Automatic 3D modeling from reality
Results | Accuracy
Thesis Research
Robotics Institute
Daniel Huber
Carnegie Mellon University

Overview
 

Evaluating hypotheses
Finding the solution

Results
  Large DB tests
Generality
Accuracy
Publications
Video

   

To test the geometric accuracy of the models constructed using automatic modeling, we used controlled experiments with a synthetic range sensor. The reconstructions are accurate to within 0.02% of the model size (0.04 mm for these objects). The pictures below show close-ups of three test objects and their reconstructed models. The reconstructions have a slight loss of detail when compared to the originals. The main sources of error are the sensor resolution, sensor noise, and voxelization in the surface reconstruction algorithm.

Input objects

 

Reconstructed models

 

 

 

 

 

Next: Publications