Diplomarbeiten, Studienarbeiten, Praktika

Next-Best-View Planning for Object Recognition within Cluttered Scenes. Bachelor/Master Thesis, Internship or Student Job.

24. Oktober 2013

Project Description

Real world tabletop scenes are usually partially known, which means that models are available for some, but not all, of the objects in the scene. Regarding robotic tasks such as grasping or manipulating objects, at least the objects that should be interacted with need to be known a priori. Other objects that may be occluded or are not in the field of view (FOV), typically remain unrecognized by an autonomous system. These can be resolved by multiple view points. For tackling the analysis of partially known scenes in an autonomous way, recognition and exploration have to cooperate as a single scene exploration system. Thereby, exploration can provide useful views from the global model for multi-view recognition, improving both the pose estimation and amount of recognized objects.

Goal

For multi-view recognition of cluttered scenes, the robot needs to decide from which next viewpoint (a so-called Next-Best-View) the maximal information of the given scene can be gained. At the Institute of Robotics and Mechatronics, various methods for Next-Best-View planning in the context of 3D modeling and exploration have been developed [1]. Now, a new Next-Best-View algorithm needs to be implemented which aims at improving the accuracy of the recognized objects in the scene. The setup from [2] must be changed based on informationtheoretic ideas from [3] (extended to 6 degrees-of-freedom pose estimation tasks).

Requirements

  • You are currently studying computer science, electrical engineering, mathematics or similar
  • Programming skills in C++ (knowledge of Eigen or PCL is an advantage)
  • Fluent in German or English
  • Computer vision or 3D modeling knowledge would be beneficial

Duration

We are looking for someone for at least 5 months, starting as soon as possible.

References

[1] Simon Kriegel, Christian Rink, Tim Bodenmüller, Alexander Narr, Michael Suppa, and Gerd Hirzinger. ”Next-Best-Scan Planning for Autonomous 3D Modeling”, In Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems IROS, Villamoura, Portugal, October 2012.

[2] Simon Kriegel, Manuel Brucker, Zoltan-Csaba Marton, Tim Bodenmüller, and Michael Suppa. ”Combining Object Modeling and Recognition for Active Scene Exploration”, In Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems IROS, Tokyo, Japan, November 2013.

[3] Tianshi Gao, and Daphne Koller. ”Active Classification based on Value of Classifier”, In Advances in Neural Information Processing Systems NIPS, 2011.

 

To apply send your complete application (cover letter, CV, transcript of grades, certificates, references) to:


URL dieses Artikels
http://www.dlr.de/rmc/rm/desktopdefault.aspx/tabid-3759/6082_read-38059/