Multi-Resolution 3D Rendering for High-Performance Web AR.

Area of Science:

• Computer Science
• Computer Graphics
• Human-Computer Interaction

Background:

• Web augmented reality (AR) faces challenges in rendering large 3D models with high visual quality and frame rates.
• Existing 3D formats often degrade original data quality, negatively impacting user experience.

Purpose of the Study:

• To investigate web-streamable, view-dependent representations for large, high-resolution 3D models in web AR.
• To develop and evaluate a dedicated, lightweight web AR format for improved 3D visualization and performance.

Main Methods:

• Developed a cross-platform prototype using Nexus.js, AR.js, and Three.js for batched multi-resolution 3D model handling.
• Implemented real-time view-dependent rendering, geometric instancing, and 3D pose regression for natural feature tracking and location-based AR.
• Tested the prototype against common formats and compression techniques.

Main Results:

• Achieved up to a 46% speedup in rendering time compared to optimized glTF models.
• Rendered a 34 million vertex 3D model in under 4 seconds with no visual quality degradation on slow networks.
• Demonstrated high-quality visualization and real-time performance across various scenes and devices.

Conclusions:

• The proposed multi-resolution scheme using Nexus.js offers an efficient solution for web AR.
• This approach significantly improves rendering speed and visual fidelity for large 3D models in web AR applications.
• Enables high-quality, real-time 3D visualization in web AR, overcoming current limitations.