Automatic Reconstruction of Multi-Level Indoor Spaces from Point Cloud and Trajectory

Area of Science:

• Computer Vision
• 3D Reconstruction
• Computational Geometry

Background:

• Existing indoor modeling methods struggle with complex environments like multi-room or multi-level buildings.
• Current approaches often segment spaces using room segmentation or horizontal slicing.

Purpose of the Study:

• To propose an automatic method for reconstructing multi-level indoor spaces.
• To develop unique models that capture inter-room and inter-floor connections.
• To enable accurate 3D modeling from point cloud and trajectory data.

Main Methods:

• Constructing structural points from registered point clouds.
• Extracting piece-wise planar segments.
• Performing 3D space decomposition and generating watertight meshes using graph cut with energy minimization.
• Defining the energy function based on visibility differences between decomposed spaces and trajectory.

Main Results:

• Successful reconstruction of complex indoor environments, including multi-room, room-less, and multi-level buildings.
• Generation of watertight meshes representing detailed indoor structures.
• Demonstrated performance across seven diverse indoor space datasets.

Conclusions:

• The proposed method offers an effective solution for automatic 3D indoor space reconstruction.
• It accurately models complex architectural features, including inter-floor connections.
• The approach advances the state-of-the-art in indoor modeling for challenging environments.