CVIDS: A Collaborative Localization and Dense Mapping Framework for Multi-Agent Based Visual-Inertial SLAM.

Area of Science:

• Robotics
• Computer Vision
• Simultaneous Localization and Mapping (SLAM)

Background:

• Traditional visual SLAM is limited to single agents, restricting mobility and mapping capabilities.
• Existing multi-agent SLAM systems often produce sparse maps or rely on depth sensors for dense reconstruction.
• A gap exists for dense mapping with monocular cameras in multi-agent scenarios.

Purpose of the Study:

• To develop a novel collaborative SLAM system, CVIDS, for multi-agent co-localization and dense reconstruction.
• To address the limitations of existing systems in terms of map density and sensor requirements.
• To enable dense 3D environment mapping using only monocular cameras in a multi-agent setup.

Main Methods:

• CVIDS employs a centralized, loosely coupled framework, integrating with existing Visual-Inertial Odometry (VIO) systems.
• A robust loop closure detection module and a two-stage pose-graph optimization pipeline facilitate efficient co-localization.
• A motion-based dense mapping module recovers 3D structures and fuses depth information from keyframes for global map reconstruction.

Main Results:

• CVIDS successfully achieves efficient co-localization of multiple agents within a unified coordinate system.
• The system enables dense 3D reconstruction of the environment using only monocular camera data.
• Experimental results demonstrate the superior performance of CVIDS in both quantitative and qualitative assessments.

Conclusions:

• CVIDS effectively fills the research gap for dense, multi-agent SLAM using monocular cameras.
• The system offers a flexible and efficient solution for collaborative localization and mapping.
• The open-source release of CVIDS promotes reproducibility and further research in the field.