PolyGraph: A Graph-Based Method for Floorplan Reconstruction From 3D Scans.

Area of Science:

• Computer Vision
• Computational Geometry
• Artificial Intelligence

Background:

• Indoor floorplan reconstruction is crucial for applications like interior design, VR, and robotics.
• Existing methods struggle with high computational costs and primitive detection errors.
• There is a need for robust and efficient floorplan reconstruction techniques.

Purpose of the Study:

• To introduce PolyGraph, a novel computational framework for high-quality indoor floorplan reconstruction.
• To address the limitations of existing methods, particularly computational cost and error sensitivity.
• To improve the accuracy and structural integrity of reconstructed floorplans.

Main Methods:

• Developed a guided wall point primitive estimation network for dense boundary sampling and robust detection.
• Formulated floorplan reconstruction as a subgraph optimization problem using a graph-based representation.
• Integrated structural information via "structural weight" into the graph representation.

Main Results:

• PolyGraph demonstrates superior effectiveness and advantages over existing optimization-based approaches.
• Achieved significant reduction in search space compared to pixel-level optimization.
• Experimental results show computational efficiency, structural integrity preservation, and fine detail capture.

Conclusions:

• PolyGraph offers a computationally efficient and robust solution for indoor floorplan reconstruction.
• The framework successfully integrates deep learning for primitive detection with graph optimization.
• The method preserves structural integrity and captures fine details, outperforming existing approaches.