How to Effectively Monitor SD-WAN and SASE Environments with ThousandEyes
ISMAR 2010
1. Interactive Modelling
for AR Applications
John Bastian, Ben Ward, Rhys Hill,
Anton van den Hengel, Anthony Dick
Australian Centre for Visual Technologies
School of Computer Science
University of Adelaide
2. Introduction
Our method allows users to rapidly build
3D models for
Manipulation within AR workspace
Use as occlusion masks
Our approach is characterised by
An interactive modelling process
Modelling by segmentation and silhouette
carving
3. Overview
Video an object with a hand-held webcam
Track camera movement with PTAM
User selects object in one frame
System segments object in subsequent frames
Observations are combined to construct the
3D model
4. Segmentation
User marks the object, using a
mouse or the camera
Build foreground and background
colour models from user marking
Apply graph cut segmentation
Optimal segmentation given
colour models and image
edges
Additional marking refines the
segmentation
6. Silhouette Carving
Use the silhouettes
to generate 3D
models
Start with a volume
around the object
Remove voxels with
projection outside
the silhouette
7. Feedback
Current shape estimate is used to
predict subsequent silhouettes
Predicted silhouette is used to:
Define cut region
Update colour models
Reduce ambiguity when foreground has
similar colour to background
8. 3D Prior
Silhouette is predicted from
current shape estimate
Initial mesh used until volume
is defined
Volume model is used to
generate a confidence map for
inclusion of each pixel in the
silhouette
13. Future Work
Register models from a
database to the images
Incorporate photoconsistency
Use more sophisticated
segmentation
Increased automation