A Plane Extraction Approach in Inverse Depth Images Based on Region-Growing

Area of Science:

• Computer Vision
• Robotics
• 3D Scene Understanding

Background:

• Planar surfaces are crucial in indoor scenes for applications like scene understanding and mobile manipulation.
• Existing plane extraction methods often rely on complex 3D scene reconstruction.
• Direct plane extraction from images offers a more efficient alternative.

Purpose of the Study:

• To develop a fast and accurate plane extraction method directly from images.
• To formulate plane representation in inverse-depth images for direct extraction.
• To enable real-time applications in computer vision and robotics.

Main Methods:

• A novel plane extraction approach utilizing region growing in inverse-depth images.
• Key components include efficient seed selection within grid cells and a region growing process.
• Accurate boundary detection using a greedy policy and normal coherence check, with merging of coplanar planes to prevent over-segmentation.

Main Results:

• Achieved 80.2% Correct Detection Rate (CDR) on the ABW SegComp Dataset, demonstrating state-of-the-art comparable performance.
• The method operates at 5 Hz on typical 680 × 480 images, indicating strong potential for real-time applications.
• Experimental validation on public datasets and synthetic images confirmed the approach's effectiveness.

Conclusions:

• The proposed inverse-depth based plane extraction method is efficient and accurate.
• It offers a viable solution for real-time scene understanding and robotic applications.
• Further improvements could enhance its applicability in complex real-world scenarios.