Image-based 3D reconstruction of urban scenes

Contact person: Anton S. Konushin (ktosh@graphics.cs.msu.ru)

Introduction

This project is devoted to reconstruction of photo-realistic 3D model of an urban scene from a number of images of streets and crossroads. Our main intention was the createion of a system for construction high-quality models with very simple and intuitive user interface.

Main goal of the project is to create a 3D content authoring tool which:

  • is easy to use and makes the modeling process very fast
  • does not require 3D modeling skills for creating realistic scenes
  • produces realistic models of urban scenes

Current modeling process

Reconstruction process includes the following steps:

  1. Automatic tilt correction
  2. Original image Automaticaly tilt corrected image
  3. Semi-automatic segmentation of buildings

  4. Building segmentation

  5. Semi-automatic texture correction
  6. Before texture reconstruction After texture reconstruction
  7. Automatic 3D model construction
  8. Vrml model screenshot (1) Vrml model screenshot (2)

Multiview modeling

We have developed an approach for semi-automatic registration of images for reconstruction of buildings from multiple views.

One of model #1 source images Multiview model #1 video
              
                One of model #2 source images Multiview model #2 video

Download current model examples

You can download several examples of reconstructed models (it took about 3-10 minutes to create these examples):

Download VRML model 1 (singleview) 1.80 Mb

Download VRML model 2 (multiview) 2.85 Mb

Download VRML model 3 (animated multiview) 1.83 Mb

Team

Publications

Acknowledgements

We are grateful to Samsung Advanced Institute of Technology (SAIT) for financial support of this project.