RGB-D SLAM Based on Extended Bundle Adjustment with 2D and 3D Information

Area of Science:

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

Background:

• Traditional RGB-D SLAM methods often fail to tightly couple depth and visual information during pose estimation.
• Existing approaches have limitations in leveraging both 2D and 3D data effectively for accurate camera tracking.

Purpose of the Study:

• To propose a new RGB-D camera SLAM method that integrates 2D and 3D information for enhanced pose estimation.
• To develop a novel projection model for tightly coupling visual and depth data within an extended bundle adjustment framework.

Main Methods:

• RGB-D camera calibration to establish geometric relationships between image coordinates and depth values.
• Automatic extraction and matching of 2D and 3D feature points to construct an image network.
• Extended bundle adjustment utilizing a new projection model that incorporates both image and depth measurements.

Main Results:

• The proposed method demonstrates significantly better performance than traditional SLAM techniques in field experiments.
• Experimental results validate the effectiveness of the integrated 2D and 3D approach in improving localization accuracy.
• The novel projection model successfully leverages both visual and depth data for precise camera pose refinement.

Conclusions:

• The developed RGB-D SLAM method offers a robust solution for accurate camera pose estimation by effectively integrating 2D and 3D data.
• The extended bundle adjustment with the new projection model represents a significant advancement in RGB-D SLAM.
• The findings highlight the importance of tightly coupling multimodal sensor data for high-precision localization in robotics and computer vision applications.