3D from two images.

A Qt based gui for extracting 3D from two images using OpenCV and PCL library. The program builds 3D from two images only. Eventually it is supposed to build 3D from multiple images.

3D point cloud generation without texture mapping

This project is for buidling 3D point cloud from multiple images of a rigid scene. The method for implementing SfM is described in paper [1]. The basic idea is to first obtain 3D points from a pair of images. SURF feature matching is done between two images having widest baseline. Corresponding points are used to estimate the motion (or transformation) between two images based on epipolar geometry. Transformation between two images are used to compute 3D points by triangulation method. This pairing up of images based on width of baseline is done for rest of the images. This yields multiple 3D point clouds but have different scale.

So as to merge these point clouds one way is to built ratio histogram. For example consider two point cloud and that they share some number of points that are common to both point clouds. A ratio of distance of a common point is computed. This is done for all available common points. And finally we built a histogram of these ratio. The ratio which is repeated most is considered as a scaling factor for the two point clouds. Currently I am working on this part , that is multiple cloud merging.

Tools and Libraries Required:

• Qt
• OpenCV
• Point Cloud Library (PCL)

Usage:

Built the project in Qt.

To run SfM project, images and intrinsic camera parameters are required. For the Camera calibration matrix, xml files of OpenCV format is used. The calibration matrix can also be entered through dialogbox.

To run a demo, the folder Demo_data has a set images and the camera calibration file in xml format.

Datasets:

Standard datasets for SfM can be found here: http://cvlabwww.epfl.ch/~strecha/testimages.html

References:

1. Modeling the World from Internet Photo Collections, N. Snavely, S. M. Seitz, R. Szeliski, IJCV 2007 http://phototour.cs.washington.edu/ModelingTheWorld_ijcv07.pdf

2. Multiple View Geometry in Computer Vision, Hartley, R. I. and Zisserman, A., 2004, Cambridge University Press http://www.robots.ox.ac.uk/~vgg/hzbook/

3. Mastering OpenCV with Practical Computer Vision Projects https://github.com/royshil/SfM-Toy-Library

Acknowledgements:

1. The numerical computation parts including, triangulation and finding camera matrix has been obtained from here