MOPED: Object Recognition and Pose Estimation for Manipulation

Alvaro Collet, Manuel Martinez, Siddhartha Srinivasa.

Overview

MOPED is a real-time Object Recognition and Pose Estimation system. It recognizes objects from point-based features (e.g. SIFT, SURF) and their geometric relationships extracted from rigid 3D models of objects. The global MOPED framework requires seven steps to recognize objects:

1) Feature extraction

2) Feature matching

3) Feature clustering

4) Hypothesis generation

5) Pose clustering

6) Hypothesis refinement

7) Pose recombination

This code is structured in two modules: first, a ROS-agnostic library called libmoped, in which all code for the 7-step algorithm is implemented; and second, ROS-enabled wrapper code that utilizes libmoped to read images from the network and to publish the detected objects.

Source Code

You can download the moped2 source code from our subversion repository, at:
https://svn.personalrobotics.ri.cmu.edu/public/latest/moped

This repository contains:
MOPED2 - Object recognition code. The documentation is in a file called mainpage.dox within the moped2 folder. This subversion repository provides read-only access to the source code. If you wish to develop and commit code for MOPED, please contact us to get read-write access.

Examples on how to use MOPED in conjunction with ROS are available in the 'imagesender' and 'moped-example' subfolders. There are two modeling tools for MOPED, one in matlab (subfolder moped2/modeling) and one in python (moped-modeling-py). The matlab version is more stable, but both are usable already. The folder 'BundlerPy' is necessary to use the python modeling tool. The ROS version of MOPED depends on our own set of PersonalRobotics messages, available here:
https://svn.personalrobotics.ri.cmu.edu/public/trunk/latest/pr_msgs

IMPORTANT: Our repository changed on Jan 20, 2012! If you have the old repo pointing to an intel-research domain, you need to update. To update your local copy, cd to your MOPED folder and type: svn switch --relocate https://svn.pittsburgh.intel-research.net/repos/pr https://svn.personalrobotics.ri.cmu.edu/

Videos

MOPED processing HD Video in Real Time

Multi-Camera MOPED on HERB

Papers

The MOPED framework: Object Recognition and Pose Estimation for Manipulation

Alvaro Collet, Manuel Martinez, Siddhartha Srinivasa
The International Journal of Robotics Research (IJRR), April, 2011
Details | pdf (5MB) | Copyrighted

@article{Collet2011,
  title={{The MOPED framework: Object Recognition and Pose Estimation for Manipulation}},
  author={Collet, Alvaro and Martinez, Manuel and Srinivasa, Siddhartha S.},
  booktitle={The International Journal of Robotics Research},
  year={2011}
}

Efficient Multi-View Object Recognition and Full Pose Estimation

Alvaro Collet and Siddhartha Srinivasa
IEEE International Conference on Robotics and Automation (ICRA '10), May, 2010.
Details | pdf (1MB) | Copyrighted

@inproceedings{Collet2010,
title = {{Efficient Multi-View Object Recognition and Full Pose Estimation}},
author = {Collet, Alvaro and Srinivasa, Siddhartha S.},
booktitle = {IEEE International Conference on Robotics and Automation},
publisher = {IEEE},
year = {2010}
}

MOPED: A Scalable and Low Latency Object Recognition and Pose Estimation System

Manuel Martinez, Alvaro Collet, and Siddhartha Srinivasa
IEEE International Conference on Robotics and Automation (ICRA '10), May, 2010.
Details | pdf (1MB) | Copyrighted

@inproceedings{Martinez2010,
title = {{MOPED: A Scalable and low Latency Object Recognition and Pose Estimation System}},
author = {Martinez, Manuel and Collet, Alvaro and Srinivasa, Siddhartha S.},
booktitle = {IEEE International Conference on Robotics and Automation},
publisher = {IEEE},
year = {2010}
}

Object Recognition and Full Pose Registration from a Single Image for Robotic Manipulation

Alvaro Collet, Dmitry Berenson, Siddhartha Srinivasa, and David Ferguson
IEEE International Conference on Robotics and Automation (ICRA '09), May, 2009.
Best Vision Paper Award Finalist in ICRA '09.
Details | pdf (1MB) | Copyrighted

@inproceedings{Collet2009,
title = {{Object recognition and full pose registration from a single image for robotic manipulation}},
author = {Collet, Alvaro and Berenson, Dmitry and Srinivasa, Siddhartha S. and Ferguson, Dave},
booktitle = {IEEE International Conference on Robotics and Automation},
publisher = {IEEE},
year = {2009}
}

Personal Robotics Lab