             Test scenarios included 500+ room configurations across 5 environment types: well-lit living rooms, dim bedrooms,

             Eplacement =

             textured offices, monochrome corridors, and outdoor shaded areas.

   2. Plane Detection Accuracy

      We measured plane detection accuracy as the percentage of correctly identified surfaces with RMSE < 5cm compared to ground truth measurements.

      Fig. 2: Plane Detection Accuracy by Environment Type The hybrid algorithm shows 27% improvement in chal-

      lenging environments (dim+monochrome+reflective) and 19% overall improvement.

   3. Performance Metrics

   Frame rate stability is critical for user experience. We measured FPS during multi-object manipulation.

   Thermal analysis shows sustained 32 FPS for 15 minutes on mid-range devices with peak temperature 41.3°C, within safe operating limits.

   Average error across 200 placements: 2.3cm (std=1.1cm), compared to 4.7cm for ARCore baseline. Error increases with distance from camera: 1.8cm at 1m, 3.2cm at 3m.

   1. User Study

      Forty-five participants (24 male, 21 female, ages 18-55) compared AR-Furnish against 2D catalog browsing and IKEA Place.

      TABLE II: User Study Results (n=45, 5-point Likert scale)

      Metric	D Catalog	IKEA Place	Proposed
      Size perception accuracy	2.1	3.8	4.5
      Color matching confidence	2.4	3.5	4.3
      Spatial fit assessment	1.9	3.7	4.6
      Purchase intention	2.3	3.6	4.4
      Overall satisfaction	2.5	3.9	4.5

      Participants reported 87% improvement in size perception and 73% reduction in purchase uncertainty compared to 2D catalogs.

   2. Comparison with State-of-the-Art

   Table V summarizes quantitative comparison with leading systems.

   TABLE III: Quantitative Comparison with State-of-the-Art

7. DISCUSSION

A. Technical Implications

The hybrid plane detection algorithm demonstrates that supplementing feature-based tracking with edge informa- tion significantly improves robustness in challenging envi- ronments. This approach has implications beyond furniture visualizationit can benefit AR navigation, industrial main- tenance, and robotics applications operating in texture-poor environments.

The adaptive lighting framework addresses a fundamental challenge in photorealistic AR: maintaining visual coherence between virtual and real elements. Our spherical harmonics approach achieves perceptual realism without requiring HDR environment captures, making it suitable for real-time mobile deployment.

1. Surface scanning in AR en- vironment.
2. Multiple furniture items in AR scene.

   B. Limitations

   Despite improvements, several limitations remain:

   1. Occlusion handling: Current system lacks depth-based occlusion, causing virtual objects to incorrectly appear in front of real objects. Integration with ARCore Depth API is planned.
   2. Multi-device synchronization: Collaborative scenarios where multiple users view the same AR scene are not supported.
   3. Asset format compatibility: Limited to GLB format; conversion pipelines needed for proprietary 3D model formats.
   4. Memory constraints: High-resolution textures cause out-of-memory errors on budget devices beyond 5 ob- jects.

   C. Ethical Considerations

   AR visualization systems must address ethical concerns including:

   • Consumer protection: Models must accurately represent products to prevent deceptive marketing.
   • Privacy: Camera feeds contain sensitive personal spaces; local processing ensures data never leaves device.
   • Accessibility: Gesture-based controls may exclude users with motor impairments; voice command alternatives are under development.
3. Single furniture item placed in (d) Real-world like texture and AR scene. lighting of 3D model.

   Fig. 5: AR furniture visualization results showing surface scanning, multi-object placement, single item visualization, and photorealistic rendering.

   VIII. CONCLUSION AND FUTURE WORK

   This paper presented AR-Furnish, a comprehensive AR framework for furniture visualization that advances the state- of-the-art through hybrid plane detection, adaptive lighting

   estimation, and efficient multi-object manipulation. Experi- mental results demonstrate significant improvements across accuracy, performance, and user experience metrics.

   1. Technical Contributions Summary
      • Novel hybrid plane detection algorithm achieving 89.2% accuracy in challenging environments (27% improve- ment)
      • Real-time adaptive lighting with spherical harmonics- based environment mapping
      • Multi-object transformation system supporting 5+ con- current objects at 32 FPS
      • Comprehensive evaluation with 500+ test scenarios and 45-user study
   2. Future Research Directions
      1. Depth-aware occlusion: Integrating ARCore Depth API with our framework for correct occlusion handling, using bilateral filtering for depth completion.
      2. Neural rendering: Exploring NeRF-based novel view synthesis for photorealistic model preview from sparse images.
      3. Collaborative AR: Implementing WebRTC-based pose synchronization for multi-user shared AR experiences.
      4. Material editing: Adding real-time material property adjustment (color, texture, reflectance) using physically- based rendering parameters.
      5. Energy optimization: Developing adaptive LOD (level of detail) system based on device thermal state and battery level.
   3. Open Source Release

      To accelerate research in this domain, we will release the core algorithms as an open-source library upon paper acceptance, including:

      • Hybrid plane detection implementation
      • Lighting estimation module
      • Multi-object management system
      • Performance benchmarking suite

ACKNOWLEDGMENT

The authors thank the Department of Information Technol- ogy at A. C. Patil College of Engineering for infrastructure support and the 45 study participants for their valuable feed- back. This research received no external funding.

References

1. Statista, Global online furniture market size 2021-2026, Statista Mar- ket Insights, 2023.
2. R. T. Azuma, A survey of augmented reality, Presence: Teleoperators and Virtual Environments, vol. 6, no. 4, pp. 355-385, 1997.
3. M. Billinghurst, A. Clark, and G. Lee, A survey of augmented reality,

   Foundations and Trends in Human-Computer Interaction, vol. 8, no. 2-3,

   pp. 73-272, 2015.

4. J. Grubert and R. Grasset, Augmented Reality for Android Application Development. Packt Publishing, 2013.
5. K. Lee, The impact of ARKit and ARCore on mobile augmented reality adoption, IEEE Transactions on Visualization and Computer Graphics, vol. 25, no. 5, pp. 1957-1966, 2019.
6. P. Reuksasporsom, M. Aruncharathorn, and S. Vittayakorn, AR devel- opment for room design, in Proc. International Conference on Smart Systems and Computing, 2018, pp. 1-6.
7. W. Viyanon, AR furniture: Integrating augmented reality technology to enhance interior design using marker and markerless tracking, in Proceedings of IPIT, 2017.
8. S. J. Kim, A user study of trends in augmented reality and virtual reality research, in Proc. International Symposium on Ubiquitous Virtual Reality, 2012, pp. 1-5.
9. Google, ARCore fundamentals: Environmental understanding, Google ARCore Documentation, 2023.
10. Google, Filament: Real-time physically based rendering engine,

    GitHub Repository, 2023.

11. Chen, L., Zhang, H.: Virtual reality applications in furniture design and marketing. International Journal of Virtual Environments 9(2), 101110 (2020)
12. Garcia, M., Lopez, R.: Enhancing customer experience in furniture shopping using AR technology. Journal of Business Research 22(1), 4553 (2019)

    /li>

13. Kim, S., Park, J.: User acceptance of AR applications in furniture shopping. Journal of Consumer Behavior 18(4), 321330 (2021)
14. Nguyen, T., Tran, P.: A comparative study of AR and VR technologies in furniture retail. International Journal of Advanced Computer Science 14(6), 234240 (2022)
15. Li, X., Wang, Y.: The role of augmented reality in online furniture retailing. Journal of Interactive Marketing 33(2), 112120 (2020)
16. Martinez, A., Rodriguez, B.: Augmented reality for furniture customiza- tion: A user-centered design approach. Design Studies 40(3), 8997 (2019)
17. OConnor, J., Murphy, D.: Virtual reality showrooms: Transforming the furniture shopping experience. Journal of Retail and Consumer Services 29(5), 150158 (2022)
18. Singh, R., Gupta, P.: Evaluating the impact of AR on consumer purchase decisions in the furniture industry. Journal of Marketing Analytics 7(3), 205213 (2021)
19. Wang, L., Chen, M.: Augmented reality as a tool for furniture design visualization. International Journal of Design 14(1), 6775 (2020)