RGB-D SLAM Combining Visual Odometry and Extended Information Filter.

Area of Science:

• Robotics
• Computer Vision

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous systems.
• Existing SLAM methods often rely on additional sensors or complex graph optimization, limiting real-time performance.

Purpose of the Study:

• To develop a novel RGB-D SLAM system that is efficient and suitable for online applications.
• To improve motion estimation and landmark representation in SLAM.

Main Methods:

• Utilized visual odometry and an extended information filter, eliminating the need for external sensors.
• Developed a visual dead reckoning algorithm based on visual residuals for motion estimation.
• Introduced the Binary Robust Appearance and Normals Descriptor (BRAND) for feature extraction and landmark representation.
• Implemented an observation model that marginalizes observation likelihood to avoid explicit data association.

Main Results:

• The proposed RGB-D SLAM system demonstrated faster performance compared to a graph-based RGB-D SLAM algorithm.
• Experimental validation on a public RGB-D dataset confirmed the system's efficiency.

Conclusions:

• The novel RGB-D SLAM system offers a faster and more efficient alternative for online applications.
• The system effectively handles motion estimation and landmark representation without explicit data association.