RGB-D SLAM with Manhattan Frame Estimation Using Orientation Relevance

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Existing sparse Red Green Blue-Depth (RGB-D) Simultaneous Localization And Mapping (SLAM) algorithms suffer from poor accuracy and robustness.
• Issues stem from image noise, blur, and inconsistencies between depth and color data, affecting pairwise spatial transformation accuracy.

Purpose of the Study:

• To propose a novel RGB-D SLAM algorithm that enhances accuracy and robustness.
• To leverage the Manhattan World assumption for improved spatial transformation computation in indoor environments.

Main Methods:

• Developed a new RGB-D SLAM algorithm incorporating Manhattan Frame Estimation.
• Introduced the concept of 'orientation relevance' for Manhattan Frame Estimation.
• Computed pairwise spatial transformation using the estimated Manhattan Frame.

Main Results:

• The proposed algorithm demonstrates significant improvements in accuracy and robustness.
• Enhanced performance in simultaneous localization and mapping tasks.
• Reduced runtime compared to existing methods.

Conclusions:

• The novel RGB-D SLAM algorithm effectively addresses limitations of previous methods.
• Manhattan Frame Estimation using orientation relevance is a viable approach for improving SLAM performance.
• The algorithm offers a more accurate, robust, and efficient solution for indoor mapping.