Readspeaker menu



Benchmarking 3D Reconstructions from Next Best View Planning

 



The following results were achieved during the DFG project "Aktive daten- und modellbasierte Sensorpositionierung zur 3-D-Rekonstruktion" (Active data- and model-driven sensor placement for 3D reconstruction, funded by the Deutsche Forschungsgemeinschaft).

 

Motivating Next Best View Planning:

 

Aiming at the 3D reconstruction of an object, in general, different views of the object yield different gains of the reconstruction result (cf. concavities, self occlusions etc.). For the reconstruction (especially within a controlled environment) it is useful to calculate next best views automatically.

 

Benchmarking:

 

To evaluate their results, scientist use different test objects and different criteria. Hence, a comparison of the existing methods is not possible. As a common measure this work proposes a reference object and common criteria to benchmark 3D reconstructions from different methods and sensor equipments, cf. Benchmarking 3D Reconstructions from Next Best View Planning (full citation and paper download at List of Publications).

 

Application:

 

An exemplar of the reference object can be ordered at the Fraunhofer IOF Jena (contact: Christoph Munkelt). Reference data (from the CAD model) and C functions (construction yard!) will be given at the bottom of this site.

 

Persons in charge:

 

Christoph Munkelt, Fraunhofer-Institut für Angewandte Optik und Feinmechanik Jena

Michael Trummer, Friedrich-Schiller-Universität Jena

 

 

 

 


Views of the proposed reference object







Download POV-Ray model

GroundTruthBody.pov
position.inc
render.ini

Surface pattern

pattern.jpg

Dense point cover of CAD surface

nbv_test_object_basic_200k.asc

last modification: 02/13/2008