Floor Plan Generator

This project generates a 2D floor plan from a video input or existing 3D reconstruction data. It utilizes 3D reconstruction, point cloud processing, and object detection to create a labeled floor plan.

Prerequisites

Python Dependencies

The project requires Python 3.11 or higher. Install the dependencies using Poetry or pip:

pip install open3d opencv-python ultralytics pycolmap trimesh pillow

OpenMVS

This project relies on OpenMVS for dense reconstruction and texturing. You need to build OpenMVS or install it via vcpkg.

Important: The OpenMVS binaries must be placed in the following directory: src/thirdparty/openmvs/make

Ensure the following executables are present in that folder:

• InterfaceCOLMAP
• DensifyPointCloud
• ReconstructMesh
• RefineMesh
• TextureMesh

Usage

The main entry point for the pipeline is src/main_pipeline.py.

Running from Video

To run the full pipeline starting from a video file:

python src/main_pipeline.py --reconstruct --input_video path/to/video.mp4 --output_dir output

Running from Existing Reconstruction

If you already have a 3D reconstruction (point cloud, sparse model, and images), you can skip the reconstruction step:

python src/main_pipeline.py \
    --point_cloud output/final.ply \
    --sparse_dir output/sparse \
    --images_dir output/images \
    --texture_path output/textured0.png \
    --output_dir output

Arguments

Argument	Description	Default
--reconstruct	Flag to run 3D reconstruction from video.	False
--input_video	Path to the input video file.	input/video.mp4
--output_dir	Directory where all output files will be saved.	output
--images_dir	Directory containing extracted images.	output_dir/images
--sparse_dir	Directory containing COLMAP sparse reconstruction.	output_dir/sparse
--point_cloud	Path to the dense point cloud file (.ply).	output_dir/final.ply
--texture_path	Path to the texture map image.	output_dir/textured0.png
--floor_plan_output	Path for the final floor plan image.	output_dir/floor_plan.png

Input & Output

Input

• Video: A video file (e.g., .mp4) scanning the room.
• (Alternative) Reconstruction Data:
  • Dense Point Cloud (.ply)
  • Sparse Reconstruction (COLMAP format)
  • Extracted Images

Output

• Floor Plan: A 2D image (floor_plan.png) showing the room layout and detected furniture.
• Intermediate Files:
  • Extracted frames from video
  • Sparse reconstruction data
  • Dense point cloud
  • Textured mesh