TIMA SLAM: Tracking Independently and Mapping Altogether for an Uncalibrated Multi-Camera System

Area of Science:

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

Background:

• Multi-camera systems offer richer environmental perception for SLAM.
• Traditional multi-camera SLAM requires laborious and time-consuming precalibration of extrinsic camera parameters.
• Existing SLAM systems like ORB-SLAM2 are primarily designed for single or precalibrated multi-camera setups.

Purpose of the Study:

• To develop a novel simultaneous localization and mapping (SLAM) system that extends ORB-SLAM2 for multi-camera usage without requiring precalibration.
• To enable online estimation of extrinsic camera parameters, reducing system setup complexity.
• To improve the accuracy and robustness of multi-camera SLAM by jointly optimizing map, keyframes, and relative camera poses.

Main Methods:

• Extension of the ORB-SLAM2 framework to handle multiple independent camera tracking on a shared map.
• Online estimation of extrinsic parameters for each camera within the multi-camera system, with scalar ambiguity for RGB cameras.
• Simultaneous optimization of the map, keyframe poses, and relative poses of the multi-camera system.
• Compatibility with both RGB and RGB-D camera sensors.

Main Results:

• The proposed system successfully performs multi-camera SLAM without the need for precalibration.
• Experimental validation on EuRoC/ASL and KITTI datasets (RGB stereo) demonstrated accurate performance in indoor and outdoor environments.
• Tests on a custom three-camera dataset with small overlapping regions and an RGB-D dataset confirmed system robustness.
• The system achieved accurate shared map generation and robust camera tracking.

Conclusions:

• The developed multi-camera SLAM system effectively removes the necessity for precalibration, significantly simplifying deployment.
• The online estimation and joint optimization approach leads to robust performance and improved mapping accuracy.
• The system demonstrates versatility by supporting both RGB and RGB-D cameras for various applications.