Reconstruction of 3D Scene from Stereo Images

Aim

Given a pair of stereo images, reconstruct a 3D model of the scene.

Platform

The system was implemented in C++, using the OpenCV library, in Linux environment. The 3D models are represented in VRML.

System Overview

The input of the system is a pair of stereo images, like the ones below:

Stereo Images

A disparity map was estimated, based on the function cvFindStereoCorrespondence, which implements a feature correspondence algorithm by Birchfield and Tomasi. (More information on [Stan Birchfield and Carlo Tomasi, "Depth Discontinuities by Pixel-to-Pixel Stereo" International Journal of Computer Vision, 35(3): 269-293, December 1999]).

Disparity Map

Then, a dense 3D model is built, based on the disparity map.

3D Model

Finally, the texture from the original images is used to render the 3D model.

3D Rendered Model

About this work

The above system was implemented by Spyridon Bakas for his final year project at Kingston University, under the supervision of Dimitrios Makris and the assistance of Jean-Christophe Nebel, during the academic year 2004-2005.