pf-localization

Localization using a Particle Filter (and random walk model)

What is this?

This is a Cuda-C++ library to localize a camera, pointed at a checkerboard pattern without solving the PnP problem and using a particle filter. Solving the PnP problem is not always feasible (specifically in case of real-time SLAM), therefore this project aims at localizing the camera rotation and translation with a Random Walk model.

Building

In order to build this library, you need a system with CMake and Cuda SDK 7+ installed.

cd <project root>
mkdir build;
cd build;
cmake .. -G "Visual Studio 14 Win64"
cmake --build . --config Release
cmake --build . --target INSTALL

Running

Matlab is required (but not necessary) to run this library with a sample dataset (in data/board.mp4). The Matlab script that uses this library is main.m. The first time you run the Matlab file, it'll take some time to build a cache of video's ground truth data using PnP solver of Matlab.

Matlab code also ships with a Matlab implementation of the project. However, If you like to run this code with Cuda, you must have a Nvidia Cuda-capable graphics card.

Reference

This code is an adaptation of the theories provided in the following paper:

@article{doi: 10.1117/1.JEI.23.1.013029,
author = { Seok-Han  Lee},
title = {Real-time camera tracking using a particle filter combined with unscented Kalman filters},
journal = {Journal of Electronic Imaging},
volume = {23},
number = {},
pages = {23 - 23 - 19},
year = {2014},
doi = {10.1117/1.JEI.23.1.013029},
URL = {http://dx.doi.org/10.1117/1.JEI.23.1.013029},
eprint = {}
}