Accurate and Robust Monocular SLAM with Omnidirectional Cameras

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous systems.
• Existing SLAM methods often struggle with wide-angle or omnidirectional cameras due to image distortion.
• The ORB-SLAM framework provides a robust foundation for visual SLAM.

Purpose of the Study:

• To develop an improved monocular visual SLAM system utilizing omnidirectional cameras.
• To enhance the ORB-SLAM framework to accommodate the unique properties of wide-angle and fisheye lenses.
• To improve the accuracy and robustness of SLAM systems in environments with significant image distortion.

Main Methods:

• Extended the ORB-SLAM framework with an enhanced unified camera model for projection.
• Implemented a novel map initialization method specifically for omnidirectional cameras.
• Analytically derived Jacobian matrices for reprojection errors concerning camera pose and 3D point positions.

Main Results:

• The system demonstrated real-time performance.
• Achieved positioning errors below 0.1% in small indoor environments and under 1.5% in large environments.
• Showcased increased accuracy and robustness compared to traditional pinhole model-based SLAM systems.

Conclusions:

• The proposed SLAM system effectively leverages omnidirectional cameras, even with extreme distortion.
• The enhanced unified camera model and initialization method improve SLAM performance with wide-angle lenses.
• This work offers a more accurate and robust SLAM solution for applications requiring a wider field of view.