Documentation of the Active Real-Time Segmentation Project

Introduction

This is the documentation for the implementation of the real-time active 3D scene segmentation presented and described in the following publication:

@InProceedings{bjorkman10,
author = {Mårten Björkman and Danica Kragic},
title = {{Active 3D Scene Segmentation and Detection of Unknown Objects}},
booktitle = "IEEE International Conference on Robotics and Automation (ICRA)",
year = "2010",
pages= {3114 - 3120}
}


It is based on an implementation of Belief Propagtion on the GPU. Therefore CUDA should preferably be installed. However, the algorithm can also run on the CPU.

Installation

Dependencies

The following dependencies are optional but recommended:

• CUDA
• TBB (Intel ® Threading Building Blocks)

The following library can be included for compatability, but is not required

• OpenCV

Compilation

cd [YOUR_SOURCE_DIRECTORY]
mkdir build
cd build
cmake ..
make


Compiling the Documentation

cd [YOUR_SOURCE_DIRECTORY]/build
make doc


Usage

fgbgtable <image format> <disparity format> <first index> <last index> <disparity range>

Example:
cd [YOUR_SOURCE_DIRECTORY]/build
bin/fgbgsegment ../segpics/clim04d.pgm ../segpics/dimg04d.pgm 1 9 64