Polyora Documentation

Introduction

Polyora is a library to retrieve images and videos from tracked feature points.

License

You are allowed to use polyora under the terms of the GNU General Public License, version 2 or above. In short, it means that any distributed project linking with polyora or that includes any portion of its source code should be released under the same license, with the source code. The details of the license can be found there: http://www.gnu.org/licenses/old-licenses/gpl-2.0.html

If the terms of the GPL license do not meet your requirements, you are welcome to contact Julien Pilet (julien.pilet(at)calodox.org) to find a solution.

Download

• Sources: http://www.hvrl.ics.keio.ac.jp/~julien/polyora/polyora-1.0.2-Source.zip
• Windows binaries: http://www.hvrl.ics.keio.ac.jp/~julien/polyora/polyora-1.0.1-win32.exe
• Sample Data: http://www.hvrl.ics.keio.ac.jp/~julien/polyora/polyora-data.zip

Installation

Polyora depends on a number of software and libraries to compile properly:

• cmake (http://www.cmake.org/)
• opencv (http://opencv.willowgarage.com/)

Additionally, you may want to use QT (http://qt.nokia.com/) for the GUI.

If you prefer to use SIFT rather than the provided feature detector/descriptor, polyora can use SiftGPU (see Using Polyora with SiftGPU).

Polyora has been compiled successfully on Windows, Linux, and Mac OS.

Quick installation instructions:

1. run cmake or cmake-gui on CMakeLists.txt
2. type 'make' or use the appropriate tool to build the project created by cmake.

See also Compiling Polyora with Microsoft Visual Studio .

Usage

Using polyora is not difficult but not straightforward. Usage is explained there: Basic usage instructions.

Examples:

• simpletrack::cpp : A basic example showing how to track features.
• vobj_tracker_view : The full application that detect and tracks objects.

Internals

Polyora is organized in several levels:

1. Sparse point track structures - Structures for sparse storage of tracked keypoints.
2. Keypoint detection and tracking - Algorithms to fill the structured already described.
3. Visual objects retrieval - Indexing and retrieval without any geometric knowledge, based on quantized point track description.
4. Object level tracking - RANSAC and other algorithms for multiple planar object tracking.

Author

Most of polyora was written by Julien Pilet, at Keio University, Japan, from Dec. 2008 to March 2010.

Acknowledgements

I sincerely thank Hideo Saito for supporting me on this project. I warmly thank Vincent Lepetit and Pascal Fua who taught me so much during my PhD studies. Vincent, even when you're on the other side of the world, you keep inspiring me !