Collaborative VR-Based 3D Labeling of Live-Captured Scenes by Remote Users.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Virtual Reality

Background:

• Previous interactive 3D labeling systems focused on single-user experience in VR/AR to accelerate scene labeling.
• Existing methods often lack efficient collaboration for large-scale or real-time data.

Purpose of the Study:

• To develop a novel, collaborative VR-based 3D labeling system for live-captured scenes.
• To enable multiple remote users to contribute to 3D scene labeling simultaneously and efficiently.

Main Methods:

• A system integrating RGB-D capture, a reconstruction client for 3D model generation, and a central server for scene management and label propagation.
• Independent VR labeling clients for users to explore and label scene parts from optimal viewpoints.
• Utilizing sparse multi-user input for automatic label propagation and prediction across the 3D scene.

Main Results:

• Demonstrated an intuitive and collaborative 3D labeling experience for multiple remote users.
• The system efficiently handles reconstruction speed, data streaming, visualization, and labeling tasks.
• Successful automatic label propagation and completion based on sparse user input.

Conclusions:

• The proposed VR system significantly enhances collaborative 3D scene labeling efficiency.
• It effectively addresses the challenges of real-time data processing and multi-user interaction in 3D environments.
• Offers a scalable solution for large-scale 3D data annotation through parallel, remote contributions.