3D Reconstruction of a Scene from a Pair of Images

In this project we designed a system for 3D reconstruction from a given pair of images, based on A-Priori information about the parameters of the cameras.

Abstract

In this project we designed a system for 3D reconstruction from a given pair of images, based on A-Priori information about the parameters of the cameras, the distance between the two points from which the images were taken, and a list of correspondences between the two images.

The background
3D Reconstruction of scene from a pair of Images taken from two locations is a fundamental process in Computer Vision. Uses for such system are numerous, such as:

  • Possible expansion of the system for reconstruction of scene from a series of images (Video)
  • Development of a computer vision based Tracking system, that tracks an objects in 3D world
  • Creation of a 3D model of an object and by that to allow new synthetic views to be generated

In Ideal state, the rays connecting the points in each image with the image’s center will meet in the object point in space, which need to be reconstructed:
1

However, since this never happens in real life because of measurement errors etc., simple reconstruction cannot be used here.

The solution
The system includes several parts:
Calibration of the given points
Initial guess – of both the parameters of the cameras and the points in space
Optimization of the Initial guess using the “fminunc” function of Matlab

Tools
We used Digital Camera to accuire the images, and PC with Matlab 6.5 for the program

Conclusions

As can be seen in the images below, from the two images taken, we managed to produce accurate 3D reconstruction of the scene:

2 3
Original Images

54
Images with corresponding points before and after optimization

6
3D View of the reconstructed scene

Acknowledgment
We Would like to thank Netanel for guiding and helping us throughout the whole process.
We would also like to thank the Lab staff for helping us with equipment and administrative issues.
We are also grateful to the Ollendorff Minerva Center Fund for supporting this project.