Interactive Hierarchical Space Carving

1
Interactive Hierarchical Space Carving with
Projector-based Calibrations
Martin Granger-Piché, Emric Epstein, Pierre
Poulin grangem epsteine poulin _at_
iro.umontreal.ca
Results
Interactive System Pipeline
normal feedback loop
move camera or move object or zoom camera on
object
silhouette carving on dynamic image list
visualization of current 3D model

• image
• acquisition
• cleanup
• calibration

continuous intervention

system initialization
color carving on static and dynamic image
structures
increase subdivision on entire model or selected
region
Abstract We present an interactive
reconstruction system based on space carving. The
user controls the camera and object positions and
their impact on the reconstruction is immediately
displayed. The flow of images is used first for
silhouette carving, and then representative
images are stored in a hemispherical structure
for balanced color carving. Projected color
points on the measured background or on the
object (acquired by structured light) allow for
automatic and adapted calibration of the camera.
The octree-based object repre-sentation is
central to our space carving algorithm voxels
are efficiently carved out at their largest size,
images are treated at their appropriate level,
and 3D regions are subdivided only when
necessary.
on user request
occasional intervention
threshold modification for color comparison

generation of structured light calibration points
static
dynamic
save current 3D model
Standard Space Carving
Projector-based Calibrations
Octree-based Space Carving
Our Technique

• Contributions
• 1. Interactive system
• tools to give control to the user
• visual feedback loop
• real-time simultaneous image acquisition,
  segmentation,
• calibration, and reconstruction display
• 2. Projector-based calibrations
• automatic camera calibration with projected
  color points
• zoomed camera calibration with projected points
  on the
• object, which are obtained with structured
  light
• automatic stand tracking
• 3. Octree-based space carving
• compactness subdivision of only external
  boundary voxels
• efficiency largest voxels are carved before
  subdivided
• hardware visible voxels detection
• appropriate octree level for each image
• 4. Image processing
• RGB ellipsoid fast and robust color detection
• automatic background segmentation

32x32x32
16x16x16
64x64x64
256x256x256
Selected valid points to allow zoomed in camera
calibration
All structured light points