Robust and Efficient CPU-Based RGB-D Scene Reconstruction.

Area of Science:

• Computer Vision
• Robotics
• 3D Scene Reconstruction

Background:

• Accurate 3D scene reconstruction is crucial for computer vision and robotics.
• Tracking in textureless environments presents significant challenges for camera tracking systems.
• Existing systems face difficulties in achieving both speed and accuracy in 3D model generation.

Purpose of the Study:

• To propose a robust and efficient CPU-based approach for indoor 3D scene reconstruction using consumer RGB-D cameras.
• To address the limitations of existing methods in textureless scenes and improve reconstruction speed.
• To develop a novel system for robotics applications that combines feature-based tracking and volumetric integration.

Main Methods:

• A hybrid camera tracking method combining point and edge features for enhanced stability in textureless scenes.
• An efficient data fusion strategy that selects camera views and integrates multi-scale RGB-D images.
• A novel RGB-D scene reconstruction system optimized for implementation on standard CPUs.

Main Results:

• The proposed approach demonstrates improved tracking stability in textureless environments.
• Efficient data fusion enhances the speed of volumetric integration.
• The system achieves high robustness and efficiency in 3D scene reconstruction compared to state-of-the-art methods.

Conclusions:

• The developed CPU-based approach offers a robust and efficient solution for 3D indoor scene reconstruction.
• The method effectively bridges feature-based tracking and volumetric integration for improved performance.
• This system provides a practical and fast solution for robotics applications requiring accurate 3D models.