The Robotics Institute

3-D Recognition and Model Construction

My group is investigating a number of algorithms and applications in the area of 3-D data manipulation and interpretation. This includes developing techniques for the automatic construction of three-dimensional models from large sets of views with applications to object model generation, interior mapping, and terrain map registration; and the recognition of 3-D objects in complex scenes from large databases of models.

Object Recognition in Images

We are also developing algorithms for recognizing objects in images. The approaches include nearest neighbor techniques for recognition of objects in general position in images, edge-based decision tree techniques specialized for recognizing wiry objects, and classification techniques for the identification of large classes of objects and for image segmentation in the context of natural scenes. All the approaches are based on learning classifiers from training exemplars; we are also exploring techniques for weakly-labeled training in order to reduce training overhead.

Mobile Robots

• Registration of terrain maps with sensor data for localization and navigation
  
• Processing of high-resolution 3-D terrain maps for traversability evaluation applied to route planning
  
• Fusion of algorithms and sensors for terrain characterization
  
• LADAR processing
  
• Scene analysis from LADAR and color data for landmark detection with applications to autonomous navigation in urban environments